Click Here
home features news forums classifieds faqs links search
6071 members 
Amiga Q&A /  Free for All /  Emulation /  Gaming / (Latest Posts)
Login

Nickname

Password

Lost Password?

Don't have an account yet?
Register now!

Support Amigaworld.net
Your support is needed and is appreciated as Amigaworld.net is primarily dependent upon the support of its users.
Donate

Menu
Main sections
» Home
» Features
» News
» Forums
» Classifieds
» Links
» Downloads
Extras
» OS4 Zone
» IRC Network
» AmigaWorld Radio
» Newsfeed
» Top Members
» Amiga Dealers
Information
» About Us
» FAQs
» Advertise
» Polls
» Terms of Service
» Search

IRC Channel
Server: irc.amigaworld.net
Ports: 1024,5555, 6665-6669
SSL port: 6697
Channel: #Amigaworld
Channel Policy and Guidelines

Who's Online
22 crawler(s) on-line.
 61 guest(s) on-line.
 2 member(s) on-line.


 kolla,  pavlor

You are an anonymous user.
Register Now!
 pavlor:  6 secs ago
 kolla:  3 mins ago
 OlafS25:  10 mins ago
 Classifieds:  17 mins ago
 pixie:  36 mins ago
 Kronos:  51 mins ago
 amigang:  2 hrs 45 mins ago
 RobertB:  3 hrs 26 mins ago
 agami:  4 hrs 20 mins ago
 matthey:  4 hrs 34 mins ago

/  Forum Index
   /  Amiga General Chat
      /  How good or bad was the AGA chipset in 1992/1993.
Register To Post

Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 Next Page )
Poll : How good or bad was the AGA chipset in 1992/1993.
10p Excellent (Best at 2D/3D, colors, and resolution, frame rate etc.)
5p Good / better than most computer.
0p Barely hanging in there.
-5p Below average / slow but usable
-10p useless / horrible
 
PosterThread
Karlos 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 9:21:40
#341 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4403
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@bhabbott

The first time I saw doom it was on a friend's 486SX 25MHz. It wasn't exactly stellar on that, either. Playable but not particularly smooth.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
kolla 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 9:23:17
#342 ]
Elite Member
Joined: 21-Aug-2003
Posts: 2888
From: Trondheim, Norway

@bhabbott

Quote:


Quote:
I have a CyberSCSI in my A4000 connected to my CyberStorm 060. So a CPU card with local SCSI. Even so, without Zorro limitations, I recall it still needed particular max transfer.

This was a legacy issue, with the latest OS3 versions you don't need it.


Drrt. Latest OS3.2.1 has this attempted fixed only for Gayle (A600/A1200) scsi.device - CyberSCSI from Phase5 has its own scsi.device with its own issues.

And secondly, no. The fix for Gayle scsi.device turned out to not cover all cases, so manually setting max transfer is still needed to be safe.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

 Status: Online!
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 10:28:30
#343 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@bhabbott

Quote:

bhabbott wrote:
@cdimauro

Quote:

cdimauro wrote:

That's plainly false. Intel did OTHER projects as well. A notable example: https://en.wikipedia.org/wiki/Intel_i860

Did I say they didn't? No.

But...
Quote:
The i860 never achieved commercial success and the project was terminated in the mid-1990s.

But... nothing. The success of something means NOTHING in this context. It's a logical fallacy.

BTW, the i860 processor was reused and found on several graphic cards with hardware acceleration.
Quote:
Quote:
Again, on your imagination...

You refute Intel's 486 overdrive chips and the cooling required to get higher Pentium speeds?

Do you understand that you're talking about the END of the Motorola 68k family?
Quote:
Later revision 060's needed less cooling, while later Pentiums needed more. That is a fact.

As it's a fact that Motorola removed A LOT of legacy stuff with its processors, already starting with the 68030 and completing the "opera" with the 68060 (which not even had 64-bit multiplications and divisions).

So, they used much less transistors. Guess what: they consumed less power...
Quote:
Quote:
According to Motorola (so, not me), 68060 matched Pentium performance for integer workloads.
But never published information about FP workloads: guess why...

Irrelevant. What matters is real-world performance.

Which was the case. So, it was relevant.
Quote:
Lightwave 3D rendering test, Pentium 90 MHz vs 68060 50 MHz
Quote:
Test performed in CeBIT 1996

Pentium 90 MHz: 1:30
68060 50 MHz: 0:56

Notes:
Amiga Lightwave rendered the test picture faster than the P90 even if it is not really optimized for the 68060 yet. ;)

Ah, cherry picking. Then I can do it as well: https://groups.google.com/g/comp.sys.amiga.misc/c/jWErDssEr5A

Benchmarks printed in Video Toaster User magazine show the '060 to be about half the speed of a P90.
Quote:
Quote:
I agree, because... Commodore never had a roadmap for producing... PROCESSORS.

And good that they didn't. Imagine Commodore trying to produce CPUs rivaling Motorola and Intel. That would be stupid.

In fact it was so stupid that Commodore was DEPENDING on Motorola.

So, depending on the LATE DELIVERIES of the new chips and on... ABANDONING the 68k family.
Quote:
Quote:
That was ANOTHER reason.

In fact, the 80386, SX version included, still had the Protected Mode introduced with the 80286 AND improved it. In fact, the 386 added the virtual 8086 mode AND it had a 48-bit virtual address space in protected mode.

It's pointless discussing anything with someone who keeps (deliberately?) misinterpreting what I say. I was talking about the 386SX specifically, not the 386DX. That's why I wrote '386SX'.

Do you understand the 386SX had the SAME FEATURES of its big brother 386DX?
Quote:
Quote:
Many DOS applications were perfectly suitable for running on an 8086 virtual machine.

Many, but not all. Specifically, not DOS programs that accessed hardware directly. That includes games, and some apps that did direct hardware access (eg. to program a device through the printer port - a task that I often used the PC for).

This ONLY depends on how the virtual 8086 machines were handled by the host o.s..

The 80386 had everything needed to fully virtualize an 8086. Which means that you had full control even for applications (and games) directly accessing the hardware.

Have you had a look at how the virtual 8086 mode worked? I don't think so.
Quote:
But this is getting way off topic. How well Windows handled DOS programs on a 386 is not relevant to how good the AGA chipset was.

Correct. The only thing, actually.
Quote:
I'll say one thing though. In 1993 I was running DOS programs on my A3000 with full multitasking, the virtual DOS machine running in its own screen 'behind' the Workbench which could be pulled down to reveal it. Try that on a PC!

Why not? In fact, you only needed the right o.s..

It wasn't the PC which was preventing it by itself. Sorry for repeating an obvious thing...
Quote:
With this I was able to develop DOS programs more conveniently than on a real PC.

A real PC with similar hardware would have allowed to do it much better and faster.
Quote:
Recently I ran the DOS version of Tomb Raider on my A1200 in 256 colors, which looked wonderful on my TV. AGA made it possible.

Sure. As long as you've proper resources. Which, in this case, it meas much more resources, since you're emulating an entire PC.
Quote:
Quote:
Are you invoking censorship because you don't like what people report to reply at your false statements?

My false statements?

Yes. See all previous posts.
Quote:
No, I'm not stopping you from saying whatever you like. The off-topic PC crap and Amiga bashing is starting to get boring though.

Which is NOT off-topic, because you had to take a look at what competitors did.

If not, this whole discussion made no sense at all. In fact, we got AGA (and ECS) and we used them. As they are. Be happy with that and everything is perfectly fine.
Quote:
So hey, keep it up and maybe you will censor me. That's your goal, right?

I don't need to censor anyone: I'm perfectly able to sustain a discussion without it, as you can see.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 10:31:21
#344 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@bhabbott

Quote:

bhabbott wrote:
Since comparing the Amiga to a PC seems to be necessary part of deciding how good AGA was, I'll drop in another datapoint.

According to thandor.net, an AMD 386 DX40 got 8.24fps in the Doom timedemo 3 test. I tested it on my A1200 with 50MHz 030 and got 9.9fps, 1.2 x times faster. This exactly matches the difference in CPU clock speeds.

Sure: if you bundle it with a crap 8-bit ISA graphic card.
Quote:
In 1992 386DX systems were considered 'mid-range'. They usually came with some unspecified ISA bus VGA card. Considering the AGA chipset's clear advantage for 2D games, the fact that with an equivalent CPU the A1200 could match a 386DX40 shows that AGA was not inferior to a 'typical' PC of the day, and in some ways was superior.

Do you understand that the your AGA had a much better graphic subsystem compared to a crap 8-bit ISA card?

 Status: Offline
Profile     Report this post  
Karlos 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 11:00:47
#345 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4403
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

The original question is obviously intended to be stoke discussion and comparison with other systems at the time. There are plenty of things AGA could do more quickly, efficiently or just "better" than VGA at the time, especially VGA attached by an 8-bit data bus.

However, we have to acknowledge that drawing 256 colour displays one pixel at a time was definitely not one of them. Had it not been for Doom, that wouldn't have bothered anyone at the time since that's just not how graphics were done.

Suddenly, the hot new thing was software rendering the entire display every frame which in the Amiga tradition was a huge retrograde step. For a time the relentless increase in processing power and improvements in bus width and speed made this irrelevant in the PC scene. Everything was going 3D and suddenly 2D, no matter how efficiently implemented or how smoothly animated was no longer a thing anyone cared about. They'd rather blow up zombies in 3D at an varying average of 15fps than side scroll blast sprites at a rock solid 50fps.

However, the Amiga philosophy won in the end and we have dedicated graphics accelerators for all things 2D and 3D.

Last edited by Karlos on 15-Oct-2022 at 11:01 AM.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
bhabbott 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 17:51:26
#346 ]
Regular Member
Joined: 6-Jun-2018
Posts: 334
From: Aotearoa

@cdimauro

Quote:

cdimauro wrote:


Sure: if you bundle it with a crap 8-bit ISA graphic card.

I doubt that many 386DX systems came with 8 bit ISA cards, that would be stupid.


Quote:
Do you understand that the your AGA had a much better graphic subsystem compared to a crap 8-bit ISA card?

In 1992 the majority of PCs sold were 386SX and 386DX models, with 'crappy' 16 bit ISA bus graphics cards (not the fabled ET4000). Whether that qualifies AGA as 'better than most' depends on how you count them, but in terms of how AGA compareed to the average PC graphics card...

And that is all we should be discussing here, not 68060 vs Pentium etc.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 21:30:29
#347 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@bhabbott

Quote:

bhabbott wrote:
@cdimauro

Quote:

cdimauro wrote:


Sure: if you bundle it with a crap 8-bit ISA graphic card.

I doubt that many 386DX systems came with 8 bit ISA cards, that would be stupid.

Have you checked the graphics cards used on the Doom benchmark tests? Some were 8-bit.

For example: https://thandor.net/object/529
https://archiwum.allegro.pl/oferta/karta-graficzna-tseng-et3000ax-8-bit-isa-i9410160654.html
Quote:
Quote:
Do you understand that the your AGA had a much better graphic subsystem compared to a crap 8-bit ISA card?

In 1992 the majority of PCs sold were 386SX and 386DX models, with 'crappy' 16 bit ISA bus graphics cards (not the fabled ET4000). Whether that qualifies AGA as 'better than most' depends on how you count them, but in terms of how AGA compareed to the average PC graphics card...

Many 16-bit ISA cards were also super-crap. Take a look at the bottom of the chart: https://thandor.net/benchmark/32

Whereas AGA had almost double the bandwidth of the best 16-bit ISA card.

The graphic card was very important for Doom, as it was highlighted on the test:

Be sure to keep in mind the graphic card will play an important role so use a card with fast DOS performance like a Matrix Millenium II PCI or Voodoo3 2000/3000 PCI.
Quote:
And that is all we should be discussing here, not 68060 vs Pentium etc.

Why don't you take a look at the written posts and see who introduced the processors on this discussion?

 Status: Offline
Profile     Report this post  
bhabbott 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 15-Oct-2022 21:45:57
#348 ]
Regular Member
Joined: 6-Jun-2018
Posts: 334
From: Aotearoa

Sorry, I just have to address a couple of points...

@cdimauro

Quote:

cdimauro wrote:

So, they used much less transistors. Guess what: they consumed less power...

This is what I was talking about. I said Motorola didn't follow the same path as Intel did towards more power-hungry CPUs, and you attempt to to counter it with this. What you don't seem to realize is that you are agreeing with me.

When chips are manufactured they have process variations, mostly differences in doping that can't be precisely controlled. This results in some chips being faster but using more power. The manufacturer may then grade the chips according to how fast they can go, and also adjust the specs to account for the different current draw. Initially Motorola only produced one version of the 060 at 50MHz. Any chips that failed to meet that speed were discarded. Later they moved to a finer process that reduced the die size and power, which allowed for faster speeds. Motorola officially only rated the faster 060's at 75MHz, but with a good heat sink and fan many of them could go faster - up to 133MHz in some cases.

MC68060
Quote:
The 68060 is the last development of the 68000 family for general purpose use, abandoned in favor of the PowerPC chips. It saw use in some late-model Amiga machines and Amiga accelerator cards as well as some Atari ST clones and Falcon accelerator boards (CT60/CT63/CT60e, the latter of which was created in 2015), and very late models of the Alpha Microsystems multiuser computers before their migration to x86, but Apple Inc. and the Unix world had moved onto various RISC platforms by the time the 68060 was available. The 68060 was introduced at 50 MHz on Motorola's 0.6 µm manufacturing process. A few years later it was shrunk to 0.42 µm and clock speed raised to 66 MHz and 75 MHz. Some users managed to overclock rev6. 68060 CPU-s (mask: 71E41J) up to 120 or 133 MHz.


Contrast this with Intel's Pentium. In 1994 they introduced the P54C with clock speeds of 75 to 120MHz. The different speeds weren't manufactured separately, they were the same chip sorted according to maximum operating speed (and therefore power consumption). But in order to get the highest speed without burning up they needed a large heat sink and fan. Another issue was only a few of them worked at the higher speed. So the price was set accordingly.

IOW, Intel deliberately chose to chase the high end by making power hungry chips. That's not a bad thing, in fact it proved to be a winning strategy. Users who could afford 'bleeding edge' CPUs set the standard that others would follow as processes got better and prices came down.

Motorola could have selected faster 060's, stuck a big fan and heatsink on them, and sold them at a premium price as 133MHz devices. Amiga fans would have loved that. They could also have put more transistors on the die to improve performance, but they didn't because without Apple the desktop market for 060's was practically dead and they didn't want to follow Intel's path of maximizing performance at the expense of power consumption.

Quote:
Ah, cherry picking. Then I can do it as well: https://groups.google.com/g/comp.sys.amiga.misc/c/jWErDssEr5A

Benchmarks printed in Video Toaster User magazine show the '060 to be about half the speed of a P90.

I'll take actual numbers over some wishy-washy 'about half' without specifying the conditions, thanks.

But - as several people in that thread pointed out - if the 060 and Pentium had equivalent processing power then you would expect a 50MHz 060 to be 'about half' the speed of a 90MHz Pentium (50/90 = 56%).

Quote:
In fact it was so stupid that Commodore was DEPENDING on Motorola.

So, depending on the LATE DELIVERIES of the new chips and on... ABANDONING the 68k family.

You do realize that Commodore went bankrupt in in 1994, right? If they had been able to commit to a large quantity of 060's and could show Motorola that they would be sticking around (and paying their bills) you can bet Motorola would have been quite happy to keep the production lines open. But with Commodore gone and Apple switching to PPC there wasn't enough market to keep the 060 going. So they (sensibly) dropped it and developed 68k for the embedded market instead.

This marked a widening divide between Motorola and Intel's strategies. Previously Intel had dabbled in the high-end embedded market with the 80186 etc., but they could see that the big money was in PCs.

As for DEPENDING on a single manufacturer, Intel intended to be that manufacturer for PC makers. That is why they chose the name 'Pentium' for their 586 line, because they couldn't stop other manufacturers using a number. They also sued any chip maker who produced a compatible CPU.

All this is post AGA of course, which came out 1-2 years before 060 and Pentium CPUs. But the pace of change in the 90's was staggering. The AGA chipset was good when it came out, but 'falling behind' rapidly. To keep up Commodore would have to bring new out machines every couple of years, making Amiga fans' heads spin. IMO we as retro fans are fortunate that Commodore exited the scene when they did, since all the Amiga's chipsets are part of one 'family' that isn't hard to get familiar with. Developers only have to consider 'AGA' and 'not AGA' (OCS/ECS) and the differences aren't huge. This means AGA won't be forgotten, and I expect to see great things done with it in coming years.

Last edited by bhabbott on 15-Oct-2022 at 09:48 PM.

 Status: Offline
Profile     Report this post  
agami 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 16-Oct-2022 2:22:58
#349 ]
Super Member
Joined: 30-Jun-2008
Posts: 1651
From: Melbourne, Australia

@bhabbott

Quote:
bhabbott wrote:
Since comparing the Amiga to a PC seems to be necessary part of deciding how good AGA was,
...
the fact that with an equivalent CPU the A1200 could match a 386DX40 shows that AGA was not inferior to a 'typical' PC of the day, and in some ways was superior.

We tend to compare the Amiga to the PC of its day, just as those of us who experienced it directly back then also compared it to the growing trend of PCs among our friends and colleagues, mostly because the PC was increasingly becoming a target for gaming, which of course was Amiga's territory.
But the Amiga should not be compared to the PC, and here's why:

While in late '92, the dawn of the mayfly AGA era, the Amiga A1200 and A4000 were personal computers, sold among other personal computers such as those from Apple, Atari, and a small parade of IBM PC Compatible makers; the IBM PC Compatible, a.k.a "The PC", is not in the same category.

Apple, Atari, and Commodore sold products. Specific SKUs of personal computers.
The PC was, and still is to a large degree, a Platform.

Product marketing is very different to platform marketing in many areas, but when it comes to personal computers and consumer electronics, the key differentiation area is that products have "handles", whereas platforms don't.

Commodore struggled with this, as can be seen in the Amiga 500 TV ad with "Stevie", where the Amiga 500 was being positioned as a "Jack of all trades".
That's platform marketing.
In reality, the A500 does not have that many handles, even with expansions.

The other key area is in how SKUs are managed in Products, specifically around pricing (there's an excellent chapter on this in Dan Ariely's "Predictably Irrational"), and how companies like Commodore, Atari, and Apple, produce SKU's to a specific price target.
Platforms, especially open platforms, don't operate this way, because there is not one company that owns and manages the entire platform.

The fact remains, that in late 1992 in Australia, especially for a person looking to upgrade from an A500, the A1200HD for $1,199 (AUD) represented the best bang for buck of any available personal computer SKU of that price, and even any IBM PC Compatible of the same price, largely due to the AGA chipset.

I agree that in the big box arena, where the A4000 was less of a Product, and more a computer of the coalescing "Amiga Platform", was less bang for buck when compared to an IBM PC Compatible, but better value than a similarly spec'd Apple Mac.
This of course is irrelevant, as the AGA chipset in the A4000 was there for compatibility reasons, and the target audience for A4000 buyers were semi-pro and pro creative users in the growing digital video and 3D graphics arena, which had already bought into specific software that did not have equivalents on the PC, and would be using add-on cards, including RTG.

This is why you and @cdimauro will never reconcile. Furthermore, you are talking about your perception of real events, as they transpired in your neck of the woods, and he is focusing on spec sheets and dates of ratified standards and official technology releases, ignoring the cost and incubation period of such.

Last edited by agami on 16-Oct-2022 at 08:52 AM.
Last edited by agami on 16-Oct-2022 at 02:23 AM.

_________________
All the way, with 68k

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 16-Oct-2022 6:55:10
#350 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@bhabbott

Quote:

bhabbott wrote:
Sorry, I just have to address a couple of points...

@cdimauro

Quote:

cdimauro wrote:

So, they used much less transistors. Guess what: they consumed less power...

This is what I was talking about. I said Motorola didn't follow the same path as Intel did towards more power-hungry CPUs, and you attempt to to counter it with this. What you don't seem to realize is that you are agreeing with me.

The problem here is that you don't understand the reason for this, which is different from what you think.

Motorola was serving workstations, servers, and personal computer, and then also arcade systems at its best period, so performance was a key feature for its 68k processors family.

However they were complex chips to design and manufacture and this caused production delays. In fact and except for the 68020, Motorola was always late compared to Intel (which was late only with its 80386, but for very good reasons: it was a brand new ISA compared to its previous 80286, bringing A LOT of stuff. And complex features, of course), which was its first competitor.

So, to speed-up the time-to-market and get better performances Motorola was substantially forced to cut existing features from its 68k family.

The embedded systems are completely different chips because usually they don't need MMU and/or FPU. So, cutting those units was OK and Motorola produced specific chip versions, in fact:
https://en.wikipedia.org/wiki/Motorola_68060#68EC060
The 68EC060 is a version of the Motorola 68060 microprocessor, intended for embedded controllers (EC). It differs from the 68060 in that it has neither an FPU nor an MMU. This makes it less expensive and it draws less power.

But this does NOT justify what Motorola did on the processors for the above mentioned markets.
Quote:
When chips are manufactured they have process variations, mostly differences in doping that can't be precisely controlled. This results in some chips being faster but using more power. The manufacturer may then grade the chips according to how fast they can go, and also adjust the specs to account for the different current draw. Initially Motorola only produced one version of the 060 at 50MHz. Any chips that failed to meet that speed were discarded. Later they moved to a finer process that reduced the die size and power, which allowed for faster speeds. Motorola officially only rated the faster 060's at 75MHz, but with a good heat sink and fan many of them could go faster - up to 133MHz in some cases.

But this happened to Intel has well.
Quote:
MC68060
Quote:
The 68060 is the last development of the 68000 family for general purpose use, abandoned in favor of the PowerPC chips. It saw use in some late-model Amiga machines and Amiga accelerator cards as well as some Atari ST clones and Falcon accelerator boards (CT60/CT63/CT60e, the latter of which was created in 2015), and very late models of the Alpha Microsystems multiuser computers before their migration to x86, but Apple Inc. and the Unix world had moved onto various RISC platforms by the time the 68060 was available. The 68060 was introduced at 50 MHz on Motorola's 0.6 µm manufacturing process. A few years later it was shrunk to 0.42 µm and clock speed raised to 66 MHz and 75 MHz. Some users managed to overclock rev6. 68060 CPU-s (mask: 71E41J) up to 120 or 133 MHz.


Contrast this with Intel's Pentium. In 1994 they introduced the P54C with clock speeds of 75 to 120MHz. The different speeds weren't manufactured separately, they were the same chip sorted according to maximum operating speed (and therefore power consumption). But in order to get the highest speed without burning up they needed a large heat sink and fan. Another issue was only a few of them worked at the higher speed. So the price was set accordingly.

IOW, Intel deliberately chose to chase the high end by making power hungry chips. That's not a bad thing, in fact it proved to be a winning strategy. Users who could afford 'bleeding edge' CPUs set the standard that others would follow as processes got better and prices came down.

Motorola could have selected faster 060's, stuck a big fan and heatsink on them, and sold them at a premium price as 133MHz devices. Amiga fans would have loved that. They could also have put more transistors on the die to improve performance, but they didn't because without Apple the desktop market for 060's was practically dead and they didn't want to follow Intel's path of maximizing performance at the expense of power consumption.

Unfortunately this doesn't justify why the 68080 was stuck with the frequencies (and before it the 68040 as well).

We know that, compared to the Pentium, it:
- had less transistors;
- consumed less power;
- had a much longer pipeline.

All those factors should have made it able to reach much higher frequencies. Yes, consuming more, of course, but nevertheless: it should have reached higher frequencies. It also means that, despite how Motorola was selling those chips, provided a good heat sink and maybe a fan, people should have been able to overclock it and reach considerable higher frequencies; even compared to Pentium, at the same or similar product processes.

Unfortunately the reality was diametrically opposite.

Let's see the specs of both.
68060
https://en.wikipedia.org/wiki/Motorola_68060#Architecture
The 68060 was introduced at 50 MHz on Motorola's 0.6 µm manufacturing process. A few years later it was shrunk to 0.42 µm and clock speed raised to 66 MHz and 75 MHz. Some users managed to overclock rev6. 68060 CPU-s (mask: 71E41J) up to 120 or 133 MHz.[5][6]

https://en.wikipedia.org/wiki/Motorola_68060#Technical_data
Officially: 50, 66, 75 MHz
Overclocked: 66 (rev1-2), 80 (rev3-4), 110, 120 and 133 MHz (rev5-6)
CMOS 0.6 μm and later 0.42 μm


But it's more interesting what was reported on one of the sources for these information: https://www.amigawiki.org/doku.php?id=de:parts:68060_mask
From 1994 to 1996, the 060 (at 50, 60 & 66 MHz) are produced with 3 versions and manufactured with 0.60um process.
These first 060 are manufactured with the 1F43G mask (Rev.1) and accept up to 66 MHz (70% of them).
In 1998, the new 0.42um manufacturing process is used
Ab 1999 wurde die letzte Revision 6 im 0.32um-Prozess hergestellt


So, it was delivered using 3 productive processes: 0.6um, 0.42um and 0.32um. The most important table is at the bottom of the page and reports information about all produced processors with maximum frequencies reached and at which conditions:

1F43G	60 MHz	nur mit guter Kühlung.
1G65V 66 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
2G59Y 75 MHz nur mit guter Kühlung.
3G59Y 75 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
1E41J 90 MHz nur mit guter Kühlung.
71E41J 100 MHz nur mit guter Kühlung. Mit aufwändigem Hack auf CSMKIII möglich. Mit geringem Aufwand auf Apollo 4060 möglich.
71E41J 133 MHz nur mit guter Kühlung auf speziellen Boards, nicht am Amiga verfügbar.

It's in German but it's easy to translate and understand.

So, with 0.6um it was able to reach 66Mhz in overclocking. With 0.42um arrived to 75Mhz. And to 133Mhz with the 0.32um process. A good cooling system was a requirement and for the best models it wasn't possible to have it on an Amiga.

It's quite evidente that, despite the premises (less transistors & power, longer pipeline) the 68060 had several issues and wasn't able to scale in frequencies even using very good cooling systems.

Now take a look at the Pentium:
https://en.wikipedia.org/wiki/List_of_Intel_Pentium_processors#P5_based_Pentiums
"P54C" (0.6 μm)    75 MHz-100 MHz	
"P54CS" (0.35 μm) 133 MHz-200 MHz
"P55C" (0.35 μm) 167 MHz-233 MHz

Yes, it consumed more and required a good cooling system, but it literally obliterated the 68060 without any overclock.

Do you spot the differences?
Quote:
Quote:
Ah, cherry picking. Then I can do it as well: https://groups.google.com/g/comp.sys.amiga.misc/c/jWErDssEr5A

Benchmarks printed in Video Toaster User magazine show the '060 to be about half the speed of a P90.

I'll take actual numbers over some wishy-washy 'about half' without specifying the conditions, thanks.

But - as several people in that thread pointed out - if the 060 and Pentium had equivalent processing power then you would expect a 50MHz 060 to be 'about half' the speed of a 90MHz Pentium (50/90 = 56%).

Yes for gp/integer. However for FP workloads the Pentium ad up to 3 times performances compared to the 68060 (at the same clock).
Quote:
Quote:
In fact it was so stupid that Commodore was DEPENDING on Motorola.

So, depending on the LATE DELIVERIES of the new chips and on... ABANDONING the 68k family.

You do realize that Commodore went bankrupt in in 1994, right? If they had been able to commit to a large quantity of 060's and could show Motorola that they would be sticking around (and paying their bills) you can bet Motorola would have been quite happy to keep the production lines open.

Well, Motorola kept the production lines open even after that Commodore went in bankrupt, since the 68060 was available for very long time and even with better productive processed (see above).
Quote:
But with Commodore gone and Apple switching to PPC there wasn't enough market to keep the 060 going.

The plan was already defined for Motorola.
Quote:
So they (sensibly) dropped it and developed 68k for the embedded market instead.

Correct, but see above: the 68060 had a long life. Not Commodore, to take advantage of it...
Quote:
This marked a widening divide between Motorola and Intel's strategies. Previously Intel had dabbled in the high-end embedded market with the 80186 etc., but they could see that the big money was in PCs.

Intel continued also on the embedded market. Actually it never stopped. Even with its very old 8085 microcontrollers family, just to let you understand its strategy.
Quote:
As for DEPENDING on a single manufacturer, Intel intended to be that manufacturer for PC makers. That is why they chose the name 'Pentium' for their 586 line, because they couldn't stop other manufacturers using a number. They also sued any chip maker who produced a compatible CPU.

Nevertheless, there were other x86 vendors.
Quote:
All this is post AGA of course, which came out 1-2 years before 060 and Pentium CPUs. But the pace of change in the 90's was staggering. The AGA chipset was good when it came out, but 'falling behind' rapidly.

It was good only for what it offered at that very low price.
Quote:
To keep up Commodore would have to bring new out machines every couple of years, making Amiga fans' heads spin.

Why? Competitors did it, sometimes even after one year.

Not even talking about PCs here.
Quote:
IMO we as retro fans are fortunate that Commodore exited the scene when they did, since all the Amiga's chipsets are part of one 'family' that isn't hard to get familiar with. Developers only have to consider 'AGA' and 'not AGA' (OCS/ECS) and the differences aren't huge. This means AGA won't be forgotten, and I expect to see great things done with it in coming years.

No comment...

@agami

Quote:

agami wrote:
@bhabbott

[...]

This is why you and @cdimauro will never reconcile.

You "forgot" the A1000, A2000, A2500, and A3000 in your analysis. Lack of memory or on purpose?

Because those models (the A1000 was the first!!!) show a very different picture from what you've built. Actually, a completely different (hi)story.
Quote:
Furthermore, you are talking about your perception of real events, as they transpired in your neck of the woods, and he is focusing on spec sheets and dates of ratified standards and official technology releases,

Right. Data and facts are a little better than personal feelings, eh!
Quote:
ignoring the cost and incubation period of such.

I never did it, my friend: please don't put words that I haven't pronounced in my mouth, thanks!

 Status: Offline
Profile     Report this post  
Massi 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 16-Oct-2022 16:03:59
#351 ]
Cult Member
Joined: 2-Feb-2011
Posts: 627
From: Rome, Italy

@All

I am in line with those who think that the AGA chipset was competitive in 92 / 93.

I am more of an OCS guy, but in the retro demoscene, people are still exploring it and still pushing it to its limit.

AGA is vast and it took quite a long time back in the days for the developers to master it.

_________________
SAM440EP-FLEX @ 733 Mhz, AmigaOS 4.1 Update 1

 Status: Offline
Profile     Report this post  
matthey 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 16-Oct-2022 22:12:44
#352 ]
Elite Member
Joined: 14-Mar-2007
Posts: 2003
From: Kansas

bhabbott Quote:

Which was the case. So, it was relevant.
Quote:
Lightwave 3D rendering test, Pentium 90 MHz vs 68060 50 MHz
[quote]Test performed in CeBIT 1996

Pentium 90 MHz: 1:30
68060 50 MHz: 0:56

Notes:
Amiga Lightwave rendered the test picture faster than the P90 even if it is not really optimized for the 68060 yet. ;)


cdimauro Quote:

Ah, cherry picking. Then I can do it as well: https://groups.google.com/g/comp.sys.amiga.misc/c/jWErDssEr5A

Benchmarks printed in Video Toaster User magazine show the '060 to be about half the speed of a P90.


The huge discrepancy in 68060 floating point performance can partially be explained by OxyPatcher. Lightwave is written in SAS/C which has poor support for the 68040 and even less for the 68060. The 68k FPU code is basically 6888x code using many trapped instructions on the 68040 and 68060 even in the last versions of Lightwave. Without OxyPatcher to avoid the traps, the performance is going to be poor. The CeBIT 1996 performance is amazing and likely demonstrates compiler issues and lack of maturity with early Pentium versions of Lightwave. The ByteMark benchmarks show the 68060 FPU was nearly on par with the Pentium and had a 40% integer performance advantage at the same clock speed. Lightwave may not be as FPU intensive as the ByteMark FPU benchmarks and the 68060 likely has the advantage for mixed code, albeit an instruction scheduler would help exploit it which SAS/C also lacks. Another difference can be what Pentium CPU was used as they changed. More on that later.

Cdimauro's link has a lot of claims and boasting without much benchmarking results. Some people say a 68060@66MHz beats a Pentium@90MHz which just happens to be very close to the 40% integer performance advantage of the 68060 that ByteMark shows. Another interesting claim by someone is the following.

Philipp Boerker Quote:

>Both CPU are superscalar which means that they can execute more than one operation
>per clockcycle. The average superscalar factors for 1-cycle-ops are:
>P5 : 1.2-1.3 ops/cycle
>060: 1.6 ops/cycle !


This is likely talking about the superscalar multi-instruction issue rate. A scalar CPU issues 1 op/cycle when it is not stalled. The 68060 multi-issues 1.6 ops/cycle which is roughly inline with Motorola claims for 68060 optimized code.

https://old.hotchips.org/wp-content/uploads/hc_archives/hc06/3_Tue/HC6.S8/HC6.8.3.pdf Quote:

Measured Performance
o 1.2 CPI measured on a range of desktop and embedded applications
o 45-55% of instructions issued as pairs/triplets (existing 680X0 code)
o 50-65% of instructions issued as pairs/triplets (targeted 68060 code)


These are very impressive stats for an in-order superscalar CPU with only 8kiB I+D caches. These superscalar issue stats don't seem to be common anymore and I couldn't find any hits when searching for the Pentium P5 1.2-1.3 ops/cycle but the Pentium has much more restrictive multi-issue capabilities. There are fewer multi-issue capable instructions, more restrictions on what superscalar order instructions fall in and basic CISC reg-mem instructions like OP mem,reg is 2 cycles and OP reg,mem 3 cycles while these are 1 cycle on the 68060. This is a huge difference and explains why Intel quickly moved to OoO for x86. While searching, I found some old superscalar stats from the Motorola 88k.

http://cjat.ir/images/PDF_English/20143.pdf Quote:

A significant characteristic of the 88110 is that it makes parallel instruction execution fairly easy to achieve in practice. Relatively simple compilers can produce effective code schedules for the 88110; in fact, the processor realizes substantial parallelism even on code originally generated for the 88100 single-issue CPU. The efficiency of superscalar issue ranges from 20 percent to over 50 percent, depending on the benchmark and memory configuration. Currently, over the SPEC benchmark suite, we find that two instructions issue on roughly half (ranging from about 35-70 percent) of the clock cycles on which an instruction executes at all. Of course, we expect these results to continually improve with advances in our compilers.


The 88110 is able to multi-issue 35%-70% (~53% on average) of the time for optimized code. This is very good as there are few restrictions on superscalar issue but this is a dual issue OoO CPU. The 68060 multi-issues at a comparable if not higher rate than the OoO 88110. The multi-issue of non-scheduled code is very good for both of these CPUs but again the in-order 68060 may be better than the OoO 88110 CPU. I would love to see superscalar multi-issue rates for modern RISC CPUs like the popular ARM Cortex-A7 or Cortex-A53 which I expect are low due to restrictions like the Pentium. I expect non-scheduled code to be even lower and barely superscalar like the Pentium. Too often the old technology is forgotten and reinvented wheels are inferior to the past.

88110 April 1992 Quote:

Base architecture extensions. We made the following minor enhancements of the base instruction set:

Extensions of integer multiply and divide to improve support for signed multiplication and for arithmetic on higher precision integers. Instructions permit multiplication of two 32-bit numbers returning a full 64-bit result, and division of a 64.bit number by a 32-bit number, returning a 64-bit quotient.


Motorola removed the same instructions from the 68060 shortly after adding them to the 88110.

The 68060 had other advantages to the Pentium. One of the most significant performance advantages was the longer pipeline which allows more Instruction Level Parallelism (ILP). Even without taking advantage of clocking up the 68060, the 68060 could process 8 instructions in parallel instead of 5. The 68060 had good branch prediction, register renaming and instruction bypass/forwarding to take advantage of this amount of pipelining. The 3rd most influential performance advantage was likely better cache efficiency from better code density, better code alignment and more associative ways. The 68060 was the Pentium killer that was chained in the basement at 50MHz for embedded use until newer and cheaper CPUs surpasses it.

cdimauro Quote:

The problem here is that you don't understand the reason for this, which is different from what you think.

Motorola was serving workstations, servers, and personal computer, and then also arcade systems at its best period, so performance was a key feature for its 68k processors family.

However they were complex chips to design and manufacture and this caused production delays. In fact and except for the 68020, Motorola was always late compared to Intel (which was late only with its 80386, but for very good reasons: it was a brand new ISA compared to its previous 80286, bringing A LOT of stuff. And complex features, of course), which was its first competitor.

So, to speed-up the time-to-market and get better performances Motorola was substantially forced to cut existing features from its 68k family.


x86 CPUs were much more complex to design compared to 68k CPUs. Intel had higher margins and increasing revenue from desktop CPUs while the 68k was losing the highest margin but smaller workstation market. It didn't help that the PPC markets were competing with the 68k and everyone but low end embedded customers were being told to switch to PPC which was receiving all the development funding from the growing 68k embedded market. Motorola management chose to cut existing features to lower prices for the embedded market where price is very important. It probably saved some time but I doubt that was the primary reason. Cutting too much sometimes affected performance and compatibility to their detriment.

cdimauro Quote:

The embedded systems are completely different chips because usually they don't need MMU and/or FPU. So, cutting those units was OK and Motorola produced specific chip versions, in fact:
https://en.wikipedia.org/wiki/Motorola_68060#68EC060
The 68EC060 is a version of the Motorola 68060 microprocessor, intended for embedded controllers (EC). It differs from the 68060 in that it has neither an FPU nor an MMU. This makes it less expensive and it draws less power.

But this does NOT justify what Motorola did on the processors for the above mentioned markets.


Sometimes, Motorola chips supposedly without MMU and/or FPU were just remarked full chips even in the case of the 68060. I'm surprised Motorola didn't always produce a full 68060 and work on improving economies of scale to lower the price. It is for high end embedded where the MMU and FPU are more likely to be used, production isn't as high as say a 68EC020 where it can be split and the 68060 uses power gating to turn off unused units so they draw very little power.

cdimauro Quote:

Unfortunately this doesn't justify why the 68060 was stuck with the frequencies (and before it the 68040 as well).

We know that, compared to the Pentium, it:
- had less transistors;
- consumed less power;
- had a much longer pipeline.

All those factors should have made it able to reach much higher frequencies. Yes, consuming more, of course, but nevertheless: it should have reached higher frequencies. It also means that, despite how Motorola was selling those chips, provided a good heat sink and maybe a fan, people should have been able to overclock it and reach considerable higher frequencies; even compared to Pentium, at the same or similar product processes.

Unfortunately the reality was diametrically opposite.

Let's see the specs of both.
68060
https://en.wikipedia.org/wiki/Motorola_68060#Architecture
The 68060 was introduced at 50 MHz on Motorola's 0.6 µm manufacturing process. A few years later it was shrunk to 0.42 µm and clock speed raised to 66 MHz and 75 MHz. Some users managed to overclock rev6. 68060 CPU-s (mask: 71E41J) up to 120 or 133 MHz.[5][6]

https://en.wikipedia.org/wiki/Motorola_68060#Technical_data
Officially: 50, 66, 75 MHz
Overclocked: 66 (rev1-2), 80 (rev3-4), 110, 120 and 133 MHz (rev5-6)
CMOS 0.6 μm and later 0.42 μm


But it's more interesting what was reported on one of the sources for these information: https://www.amigawiki.org/doku.php?id=de:parts:68060_mask
From 1994 to 1996, the 060 (at 50, 60 & 66 MHz) are produced with 3 versions and manufactured with 0.60um process.
These first 060 are manufactured with the 1F43G mask (Rev.1) and accept up to 66 MHz (70% of them).
In 1998, the new 0.42um manufacturing process is used
Ab 1999 wurde die letzte Revision 6 im 0.32um-Prozess hergestellt


So, it was delivered using 3 productive processes: 0.6um, 0.42um and 0.32um. The most important table is at the bottom of the page and reports information about all produced processors with maximum frequencies reached and at which conditions:
1F43G	60 MHz	nur mit guter Kühlung.
1G65V 66 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
2G59Y 75 MHz nur mit guter Kühlung.
3G59Y 75 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
1E41J 90 MHz nur mit guter Kühlung.
71E41J 100 MHz nur mit guter Kühlung. Mit aufwändigem Hack auf CSMKIII möglich. Mit geringem Aufwand auf Apollo 4060 möglich.
71E41J 133 MHz nur mit guter Kühlung auf speziellen Boards, nicht am Amiga verfügbar.

It's in German but it's easy to translate and understand.

So, with 0.6um it was able to reach 66Mhz in overclocking. With 0.42um arrived to 75Mhz. And to 133Mhz with the 0.32um process. A good cooling system was a requirement and for the best models it wasn't possible to have it on an Amiga.

It's quite evidente that, despite the premises (less transistors & power, longer pipeline) the 68060 had several issues and wasn't able to scale in frequencies even using very good cooling systems.

Now take a look at the Pentium:
https://en.wikipedia.org/wiki/List_of_Intel_Pentium_processors#P5_based_Pentiums
"P54C" (0.6 μm)    75 MHz-100 MHz	
"P54CS" (0.35 μm) 133 MHz-200 MHz
"P55C" (0.35 μm) 167 MHz-233 MHz

Yes, it consumed more and required a good cooling system, but it literally obliterated the 68060 without any overclock.

Do you spot the differences?


I spot the differences you don't mention.

Pentium
P5 0.8um 60-66MHz 3.1 million transistors
P54C 0.6um 75-100MHz 3.3 million transistors
P54CS 0.35um 133-200MHz 3.3 million transistors
P55C 0.35um 167-233MHz 4.5 million transistors

68060
1F43G 0.6um 50-60MHz 2.5 million transistors
1G65V 0.6um 50-60MHz 2.5 million transistors
2G59Y 0.6um 50-60MHz 2.5 million transistors
3G59Y 0.6um 50-60MHz 2.5 million transistors
1E41J 0.42um 50-60MHz 2.5 million transistors
71E41J 0.42um 50-60MHz 2.5 million transistors
71E41J 0.32um 50-60MHz 2.5 million transistors

I've never seen a full 68060 marked above 60MHz. There is a better chance it is a fake than real if you find one. Note that the 0.6um and 0.5um BICMOS process are the same as bhabbott used the latter earlier which Motorola often favored.

The Pentium was significantly improved with each process upgrade. Intel was obviously putting many man hours of labor into optimizing each upgrade while Motorola was doing the minimum effort to upgrade to new processes and was making zero effort to increase the clock rating. At the same time, Motorola was putting maximum effort into optimizing and raising PPC clock ratings which boosted performance until their large cache size became the limiting factor to increasing clock ratings. That's when x86 CPUs caught up and surpassed PPC as they were more cache efficient. The most cache efficient CPU was not even allowed in the competition.


cdimauro Quote:

Yes for gp/integer. However for FP workloads the Pentium ad up to 3 times performances compared to the 68060 (at the same clock).


Is theoretical performance real performance? The Pentium had much better compiler support than the 68060 yet I haven't seen any benchmarks showing 3 times the FP performance. Intel was good at making theoretically high performance CPUs like the i860 mentioned earlier. Motorola used to value real performance which was easy to obtain but then they traded in the 68k and 88k for PPC.

cdimauro Quote:

Well, Motorola kept the production lines open even after that Commodore went in bankrupt, since the 68060 was available for very long time and even with better productive processed (see above).


Despite not clocking up the 68060, the last process upgrade was 1999 and EOL was 2015 according to your link. The process upgrades would not have happened without enough demand showing how successful the 68060 was, likely being profitable for a decade. The in-order original Pentiums are shown as 1993-1999 by Wiki but they were really several different CPU designs with an even shorter lifespan. Intel quickly lost embedded market share due to higher power x86 CPUs but higher performance CPUs had good margin despite being short lived.

cdimauro Quote:

Correct, but see above: the 68060 had a long life. Not Commodore, to take advantage of it...


We agree. The margin on high end embedded market CPUs is good. The lifespan of an upgraded 68060 could have been longer but they replaced it with PPC and then gave the market to ARM.

cdimauro Quote:

Intel continued also on the embedded market. Actually it never stopped. Even with its very old 8085 microcontrollers family, just to let you understand its strategy.


Intel lost a lot of the 32 bit embedded market share to the 68k and then they both lost it to SuperH and ARM with improved code density and prices while the 68k was practically dropped for PPC. Intel also lost a lot of their early memory market to Japan. It looked pretty bad for Intel until IBM made the fateful decision to choose the lackluster 8088 for the IBM PC.

cdimauro Quote:

Nevertheless, there were other x86 vendors.


There were multiple 68k chip producers too but they made lower end CPUs under license for embedded use. Many customers chose to buy from Intel and Motorola as both had good reputations.

cdimauro Quote:

It was good only for what it offered at that very low price.


AGA was good for being low end. The problem is that low end is only one step above obsolete. It's probably better to release a chipset at mid to high end and discontinue at low end to obsolete. This doesn't matter for a C64 replacement toy according to CBM though.

Last edited by matthey on 16-Oct-2022 at 10:17 PM.

 Status: Offline
Profile     Report this post  
bhabbott 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 16-Oct-2022 23:33:00
#353 ]
Regular Member
Joined: 6-Jun-2018
Posts: 334
From: Aotearoa

@cdimauro

Quote:

cdimauro wrote:

We know that, compared to the Pentium, it:
- had less transistors;
- consumed less power;
- had a much longer pipeline.

All those factors should have made it able to reach much higher frequencies. Yes, consuming more, of course, but nevertheless: it should have reached higher frequencies. It also means that, despite how Motorola was selling those chips, provided a good heat sink and maybe a fan, people should have been able to overclock it and reach considerable higher frequencies; even compared to Pentium, at the same or similar product processes.

Unfortunately the reality was diametrically opposite.

I overclocked my 060 from 50MHz to 66MHz (because Quake was too slow at 50MHz). To run reliably at this speed it needed active cooling. However there was very little room between the CPU and floppy drive tray in the A3000, so I cut a hole in the drive tray to get a fan onto the CPU. This took ages because the metal was thick and and I only had hand tools.

It wouldn't work reliably over 66MHz, but this may have been partly due to timing issues on the Cyberstorm card (it wasn't designed to go faster then 50MHz). When you say 'people should have been able to overclock it' that assumes they had the resources to do it. Accelerator card manufacturers would most likely design their card to work at the rated clock speed. These cards were expensive already without making them faster to cope with users overclocking their CPUs.

Unfortunately all the better 060's have now been snapped up by retro enthusiasts and/or scalpers, so we may never get an opportunity to find out how fast selected chips could go with a modern board optimized for highest speed. However we do have anecdotal evidence of later revision 060's going as fast as 133MHz, suggesting that reality was NOT 'diametrically opposite'.


Last edited by bhabbott on 16-Oct-2022 at 11:33 PM.

 Status: Offline
Profile     Report this post  
Kremlar 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 0:53:51
#354 ]
Regular Member
Joined: 12-Aug-2010
Posts: 108
From: Milford, MA

Huge letdown. Diehard Amiga fans were looking for something to propel Commodore/Amiga back into the forefront. That didn't happen, and may bailed shortly after AGA.

 Status: Offline
Profile     Report this post  
ferrels 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 3:54:24
#355 ]
Cult Member
Joined: 20-Oct-2005
Posts: 922
From: Arizona

@Kremlar

Quote:
Huge letdown. Diehard Amiga fans were looking for something to propel Commodore/Amiga back into the forefront. That didn't happen, and may bailed shortly after AGA.


That was my exact experience. I spent a small fortune on an A1200 with an 030/50 Mhz accelerator with FPU and 16MB of FastRAM. Even with the accelerator my A1200 was anemic and CPU benchmarks put it in the same class as a 486-33Mhz and it ran so hot that it nearly melted the case. Lack of chunky graphics was the nail in the coffin for me. My work PCs literally ran rings around my A1200 graphically as well as with CPU intensive tasks so I quickly boxed the A1200 and sent it off to be stored for 12 years before I even got it out again for the sake of nostalgia.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 5:17:21
#356 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@Massi

Quote:

Massi wrote:
@All

I am in line with those who think that the AGA chipset was competitive in 92 / 93.

It's was cheap: that was its major feature. But definitely not competitive.
Quote:
I am more of an OCS guy, but in the retro demoscene, people are still exploring it and still pushing it to its limit.

AGA is vast and it took quite a long time back in the days for the developers to master it.

Better to look to something else IMO. Like fixing the pile of badly written games or applications that weren't able to run or had issues running of newer machines, AGA included.

AGA as it is it's really a horrible patch over ECS and very castrated (4x memory bandwidth but the exactly the same Blitter): it doesn't look attractive.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 5:28:21
#357 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@bhabbott

Quote:

bhabbott wrote:
@cdimauro

Quote:

cdimauro wrote:

We know that, compared to the Pentium, it:
- had less transistors;
- consumed less power;
- had a much longer pipeline.

All those factors should have made it able to reach much higher frequencies. Yes, consuming more, of course, but nevertheless: it should have reached higher frequencies. It also means that, despite how Motorola was selling those chips, provided a good heat sink and maybe a fan, people should have been able to overclock it and reach considerable higher frequencies; even compared to Pentium, at the same or similar product processes.

Unfortunately the reality was diametrically opposite.

I overclocked my 060 from 50MHz to 66MHz (because Quake was too slow at 50MHz). To run reliably at this speed it needed active cooling.

Do you understand that then you're debunking yourself what you stated before about the marvelous 68060 which didn't even required a heat sink?
Quote:
However there was very little room between the CPU and floppy drive tray in the A3000, so I cut a hole in the drive tray to get a fan onto the CPU. This took ages because the metal was thick and and I only had hand tools.

It wouldn't work reliably over 66MHz, but this may have been partly due to timing issues on the Cyberstorm card (it wasn't designed to go faster then 50MHz). When you say 'people should have been able to overclock it' that assumes they had the resources to do it. Accelerator card manufacturers would most likely design their card to work at the rated clock speed. These cards were expensive already without making them faster to cope with users overclocking their CPUs.

But seems that several people were able to overclock their cards.
Quote:
Unfortunately all the better 060's have now been snapped up by retro enthusiasts and/or scalpers, so we may never get an opportunity to find out how fast selected chips could go with a modern board optimized for highest speed. However we do have anecdotal evidence of later revision 060's going as fast as 133MHz, suggesting that reality was NOT 'diametrically opposite'.

Bruce, you're really a desperate case.

I've already reported information about those revision v6 chips and what they required to run (IF able) at those frequencies. Here is again:

71E41J 100 MHz nur mit guter Kühlung. Mit aufwändigem Hack auf CSMKIII möglich. Mit geringem Aufwand auf Apollo 4060 möglich.

which translated is:

only with good cooling. Possible with complex hack on CSMKIII. Possible with little effort on Apollo 4060.

But:

71E41J 133 MHz nur mit guter Kühlung auf speziellen Boards, nicht am Amiga verfügbar.

which translated is:

only with good cooling on special boards, not available on the Amiga.

Do you understand it?

Anyway, and what's even worse, you don't understand why I've said "diametrically opposite", even whey I've reported the data. Which I do it again and for the last time for your benefit (IF any).

Those chips were using the best productive process: 0.32um. And to reach such frequencies they required super special cooling.

Now take a look at the Pentiums:

"P55C" (0.35 μm) 167 MHz-233 MHz

They were able to reach 233Mhz. On a WORSE productive process. With MANY MORE transistors (those were the Pentium MMX chips!). And WITHOUT requiring a super special cooling.

Do you understand now why the "diametrically opposite"? Because FACTs are "diametrically opposite" compared to what YOU have reported before about this discussion.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 6:00:02
#358 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3650
From: Germany

@matthey

Quote:

matthey wrote:
cdimauro Quote:

Ah, cherry picking. Then I can do it as well: https://groups.google.com/g/comp.sys.amiga.misc/c/jWErDssEr5A

Benchmarks printed in Video Toaster User magazine show the '060 to be about half the speed of a P90.


The huge discrepancy in 68060 floating point performance can partially be explained by OxyPatcher. Lightwave is written in SAS/C which has poor support for the 68040 and even less for the 68060. The 68k FPU code is basically 6888x code using many trapped instructions on the 68040 and 68060 even in the last versions of Lightwave. Without OxyPatcher to avoid the traps, the performance is going to be poor.

Possible. But a guy in a post said that Lightwave increased the speed by 300% when it was properly compiled for the 68040. So, I assume that OxyPatcher wasn't required at least for this processor and which also means that it was good for the 68060 as well, since it had an extra instruction (which was removed on the 68040).

What was left I think it was the instructions ordering for the 68060's superpipeline.
Quote:
The CeBIT 1996 performance is amazing and likely demonstrates compiler issues and lack of maturity with early Pentium versions of Lightwave.

Reasonable. Lightwave was developed on the Amiga and optimized for it (see above). Only late it was ported to the PC.
Quote:
The ByteMark benchmarks show the 68060 FPU was nearly on par with the Pentium and had a 40% integer performance advantage at the same clock speed. Lightwave may not be as FPU intensive as the ByteMark FPU benchmarks and the 68060 likely has the advantage for mixed code, albeit an instruction scheduler would help exploit it which SAS/C also lacks. Another difference can be what Pentium CPU was used as they changed. More on that later.

As you know, I don't like those synthetic benchmarks: I always prefer real world applications. A rich set, if possible.
Quote:
Cdimauro's link has a lot of claims and boasting without much benchmarking results. Some people say a 68060@66MHz beats a Pentium@90MHz which just happens to be very close to the 40% integer performance advantage of the 68060 that ByteMark shows. Another interesting claim by someone is the following.

Philipp Boerker Quote:

>Both CPU are superscalar which means that they can execute more than one operation
>per clockcycle. The average superscalar factors for 1-cycle-ops are:
>P5 : 1.2-1.3 ops/cycle
>060: 1.6 ops/cycle !


This is likely talking about the superscalar multi-instruction issue rate. A scalar CPU issues 1 op/cycle when it is not stalled. The 68060 multi-issues 1.6 ops/cycle which is roughly inline with Motorola claims for 68060 optimized code.

https://old.hotchips.org/wp-content/uploads/hc_archives/hc06/3_Tue/HC6.S8/HC6.8.3.pdf Quote:

Measured Performance
o 1.2 CPI measured on a range of desktop and embedded applications
o 45-55% of instructions issued as pairs/triplets (existing 680X0 code)
o 50-65% of instructions issued as pairs/triplets (targeted 68060 code)


These are very impressive stats for an in-order superscalar CPU with only 8kiB I+D caches.

They look strange to me. On the usual code (not fully optimized for the 68060) 16-bit instructions didn't even reached 50% of the total (on average).

The 68060 was able to execute a pair of instructions only if they were both 16-bit.

Let's simplify the calculations and say that 16-bit instructions were exactly 50% of the total. It means that the probability to have (and execute) a pair of them is 50% * 50% = 25%. So, very very distant from the above numbers.
Quote:
These superscalar issue stats don't seem to be common anymore and I couldn't find any hits when searching for the Pentium P5 1.2-1.3 ops/cycle but the Pentium has much more restrictive multi-issue capabilities. There are fewer multi-issue capable instructions, more restrictions on what superscalar order instructions fall in

Really? I was recalling differently. I'll check again the optimization manual once I've time.
Quote:
and basic CISC reg-mem instructions like OP mem,reg is 2 cycles and OP reg,mem 3 cycles while these are 1 cycle on the 68060. This is a huge difference and explains why Intel quickly moved to OoO for x86.

Yes, definitely this is an advantage for the 68060.
Quote:
While searching, I found some old superscalar stats from the Motorola 88k.

http://cjat.ir/images/PDF_English/20143.pdf Quote:

A significant characteristic of the 88110 is that it makes parallel instruction execution fairly easy to achieve in practice. Relatively simple compilers can produce effective code schedules for the 88110; in fact, the processor realizes substantial parallelism even on code originally generated for the 88100 single-issue CPU. The efficiency of superscalar issue ranges from 20 percent to over 50 percent, depending on the benchmark and memory configuration. Currently, over the SPEC benchmark suite, we find that two instructions issue on roughly half (ranging from about 35-70 percent) of the clock cycles on which an instruction executes at all. Of course, we expect these results to continually improve with advances in our compilers.


The 88110 is able to multi-issue 35%-70% (~53% on average) of the time for optimized code. This is very good as there are few restrictions on superscalar issue but this is a dual issue OoO CPU. The 68060 multi-issues at a comparable if not higher rate than the OoO 88110. The multi-issue of non-scheduled code is very good for both of these CPUs but again the in-order 68060 may be better than the OoO 88110 CPU. I would love to see superscalar multi-issue rates for modern RISC CPUs like the popular ARM Cortex-A7 or Cortex-A53 which I expect are low due to restrictions like the Pentium. I expect non-scheduled code to be even lower and barely superscalar like the Pentium. Too often the old technology is forgotten and reinvented wheels are inferior to the past.

88110 April 1992 Quote:

Base architecture extensions. We made the following minor enhancements of the base instruction set:

Extensions of integer multiply and divide to improve support for signed multiplication and for arithmetic on higher precision integers. Instructions permit multiplication of two 32-bit numbers returning a full 64-bit result, and division of a 64.bit number by a 32-bit number, returning a 64-bit quotient.


Motorola removed the same instructions from the 68060 shortly after adding them to the 88110.

Thanks for the above data. Which confirms the stupid decision to drop the 68k family in favor of RISCs. Bah...
Quote:
The 68060 had other advantages to the Pentium. One of the most significant performance advantages was the longer pipeline which allows more Instruction Level Parallelism (ILP). Even without taking advantage of clocking up the 68060, the 68060 could process 8 instructions in parallel instead of 5.

But it had a bigger performance hit when it had to refill it with fresh new instructions.
Quote:
The 68060 had good branch prediction, register renaming and instruction bypass/forwarding to take advantage of this amount of pipelining. The 3rd most influential performance advantage was likely better cache efficiency from better code density, better code alignment and more associative ways. The 68060 was the Pentium killer that was chained in the basement at 50MHz for embedded use until newer and cheaper CPUs surpasses it.

Yes, it had many advantages.

But Pentium had also its own: 64-bit data bus, larger instructions fetch buffer (so, being able to pair longer instructions), all instructions supported (no trapping and/or software patches or special library versions needed).
Quote:
cdimauro Quote:

Unfortunately this doesn't justify why the 68060 was stuck with the frequencies (and before it the 68040 as well).

We know that, compared to the Pentium, it:
- had less transistors;
- consumed less power;
- had a much longer pipeline.

All those factors should have made it able to reach much higher frequencies. Yes, consuming more, of course, but nevertheless: it should have reached higher frequencies. It also means that, despite how Motorola was selling those chips, provided a good heat sink and maybe a fan, people should have been able to overclock it and reach considerable higher frequencies; even compared to Pentium, at the same or similar product processes.

Unfortunately the reality was diametrically opposite.

Let's see the specs of both.
68060
https://en.wikipedia.org/wiki/Motorola_68060#Architecture
The 68060 was introduced at 50 MHz on Motorola's 0.6 µm manufacturing process. A few years later it was shrunk to 0.42 µm and clock speed raised to 66 MHz and 75 MHz. Some users managed to overclock rev6. 68060 CPU-s (mask: 71E41J) up to 120 or 133 MHz.[5][6]

https://en.wikipedia.org/wiki/Motorola_68060#Technical_data
Officially: 50, 66, 75 MHz
Overclocked: 66 (rev1-2), 80 (rev3-4), 110, 120 and 133 MHz (rev5-6)
CMOS 0.6 μm and later 0.42 μm


But it's more interesting what was reported on one of the sources for these information: https://www.amigawiki.org/doku.php?id=de:parts:68060_mask
From 1994 to 1996, the 060 (at 50, 60 & 66 MHz) are produced with 3 versions and manufactured with 0.60um process.
These first 060 are manufactured with the 1F43G mask (Rev.1) and accept up to 66 MHz (70% of them).
In 1998, the new 0.42um manufacturing process is used
Ab 1999 wurde die letzte Revision 6 im 0.32um-Prozess hergestellt


So, it was delivered using 3 productive processes: 0.6um, 0.42um and 0.32um. The most important table is at the bottom of the page and reports information about all produced processors with maximum frequencies reached and at which conditions:
1F43G	60 MHz	nur mit guter Kühlung.
1G65V 66 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
2G59Y 75 MHz nur mit guter Kühlung.
3G59Y 75 MHz ohne Kühlung, passiv mit Kühlkörper kann zur eigenen Beruhigung gemacht werden.
1E41J 90 MHz nur mit guter Kühlung.
71E41J 100 MHz nur mit guter Kühlung. Mit aufwändigem Hack auf CSMKIII möglich. Mit geringem Aufwand auf Apollo 4060 möglich.
71E41J 133 MHz nur mit guter Kühlung auf speziellen Boards, nicht am Amiga verfügbar.

It's in German but it's easy to translate and understand.

So, with 0.6um it was able to reach 66Mhz in overclocking. With 0.42um arrived to 75Mhz. And to 133Mhz with the 0.32um process. A good cooling system was a requirement and for the best models it wasn't possible to have it on an Amiga.

It's quite evidente that, despite the premises (less transistors & power, longer pipeline) the 68060 had several issues and wasn't able to scale in frequencies even using very good cooling systems.

Now take a look at the Pentium:
https://en.wikipedia.org/wiki/List_of_Intel_Pentium_processors#P5_based_Pentiums
"P54C" (0.6 μm)    75 MHz-100 MHz	
"P54CS" (0.35 μm) 133 MHz-200 MHz
"P55C" (0.35 μm) 167 MHz-233 MHz

Yes, it consumed more and required a good cooling system, but it literally obliterated the 68060 without any overclock.

Do you spot the differences?


I spot the differences you don't mention.

Pentium
P5 0.8um 60-66MHz 3.1 million transistors
P54C 0.6um 75-100MHz 3.3 million transistors
P54CS 0.35um 133-200MHz 3.3 million transistors
P55C 0.35um 167-233MHz 4.5 million transistors

68060
1F43G 0.6um 50-60MHz 2.5 million transistors
1G65V 0.6um 50-60MHz 2.5 million transistors
2G59Y 0.6um 50-60MHz 2.5 million transistors
3G59Y 0.6um 50-60MHz 2.5 million transistors
1E41J 0.42um 50-60MHz 2.5 million transistors
71E41J 0.42um 50-60MHz 2.5 million transistors
71E41J 0.32um 50-60MHz 2.5 million transistors

I didn't mentioned it because this is another advantage for the Pentiums.

They were able to achieve those much higher frequencies despite packing many more transistors. Especially for the MMX versions.
Quote:
cdimauro Quote:

Yes for gp/integer. However for FP workloads the Pentium ad up to 3 times performances compared to the 68060 (at the same clock).


Is theoretical performance real performance?

Theoretical, due to the pipelined FPU.
Quote:
The Pentium had much better compiler support than the 68060 yet I haven't seen any benchmarks showing 3 times the FP performance.

That's why I liked to have results for SPEC FP for both Pentium and 68060.
Quote:
It looked pretty bad for Intel until IBM made the fateful decision to choose the lackluster 8088 for the IBM PC.

It looks like that IBM had no other choice, since the 68k wasn't able to fit its requirements (especially the missing second supplier).
Quote:
cdimauro Quote:

It was good only for what it offered at that very low price.


AGA was good for being low end. The problem is that low end is only one step above obsolete. It's probably better to release a chipset at mid to high end and discontinue at low end to obsolete.

Exactly. AGA was low-end but... it was used on the top notch machine: the Amiga 4000. Only Commodore could have done it...
Quote:
This doesn't matter for a C64 replacement toy according to CBM though.

Sad but true. Commodore was a disaster. At all levels...

 Status: Offline
Profile     Report this post  
Hammer 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 7:14:58
#359 ]
Elite Member
Joined: 9-Mar-2003
Posts: 5275
From: Australia

@bhabbott

Quote:

bhabbott wrote:
@cdimauro

Quote:

cdimauro wrote:


Sure: if you bundle it with a crap 8-bit ISA graphic card.

I doubt that many 386DX systems came with 8 bit ISA cards, that would be stupid.


Quote:
Do you understand that the your AGA had a much better graphic subsystem compared to a crap 8-bit ISA card?

In 1992 the majority of PCs sold were 386SX and 386DX models, with 'crappy' 16 bit ISA bus graphics cards (not the fabled ET4000). Whether that qualifies AGA as 'better than most' depends on how you count them, but in terms of how AGA compareed to the average PC graphics card...

And that is all we should be discussing here, not 68060 vs Pentium etc.

That's a flawed argument when PC DOS Doom sales smashed under a million A1200 install base.

https://doom.fandom.com/wiki/Sales

Based on these data, one may place the sales until 1999 in the range 1.9–2.1+ million for Doom (including Ultimate Doom), and 1.5–1.8+ million for Doom II. It is probably safe to deduce that as of 2005, both Doom and Doom II have sold well over 2 million copies each





_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

 Status: Offline
Profile     Report this post  
agami 
Re: How good or bad was the AGA chipset in 1992/1993.
Posted on 17-Oct-2022 7:46:36
#360 ]
Super Member
Joined: 30-Jun-2008
Posts: 1651
From: Melbourne, Australia

@cdimauro

Quote:
It's was cheap: that was its major feature. But definitely not competitive.

Exactly, nothing else could compete with it in its price class.

_________________
All the way, with 68k

 Status: Offline
Profile     Report this post  
Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 Next Page )

[ home ][ about us ][ privacy ] [ forums ][ classifieds ] [ links ][ news archive ] [ link to us ][ user account ]
Copyright (C) 2000 - 2019 Amigaworld.net.
Amigaworld.net was originally founded by David Doyle