@OlafS25

Quote:

when they had earned more money that also would helped the 4.X line. Ignoring customer wishes is never a good idea for a business.

I don´t think so. It would lead to further delay and even OS4 "cash cow" "BAF" users don´t have unlimited patience.

@pavlor

But 68k users outnumber "BAF" multiple times

what you write shows how far this community is away from real markets

@OlafS25

Quote:

Wintel hate still in 2002/2003?

The first mention of a PowerAmiga goes back to 1995. It was a plan of Amiga Technologies that of course never came to fruition. It was planned to released in 1997. Still plenty of Wintel hate in that period.

Petro famously told a story about asking a person at an expo to demonstrate the multitasking capabilities of Win 95. After getting the response "Just let me finish formatting this floppy disk", Petro replied "Thanks, that's OK."

And during the Gateway era and immediately after it, a plan was set to transition from 68k to PowerPC, and to a portable OS. Haage & Patner were going to bridge the gap by delivering OS 3.5 and 3.9. Then EyeTech was selected to deliver PowerPC hardware that would run AmigaOS 4 which was outsourced to Hyperion Entertainment so that the newly formed Amiga Inc. could focus on AmigaOS 5.
The Pentium III of the late '90s was an OK CPU but I wasn't blowing anyone's hair back. And neither was Win 98 SE. Actually, it was in early 2000 that AMD beat Intel to the 1GHz mark, and I remember many people revelling in the triumph. We didn't care who won as long as Intel lost.

But yes, by the mid '00s most people stopped caring about the CPU. Alas, for me it's not about the hate, rather it's about cheering for the underdog. I use Intel CPUs but I still cheer for AMD when they win lucrative console part contracts.

And when Microsoft with all its billions of dollars can't bring to market a smartphone platform, I feel vindicated. These people are not brilliant geniuses, they lucked out at a crucial point in computing history and have been riding that wave ever since. Though the Xbox division has done well, I have to wonder if they would have done so if SEGA never faltered.

_________________
All the way, with 68k

@OlafS25

Quote:

But 68k users outnumber "BAF" multiple times

If you mean users willing to pay for software, then certainly not...
If you mean gamers demanding everything for free, then these aren´t real customers...

Quote:

what you write shows how far this community is away from real markets

Amiga is not driven by market logic at least since 1994.

@pavlor

Quote:

If you mean users willing to pay for software, then certainly not... If you mean gamers demanding everything for free, then these aren´t real customers...

there must be some balance calculation between paying customers and the sharing community, enterprises usually base their policies on. a restrictive one may not always the one that gains most return. among others huge community, whether they all pay or not generates considerable buzz around a product, a marketing value in itself.

funny enough they started to realize, or rather take advantage of that, even if much too late with the release of 4.1fe just in time for winuae ppc extension. im pretty sure there must have been people who allowed themselves to make a "free" copy of os4 along with cyberstorm firmware. but the loyal usership remained a stable base and no matter what, the comparatively liberal policy attracted additional "backers".

@wawa

Commodore wanted to use the PA RISC, so they did intend to move on from 68K.
So if Commodore was still around to further development of the Amiga, we would have seen Commodore abandoning the 68K cpu a long time ago.

_________________

@OlafS25

Quote:

what you write shows how far this community is away from real markets

Yes, great isn't it? What a small entity is capable of where the big ones fail miserably!

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@Yssing

Quote:

Commodore wanted to use the PA RISC, so they did intend to move on from 68K.

For chipset in CD32 successor, not as general purpose CPU in Amiga desktop range.

Quote:

So if Commodore was still around to further development of the Amiga, we would have seen Commodore abandoning the 68K cpu a long time ago.

Highly probably.

@NutsAboutAmiga

Quote:

Hex decimal numbers on PowerPC and 680x0 are lot easier to read then on Intel.

That's where THIS comes in quite handy...

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@cdimauro

Quote:

In short: it's just another example of very bad management of Commodore's heritage...

Glad we got used to it as we endured it for so long and by now know how to deal with it, no?

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@pavlor

Quote:

For chipset in CD32 successor, not as general purpose CPU in Amiga desktop range.

I've read quotes from people working at C= at that time that PA-RISC was much preferred to anything else(including PPC) because HP was already manufacturing C= chips in their foundries and they were OK with C= making custom PA-RISC chips for it's own use.
The official decision was never made because of the bankruptcy and lack of money in that period.

_________________

@WolfToTheMoon

Exactly, don't forget Hombre chipset.

_________________

@WolfToTheMoon

Quote:

I've read quotes from people working at C= at that time that PA-RISC was much preferred to anything else

Depends on people you quote - nearly everyone at Commodore had distinct idea of future...

Quote:

The official decision was never made because of the bankruptcy and lack of money in that period.

To paraphrase Dave Haynie: They weren´t able to port 3 libraries to AAA, how they can even dream about new CPU ISA.

@OldFart

Quote:

we got used to it

No, we came to expect it and accepted it as normal.
Reminds me on the last time I saw box turtle in the middle of the road with a severely cracked shell.

I ran back over it.

@pavlor

Quote:

To paraphrase Dave Haynie: They weren´t able to port 3 libraries to AAA, how they can even dream about new CPU ISA

And yet it GOT done without Dave and company...twice.
And will soon move to a third generation ISA.

Because Amiga was never just Commodore (or whoever owned it), the community and third party developers played a big role.

"how can they dream...' indeed, when did you stop Dave?

Mind you, either PA-RISC or PPC was a solid choice in mid 90s However, in only 4-5 years they would have been lagging behind x86 CPUs... Pretty much any choice other than x86 was a dead end on desktop by 2000.

_________________

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: YOU also reported that the preliminary OS4 version was mostly 68K on 2004... I wrote: non-kernel components ExecSG was of course PowerPC with many OS parts still runing under 68k emulation.

The kernel is essentially Exec, which occupied 18KB circa for Kickstart 1.2, and around double for Kickstart 3.0.

So, it's a small, albeit the most important, part of the entire o.s., utilities et similia included.
Quote:

Quote: If you look at the SINGLE G5, the situation is simply embarrassing: it was obliterated! I fear I don´t share your passion for overblown words: 30 % is not obliterated in my point of view.

Now I know why you preferred to go for history: because your mathematical skills are as embarrassing as the G5 situation.

I made some stats and created a nice chart for you, hoping that your glasses don't make you fail again:



As you can see, comparing the top x86 machine with the SINGLE G5 one, the MINIMUM difference (worse, of course) is 25%, up to 138% more time spent.

Regarding Quake, they are FPS, so have a different meaning: it's home many frames, in percentage, the G5 lost compared to the x86 machine. Of course, it's negative: it's doing around 50% LESS frames per seconds.

For your joy, I also added the worst x86 machine of the time: the Pentium 4. Which, like the Athlon, never lose a single test against the G5.

Taking the DUAL G5, this machine was able to win only on two tests, and by at most 15%. But in 3 tests it around double slower than the Athlon. The P4 has a similar trend, with only 2 tests lost of maximum 25%.

What's clear, at this point, is that a new pair of glasses can solve your problems...
Quote:

Quote: I never saw a dual G4 on 2001 losing almost all tests against a single x86. What of benchmarks you linked use both CPUs in G5? Premiere? judging by result not.

Yes, it uses the dual CPUs, albeit little.
Quote:

Photoshop? Seems so, but there the difference is not big.

Not big, but it's quite clear the it uses the dual CPUs.
Quote:

Word? No.

Correct. The only thing that you spot-on.
Quote:

Quake 3? No.

Here you made a mistake which only a blind can do, since Quake 3 shows DOUBLE the performance, comparing the DUAL G5 with the single one: 294 vs 147 FPS is exactly 100% more performance of the former over the latter.

So, either you aren't able to read, or to understand simple data that even a kid can see.

Seriously pavlor: the next time take care of History, and leave technical things far away from you.
Quote:

Quote: It was, with the 68060, because Motorola removed even some useful instructions in user mode (so, no MMU involved). Sure, but was this real problem on Amiga?

Yes, it was, because usually games killed the o.s., so it wasn't possible to emulate the missing instructions.

Some games and demos used 68020+'s long multiply, and some the very weird games used the 68000's MOVEP instruction, and they cannot work.

Of course, WHDLoad solves such issues, but the problem is that, at the time, such software stopped working on 68060 systems.

@matthey

Quote:

matthey wrote: Quote: cdimauro wrote: That was part of the reason, but history shown that Motorola wasn't able to keep-up the contenders, RISC included, with its 68K family. Take a look a what she released, when, and the drastic decisions that she made to try to sustain it. R&D was going to more promising but flawed RISC research when Motorola saw greener pastures on the other side of the fence and was drawn into the AIM Consortium. Motorola/Freescale has always tried to reduce and simplify to the point of performance degradation. Look no further than ColdFire and MCore for example.

Intel was also involved with RISCs: 80960 and 80860 in the Commodore period, and Itanium at the beginning of '00s.
Quote:

Some of the Intel advantage was as much luck that FPS games worked better on PCs and lack of companies like C= to quickly add chunky modes.

Looking only at the graphic, it wasn't an Intel advantage, but a PC (IBM) advantage, thanks to the VGA's mode 13h.
Quote:

Management probably also played a role. Motorola designs were generally more professionally done with less bugs while Intel designers seemed to be given more room to be creative but they made more mistakes. Some of the mistakes came from pressure from the marketing department to increase clock speeds which sold processors.

Which is normal/human, I think. Motorola did mistakes too, of other kind.
Quote:

I have heard rumors that is how the Pentium 4 space heater design came about. Both companies made mistakes along the way but Intel won the cash flow battle.

That happened only because PCs became selling more than the home/personal computers of competitors, and by consequence Intel gained a lot of market.
Quote:

Quote: Last but not really least, the 68060 only allowed to pair simpler and the shortest instructions, whereas the "inferior" Pentium allowed to pair much more complex ones, with an certain advantage with the executed code. The 68060 can handle very powerful instructions in each integer pipe but it doesn't have enough fetch to feed it. This should not have been a problem to fix. I don't believe instruction folding/fusing would have been a problem for the 68060 design either.

Instructions fusion appeared with Pentium M/Banias/Centrino. I never heard of something similar before that. It was a brilliant idea of Intel's Israel research center (under the great Dadi Perlmutter).
Quote:

It may already do this for decrement+branch into DBcc.

It doesn't need anything like that, because it's a single instruction which is 4 bytes long, so it should be handled without problems by the 68060 (no dependency by other instructions, unless if you want to look at the condition code with the DBcc. DBRA, which is the more frequent one, has not such problems).
Quote:

I expect there was just not time to finish optimizing for the 68060 and it was already stronger per clock than the Pentium so they needed to get it out the door.

The history is not made of what if, Matt.

Motorola was very late with the 68060, even:
- removing some instructions;
- with the superpipeline limited to execute only 2 simple instructions of one word each;
- running at lower clock speeds.

How much time you wanted to give Motorola to fix its design? Another two years?

From 1993 (the first Pentium) to 1996 Intel released the 200Mhz Pentium MMX and the 200Mhz PentiumPro...

We have to accept the reality, Matt: the 68060 architecture was too complex for that time, and that lead to the limits and design decisions made by Motorola.
Quote:

Quote: Again, that's because Motorola was constrained by the transistor (and power too: its chips, 68040 included, generated a lot of heat) budget, cutting features of its "beauty" but too much complex architecture. The 68040 was not a good design because it was not the direction they needed to go for the future.

For which reason? Because I see similarities with other designs of the time.
Quote:

The 68060 design had to be largely created from scratch. The 68060 was a good base design which later designs could have improved on. It should have been possible to clock it up but the EA before ALU designs are not high clockers (but efficient) and a new OoO design for the highest performance would have been needed at some point.

OoO means going even later. Motorola already suffered a lot for being late with the 68040 first, and especially with the 68060 after.
Quote:

The complex 68k addressing modes should be easier to handle with OoO although the 68060 and Apollo core figured out how to keep them from being much of a bottleneck.

Apollo is (will be) a ultra modern design, which makes use of much-easier-to-program FPGAs. It cannot be taken as a benchmark, comparing it with what was possible at the time.
Quote:

There are some complexities of the 68020 ISA like this which limit 68k clock speeds but the powerful addressing modes potentially make it more powerful per clock than x86

The problem is that you can make a more powerful design, for sure, but if it's limited by the clock frequency, and such powerfulness is rarely used in code, at the end you can be surpassed by inferior designs, which are more focused on simpler and more common tasks. And that's what happened with x86.
Quote:

(68k's shorter average instruction length and simpler encoding are also small advantages for parallel processing).

Why?

Regarding the encoding, it's questionable that the 68K is easier to decode than the x86.

x86's prefixes is its "black beast", but they can be looked-up in parallel (in a couple of pipeline stages), and after that you have the byte opcode plus a simpler EA extension byte to care of.

68K has no prefixes (very good thing!), but 16-bit opcodes require more logic for decoding, especially for the many exceptions to handle. Last but not least, the more advanced 68020+'s EA extension word is a nightmare for calculating the final instruction length (and the full list of EA arguments).

I think that, from this point of view, both have their big problems for decoding instructions, but I don't see a real winner here.

Old CISCs like them had the fortune to be pioneers and be wide spread, but unfortunately their engineers made a lot of mistakes with their designs, dictated by the lack of experience and no vision of the future.
Quote:

Each processor ISA gives some innate traits as we know ;).

Absolutely. That's why I prefer CISCs instead of RISCs: they give more creative and programmer-friendly toys to play with.

@agami

Quote:

agami wrote: @Sky7 2. In all my real-world use cases of ripping CDs to MP3, ripping DVDs to MP4, and using software which was available on both Windows and Mac OS X e.g. Handbrake, the PowerPC G5 was on par, and in some cases faster than an equally clocked Intel processor.

Do you have any benchmark for it?
Quote:

2.1 PowerMac G5 single core 1.8GHz vs. Pentium 4 1.8GHz. Heck, even the Pentium M 1.8GHz outperformed the P4.

Clock cannot be taken into account when comparing different architectures. And even with the same family, it doesn't make sense.

You can only measure the final results of the machines.
Quote:

2.2 Later a quad core PowerMac 2.5 GHz easily beat the the early Intel Core2 Quad in the above tasks. It wasn't until about 2010-11 that a quad core i5 at 2.5GHz outperformed the 2005 G5.

Then you have to ask Apple why up to the day BEFORE the announcement of the switch to Intel they claimed that their G5 were up to 2x faster the Intel's Core, and the day AFTER they claimed that their Intel's Core machine were up to 4x their old G5 machines.

In other words: I prefer to see concrete results, with a wide range of different applications used.


@CodeSmith

Quote:

CodeSmith wrote: One important thing to remember when looking at G5 vs x86 is that most of the PPC data we have from the time comes from Apple marketing, and at the time Apple was in a full PR offensive against x86 (ironically, considering the timing when OSX first came out, they must have had a full team working on it by then). It was pretty common to see really dubious benchmarking going on, so one needs to be careful when looking at sources from that time period.

Absolutely! See above my reply to agami.


@tlosm

Quote:

tlosm wrote: @pavlor some one dont wanna understand the g5 start be optimized in late 2007/2009 in the apple compilers... before was not . only in xcode 3.1.4 for example the gcc was doing right it job on g5

The same happens with the compilers for other architectures: they improved over the time.
Quote:

all the stupid benchmarks did on programs optimized for g4 and make it running on g5 was making all worst results compared the low classes cpus and compared with p4 . this is the reality of the facts ...

See above: the same happened for the x86 too. It took a lot of time to use the SSE after their introduction, just to give an example.

And the above benchmarks that I reported only used 32-bit applications, where just recompiling them for x64 immediately gained 10-15% of performance.


@pavlor

Quote:

pavlor wrote: @OlafS25 Quote: when they had earned more money that also would helped the 4.X line. Ignoring customer wishes is never a good idea for a business. I don´t think so. It would lead to further delay and even OS4 "cash cow" "BAF" users don´t have unlimited patience.

Correct me if I'm wrong, but Amiga OS 3.9 sold 50K copies, if I remember correctly. That's a lot of cache.

So, supporting also the 68K line made sense, IMO.


@agami

Quote:

agami wrote: @OlafS25 But yes, by the mid '00s most people stopped caring about the CPU. Alas, for me it's not about the hate, rather it's about cheering for the underdog. I use Intel CPUs but I still cheer for AMD when they win lucrative console part contracts.

It's the exact contrary: the console contracts aren't lucrative for AMD, because its gain is around half of what it gets from its regular processors line.

But they are good for AMD, because it's selling tenths of millions of pieces, and that's the reason why it's surviving.
Quote:

And when Microsoft with all its billions of dollars can't bring to market a smartphone platform, I feel vindicated.

Don't you feel a little childish saying that? It doesn't make sense for me.
Quote:

These people are not brilliant geniuses, they lucked out at a crucial point in computing history and have been riding that wave ever since.

Not brilliant? Have you took at look at Windows Presentation Foundation? Can you find a comparable framework for building GUI applications?
Quote:

Though the Xbox division has done well, I have to wonder if they would have done so if SEGA never faltered.

Well, Microsoft provided its o.s. for SEGA's Dreamcast console, included its DirectX...


@OldFart

Quote:

OldFart wrote: @cdimauro Quote: In short: it's just another example of very bad management of Commodore's heritage... Glad we got used to it as we endured it for so long and by now know how to deal with it, no? OldFart

With resignation...

@cdimauro

Quote:

Looking only at the graphic, it wasn't an Intel advantage, but a PC (IBM) advantage, thanks to the VGA's mode 13h.

Oh, that helps explain why I helped write an ET4000 driver for our company's 68K systems (not Amigas btw).
Yeah, that was a killer, THEN 2D and finally 3D acceleration.

Could we have kept up,?
Well some of us did, but you had to design your systems with more modularity.