@Hammer

I only quote this, because for the rest I I fully agree with matthey.
Quote:

Hammer wrote: You're not factoring in microarchitecture's complexity that consumes higher transistor budgets. Microarchitecture's complexity needs to be traded with the L1 cache's transistor budgets.

The context is important here.

First of all, the core of a processor remains more or less the same in terms of decoders size & their complexity. It happens from time to time that there are jumps (e.g.: from 4 to 6 decoders), but those are being seen occasionally.
BTW, a similar things happens to the backend (execution units) about the support to legacy stuff (complex instructions): it remains the same (usually... exactly the same).

This means that the higher transistors budget of a CISC has less and less impact when moving to new processes (e.g.: those areas reduce in size / number of transistors).

So, you can see it as an "investment" for a CISC which is given a "ROI" in future microarchitectures (read: the benefits stay whereas the burden reduces).


Second and equally important, you're comparing processors which were initially designed in a certain way (e.g.: decisions which introduced complexity) with much newer ones which were designed with that hindsight and avoided such complexity.

x86s and 68ks are very old architectures which carry a lot of burden and that's why they are more complex.

However, if you design a modern CISC nowadays you can still introduce more complex things (e.g.: mostly about the number of instructions/opcodes formats to be decoded and the richer addressing modes) whilst keeping the overall complexity (in terms of implementation / transistors budget) of such decisions very very low.

A modern CISC simply obliterates any modern "RISC" (note the quotes, since they are no RISCs from very long time: you can only call them L/S -> Load/Store architectures) on every area: code density, "data density" (!), number of executed instructions, transistors budget and even on power consumption / efficiency (e.g.: overall power consumed over the entire execution time).

Last edited by cdimauro on 21-Jul-2024 at 06:01 AM.

@Hammer

I see you're still unwilling to guess at a price for an A500 hardware add-on to enable it to play the 1993 Doom at a respectable clip.

Let's then reverse engineer the problem:
The retail price of the game appears to have been $40 USD, and in September of 1995 on SNES it was $65 USD.
How much would you say, an A500 gamer would be willing to pay for a bundle that included Doom + Special HW Expansion in early 1994?
$99 USD? ($59 delta)

Some might be tempted to apply the PS VR Farpoint + Aim controller marketing and logistics. The idea that the controller is not just for Farpoint, but there will be future games that will leverage the controller.

Given that C= wasn't a game publishing house like SONY, and the A500 was the 'Field of Dreams' equivalent of a gaming computer, there is less of a guarantee that future games would be released to take advantage of the expansion.

But let's for argument's sake say that consumer sentiment surrounding C= was very positive, and most A500 owners would buy into the proposition that while initially bundled with Doom, it will bring new exciting games to the A500. What would an A500 gamer be willing to pay for the Doom + Expansion bundle in this scenario? $199 USD? ($159 Delta)

Of course, the more that the expansion costs, the fewer people would buy it. How many of the 4M+ A500 owners would have to buy this bundle to make it worthwhile? Let's not forget that a decent portion of the A500 market are budget buyers who did so because they knew they could get free games (blank floppy pricing excl.), so they are not the target demographic for this bundle. How many are even interested in this type of game?
Not to mention that by December 1993 every single A500 is out of warranty, many of which have undergone some form of expansion, and even some hacking. How much of a support call burden would there be supporting A500 users who purchased such an expansion and experienced problems or outright machine failures?

So let's say that our sweet spot is $40 for the game + $99 for the hardware = $139 for the bundle, with the aim to sell 200k units in 1 year. A number which sufficiently distributes R&D costs, and one that should get the interest of id Software for a port.

Could Commodore, between mid 1993 and through 1994, design, produce, and support such an expansion at such a volume for a retail price of $99 per unit?
Even if there was a BoM in 1993 that could support a hardware production cost of $50 per unit, that's an outlay of $10M for C= who at that point were already in debt. In a cutthroat scenario where C= at best makes only 5% profit.
Who in 1993 would gamble $10M USD to bring Doom to the A500?

So, Mr @Hammer.
Can you give me a wall of unrelated pricing facts which tenuously weave a narrative that an A500 expansion could be made in 1993 for $50 at 200k+ volume, which would make A500 Doom playable at 256 x 240 x 8 @ 15fps?

Last edited by agami on 21-Jul-2024 at 04:01 AM.

_________________
All the way, with 68k

@agami

There was Sega 32x back then... I think it's more comparable... It didn't sell like hot cakes or anything, but it sold for the permise a few new unobtainable games could run in it, instead of having to attach the same chip on every card like they did before.

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

@agami

Quote:

I see you're still unwilling to guess at a price for an A500 hardware add-on to enable it to play the 1993 Doom at a respectable clip. Let's then reverse engineer the problem: The retail price of the game appears to have been $40 USD, and in September of 1995 on SNES it was $65 USD. How much would you say, an A500 gamer would be willing to pay for a bundle that included Doom + Special HW Expansion in early 1994? $99 USD? ($59 delta)

Without factoring memory storage requirements, SNES doesn't have a problem with 256 colors with chunky pixels via Mode 7, hence SNES just needs compute power increase i.e. $10 SuperFX chip.

You can't say the same for A500. The new technology injection for A500 is greater than A1200/CD32. The Doom game accelerator for A500 is effectively a new machine.

Remember, Commodore has reduced their A1200 (399 UKP)'s "healthy profit margins" with CD32 (299 UKP). Any Doom-capable machine from Amiga's mass-produced models must be attached to the Amiga out of the box.


https://amitopia.com/interview-with-the-ex-ceo-of-commodore-uk-david-john-pleasance/
From David Pleasance
Quote:

We had determined we would need US $50 million, not only to purchase the worldwide assets (estimated we would have to pay US 15 million from the liquidator) but more importantly a further US $ 35 million to fund the running of the business and pay upfront for the components, as no supplier would give us trading terms having been “burnt� with bad debts from CIL. Through Coopers & Lybrand we raised US$ 25 Million from mostly high wealth individuals, plus we were also approached by a Chinese Manufacturing company New Star Electronics (who up to that point in time had been illegally manufacturing Sega and Nintendo products) They had been told by the Chinese Government that they must become legitimate and so had committed to us the second US$ 25 million --- 36 Hours before the Auction of the Commodore worldwide assets, New Star Electronics without warning pulled out of the deal. This meant that even though we could have bought those assets,(sold for less than the US $15 million we had estimated) knowing we did not have sufficient funds to maintain the business until we reached profitability, meant it would almost be certain we would lose the US 25 million provided by the investors, so we had no option but to withdraw.

Commodore UK was able to gain $25 million dollars from investors outside of New Star Electronics' $25 million component.


--------------------------------------------

I used PiStorm-Emu68 with RPi 4B @ 1.8Ghz on A500, running Quake demo1 benchmark with A500's HAM6 mode resulted in 25 fps average. Using the same RPi 4B with A1200's AGA, it nearly doubled the frame rate i.e. 40 fps HAM6/HAM8. 60 fps with AGA's 256 color mode with Clickboom's Quake port.

With higher compute power and optimized C2P software, AGA is pretty good for texture-mapped 3D games.

Amiga OCS's HAM6 mode and slower CPU link with Chip RAM imposes a large bottleneck.

Without a proper 256-color display capability, PPS (Progressive Peripherals & Software) 040-500's 68040 @ 28Mhz or 33 Mhz accelerator on the A500 is nearly a waste of time. This is like a PC with a 486DX-33 CPU with non-upgradeable Tandy++ graphics.

I don't have Indivision ECS to display 256 colors Graffiti for A500.

Last edited by Hammer on 21-Jul-2024 at 01:46 PM.
Last edited by Hammer on 21-Jul-2024 at 01:11 PM.
Last edited by Hammer on 21-Jul-2024 at 01:08 PM.
Last edited by Hammer on 21-Jul-2024 at 01:00 PM.
Last edited by Hammer on 21-Jul-2024 at 12:48 PM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@pixie

Sega's 32X vs Saturn messaging is confusing and the X32 add-on looked bad since the design didn't blend with the Genesis as a single unit look and feel.

From https://en.wikipedia.org/wiki/32X

Despite the lower price console's positioning as an inexpensive entry into 32-bit gaming, Sega had a difficult time convincing third-party developers to create games for the new system. Top developers were already aware of the coming arrival of the Sega Saturn, Nintendo 64, and PlayStation, and did not believe the 32X would be capable of competing with any of those systems

Last edited by Hammer on 21-Jul-2024 at 09:08 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@cdimauro

Quote:

First of all, the core of a processor remains more or less the same in terms of decoders size & their complexity. It happens from time to time that there are jumps (e.g.: from 4 to 6 decoders), but those are being seen occasionally. BTW, a similar things happens to the backend (execution units) about the support to legacy stuff (complex instructions): it remains the same (usually... exactly the same). This means that the higher transistors budget of a CISC has less and less impact when moving to new processes (e.g.: those areas reduce in size / number of transistors). So, you can see it as an "investment" for a CISC which is given a "ROI" in future microarchitectures (read: the benefits stay whereas the burden reduces).

Intel's 486SX2-50 wasn't involved in the game console market when Sony established the PlayStation platform.

The A500/A1200's target price is near game consoles.

Fat X86 CPUs entered the game console market during the original Xbox's development. The PC market has effectively subsidized Xbox's X86 CPU and NVIDIA's GeForce 3.5 Ti NV2A GPU variant.

Both Amiga Hombre and Sony PS1 have 1 million transistors budgets in the 1992 to 1994 date range context.

Prove Amiga's target audience has shown to spend 1.2 million PowerMac units level spending within 1 year.

Motorola's 68030-25 was following Intel's 386DX-25 prices in 1992-1993.

Good luck with designing an X86 (or 68K) performance competitive games machine in a game console price range in the 1992 to 1994 date range.

Last edited by Hammer on 21-Jul-2024 at 02:01 PM.
Last edited by Hammer on 21-Jul-2024 at 01:23 PM.
Last edited by Hammer on 21-Jul-2024 at 01:19 PM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@Hammer

Quote:

Hammer wrote: @cdimauro Quote: First of all, the core of a processor remains more or less the same in terms of decoders size & their complexity. It happens from time to time that there are jumps (e.g.: from 4 to 6 decoders), but those are being seen occasionally. BTW, a similar things happens to the backend (execution units) about the support to legacy stuff (complex instructions): it remains the same (usually... exactly the same). This means that the higher transistors budget of a CISC has less and less impact when moving to new processes (e.g.: those areas reduce in size / number of transistors). So, you can see it as an "investment" for a CISC which is given a "ROI" in future microarchitectures (read: the benefits stay whereas the burden reduces). Intel's 486SX2-50 wasn't involved in the game console market when Sony established the PlayStation platform.

And? The Playstation arrived at the (very) end of '94 and the 486 already moved to the DX4.
Quote:

The A500/A1200's target price is near game consoles.

Not so near, unless your benchmarks are the Sharp X68000 and the NeoGeo, respectively.
Quote:

Fat X86 CPUs entered the game console market during the original Xbox's development. The PC market has effectively subsidized Xbox's X86 CPU and NVIDIA's GeForce 3.5 Ti NV2A GPU variant.

Yes, and? x86s are also dominating the consoles market since many years now.

Plus... the "fat" x86 on the Xbox provided the best performances, compared to the "RISC" (read: L/S processors) of the competition.
Quote:

Both Amiga Hombre and Sony PS1 have 1 million transistors budgets in the 1992 to 1994 date range context.

Irrelevant? It depends on how you use the transistors on a device.

Do you recall the comparison of SIMD units? SIMD units aren't all the same and can't be compared by checking the total number of "bits". What's important is how they are organized / used.

It's the same here: you can't just compare transistors vs transistors.
Quote:

Prove Amiga's target audience has shown to spend 1.2 million PowerMac units level spending within 1 year.

?!? Non-sense...
Quote:

Motorola's 68030-25 was following Intel's 386DX-25 prices in 1992-1993.

Why a 68030? You don't need it at the time: a 68EC020 was enough.
Quote:

Good luck with designing an X86 (or 68K) performance competitive games machine in a game console price range in the 1992 to 1994 date range.

Well, maybe you need to take a look at the consoles of the time: even the 68EC020 of the Amiga 1200 (read: 1992) was way more than enough of what needed.

@Hammer

Quote:

Hammer wrote: @agami ...The Doom game accelerator for A500 is effectively a new machine.

Thank you.

So we agree that in 1993, there was no commercial or logistical way that C= could've worked with id Software, to get them to port Doom for the global Amiga gaming market, because:
a) The only addressable market with the justifiable numbers was the A500 install base, and
b) The A500 Doom hardware accelerator would effectively be a new machine, using the A500 as little more than a game controller, and would cost too much to make it marketable.

I agree, if the A1200 had an install base in 1M+ by mid 1993, then and accelerator could've been made as a loss leader that would be palatable enough to those looking to play Doom, and potentially drive new A1200 sales. But even combined with CD32 numbers, the economies of scale just weren't there.

_________________
All the way, with 68k

@cdimauro

Quote:

And? The Playstation arrived at the (very) end of '94 and the 486 already moved to the DX4.

What's the cost? Playstation's cost model is close to CD32/A1200's.

http://archive.computerhistory.org/resources/access/text/2013/04/102723312-05-01-acc.pdf
Page 323 of 356 for Dataquest's package cost estimate
Intel 486DX4 @ 100Mhz = $28
AMD 486-100 @ 100Mhz = $23

During 1995, it was unlikely Intel and AMD would offer CPUs with near-zero profit margins.



CPU product releases and wholesale prices for 1994
http://kpolsson.com/micropro/proc1994.htm

March:

(month unknown)
Intel announces the 486SX2 processor. It can replace the 486SX in 16-25 MHz systems with a clock-doubled processor. Price is US$249.


----
April:

Motorola formally introduces the 50 MHz 68060 processor. Price is US$263 in 10,000 unit quantities. A 66 MHz version is scheduled to ship near the end of the year.


----
May:

Motorola ships sample copies of the PowerPC 603 processor. Volume pricing is set at US$160 (66 MHz) and US$199 (80 MHz) in 20,000 unit quantities.

----
July:

IBM makes available sample quantities of the PowerPC 603 processor. High quantity pricing is US$165 for the 66 MHz chip, and US$195 for the 80 MHz version.

----
September:
Advanced Micro Devices ships its Am486DX2-80 40/80 MHz processor. Price is US$266 in 1000 unit quantities.

(month unknown)
Intel announces the Write-Back Enhanced DX2 processor, a 486DX2 with the support of write-back cache operation, making it feasible to implement in systems without a second-level cache. Prices are US$149 for 50 MHz and US$199 for 66 MHz, in 1000 unit quantities.


November:
Digital Equipment announces the 21066A processor, a 0.5-micron version of the Alpha 21066 processor, operating 40% faster than the previous version. Speeds announced are 166 MHz and 233 MHz. Price is US$396 in 1000 unit quantities.

Quote:

Not so near, unless your benchmarks are the Sharp X68000 and the NeoGeo, respectively.

Sharp X68000's cost structure wasn't a game console.

NeoGeo ($649.99 USD) was beaten by lower-cost SNES (US $199 USD). Mainstream gamers are looking for the delivered gaming experience and entry costs. Technical specs debate is mostly irrelevant to this audience.

The delivered gaming experience has a hardware capability foundation.

NeoGeo's USD $649.99 entry price mistake was followed by 3DO's USD $699 entry price mistake since they are single-purpose gaming platforms.

Amiga 500's 1987 $699 entry price was acceptable due to its multi-purpose potential with gaming bias. Amiga 500's price was reduced to the $500 range in 1988.

Quote:

?!? Non-sense...

It's basic market intelligence and results e.g. PowerAmiga's uncompetitive performance vs price and the resulting failure.

https://metro.co.uk/2022/04/08/a500-mini-amiga-console-interview-thats-our-passion-for-commodore-16427365/
Quote:

So we created the C64 direct to TV joystick back in 2004, it became the fastest selling toy product in QVC’s history. I think we sold 186,000 units in the first hour of trading, which was phenomenal. But it proved to us that there was still this great love for all things Commodore. So, as you rightly pointed out, the C64 then made the jump onto the Wii, and we were behind that. And again… they weren’t million sellers, but they would sell in decent numbers every month.

Raspberry Pi's success wasn't about copying yet another PC's cost structures.

Last edited by Hammer on 22-Jul-2024 at 04:42 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@agami

Quote:

agami wrote: @Hammer Thank you. So we agree that in 1993, there was no commercial or logistical way that C= could've worked with id Software, to get them to port Doom for the global Amiga gaming market, because: a) The only addressable market with the justifiable numbers was the A500 install base, and b) The A500 Doom hardware accelerator would effectively be a new machine, using the A500 as little more than a game controller, and would cost too much to make it marketable.

A500 would be audio, user input, CIA timers, a keyboard, and a primary external case.

When I dumped my 386DX-33-based PC for a Pentium 166-based PC, my Yamaha OPL3 sound card, keyboard, mouse, SVGA monitor, and AT mini-tower case were recycled.

CPU, mainboard, RAM, and SVGA were upgraded which mirrored the recent PiStorm upgrade for A500 i.e. CPU, RAM, SBC board, and IGP were upgraded.

PC SVGA PCI upgrade includes legacy VGA support. I could have recycled my old ET4000 SVGA card, but no-name OEM S3 Trio 64 UV+ is cheap.

The Amiga architecture wasn't properly partitioned.

Quote:

I agree, if the A1200 had an install base in 1M+ by mid 1993, then and accelerator could've been made as a loss leader that would be palatable enough to those looking to play Doom, and potentially drive new A1200 sales. But even combined with CD32 numbers, the economies of scale just weren't there.

For the 1 million install base target, AGA's release needs to be in H2 1991 and avoid the whole A600 debacle i.e. Commodore's $357 million loss was fatal.

Instead of spending on AGA machines, A600's ECS release has wasted customers' spending.

AA500+ and AA3000+ need to be in play in H2 1991.

AA500+ would have direct AGA replacement from A500P's design i.e.

ECS Denise to Lisa,
ECS Agnus to Alice,
Gary to Fat Gary, no wasting time on A300/A600's Gayle and A1200's AA-Gayle.
Paula remains as is.
1MB 240 ns read/write DRAM to 2MB 140 ns read/write FP DRAM.
Ramsey for 32-bit memory controller instead of A1200's Budgie.
Four TTL bridge chips need to be cost-reduced i.e. Bridgette instead of A1200's Budgie.
Two CIA chips can have their package changed.
Adv101 tripple 8bit video DAC replacing VIDIOT.

AA500+'s edge connector would be A3000's local CPU slot.

Extra PCMCIA and IDE features would be released as A1200.

AAA's chunky pixels R&D needs to be for AGA.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@Hammer

Quote:

Hammer wrote: @cdimauro Quote: And? The Playstation arrived at the (very) end of '94 and the 486 already moved to the DX4. What's the cost? Playstation's cost model is close to CD32/A1200's. http://archive.computerhistory.org/resources/access/text/2013/04/102723312-05-01-acc.pdf Page 323 of 356 for Dataquest's package cost estimate Intel 486DX4 @ 100Mhz = $28 AMD 486-100 @ 100Mhz = $23 During 1995, it was unlikely Intel and AMD would offer CPUs with near-zero profit margins.

And it was absolutely not needed. Have you took a look at what processor was used for the Playstation? It was an R3000@33Mhz.

Such 486s were MONSTERS compared to it...
Quote:

CPU product releases and wholesale prices for 1994 http://kpolsson.com/micropro/proc1994.htm March: (month unknown) Intel announces the 486SX2 processor. It can replace the 486SX in 16-25 MHz systems with a clock-doubled processor. Price is US$249. ---- April: Motorola formally introduces the 50 MHz 68060 processor. Price is US$263 in 10,000 unit quantities. A 66 MHz version is scheduled to ship near the end of the year. ---- May: Motorola ships sample copies of the PowerPC 603 processor. Volume pricing is set at US$160 (66 MHz) and US$199 (80 MHz) in 20,000 unit quantities. ---- July: IBM makes available sample quantities of the PowerPC 603 processor. High quantity pricing is US$165 for the 66 MHz chip, and US$195 for the 80 MHz version. ---- September: Advanced Micro Devices ships its Am486DX2-80 40/80 MHz processor. Price is US$266 in 1000 unit quantities. (month unknown) Intel announces the Write-Back Enhanced DX2 processor, a 486DX2 with the support of write-back cache operation, making it feasible to implement in systems without a second-level cache. Prices are US$149 for 50 MHz and US$199 for 66 MHz, in 1000 unit quantities. November: Digital Equipment announces the 21066A processor, a 0.5-micron version of the Alpha 21066 processor, operating 40% faster than the previous version. Speeds announced are 166 MHz and 233 MHz. Price is US$396 in 1000 unit quantities.

See above: not needed.
Quote:

Quote: Not so near, unless your benchmarks are the Sharp X68000 and the NeoGeo, respectively. Sharp X68000's cost structure wasn't a game console.

Neither the Amiga.
Quote:

NeoGeo ($649.99 USD) was beaten by lower-cost SNES (US $199 USD). Mainstream gamers are looking for the delivered gaming experience and entry costs. Technical specs debate is mostly irrelevant to this audience. The delivered gaming experience has a hardware capability foundation. NeoGeo's USD $649.99 entry price mistake was followed by 3DO's USD $699 entry price mistake since they are single-purpose gaming platforms. Amiga 500's 1987 $699 entry price was acceptable due to its multi-purpose potential with gaming bias. Amiga 500's price was reduced to the $500 range in 1988.

You confirmed my point...
Quote:

Quote: ?!? Non-sense... It's basic market intelligence and results e.g. PowerAmiga's uncompetitive performance vs price and the resulting failure.

You were comparing them to the PowerMacs, which were very expensive machines.

The Amiga market was VERY different. Hence, you can't compare them.
Quote:

https://metro.co.uk/2022/04/08/a500-mini-amiga-console-interview-thats-our-passion-for-commodore-16427365/ Quote: So we created the C64 direct to TV joystick back in 2004, it became the fastest selling toy product in QVC’s history. I think we sold 186,000 units in the first hour of trading, which was phenomenal. But it proved to us that there was still this great love for all things Commodore. So, as you rightly pointed out, the C64 then made the jump onto the Wii, and we were behind that. And again… they weren’t million sellers, but they would sell in decent numbers every month. Raspberry Pi's success wasn't about copying yet another PC's cost structures.

Exactly. Do you see how similar was/is the RPis to the Amiga?

This shows you which direction Commodore (all employees!) should have taken for our beloved platform...

@Hammer

Quote:

Hammer wrote: @agami Quote: agami wrote: @Hammer Thank you. So we agree that in 1993, there was no commercial or logistical way that C= could've worked with id Software, to get them to port Doom for the global Amiga gaming market, because: a) The only addressable market with the justifiable numbers was the A500 install base, and b) The A500 Doom hardware accelerator would effectively be a new machine, using the A500 as little more than a game controller, and would cost too much to make it marketable. A500 would be audio, user input, CIA timers, a keyboard, and a primary external case. When I dumped my 386DX-33-based PC for a Pentium 166-based PC, my Yamaha OPL3 sound card, keyboard, mouse, SVGA monitor, and AT mini-tower case were recycled. CPU, mainboard, RAM, and SVGA were upgraded which mirrored the recent PiStorm upgrade for A500 i.e. CPU, RAM, SBC board, and IGP were upgraded. PC SVGA PCI upgrade includes legacy VGA support. I could have recycled my old ET4000 SVGA card, but no-name OEM S3 Trio 64 UV+ is cheap. The Amiga architecture wasn't properly partitioned.

Let's keep it simple: the Amiga architecture wasn't properly EVOLVED due to the incompetence of the engineers which remained after that the original team left.
Quote:

Quote: I agree, if the A1200 had an install base in 1M+ by mid 1993, then and accelerator could've been made as a loss leader that would be palatable enough to those looking to play Doom, and potentially drive new A1200 sales. But even combined with CD32 numbers, the economies of scale just weren't there. For the 1 million install base target, AGA's release needs to be in H2 1991 and avoid the whole A600 debacle i.e. Commodore's $357 million loss was fatal. Instead of spending on AGA machines, A600's ECS release has wasted customers' spending. AA500+ and AA3000+ need to be in play in H2 1991. AA500+ would have direct AGA replacement from A500P's design i.e. ECS Denise to Lisa, ECS Agnus to Alice, Gary to Fat Gary, no wasting time on A300/A600's Gayle and A1200's AA-Gayle. Paula remains as is. 1MB 240 ns read/write DRAM to 2MB 140 ns read/write FP DRAM. Ramsey for 32-bit memory controller instead of A1200's Budgie. Four TTL bridge chips need to be cost-reduced i.e. Bridgette instead of A1200's Budgie. Two CIA chips can have their package changed. Adv101 tripple 8bit video DAC replacing VIDIOT. AA500+'s edge connector would be A3000's local CPU slot. Extra PCMCIA and IDE features would be released as A1200. AAA's chunky pixels R&D needs to be for AGA.

See above.

And I add that Commodore's (technical) VP of the time reported that the team of engineers wasted time doing NOTHING. They weren't doing any serious develpment up to a certain point in time.

When it was realized that new machines had to quickly go to the market, then they rushed like devils. But it was too late!

And we know know what happened then, right? It was the beginning of the hell.

TLDR: you're defending too much such incompetent people.

Last edited by cdimauro on 23-Jul-2024 at 10:08 PM.

@cdimauro

Quote:

And it was absolutely not needed. Have you took a look at what processor was used for the Playstation? It was an R3000@33Mhz. Such 486s were MONSTERS compared to it...

It comes down to cost vs performance.

PlayStation's 1 million transistor budget includes 33 mips R3000A CPU @ 33 Mhz, 66 mips GTE (geometry math) @ 58 Mhz, and a GPU (e.g. texture mapper).

https://www.youtube.com/watch?v=54s6ZVU6YoI
Tomb Raider running on 486DX4 @ 100Mhz example.


Tomb Raider PC uses the FPU for the performance boost.
https://youtu.be/XHqLYzqZciM?t=140
Am486SX2-66 has slide show results in Tomb Raider. This YouTube video includes Tomb Raider benchmarks for multiple 486 models.


Quote:

You were comparing them to the PowerMacs, which were very expensive machines. The Amiga market was VERY different. Hence, you can't compare them.

That's my point i.e. PowerAmiga camp attempted to treat the Amiga as a Mac.

The Amiga's demographics have more things in common with RPi and game consoles.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@cdimauro

Quote:

And I add that Commodore's (technical) VP of the time reported that the original team wasted time doing NOTHING. They weren't doing any serious develpment up to a certain point in time. When it was realized that new machines had to quickly go to the market, then they rushed like devils. But it was too late! And we know know what happened then, right? It was the beginning of the hell. TLDR: you're defending too much such incompetent people.

.
The "to be or not be" Amiga chipset upgrade drama is detailed in Commodore The Final Years by Brian Bagnall.

An example from the mentioned book,

Quote:

Bob Welland had the most diffuse jobs. He would continue working on his 68020 accelerator board for the Amiga 2000, in order to get Unix running on the system. But more importantly, Welland would begin defining the next quantum leap in display chips—the very thing the original Amiga engineers wanted. “We were hoping they would also continue advanced development of graphics,� recalls Dale Luck. Porter specifically wanted 1000 by 800 resolution with 8 bit planes and 16 million colors —something to exceed the current competition. Of importance would be keeping the new video technology compatible with existing commercial Amiga software. All of these plans would be discussed at Commodore’s worldwide engineering meeting, scheduled for September 22, 1987 at the Embassy Suites hotel in New York. The meeting was called by Henri Rubin, and those invited included the main West Chester engineers, along with engineers from Germany and Japan.

Notice Jeff Porter's "8 bit planes and 16 million colors" focus which led to AGA's "8 bit planes and 16 million colors palette". Moooore colors. What happened to 3D?


There is also the resource-wasting monochrome hi-res Denise vs color hi-res Denise drama in 1987.

Last edited by Hammer on 22-Jul-2024 at 08:18 AM.
Last edited by Hammer on 22-Jul-2024 at 08:15 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

Quote:

1000 by 800 resolution

Geh…

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla

Quote:

kolla wrote: Quote: 1000 by 800 resolution Geh…

AAA wouldn't solve the 3D problem.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@Hammer

Quote:

Hammer wrote: @cdimauro Quote: And it was absolutely not needed. Have you took a look at what processor was used for the Playstation? It was an R3000@33Mhz. Such 486s were MONSTERS compared to it... It comes down to cost vs performance. PlayStation's 1 million transistor budget includes 33 mips R3000A CPU @ 33 Mhz, 66 mips GTE (geometry math) @ 58 Mhz, and a GPU (e.g. texture mapper). https://www.youtube.com/watch?v=54s6ZVU6YoI Tomb Raider running on 486DX4 @ 100Mhz example. Tomb Raider PC uses the FPU for the performance boost. https://youtu.be/XHqLYzqZciM?t=140 Am486SX2-66 has slide show results in Tomb Raider. This YouTube video includes Tomb Raider benchmarks for multiple 486 models.

You're comparing two completely different systems.

A fair comparison would have had only the CPU changed, while keeping everything else the same. Which is clearly not possible.
Quote:

Quote: You were comparing them to the PowerMacs, which were very expensive machines. The Amiga market was VERY different. Hence, you can't compare them. That's my point i.e. PowerAmiga camp attempted to treat the Amiga as a Mac.

Well, that's due to Commodore engineers which focused on that instead of attacking Amiga's primary market.
Quote:

The Amiga's demographics have more things in common with RPi and game consoles.

Indeed. Let's tell to the above engineers...
Quote:

Hammer wrote: @cdimauro Quote: And I add that Commodore's (technical) VP of the time reported that the original team wasted time doing NOTHING. They weren't doing any serious develpment up to a certain point in time. When it was realized that new machines had to quickly go to the market, then they rushed like devils. But it was too late! And we know know what happened then, right? It was the beginning of the hell. TLDR: you're defending too much such incompetent people. . The "to be or not be" Amiga chipset upgrade drama is detailed in Commodore The Final Years by Brian Bagnall. An example from the mentioned book, Quote: Bob Welland had the most diffuse jobs. He would continue working on his 68020 accelerator board for the Amiga 2000, in order to get Unix running on the system. But more importantly, Welland would begin defining the next quantum leap in display chips—the very thing the original Amiga engineers wanted. “We were hoping they would also continue advanced development of graphics,� recalls Dale Luck. Porter specifically wanted 1000 by 800 resolution with 8 bit planes and 16 million colors —something to exceed the current competition. Of importance would be keeping the new video technology compatible with existing commercial Amiga software. All of these plans would be discussed at Commodore’s worldwide engineering meeting, scheduled for September 22, 1987 at the Embassy Suites hotel in New York. The meeting was called by Henri Rubin, and those invited included the main West Chester engineers, along with engineers from Germany and Japan. Notice Jeff Porter's "8 bit planes and 16 million colors" focus which led to AGA's "8 bit planes and 16 million colors palette". Moooore colors.

That's nice. It's the 1000x800 resolution which doesn't make sense, as also kolla reported.
Quote:

What happened to 3D?

Not needed at the time.
Quote:

There is also the resource-wasting monochrome hi-res Denise vs color hi-res Denise drama in 1987.

Another crap... -_-
Quote:

Hammer wrote: @kolla Quote: kolla wrote: [quote]1000 by 800 resolution Geh…

AAA wouldn't solve the 3D problem.[/quote]
Packed/chunky, yes, and before the AAA.

Last edited by cdimauro on 23-Jul-2024 at 06:11 AM.

@cdimauro

Quote:

You're comparing two completely different systems. A fair comparison would have had only the CPU changed, while keeping everything else the same. Which is clearly not possible.

They run a similar Tomb Raider game. There are many common PlayStation and PC game ports.

Ken Kutaragi's 1 million transistor budget is about delivering the PlayStation product within a certain budget.

The CPU is only a single aspect of the total system cost.

A1200/CD32 has certain BOM cost targets which is similar to game consoles.

Good luck with designing a $599 retail-priced A1200-like games bias machine from 486DX4.

The main reason for Amiga Hombre is CSG's lower production cost compared to buying a CPU solution which includes 3rd party CPU's profit margins.

Dr Hepler selected the PA-RISC instruction set.

Lew Eggebrecht favored the MIPS and PowerPC instruction set.

For Sony, LSi is willing to customize R3000A for Sony's target price.

In modern times, AMD customized Sony's Zen 2 design with the entire AVX pipeline being deleted for cost reduction.

Quote:

From Commodore The Final Years by Brian Bagnall Commodore’s Ted Lenthe began looking into RISC chips with the AAA chipset back in the summer of 1989, specifically the Motorola 88000. But the engineers always balked at concrete plans due to the incompatibility problems a new processor would cause. For his part, Ed Hepler favored creating his own RISC CPU on the basis that Commodore could produce it much cheaper than buying the 88000 from Motorola.

Intel X86 has license complexity issues i.e. Intel vs AMD court battles.

Pat Gelsinger's Intel is willing to customize CPU cores for game console platform customers (1,2).

Reference
1. https://www.theregister.com/2022/02/14/intel_x86_licensing/
2. https://www.tweaktown.com/news/95974/intel-wants-to-be-inside-microsofts-next-gen-xbox-console-would-built-in-the-usa/index.html

A leadership change from Intel with game focus Xbox direction. Healthy competition is good in the X86 world.

Quote:

Well, that's due to Commodore engineers which focused on that instead of attacking Amiga's primary market.

PowerAmiga refers to Escom era Amiga Technologies GmBH and Phase 5.

References
1. http://www.bambi-amiga.co.uk/amigahistory/atinfo.html
2. http://www.bambi-amiga.co.uk/amigahistory/pamiga97.html

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@cdimauro

Quote:

Packed/chunky, yes, and before the AAA.

ET4000W32i with 386DX-16 level CPU is a joke.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB

@matthey

Quote:

The 603(e) has a single precision FPU to save die area and transistors so double precision operations have a longer latency and throughput than a pipelined double precision FPU.

The longer is relative. Longer when compared to (insert your CPU).

Quote:

The 68040 has a partially pipelined extended precision FPU that is outperformed by the newer PPC 603(e) but easily outperforms the older 80486.

68040 has a single instruction issue per cycle decoder.

68040 FPU is not fast.

Quote:

Many compromises were made with restrictive transistor budgets back then and it wasn't just for the 68k. Year | CPU | Pipeline | Caches | Transistors 1990 68040 6-stage 4kiB_I+D_4-way 1,170,000 1994 68060 8-stage 8kiB_I+D_4-way 2,530,000 1994 PPC603 4-stage 8kiB_I+D_2-way 1,600,000 1995 PPC602 4-stage 4kiB_I+D_2-way 1,000,000 1995 PPC603e 4-stage 16kiB_I+D_4-way 2,600,000

IBM PPC 602 @ 66Mhz's 1 million transistor budget is designed for 3DO M2.

IBM offered two low-priced PPC 602 @ 66Mhz for 3DO M2.

A game console designed around the expensive price of 68040? LOL.

Quote:

Compared to the PPC602, the 68040 uses more transistors for a deeper pipeline, 4-way set associative caches instead of 2-way and a partially pipelined extended precision FPU instead of a fully pipelined single precision FPU which doesn't leave much for the partial microcoding wasted transistors that isn't necessary with the 68060. Comparing the PPC603e and 68060, the PPC603e has double the data cache but the performance of half the instruction cache due to the code density difference. Considering the 68060 transistors are used on the deeper pipeline to boost ILP and clock speed, branch cache/prediction unit, a 2nd barrel shifter for the 2nd execution pipe, a multi-banked data cache for both execution pipes in parallel and an instruction buffer to decouple the fetch pipeline and execution pipelines, I don't see much 68k specific tax.

68060 has 4 byte fetch per cycle from the L1 instruction cache.

68060 has a 32-bit external bus when the competition has a 64-bit external bus.

Quake demo1 on 68060 @ 50 Mhz is 10 fps.


Quote:

The 68060 outperforms the PPC603e at shift, MUL, cache/mem accesses and most integer operations. The PPC603e has a small advantage in FPU performance due to better single precision performance and more GP FPU registers while the 68060 generally has better FPU instruction latency timings for double/extended precision and CISC mem-reg operations. The 68060 uses its transistor budget for extended precision 68k software compatibility while the PPC603e targets single precision performance which is good for 3D but not C compilers which defaulted to double precision. I still don't see any major 68k architecture tax like x86. Even back then with tiny caches compared to today, the 68060 code density advantage offsets any PPC603(e) RISC pipeline savings advantage.

The major tax for fat 68K is Motorola's price.

Last edited by Hammer on 23-Jul-2024 at 08:28 AM.

_________________
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68)
Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68)
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB