Was anyone else disappointed with the A1200?

[Promilus]

Originally Posted by Gorf

Strange argument, since we actually do have this kind of acceleration with PPC datatypes for ages or now brand-new on the A600GS with ARM.

This might be a special problem of the ZZ9000 (context switches? not really on the same bus?) or due to it not being widespread to put it mildly.

And not only this:
there IS actually software using the DSP running on the reimplemented (A)A3000+ despite only a handful exist.

Ok... so is software for coldfire atari and that doesn't mean it was so good solution... It was fine experiment and so is both ZZ9000 and AA3000. Now let's see how DSP code would affect the future of Amiga if commodore once again was blown sky high...

So you did offload your 020 CPU from heavy calculation and put that workload on DSP. Your code now is relevant on DSP and not 68k. You put faster 68k which does not translate in much of the performance increase. You put much faster 68k and still won't get that much of an improvement because... performance of those apps are relying on on-board DSP from AT&T. Now how many compatible newer DSP AT&T did create afterwards? Hmmm ...

Now then let's assume DSP trick did work out and Commodore didn't die. To be backward compatible they'd have to attach that DSP as well to subsequent machines. And with each iteration they'd prove to be more of a burden. And then, at some point commodore would've introduced new architecture but no backward compatibility. Because it'd be next to impossible to emulate both 68k and dsp on PPC or ARM or PA RISC from that era. So... yeah, I love how ppl tend to jump on the first wagon they find driving to desirable location not worrying what comes after that...

[dreadnought]

Speaking from strictly gaming perspective, if you didn't have at least some degree of PC (or console) envy in 1992 I'd say it'd be a bit suspicious. In 1993, that'd be living in a bubble. Later years, full in-denial syndrome.

[TCD]

Originally Posted by dreadnought

Speaking from strictly gaming perspective, if you didn't have at least some degree of PC (or console) envy in 1992 I'd say it'd be a bit suspicious. In 1993, that'd be living in a bubble. Later years, full in-denial syndrome.

I wouldn't call it envy, but I was certainly looking elsewhere for the next gaming 'fix' by 1993. I guess a lot of people read only Amiga magazines and in them up until the middle of 1994 the world seemed a-okay for Amiga.

[dreadnought]

Originally Posted by TCD

I wouldn't call it envy, but I was certainly looking elsewhere for the next gaming 'fix' by 1993. I guess a lot of people read only Amiga magazines and in them up until the middle of 1994 the world seemed a-okay for Amiga.

Sure, but that's why I call it a "bubble". No doubt you can be happy in it but the fact remains that there are things in the wider world that would probably change your mind if you knew about them.

I don't mind the word "envy" myself, it doesn't have to be a pejorative. I'm always sort of envious if there's something amazing to play on other platforms. Eg atm my semi-powerful PC rig covers most of the bases rather well, but I'd really like to play Bloodborne and new Demon's Souls too. Next year it will be GTA 6

[Gorf]

Originally Posted by Promilus

Now then let's assume DSP trick did work out and Commodore didn't die. To be backward compatible they'd have to attach that DSP as well to subsequent machines. And with each iteration they'd prove to be more of a burden. And then, at some point commodore would've introduced new architecture but no backward compatibility. Because it'd be next to impossible to emulate both 68k and dsp on PPC or ARM or PA RISC from that era. So... yeah, I love how ppl tend to jump on the first wagon they find driving to desirable location not worrying what comes after that...

Like any other part of the chipset or eventually the 68K-line:
you have to provide some substitute in form of a compatible DSP-unit or software emulation or a mixture of both.
(same is true for legacy gfx-modes on PC as well or of course the XT mode for DOS on newer CPUs..)

DSPs do come with software libraries/drivers and the DSP in the AA3000+ is spoken to via the dsp3210.library and the dsp3210.device
https://aminet.net/package/driver/other/dsp3210

If you want to avoid that risk Amiga is the wrong platform, since it always relied on specialized hardware.

Apple did use 68K+DSP and it turned out this is not really a problem in the long run, as long as the whole ecosystem is alive and well you can switch architectures - and Apple has not done that once but three times now.

P.S.:
I am not really a strong DSP proponent and I am only arguing this case, because the objections to it are not really sound.
I would indeed prefer a better chipset, a more capable Blitter and Paula and so on.
But if the question is, what the A1200 should have been like and what simple solutions could have made it a more capable machine FastRAM + DSP are things that would have been available at the time, were already explored at Commodore, were cost effective and not difficult to implement.

[Tigerskunk]

Originally Posted by dreadnought

Speaking from strictly gaming perspective, if you didn't have at least some degree of PC (or console) envy in 1992 I'd say it'd be a bit suspicious. In 1993, that'd be living in a bubble. Later years, full in-denial syndrome.

I had a 2 years older friend who was already earning his own money in late 1991, and he bought a 386PC, and showed my Wing Commander 2, Ultima 6 and Gunship 2000.

I can safely say that my envy didn't know any bounds then.. lol..

I felt let down by Commodore and the Amiga. It was such a huge promise in 1987, only to get overtaken by 1991 already.

[sandruzzo]

Can we do some sort of wrap-up about what went wrong ?

[Gorf]

Originally Posted by sandruzzo

Can we do some sort of wrap-up about what went wrong ?

Commodore bought Amiga Inc.

[hammer]

Originally Posted by Promilus

Ok... so is software for coldfire atari and that doesn't mean it was so good solution... It was fine experiment and so is both ZZ9000 and AA3000. Now let's see how DSP code would affect the future of Amiga if commodore once again was blown sky high...

So you did offload your 020 CPU from heavy calculation and put that workload on DSP. Your code now is relevant on DSP and not 68k. You put faster 68k which does not translate in much of the performance increase. You put much faster 68k and still won't get that much of an improvement because... performance of those apps are relying on on-board DSP from AT&T. Now how many compatible newer DSP AT&T did create afterwards? Hmmm ...

If a MAME port runs Race Drivin' Arcade ROM, then it has DSP32C emulation software.

Originally Posted by Promilus

Now then let's assume DSP trick did work out and Commodore didn't die. To be backward compatible they'd have to attach that DSP as well to subsequent machines. And with each iteration they'd prove to be more of a burden. And then, at some point commodore would've introduced new architecture but no backward compatibility. Because it'd be next to impossible to emulate both 68k and dsp on PPC or ARM or PA RISC from that era. So... yeah, I love how ppl tend to jump on the first wagon they find driving to desirable location not worrying what comes after that...

Without an abstraction layer, switching from Am29K to DSP32 would be annoying for third-party developers.

DSP would be relegated to audio work with the new hardware platform. Fat PS3 has PS2 on a chip. Next PS3 revision removes parts of the PS2 chip and emulates on PS3 hardware. When the PS3 game library reaches a certain size, the backward compatibility is dropped. PS4 has an embedded PS2 software emulator for the PSN store's "buy-again" classic PS2 hits.

At some point, CPU power has increased to a level that it can emulate the missing DSP at the required performance level.

Xbox One has a cutting-edge Xbox 360 software emulator on par with Apple's Rosetta. Certain software mastery is important for the successful emulation of the previous hardware generation. Creating a fast emulator is an art in itself.

Missing nForce's SoundStorm DSP (Motorola 56300-based) was annoying at first until I purchased the Sound Blaster Audigy equivalent. SoundStorm DSP was driven by code largely derived from the 3D audio middleware company Sensaura and it has real-time Dolby Digital 5.1 encoding.

NVIDIA nForce 2 motherboards with SoundStorm are powered by Motorola 56300-based DSP, the same DSP is in the original Xbox's audio solution. SoundStorm's development was funded by Microsoft for their Xbox project.

Nvidia decided the cost of including the SoundStorm SIP (Motorola 56300) block on the dies of their chipsets was too high and was not included in nForce3 and beyond.

My K7 Athlon PC's nForce 2 motherboard had Motorola 56300 DSP under NVIDIA's SoundStorm brand. I still have my Athlon XP/nForce 2 with SoundStorm in storage.

Reference
https://en.wikipedia.org/wiki/SoundStorm

AMD's True Audio (based on Cadence Tensilica HiFi EP DSP with Tensilica Xtensa SP float support) and True Audio Next (based on AMD's GCN Polaris, GpGPU) are the next-generation game console-related DSP projects.

Cadence Tensilica is a stop-gap measure until AMD completes its DSP solution on GpGPU.

Support for True Audio Next was added to the Steam Audio API in February 2018.

https://en.wikipedia.org/wiki/AMD_TrueAudio

[Amigajay]

Originally Posted by Gorf

Thanks for the information.

Announcing an A500 CD upgrade that early might have been also a mistake that did hurt the CDTV sales quite a bit.
Why buy a CDTV if you have already a A500?
Or why not buy a A500 now and get the A570 later to have the best of both worlds?

Was it really worth to appear as a nice company that cares about its A500 customers in this case?

It seems especially odd since the CDTV was not marketed as an Amiga at first - they even tried very hard to avoid the Amiga image in the first year and only resorted to that route after the initial campaign flopped.
So why would you announce an A500 CD upgrade, if the CDTV was supposed to be a totally different animal?

That C= still kept that early promise after the CDTV production was long canned and even the A500(plus) was gone, underlines my claim, that they must have been under some contractual pressure.

Whatever it was, the plan did not really work out either:
only around 20k units of the A570 were ever sold.
(partly because you needed a A500+ with 1MB ChipRam to be really compatible with many CDTV titles)



I could not find any information if the AVM module (Advanced Video Mode) did exist even as a prototype. It was supposed to support more colours, probably in a similar way HAM-E or DCTV do, and Photo CD compatibility.

The A570/CDTV along with planned A2000/3000 CD drives should have been launched at the same time to help build publisher support and grow across all sectors.

I don’t think the A570 would have harmed CDTV sales, both being offered for totally different markets. (Same for the CD32 and the A1200CD yet Commodore were so scared it would they kept delaying it, not thinking more CD drives = more sales = more developer support = more games = more hardware sales).

I can partly see why they didn’t want/need the Amiga name on the CDTV as it aimed at people who wanted multimedia and information/education for the living room, the type of people who bought encyclopaedia’s not who bought computers or consoles.

The fact the A570 drive was only released 18 months after the CDTV which now adorned the Amiga name, should Commodore didn’t give too hoots for A500 owners, which as you say needed the A500+ to be fully compatible (and was brought out for that very reason and discontinued 6 months later) is painfully obvious in hindsight, also bringing incompatibilities in the Amiga scene.

The AVM prototype card was shown at one of the Wofc Toronto and impressed people saying it easily matched and surpassed AGA in some area’s, basically a budget DCTV card to be sold for £50.

Originally Posted by sandruzzo

Can we do some sort of wrap-up about what went wrong ?

We haven’t got time!

In all seriousness, it would be a lot quicker to list the things that went right!

[Tigerskunk]

Originally Posted by sandruzzo

Can we do some sort of wrap-up about what went wrong ?

I think it would be better to just split up the thread and let those "gifted people" who want to talk about numbers have their fun in there.

[sandruzzo]

@Tigerskunk

We're getting nowhere and we are spinning around..

[hammer]

Originally Posted by Amigajay

The A570/CDTV along with planned A2000/3000 CD drives should have been launched at the same time to help build publisher support and grow across all sectors.

I don’t think the A570 would have harmed CDTV sales, both being offered for totally different markets. (Same for the CD32 and the A1200CD yet Commodore were so scared it would they kept delaying it, not thinking more CD drives = more sales = more developer support = more games = more hardware sales).

I can partly see why they didn’t want/need the Amiga name on the CDTV as it aimed at people who wanted multimedia and information/education for the living room, the type of people who bought encyclopaedia’s not who bought computers or consoles.

The fact the A570 drive was only released 18 months after the CDTV which now adorned the Amiga name, should Commodore didn’t give too hoots for A500 owners, which as you say needed the A500+ to be fully compatible (and was brought out for that very reason and discontinued 6 months later) is painfully obvious in hindsight, also bringing incompatibilities in the Amiga scene.

The AVM prototype card was shown at one of the Wofc Toronto and impressed people saying it easily matched and surpassed AGA in some area’s, basically a budget DCTV card to be sold for £50.





We haven’t got time!

In all seriousness, it would be a lot quicker to list the things that went right!

A570 needs 1 MB Chip RAM And Agnus ECS 1 MB.

My 1989 A500 Rev 6A had Agnus ECS 1 MB.

CD1200's expansion card for A1200's internal edge connector had a 68030 socket, SIMM slot, FPGA, and Akiko. Commodore doesn't want cheapo A1200 with CD1200 to step on their profit cream A4000/030.

[Gorf]

Originally Posted by hammer

My 1989 A500 Rev 6A had Agnus ECS 1 MB.

In 89?
Or did you make an upgrade later?

[Dunny]

Originally Posted by hammer

Emu68 has quad-core(four threads) soft 68K extensions beyond increasing the single 68K thread. Emu68 exposes quad-core ARM CPUs for quad-core 68K. It needs new software for quad-core 68K usage. New features will need new or modified software.

lolwut you tripping dude

[pandy71]

Originally Posted by Thorham

Science uses whatever precision is appropriate, including 16 bit and less (minifloat).

True, nowadays FP8 are widely used for neural data representation (all that fancy NPU units and some of the GPUS's use FP8 as basic format).
Sometimes (not so rare) even FP80 is not enough for science as such we have 128 decimal and binary float, 256 floats, arbitrary precision floats etc...

[hammer]

Originally Posted by Gorf

In 89?
Or did you make an upgrade later?

My A500 Rev 6A has the ECS Agnus 8372A chip.


My A500 starter pack comes with Crazy Cars, Super Ski, Hole-in-One Miniature Golf, Kind Words 2.0 and Fusion Paint. 1084S monitor purchased along with A500 pack.

https://amiga.abime.net/games/view/starter-kit

I only need the 512KB RAM card. The two jumpers are on the Rev 6A motherboard.

[pandy71]

Originally Posted by Promilus

Now then let's assume DSP trick did work out and Commodore didn't die. To be backward compatible they'd have to attach that DSP as well to subsequent machines. And with each iteration they'd prove to be more of a burden. And then, at some point commodore would've introduced new architecture but no backward compatibility. Because it'd be next to impossible to emulate both 68k and dsp on PPC or ARM or PA RISC from that era. So... yeah, I love how ppl tend to jump on the first wagon they find driving to desirable location not worrying what comes after that...

? Why? Emulation is the same problem - if you have decent HW then it is more like proper usage of this HW in context of emulation - i have impression that emulation on PowerPC don't care whether currently emulated instruction is logic or DSP like...
Same like 68882 emulation - it is probably direct translation from 68882 instruction to native one and using data from virtual 68882 registers...

Originally Posted by Promilus

So you did offload your 020 CPU from heavy calculation and put that workload on DSP. Your code now is relevant on DSP and not 68k. You put faster 68k which does not translate in much of the performance increase. You put much faster 68k and still won't get that much of an improvement because... performance of those apps are relying on on-board DSP from AT&T. Now how many compatible newer DSP AT&T did create afterwards? Hmmm ...

Why? Let say you have generic math and dedicated math so DSP specific (like DCT/iDCT etc) - let say dsp.library, same as you are offering math????.library to shift numeric calculations to faster HW - applications may use system standard or provide their own method, system method may be slightly slower but offer higher flexibility, also app can be written in a way to fallback in system libs when own not work...

[hammer]

Originally Posted by Dunny

lolwut you tripping dude

[ Show youtube player ]
SmallPT path tracer on Emu68 with four threads. New features require new or modified software.

[hammer]

Originally Posted by pandy71

True, nowadays FP8 are widely used for neural data representation (all that fancy NPU units and some of the GPUS's use FP8 as basic format).
Sometimes (not so rare) even FP80 is not enough for science as such we have 128 decimal and binary float, 256 floats, arbitrary precision floats etc...

ADA:
Tensor core: FP64, TF32, bfloat16, FP16, FP8, INT8.
CUDA cores: FP64, FP32, FP16, bfloat16, INT8.
Gaming ADA GPUs has slow FP64.


Hopper:
Tensor cores: FP64, TF32, bfloat16, FP16, FP8, INT8.
CUDA cores: FP64, FP32, FP16, bfloat16, INT8.


Ampere:
Tensor cores: FP64, TF32, bfloat16, FP16, INT8, INT4, INT1
CUDA cores: FP64, FP32, FP16, bfloat16, INT8.
Gaming Ampere GPUs has slow FP64.

Reference
https://www.cmc.ca/wp-content/upload...ffin-Lacey.pdf

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

https://www.ign.com/articles/nvidia-...hing-announced

Meanwhile, the new tensor core will support FP8 and power DLSS 3.0.

Gaming DLSS 3.0 is only available on ADA (GeForce RTX 4000 series) due to FP8. Too bad for the older Ampere GPU owners.

IBM Power9 and some NEC vector addon card have native support for quadruple precision 128-bit FP.