new rumor about a PS3 without GPU

[archie4oz]

I always find these articles amusing since they lead to rumors being asserted as truth. Anyway, Zeross's comment is far more accurate than any article I've read so far regarding any "history" of PS3 GPU selection...

 

[patsu]

If you guys published a "Making of ..." book, it would clear things up once and for all.

 

[Deleted member 13524]

Eric lengyel who was working in the ICE team at this time challenged the accuracy of this article. From what I've gathered though it seems that there were a lot of designs evaluated but some were scrapped pretty early in the concept stage : two Cell design, Toshiba Cell-ish GPU, Toshiba VLIW GPU (Reality Synthetizer ?)...

The only people knowing the truth are not willing to share it due to NDA probably, that's a pity since nowadays it's not that important for Sony to keep it a secret but it would be interesting to learn more about it. I would love to read a book like "Xbox 360 Uncloaked" about the PS3 hardware design process.

Whatever was or wasn't scrapped early in the concept stage, there are quite some factors pointing to the inclusion of nVidia in the project being rushed or at least a bit mismanaged:

1 - PS3 came out a substantial amount of time later than the competition
2 - Later release date didn't translate into better performance, power efficiency or cost effectiveness than the competition (ok, it was more reliable, but that was a series of mistakes on microsoft's part).
3 - RSX doesn't seem to be a heavily customized part at all. In fact, it seems to be just a G70 with halved ROPs and (consequently) halved memory interface.
4 - Two different memory pools were not exactly an elegant solution, not even by 2003 standards. Two distinct memory controllers had to be there, PCB was probably more expensive, licensing had to be done for two different types of memory, purchasing all the memory from the same place would probably have come out cheaper, etc. etc. Furthermore, developers complained a lot with the dual memory subsystem in the PS3.
One can only guess that nVidia didn't have time to implement a FlexIO controller in RSX.
5 - I remember some 3D demos being shown with Cell a couple of years before the PS3 (the one with the ducks?). Many people started to think that Cell (alone) would be used for graphics in the PS3 because of that.

 

[patsu]

They were also troubled by blue diode production delay at that time.

5 - I remember some 3D demos being shown with Cell a couple of years before the PS3 (the one with the ducks?). Many people started to think that Cell (alone) would be used for graphics in the PS3 because of that.

... which it eventually did. A huge portion of the SPU workload are for graphics.

Some of the demoes were actually done on 2 Cells, not one. I believe the Toshiba magic mirror is one of them.

 

[pjbliverpool]

Being the crazy person that I am I would have loved to see the PS3 with 2 Cell processors.

about 400GFLOPS of software rendering on a console in 2006 might have changed gaming.

You're right, Sony might have dropped out of the console business altogether by now!

 

[liolio]

Cell didn't lack direction. They were too ambitious: consumer devices (Toshiba), Sony (Playstation), and IBM (HPC). It went much further than intel's Itanium and LRB efforts.

I disagree there was a disagreement on both what it should look like and the purpose:
IBM was aiming at HPC and wanted a "better Xenon"
Toshiba wanted a stream processor, and wanted a better Xenon too though based on MIPS
Sony wanted a successor of EE and so a heterogeneous chip which included vector processors.

Then there is more to Sony failure than the Cell, the management failed my belief if that they wanted both a new EE and a new GS, they focus way too much resources on the former and the latter did not see the light. Cerny for me made some serious critics of the previous management, wrapped in a smile, yet really mean, whatever the reason (ego, bad practices, etc) the approach within Sony were wrong with tiny team working in insulation mostly from the world. When it comes to tackles seriously complex issues it is not the good way to do it, you need to do what Cerny put in place, wide collaboration, taking insight from as many sources as possible, looking at what the competition are doing, etc. a more 'complex' approach.

Wrt Itanium it did not failed, it did not replaced X86 though it pretty much killed its competition, leaving IBM as the last man standing (~).
Larrabee did not shipped but was a focus, the goal was clear, design a many core X86 CPU which can handle rendering. Further more the hardware and programming model were developed hand in hand.

Anyway, I won't speak of the matter further, there has been too many discussion on the matter of the Cell success and failing.

You're right, Sony might have dropped out of the console business altogether by now!

Indeed though as I read it, it was not intended to ship without a GPU more than they mismanaged the development of the GS 2.0.

 

[patsu]

I disagree there was a disagreement on both what it should look like and the purpose:
IBM was aiming at HPC and wanted a "better Xenon"
Toshiba wanted a stream processor, and wanted a better Xenon too though based on MIPS
Sony wanted a successor of EE and so a heterogeneous chip which included vector processors.

IBM's HPC effort was PowerPC based. And they later introduced improved Cell for double precision. HPC has a lot of heterogenous, GPU based solutions these days.

Toshiba went with SPUREngine and ditched the PPU.

Even without EE influences, AMD APU is heterogenous.

At that time, they all got together to try to explore a common hardware platform. They didn't lack a vision. KK was the primary driver since the IBM lead had to take his input as requirements (hence, 8 cores). Toshiba also tagged along despite their unhappiness with the PPU.

Then there is more to Sony failure than the Cell, the management failed my belief if that they wanted both a new EE and a new GS, they focus way too much resources on the former and the latter did not see the light. Cerny for me made some serious critics of the previous management, wrapped in a smile, yet really mean, whatever the reason (ego, bad practices, etc) the approach within Sony were wrong with tiny team working in insulation mostly from the world. When it comes to tackles seriously complex issues it is not the good way to do it, you need to do what Cerny put in place, wide collaboration, taking insight from as many sources as possible, looking at what the competition are doing, etc. a more 'complex' approach.

Wrt Itanium it did not failed, it did not replaced X86 though it pretty much killed its competition, leaving IBM as the last man standing (~).
Larrabee did not shipped but was a focus, the goal was clear, design a many core X86 CPU which can handle rendering. Further more the hardware and programming model were developed hand in hand.

Anyway, I won't speak of the matter further, there has been too many discussion on the matter of the Cell success and failing.


Indeed though as I read it, it was not intended to ship without a GPU more than they mismanaged the development of the GS 2.0.

Itanium and LRB were aborted efforts. Look at the install base, or lack there of.