Why not let PowerVR build RSX?

[Rolf N]

Have they been approached at all? It puzzles me.
Now for prosperity, RSX is not the slouch certain people would reverse-logic it to be. We know it has a lot of texture samplers that go along with lots of pixel ALUs and even though its vertex oomph may not be world-beating, it is dedicated oomph that doesn't eat into other performance metrics once utilized.

It is obvious though that RSX would be well served with more bandwidth. A 128 bit GDDR3 bus may not be as much of a problem when long shaders are to be executed per pixel but still it is a significant constraint to some of the simpler, pure-fillrate intensive things the chip might be doing regularly (shadow map or volume rendering, tone mapping etc).

Okay. So we have the ideal fit. A little company that could build a graphics core with minimal bandwith requirements per pixel filled, but also per AA sample. An architecture that has been touted (in the official press releases) to be very scalable, which I assume must mean at least a 16-piper could be built. A core that is available as IP from a company with extensive experience with customization and system integration.

It puzzles me that RSX is not a PowerVR architecture. While NVIDIA would have to bend and change their ways to accept an IP deal, Img Tech has been doing this exact kind of business all along.

Is it still the fear of limited vertex throughput? Were the performance requirements out of Img Tech's reach? Have they not been considered? Politics, rivalry in other fields?

 

[Shifty Geezer]

Software development tools? Proven top-end performance (if PowerVR is so great, why don't they have top-end performing parts)? Developer familiarity with the architecture?

 

[Rolf N]

Software development tools? Proven top-end performance (if PowerVR is so great, why don't they have top-end performing parts)? Developer familiarity with the architecture?

Img Tech has enough experience with OpenGL ES to build something nice for a closed box.
If you want to swim in a PC environment you need to support lots of brain-dead API revisions and the quirks and bugs in thousands of obscure games. Robustness or conformity with one current standard alone won't get you anywhere. Plus, of course, Img Tech does not make the investments required for producing silicon and boards, they'd need a partner willing to absorb all the risks.

I'm not sure what tools NVIDIA brings to the table and how neccessary they are. Fine-grained performance profiling is the one thing that strikes me as important, though even that isn't strictly required to make a nice game. It isn't very hard to build some self-profiling code into an application. It may not be as precise, and it may not tell you why performance is like it is, but it can still be used to make effective optimization passes.

 

[tongue_of_colicab]

Maybe because of time? The idea behind ps3 was that cell would do everything. But at some point sony figured cell wasnt up to the gfx job so they needed to look for a gpu elswhere. So they went shopping at Nvidia's from who they knew they could get a high performance gpu. If they went for powervr they should have build a total new chip, wich there might not be enough time for and sony wouldnt have known how it performed befored it was build. Given that they didnt had alot of time, that might been a to big risk?

 

[Darkon]

Maybe because of time? The idea behind ps3 was that cell would do everything. But at some point sony figured cell wasnt up to the gfx job so they needed to look for a gpu elswhere. So they went shopping at Nvidia's from who they knew they could get a high performance gpu. If they went for powervr they should have build a total new chip, wich there might not be enough time for and sony wouldnt have known how it performed befored it was build. Given that they didnt had alot of time, that might been a to big risk?

Why ,why , does this rumour continue to exists.

Actually from what i've heard Sony did contact imagetech

 

[tongue_of_colicab]

I dont know why But long long ago I believe i've read a sony statement that cell would power everything in the ps3. Now im might be wrong on that. But I still get the feeling that sony didnt work with nvidia from the start.

 

[archie4oz]

I dont know why But long long ago I believe i've read a sony statement that cell would power everything in the ps3. Now im might be wrong on that. But I still get the feeling that sony didnt work with nvidia from the start.

They didn't, but nor was Cell going to be the GPU either... What's amusing to me though is that even folks from ATI^H^H^H AMD, and even BillyG himeself seems to think so...

 

[nonamer]

The original plan was for Toshiba to make the GPU. Unfortunately, Toshiba sucked at making GPUs, and I suspect Sony was very wary of having yet another unproven company making the GPU. nVidia was the obvious and perhaps only choice, given the circumstances.

 

[Fran]

Fine-grained performance profiling is the one thing that strikes me as important, though even that isn't strictly required to make a nice game.

But it helps a long long way.

 

[see colon]

it's more of a question of why shouldn't they ahve gon with nVidia? nVidia has a proven track record in the PC space and in embedded devices (with the goforce and XGPU). plus, their main competator is supplying technology to Sony's other console competators. there's sure to be a bit of politics that would compel them to work together.

 

[Capeta]

Unfortunately, Toshiba sucked at making GPUs, and I suspect Sony was very wary of having yet another unproven company making the GPU.

Which would mean SONY sucks even harder.

Anyway the rumor about a CELL based GPU arose from the CELL patents.

 

[Skrying]

Lack of time. Somewhere along the line Sony had different plans other than what amounts to off the shelf (relatively speaking here) part that RSX is now. They guessed wrong early on, it came back to bite them, they jumped to Nvidia in order to get a known competitive part from a company with lots of proven experience.

 

[Carl B]

Sony began working with NVidia at least as early as 2003, so whatever the status of 'rushed' or 'last minute,' the move to an "off-the-shelf" part had nothing to do with time constraints. For whatever reasons, be they economic or otherwise, RSX as based on GeForce 7 was simply the route Sony chose to take (and likely that NVidia pushed). Most people are aware of RSX being Plan B so to speak, but I don't think many people realize just how early in the game Sony switched to plan B (aka NVidia) to begin with.

Which is to say, that if their are shortfalls in the design of RSX, the shortfalls are of vision rather than of arrogance and time crunches; I feel folk are often too quick to envision Sony walking into "gotcha!" situations... they figured out the score with the Toshiba GPU very early on, and NVidia came soon after.

 

[Darkon]

Which would mean SONY sucks even harder.

How so

Anyway the rumor about a CELL based GPU arose from the CELL patents.

Here's some info on the " cell based gpu"
http://www.beyond3d.com/forum/showpost.php?p=931675&postcount=77

they figured out the score with the Toshiba GPU very early on, and NVidia came soon after.

Didn't Cpia say from what he heard , Sony looked around first before they went with Nvidia ?

BTW what was the name of that chip that GS was derived off

 

[patsu]

I dont know why But long long ago I believe i've read a sony statement that cell would power everything in the ps3. Now im might be wrong on that. But I still get the feeling that sony didnt work with nvidia from the start.

If so, then you must also believe that I have a PS3 4D time machine on my desk.

But seriously other than the PSU, Cell still call the shot for everything happening in PS3. RSX just handles the rendering.

 

[nonamer]

Sony began working with NVidia at least as early as 2003, so whatever the status of 'rushed' or 'last minute,' the move to an "off-the-shelf" part had nothing to do with time constraints. For whatever reasons, be they economic or otherwise, RSX as based on GeForce 7 was simply the route Sony chose to take (and likely that NVidia pushed). Most people are aware of RSX being Plan B so to speak, but I don't think many people realize just how early in the game Sony switched to plan B (aka NVidia) to begin with.

Which is to say, that if their are shortfalls in the design of RSX, the shortfalls are of vision rather than of arrogance and time crunches; I feel folk are often too quick to envision Sony walking into "gotcha!" situations... they figured out the score with the Toshiba GPU very early on, and NVidia came soon after.

Half the problem was due to nVidia's conservatism. They didn't want to make a custom GPU not because of cost, but because it was too much of a resource hog. They really wanted to get their G80 out the door before ATI's response, and they didn't want anything interfering with that.

 

[DuckThor Evil]

Sony began working with NVidia at least as early as 2003.

I don't want to stir up shit, but do we really know that as a fact. Even if they had some collaboration at that time, can we be absolutely sure that they were working on RSX at that time. To me it just seems very odd that they would work for years and then come up with what they did. I'm not saying it's a bad chip, but it doesn't seem like a custom chip, that was engineered for years separately from G70.

 

[Shifty Geezer]

I'm not saying it's a bad chip, but it doesn't seem like a custom chip, that was engineered for years separately from G70.

As I ask of everyone who raises this (which no-one answers ), what would you expect a custom chip to look like? The GPU would want to have vertex units, pixel units, texture units, etc. If Sony ruled against eDRAM, and nVidia didn't have US working in a useable form, what would they do differently? The end result of a custom GPU would look a lot like G70, no? Vertex units+pixel units+texture units.

 

[DuckThor Evil]

As I ask of everyone who raises this (which no-one answers ), what would you expect a custom chip to look like? The GPU would want to have vertex units, pixel units, texture units, etc. If Sony ruled against eDRAM, and nVidia didn't have US working in a useable form, what would they do differently? The end result of a custom GPU would look a lot like G70, no? Vertex units+pixel units+texture units.

That's a valid point, I just find the sentence "Sony worked with nVidia for years" little bit baseless when the end result is basically of the shelf GPU. To me it seems that there wasn't much collaboration between the companies at least in engineering, maybe they purchased the chip and the desing in 2003, but I doubt there was much "working" done between the companies at that time. I don't have the technical knowledge to say much about on the question should there be more difference between a PC GPU or one that works with Cell, so I can't give good answer for that...

To me it just seems that it's kind of bolted in there... with PS3s' memory structure it just kind of feels inelegant, while still performing nicely, though I must say that the GPU part of the PS3 is the least cutting edge part of the console, especially considering the late launch. But hey what do I know... Not much that's for sure :smile: I'm just saying how I see it.

 

[ban25]

When was the last time PowerVR built anything that didn't end up in a mobile phone? Kyro II? The Dreamcast? Are you serious about this thread?