Are PS3 devs using the two mem pools for textures?

Does anyone know why Cell is limited in reading from GDDR to 16MB/s? Is there some advantage in having a one way data path? Is it because the data transfer rates from Cell is throttled in order to ensure that RSX memory access is not interfered with, and because read from GDDR (16MB/s) is rarely required whereas write to GDDR (4GB/s) is much more useful?

CPU's reading from VRAM isn't something you generally do, consider it a small miracle that you can do it all. Try doing it on your PC sometime and see what happens.
 
CPU's reading from VRAM isn't something you generally do, consider it a small miracle that you can do it all. Try doing it on your PC sometime and see what happens.

Thats not strictly true (at least as of a few years ago, last time I did serious PC work), with the right restrictions and usage pattern you can do it on PC with no major performance cost.

The trick is multi buffer your surfaces, with enough delay, so that the CPU lock doesn't stall the currently rendering surface. Then the limit is only the bus transfer speed, which even for AGP isn't too bad as long as you only read a few pixels.

Perfectly usable for light occlusion tests for example.
 
CPU's reading from VRAM isn't something you generally do, consider it a small miracle that you can do it all. Try doing it on your PC sometime and see what happens.

I think it's not necessary because in this case it is more efficient to simply have the display buffer you want to work with using the CPU in main memory anyway? Also maybe it is possible to send stuff from RSX through an SPU directly?
 
Thats not strictly true (at least as of a few years ago, last time I did serious PC work), with the right restrictions and usage pattern you can do it on PC with no major performance cost.

The trick is multi buffer your surfaces, with enough delay, so that the CPU lock doesn't stall the currently rendering surface. Then the limit is only the bus transfer speed, which even for AGP isn't too bad as long as you only read a few pixels.

Perfectly usable for light occlusion tests for example.

Yeah, true you can read pixels out to RAM. Although in most of those cases it's usually the CPU reversing the bus (in the case of AGP) and the GPU pushing the data down the bus (actually Mac GPU drivers are rather adept at it considering (I'd imagine Vista drivers as well).
 
I see this sometime ago (july/2006) from guy seeking information exhaustingly on what would be the RSX :

RSX


Core Frequency - 500MHz
Memory Frequency - 650MHZ
Bus Size: 128BIT
Pixel Shaders - 24
Vertex Shaders - 8
ROPS - 8
Total Texture Cache Per Quad of Pixel Pipes (L1 & L2) - 96KB
Post Transform & Lighting Cache - 63 Max Vertices
*A few extra shader instructions - Extra Texture Lookup Logic & Fast Vector Normalize
*FLEX IO interface to CPU (Much Faster)


7800GTX


Core Frequency - 430
Bus Size: 256BIT
Memory Frequency - 600MHZ
Pixel Shaders - 24
Vertex Shaders - 8
ROPS - 16
Total Texture Cache Per Quad of Pixel Pipes (L1 & L2) - 48KB
Post Transform & Lighting Cache - 45 Max Vertices
PCI BUS interface to CPU (Much Slower)


NOTES: About the RSX.


Total Texture Cache Per Quad of Pixel Pipes - (L1 and L2) 96KB total Texture Cache - Previously 48K
(L1 only available to Pixel Shaders)

Post Transform and Lighting Vertex Cache - 63 Max Vertices - Previously 45 Vertices
(Cache located after Vertex Shader and before the triangle setup and before the Rasterizer.)
Vertex shader --> Post Transform and Lighting Vertex Cache 63MAX ---> Triangle Setup
Texture Lookup Logic to help RSX transport data from XDR


( I have heard this spec fron the others places too ...include here in this forum)

Just a question but although the G70/Nv7800GTX is 302MT .11nm versus G71/Nv7900GTX is 278MT .09nm and the latter chip was able to hit 500~550Mhz easy some months prior to PS3 production...

Would it no be impossible to rule out that RSX/G7X Sony Custom use being 302MT? .09nm be able to hit 550Mhz as Sony had initially released info to press or is it just downclocked for some reason and will later get up clocked again with a more stable firmware? (kind of like PSP)

Anyways not placing too much importance on the missing 50Mhz but making sense of it.
 
I see this reason pop up a lot for the reason why RSX has a higher TC than it's PC counterpart, why is Sony/nVidia facing this and yet I've heard nothing similar for MS/ATI in producing Xenos.

But you have heard about Xenos exceeding it's thermal rating and nuking systems.

Thats what happens when your yields are not really as good as your QA guys are claiming they are.
 
RSX has more than either 278 or 302 million transistors. Take a look at the die size. ;)

It also has four extra (disabled) pipes for redundancy.
But you have heard about Xenos exceeding it's thermal rating and nuking systems.

Thats what happens when your yields are not really as good as your QA guys are claiming they are.

I doubt that's the "problem". Both are clocked 500mhz.
 
I see this reason pop up a lot for the reason why RSX has a higher TC than it's PC counterpart, why is Sony/nVidia facing this and yet I've heard nothing similar for MS/ATI in producing Xenos.

I believe Xenos has significantly few transistors and is smaller if we leave out the eDRAM, which is on a separate chip. Its yield problems shouldn't be as great as RSX's.
 
Interesting.... From what I did read I think I received much the same impression some people did about latencies, and I can only conclude RSX is designed to mitigate the effects of latency.

If that is so, does this mean that we could see a game with PGR3 textures running on PS3 someday in the near future?

I remember reading in this very forum that UMA memory has an advantage there because of its nature, giving X360 some edge.

I don't want to get technical and all here, so to see what I mean simply click on this link:

http://www.youtube.com/watch?v=EjhhpcxrW6U

LBP, HS, Ratchet, MGS4 and GT have awesome textures and they are games that have that certain something.

GT has a great lighting system and cars models but most of the photos are close shots, while PGR3 has the cars models plus the environments (the red building towards the end of the linked video is sick).

Anyway, it's not about comparisons but a real life example and I just want to know if that's possible or not, or it's just that only Bizarre Creations could do that.

Cheers
 
Are you sure you want to ask that question?
Hmm, not so sure if you ask me, but it's one of those studios with a great personality. Like Polyphony and their lighting system, just to give you another example. I am not demanding an answer there, perhaps it brings a side question but that's pretty much there is to it.

In this forum you can read living examples of creative, talented people, some of those do that just have everything, or it seems so. I was talking about an abstract concept, though, developers that I admire for whatever reason.

It's perhaps I'm "rediscovering" PGR3 (it was my first X360 game, I've had lot of fun playing it but I criticized it because it lacked the gameplay mechanics of PGR2) and, as weird as it sounds, I am pretty impressed with the graphics after a year and a half of NG gaming. What's more, the game came out when the knowledge of the hardware was very limited in comparison with today's standards.

Cheers
 
I believe Xenos has significantly few transistors and is smaller if we leave out the eDRAM, which is on a separate chip. Its yield problems shouldn't be as great as RSX's.

Excluding the eDRAM thus only the logical components, the Xenos 'seems' to give a better utilisation of transistors or 'bang per buck' in terms of GPU functions compared to the RSX.

In other words the Xenos with less transistors in it's main core packs in (feature wise) as much if not more (give or take) as the RSX. That said, the eDRAM, though separate from the main die, is part of the entire GPU package.

What l'd like to find out is with all things equal (I know its not, but roughly speaking), Xenos including the eDRAM @330 MT and RSX @300 MT are similar in transistors ball park figure wise, where is the outstanding 70 MT in the RSX used for given it does NOT have a 10MB eDRAM consuming 100MT?

It's hard to account 23% (70 MT) of total transistors used for redundancy (yield) and FlexIO. Have I forgotten something or is this the $million question?
 
Last edited by a moderator:
Excluding the eDRAM thus only the logical components, the Xenos 'seems' to give a better utilisation of transistors or 'bang per buck' in terms of GPU functions compared to the RSX.

In other words the Xenos with less transistors in it's main core packs in (feature wise) as much if not more (give or take) as the RSX. That said, the eDRAM, though separate from the main die, is part of the entire GPU package.

What l'd like to find out is with all things equal (I know its not, but roughly speaking), Xenos including the eDRAM @330 MT and RSX @300 MT are similar in transistors ball park figure wise, where is the outstanding 70 MT in the RSX used for given it does NOT have a 10MB eDRAM consuming 100MT?

It's hard to account 23% (70 MT) of total transistors used for redundancy (yield) and FlexIO. Have I forgotten something or is this the $million question?


You can't hust discard the EDRAM like that in a comparison.
Because it has EDRAM Xenos doesn't need things like Framebuffer or Z Compression, and it probably has a much simpler interface to the framebuffer memory in generl, if it didn't have EDRAM it would have a much higher count in it's logic unit.
 
What l'd like to find out is with all things equal (I know its not, but roughly speaking), Xenos including the eDRAM @330 MT and RSX @300 MT are similar in transistors ball park figure wise, where is the outstanding 70 MT in the RSX used for given it does NOT have a 10MB eDRAM consuming 100MT?

It's hard to account 23% (70 MT) of total transistors used for redundancy (yield) and FlexIO. Have I forgotten something or is this the $million question?

You have forgotten that Xenos was the first GPU with unified shaders. Thats the reason it can have less transitors for logic and get good results.

It uses the 48 shader pipes for both pixel and vertex shading. (+16 texture filter units), this is a much more flexible design in gaming, because your rarely using enough vertex AND pixel shaders at once to bottleneck the GPU seriously. With this unified architecture your using all your shader power no matter how much vertex or pixel shading your doing.

In RSX's case it has seperate pixel and vertex pipes, so a lot of the shader power is not utilized (because you rarely have scenes that need huge amounts of pixel and vertex at the same time).
 
Yes, but if you're in the console arena, wouldn't you make games that get the most utilization of your resources? I think devs would look at unused shading units as an opportunity to do more.
 
Back
Top