Please Clear This Up - What PC GPU Does the XBOX 360 & PS3 Use?

[Acert93]

Heh..it'll be awesome to get a comparitive PC side GPU benchmarked..

Assuming that ever happens..main problem I see is if ATI retains the PC mini-ALU for R600 we'll never get a clear comparison.

The R600 will have more features overall, but also may be lacking some features Xenos has. And it will be difficult to compare in some ways because eDRAM is very central to the effeciency of the 360 (the eDRAM alleviates a lot of the framebuffer and fillrate bottlenecks and puts the weight back on the shader pipeline) and consoles have fixed designs and target resolutions where PCs are modular.

Synthetic tests could tell us a bit, but we don't have any of those for Xenos. I guess R600 will tell us some basic stuff like "do unified shader designs really work" (although R600 could hit some nasty driver issues early...). I guess I would not put too much on R600 in that it seems to be very shader focused. Yet we have seen how the R580, which is also shader focused and has excellent DB, is barely touched and underperforms in many ways because games are not really tailored to it. Hmmm... now that I put it that way, it will tell us a lot because 7 months into the 360 a lot of title are definately not tailored towards the 360's strengths and featureset!

 

[Acert93]

AFAIK
Xenos is a X1900

No :smile: Xenos could be called a hardware "preview" (but not all) of some D3D10 / R600 features though (IMO). ATI seems to have confirmed this by saying R600 would be leveraging their research in Xenos. Xenos does share some things in common with X1900 (and I believe a lot of what is in the X1900 in constrast to the X800 is there to test things out thoroughly for R600) but I think Xenos probably leans more toward R600 in design and features.

(+) plus
...
(-) minus

I would not call them pluses or minuses as they are designed for different markets with different needs and threshholds.

I would say on similarities:
* both have a high ration of math to textures
* both have solid batch sizes for dynamic branching (48 for R580, 64 for Xenos) and seem to share some of the same threading DNA
* Decoupled pipeline (including ROPs and TMUs)
* SM3.0
* 90nm process and released in the same 3 month window

Differences abound:
* Xenos has a unified shader array with load balancing, R580 has discreet PS and VS, heavy on PS; different shader array configurations
* The R580 48 Pixel Shaders have more math capability overall, Xenos goes the simple/robust route which fits the utilization/load balancing design
* Completely different memory configurations -- really not comparable and meet totally different needs due to system configuration. Xenos shares a common memory pool with the CPU, R580 does not; Xenos has enough bandwidth to sustain worse-case scenarios for the framebuffers and fillrate due to the eDRAM, the R580 does not
* Xenos has a number of extra features and/performance additions: Hardware Tesselation, Vertex Texturing from every Shader ALU, Higher Order Surface support, coherant memory reads and writes (memexport), FP10 blending and filtering, TMUs also have point-sampling, double Z samples per clock, even when MSAA is enabled * No penalty for 4xMSAA (I believe R580 does at 4xMSAA)
* Xenos has a direct link to the CPUs L2 cache and the CPU has custom instructions for datastreaming
* General differences: R580 is much, much bigger; R580 has a much higher peak ALU performance; R580 operates at 650MHz compared to 500MHz for Xenos

The X1900 is a better chip for the desktop space; likewise I think Xenos has a better balance of performance for the console space and has a lot more features that will be important in the future once DX10 kicks into gear or are valuable for performance in a closed platform (e.g. FP10). They both share a number of common features, some under utilized today (dynamic branching, a lot of math per pixel, etc).

Anyhow, I don't think we can call Xenos an X1900 with some tweaks of pluses/minuses. ATI obviously did not dumb all their research when creating Xenos, but it definately is a significant departure from the R520 and R420 series of GPUs. The workflow and featureset is substantially different.

 

[JarHead]

No :smile: Xenos could be called a hardware "preview" (but not all) of some D3D10 / R600 features though (IMO). ATI seems to have confirmed this by saying R600 would be leveraging their research in Xenos. Xenos does share some things in common with X1900 (and I believe a lot of what is in the X1900 in constrast to the X800 is there to test things out thoroughly for R600) but I think Xenos probably leans more toward R600 in design and features.



I would not call them pluses or minuses as they are designed for different markets with different needs and threshholds.

I would say on similarities:
* both have a high ration of math to textures
* both have solid batch sizes for dynamic branching (48 for R580, 64 for Xenos) and seem to share some of the same threading DNA
* Decoupled pipeline (including ROPs and TMUs)
* SM3.0
* 90nm process and released in the same 3 month window

Differences abound:
* Xenos has a unified shader array with load balancing, R580 has discreet PS and VS, heavy on PS; different shader array configurations
* The R580 48 Pixel Shaders have more math capability overall, Xenos goes the simple/robust route which fits the utilization/load balancing design
* Completely different memory configurations -- really not comparable and meet totally different needs due to system configuration. Xenos shares a common memory pool with the CPU, R580 does not; Xenos has enough bandwidth to sustain worse-case scenarios for the framebuffers and fillrate due to the eDRAM, the R580 does not
* Xenos has a number of extra features and/performance additions: Hardware Tesselation, Vertex Texturing from every Shader ALU, Higher Order Surface support, coherant memory reads and writes (memexport), FP10 blending and filtering, TMUs also have point-sampling, double Z samples per clock, even when MSAA is enabled * No penalty for 4xMSAA (I believe R580 does at 4xMSAA)
* Xenos has a direct link to the CPUs L2 cache and the CPU has custom instructions for datastreaming
* General differences: R580 is much, much bigger; R580 has a much higher peak ALU performance; R580 operates at 650MHz compared to 500MHz for Xenos

The X1900 is a better chip for the desktop space; likewise I think Xenos has a better balance of performance for the console space and has a lot more features that will be important in the future once DX10 kicks into gear or are valuable for performance in a closed platform (e.g. FP10). They both share a number of common features, some under utilized today (dynamic branching, a lot of math per pixel, etc).

Anyhow, I don't think we can call Xenos an X1900 with some tweaks of pluses/minuses. ATI obviously did not dumb all their research when creating Xenos, but it definately is a significant departure from the R520 and R420 series of GPUs. The workflow and featureset is substantially different.

AFAIK
Xenos is a X1900



(+) plus

10 MB EDRAM (32GB/s connection to GPU, internal 256GB/sec)
free AA upto 4x (???)
10-bit HDR and AA
48 ALU/Unified shaders
some possible shader capabilities not available to the X1900
UMA negating any DATA copying/replication


(-) minus

8 ROPS
128 bit bus
22.4 GB/s unified bandwidth to be shared with Xenos
no vertex shaders (opposed to 8 vertex shaders in addition to the 48 fragment pixel shaders(2 ALU+1BEU each fragment pixel shaders) that is available to the X1900)
simpler ALU configuration as opposed to the X1900
some shader abilities on X1900 not available to Xenos



Still no sign of any technical article to properly determine RSX aside from it being a 7800GTX minus 8 ROPS, 128-bit bus and possible quads being disabled for redundancy...

please feel free to correct me if I am wrong

This helps dummies like me put it into perspective at least...thanks

 

[Hardknock]

Didn't an underclocked Xenos run that r520 Ruby demo at 30fps with no optimizations at last years E3? That should tell us something right?

 

[Megadrive1988]

AFAIK
Xenos is a X1900






please feel free to correct me if I am wrong

I know you said Xenos is X1900 but with all the pluses and minuses, but...

Xenos is not structured whatsoever like the X1900 (R580) or the X1800 (R520), as you know. completely different architectures. they're really not comparible.

the upcoming R600 is based heavily on the Xenos architecture, but most likely R600 will not have any EDRAM.



R300 ==> R350 => R360 =====> R420 => R480 ========> R520 ====> R580 ==> R590/R580+?

................R400 (withheld, unreleased) ================> Xenos/C1 ===========> R600=====> ????

that is not meant to show power/performance, just the family lines of ATI GPUs.

 

[LunchBox]

I know you said Xenos is X1900 but with all the pluses and minuses, but...

Xenos is not structured whatsoever like the X1900 (R580) or the X1800 (R520), as you know. completely different architectures. they're really not comparible.

the upcoming R600 is based heavily on the Xenos architecture, but most likely R600 will not have any EDRAM.



R300 ==> R350 => R360 =====> R420 => R480 ========> R520 ====> R580 ==> R590/R580+?

................R400 (withheld, unreleased) ================> Xenos/C1 ===========> R600=====> ????

that is not meant to show power/performance, just the family lines of ATI GPUs.

i totally agree.
however, the thread starter asked what was the closest possible PC counterpart...
Eventhough, there isn't an exact architectural comparison that can be made...
I opted to compare it to the X1900....

 

[swaaye]

Yeah I think X1900 is a good current comparison. In reality though, like has been said, it is more in between X1900 (R580) and R600. 360's GPU is really way ahead of its time. Personally I find it a hell of a lot more interesting than what's going into PS3.

The straw that is going to break the PS3's back is this interview over at Ars. Sure it's from a 360 hardware guy but he's very candid. I am amazed he posted in the forum at all. Usually you don't see that. Hopefully he doesn't get himself fired lol.

PS3 is very underwhelming IMO, especially the CPU.

 

[Robert.L]

Yeah I think X1900 is a good current comparison. In reality though, like has been said, it is more in between X1900 (R580) and R600. 360's GPU is really way ahead of its time. Personally I find it a hell of a lot more interesting than what's going into PS3.

The straw that is going to break the PS3's back is this interview over at Ars. Sure it's from a 360 hardware guy but he's very candid. I am amazed he posted in the forum at all. Usually you don't see that. Hopefully he doesn't get himself fired lol.

PS3 is very underwhelming IMO, especially the CPU.

How so ? i only see FUD and misinformation from that guy about ps3.

the guy should act more like nAo , for a dev who's close to Sony he certainly seems unbiased , both pointing out the pro's and the con's of xenos for example.

 

[pjbliverpool]

Yeah I think X1900 is a good current comparison. In reality though, like has been said, it is more in between X1900 (R580) and R600. 360's GPU is really way ahead of its time. Personally I find it a hell of a lot more interesting than what's going into PS3.

The straw that is going to break the PS3's back is this interview over at Ars. Sure it's from a 360 hardware guy but he's very candid. I am amazed he posted in the forum at all. Usually you don't see that. Hopefully he doesn't get himself fired lol.

PS3 is very underwhelming IMO, especially the CPU.

It may be between R580 and R600 architecturally and feature wise but the power of Xenos compared to R580 does not fall in line with that assumption.

 

[swaaye]

How so ? i only see FUD and misinformation from that guy about ps3.

That's not what I read.

It felt to me that he has a "get the job done best" mentality and said that Xenon and Cell are similar in theoretical performance, but Xenon is far easier to work with (practical performance) and the dev tools are more developed as well. The guy sounds very educated and down to earth to me. Cell is literally prototype tech, while Xenon is a more proven design, a more understood design. It is no surprise, and I think it is expected, that Cell turn out to be hardly optimal in comparison.

He certainly did not spew shockingly biased FUD lol. He even blatantly talked about some of the difficulties with 360. Xenon isn't exactly a super-easy-to-work-with CPU either. But it's better than Cell. IMO 360 sounds like a tighter system in general, from reading the various technical overviews out there. My fav is probably Ars Technica's. Hannibal knows his stuff.

To be honest, I don't think we've heard anything really good about Cell yet at all. At least I haven't. No stunning games. And there is just a constant flow of bad news about PS3 it seems. I don't own a 360 btw.

It may be between R580 and R600 architecturally and feature wise but the power of Xenos compared to R580 does not fall in line with that assumption.

Well I've heard in a few places that R580 is faster. But it's obviously not custom-designed for a cosnole price or environment. Xenos seems to be amazingly well thought out for what it needs to do, IMO.

R600 is going to be well ahead of either console I think, in both capability and speed. Of course, it will also cost as much or more than those entire machines lol.

 

[Huh?]

How so ? i only see FUD and misinformation from that guy about ps3.

Same here, he goes on about procedural rendering and how it's really new.
PS1/PS2 ring a bell?

Then talks about XNA, which has nothing to do with the Xbox360's hardware and doesn't appear to be any better than what Criterion and other middleware makers already have.

Just sounds like PR hype talk to me.

 

[Mmmkay]

That's not what I read.

There have been a couple of B3D threads on Matt Lee's ars article, perhaps you should read them too?
http://www.beyond3d.com/forum/showthread.php?t=31446
http://www.beyond3d.com/forum/showthread.php?t=31431

His disingenuous factual reporting (cache comparison), erroneous analysis (memory pool comments), sweeping statements (done the numbers), and one sided arguments (poor branch prediction being only a Cell issue) are pretty much the same mantra that Microsoft has been chanting ever since Major Nelson performed his 'analysis' after E3 2005. At the end of the day, these are all very old topics which were brought up in his answer and a search should yield plenty of threads on the subjects if you wish to read further.

And it was very interesting to note that while he indeed was posting in the Ars forum, he did not feel it necessary to elaborate on his statements nor respond to any of the criticisms provided by several posters.

 

[scooby_dooby]

Major Nelson's comparison was given to him by MS's 'engineers', I would not be surprised at all if Matt was one of the people who originally drew up that document.

Remember though, that was created in response to a huge negative backlash 360 was recieving based on almost complete hype(2TFLOP twice the PoWah!!) from sony that the entire media bought into.

 

[swaaye]

And it was very interesting to note that while he indeed was posting in the Ars forum, he did not feel it necessary to elaborate on his statements nor respond to any of the criticisms provided by several posters.

Hey, I'm blown away that he showed up in the forum at all. The devs here are almost all anon, and for good reason. He had to measure his responses or all hell could (and would) break loose. PR is dangerous business and his responses reflect everyone at MS in the eyes of the readers. What would you do?

Of course 360's CPU isn't all that in the branch prediction area. Everyone who cares at all about the tech behind this stuff should know that by now. But, Cell is plainly worse off. Sure it has ultra-super potential speed, "potentially" (<- let's stress that) well in excess of Xenon, but it's looking to be somewhat impossible to actually "harness" that speed in any real application where the code will never, ever, in-a-billion-years, line up with the optimal set of conditions. These CPUs suck for general purpose processing. We went from nice OOE processors in Xbox and Cube to a step backward to make things "cheaper yet maybe way faster". Except now they're not single-core in-order CPUs, devs gotta make several cores run fast with tough programming on a new dimension. It's like Sega Saturn is back lol.

One of my friends has a 360. I have not been impressed by some of the chop in the games out right now. And I think it partly points to that CPU being a problem.

Wii is the only console with a CPU that's easy to work with. That's assuming it's a derivative of what's in 'Cube, of course, and not another multi-headed mosquito-on-speed (HA!).

 

[macabre]

It`s still early. The games don`t use the 360s full potential yet and the PS3 games are months away from launch.

 

[chachi]

These CPUs suck for general purpose processing. We went from nice OOE processors in Xbox and Cube to a step backward to make things "cheaper yet maybe way faster".

The OOE in the new IBM core (Xenon/PPE) is probably fairly similar to that in the Cube which is really rudimentary. The Cube CPU had a really small pipeline though (if it's similar to the 750 which I think is the case) so it's not as big a deal.

What is on the general purpose IBM core though is much better than what is on the SPU which doesn't have any branch predication at all and relies on programmers to hint branches. If you're hand tuning everything then it isn't a problem but the point of stuff like this being implemented in the first place was to make code run fast regardless of compilers or talent and time spent optimizing things. You can be sure that for the next generation of Cell they'll have that stuff added back in (along with better DP FP, etc.) because it's good to have, they just didn't have the transistor budget to do it this time around.

 

[Megadrive1988]

I like this quote by Matt Lee

http://arstechnica.com/articles/culture/mattlee.ars/4

If I handed our game developers a new console tomorrow that had one hundred times the capabilities of Xbox 360, they would figure out a way to use it immediately.

hmmm, well by some measurements, the Xenon CPU has 76x the floating point performance of the Xbox1 CPU. also, IIRC, some have said that Xenos has 100x the shader performance of NV2A. I'm not saying Xbox 360 has a hundred times the capability of Xbox, or even close to it. but, I'd expect a bigger leap with the next Xbox after the '360, and developers will just GOBBLE up that without batting an eye, especaially in terms of the GPU.

 

[sonyps35]

If I handed our game developers a new console tomorrow that had one hundred times the capabilities of Xbox 360, they would figure out a way to use it immediately.

I really liked that qoute too. I tend to think in general that, power, graphics, doesn't always get it's due. Especially with the Wii and all that creating a lot of discussion. It's almost like you have to apologize for good graphics these days. Then you also get people constantly saying things that graphics are maxed out, power is maxed out, diminishing returns, etc etc.

some have said that Xenos has 100x the shader performance of NV2A.

Hmm, well I'm a technical illiterate but I dont think it's nearly that much. If you compare PS3 to Xbox you have somewhat similar GPU's (not the case with Xenos/Xbox so it's a better comparison). In Xbox you have 4 pixel pipes at 233mhz, in RSX 24 pipes at 550 mhz. So discounting all the support structures, memory, etc, you have 6 times the pixel pipes, and then ~2x the clock, so RSX~12x NV2A.

This compares well with a note in the Takahashi book that EA tested the 360 to make sure it was up to their next gen standards I guess, and came back in the affirmative that it was around 10X as powerful as Xbox 1.

 

[Megadrive1988]

like i was trying to say, Xenos/X360 does not have nearly 100 times the raw performance of NV2A/Xbox. overall, it's more like 3-5 times. (main memory bandwidth, fillrate, polys/sec)

but in pure pixel shader performance, Xenos is a huge improvement over NV2A, from what ive read. whether that is 20, 50 or 100 times is open to debate.

. If you compare PS3 to Xbox you have somewhat similar GPU's (not the case with Xenos/Xbox so it's a better comparison). In Xbox you have 4 pixel pipes at 233mhz, in RSX 24 pipes at 550 mhz. So discounting all the support structures, memory, etc, you have 6 times the pixel pipes, and then ~2x the clock, so RSX~12x NV2A.

no, NV2A has 4 pixel pipes, the RSX has 8 pixel pipes. so RSX has twice the pixel pipelines of NV2A, not 6 times.
NV2A has, i think 4 or 8 pixel shader ALUs/units (1 or 2 per pixel pipeline, correct me NV2A experts if im wrong)
the RSX has 24 pixel shader pipes.
you're confusing pixel pipelines with pixel shader units.

 

[pjbliverpool]

Anyone know what the NV2a pixel shader capabilities where? It should be fairly easy to make a rough comparison once we have that information.