How can Nvidia be ahead of ATI but 360 GPU is on par with RSX?

[mckmas8808]

Just wondering if the RSX was just a mere modified variant that runs on Flex I/O and works with the Cell why is Nvidia taking so long to have it taped out?The Xenos without the edram has significantly less transistors than the G70 yet it easily outperforms it in a console.Nvidia's RSX is also made to work in a console environment and yet we are comparing it with the G70 and saying that it is similar by just looking at its transistors and setup?

Hugo those questions are not allowed here. We don't want to talk about the truth when it comes to RSX. Don't come here with only 80 something post and think that smart post will be answered [/sarcasm]

It's still amazing to me that without knowing much about the RSX that people can make for sure comments. The CELL vs. XeCPU is different. We have pages and pages worth of information. We don't have the samething with the RSX. When they release info and the RSX sucks in comparison then yes go ahead and say it, but can we just wait until that time happens. I mean 3 years and over 50 people worth of work from Nvidia and all the RSX will be is a G70 with Flexio to CELL? I don't believe it.

 

[Acert93]

Just wondering if the RSX was just a mere modified variant that runs on Flex I/O and works with the Cell why is Nvidia taking so long to have it taped out?

We do not know the full scale of adaptation of the G70 for PS3. It could be what you outline, more, or less. We don't know.

The Xenos without the edram has significantly less transistors than the G70 yet it easily outperforms it in a console.

Only on paper and in theoretical performance. And as for the console part, this is not a PC. The games will take everything into consideration and balance/design with each part in mind. Its not a PC where you can just change out a chip and get a relative performance within that game. Console games are not always designed that way, and even cross platform games get a bit of tweaking.

Nvidia's RSX is also made to work in a console environment and yet we are comparing it with the G70 and saying that it is similar by just looking at its transistors and setup?

The "similar" deductions are from NV's press release (adaptation of G70), transistor count (both at ~300M), and the RSX performance numbers are nearly identical to a G70 @ 550MHz.

Of course there could be changes, significant changes at that! We don't know. Yes, RSX is being made to work in a console environment, but what does that mean? It obviously is not a clean slate custom design like Xenos. There are obvious Pros to a PC part (like fillrate, 2x HDMI) but also areas where console centric design is helpful (e.g. eDRAM which has a lot of benefits but is not practical on a PC).

How all these play out and actually impact development we do not know. Expect pros and cons on both sides with a lot of blurry issues inbetween that are irrelevant, because as noted before, they are in a console. Some things will impact game IQ, some performance, and others are technicalities.

In the end the BEST devs will consider those things, plan accordingly, and then produce KILLER product (regardless of any pros and cons).

 

[j^aws]

...This of course, entirely leads into the argument of workload type eficiency...

Yep. And if you want to extend that route, then it leads you onto how the CPU effects things. Ultimately you have to look at the 'whole' architecture and not just isolate the GPUs.

...
RSX will have a fixed VS/PS ratio, meaning that to hit peak efficiency between the units the workload has to be distributed according to that ratio; if we have a VSS complexity of 1:5 in the game then thats not going to benefit RSX, if we have 5:1 thats also not going to benefit RSX....

.

And to follow on with the workload type efficiency, i.e. load balancing, this is also where the SPUs in CELL can have a load balancing effect with RSX. The interaction with CELL is where RSX differs from G70 in the PC arena. Furthermore, the varying clock domains that are present in the G70 will have a smoothing/ load balancing effect too...

Vast is the right word when you're talking about 100%+ TMU utilisation gains....

So your definition of 'vast' now is 100%+ TMU utilisation gains? Care to explain where you are digging up the 100%+ gains number from?

In RSX or any other conventional GPU, the TMU pipelines are idle when the shader for the current batch of pixels doesn't request texturing.

Xenos will find a different batch of pixels that requires texturing even though the "current batch" (there could be 3 batches current in Xenos) doesn't (in fact, can't) be texturing. The scheduler juggles batches in order to ensure that pipelines are never empty.

This counts for both ALU pipelines and TMU pipelines.

Yes, I'm aware of decoupled TMUs in Xenos and coupled TMUs in G70. That still doesn't explain where you're getting your 100%+ numbers from...

If we only look at fragment pipes for G70/RSX and the equivalent 'peak' for Xenos, i.e. 3 SIMD engines (48 ALUs) and 16 filtered texture units,

Xenos (peak fragments only),

-96 instructions/cycle + 16 texture inst./cycle ~ 112 inst.cycle
~ 112*0.5GHz ~ 56 Billion inst./sec

Or

48 Billion shader inst./sec + 8 Billion filtered texture inst./sec


RSX (24 fragments pipes only),

- 120 instructions/cycle (includes 24 filtered texture inst./cycle)
~120 * 0.55 GHz ~ 66 Billion inst./sec

Or

53 Billion shader inst./sec + 13 Billion filtered texture inst./sec


Key difference with Xenos ~ 56 Ginst./sec is that 8 Ginst./sec are always filtered texture instructions but for RSX ~ 66 Ginst./sec, the instructions are scheduled differently.

Each G70/RSX fragment pipeline can issue 5 inst./cycle. The vec4 ALUs also work on texture data with the texture ALUs. Now when I asked about the distribution of these instructions in this thread,

http://www.beyond3d.com/forum/viewtopic.php?t=24933

...no one could give me a clear answer on how these instructions are distributed...so we're covering old ground again...

Eh? Of course it can. So can RSX. It's a question of viability. 1000-pixel batches are too coarse-grained.

Err...no. It still doesn't change the fact that it doesn't do per pixel dynamic branching. Referring to different batch sizes doesn't change the fact that technically it still doesn't do this.

 

[Jawed]

Jaws, your obsession with meaningless peak statistics and your inability to understand predication and the effect of batch-sizing on efficiency all lead me to one simple conclusion.

Jawed

 

[j^aws]

Jaws, your obsession with meaningless peak statistics and your inability to understand predication and the effect of batch-sizing on efficiency all lead me to one simple conclusion.

Jawed

I don't make hyperbole claims and post random unsubstantiated numbers.

Laters...

 

[hugo]

Hugo those questions are not allowed here. We don't want to talk about the truth when it comes to RSX. Don't come here with only 80 something post and think that smart post will be answered [/sarcasm]

It's still amazing to me that without knowing much about the RSX that people can make for sure comments. The CELL vs. XeCPU is different. We have pages and pages worth of information. We don't have the samething with the RSX. When they release info and the RSX sucks in comparison then yes go ahead and say it, but can we just wait until that time happens. I mean 3 years and over 50 people worth of work from Nvidia and all the RSX will be is a G70 with Flexio to CELL? I don't believe it.

Heh my 80 something post not worthy?Perhaps my most favourite phrases in this thread were the RSX is a modified G70 and the Xenos is at least 2 generations ahead.Me like those phrase.

To all Xenos fans including myself(actually I am a fan of cartoonish/anime games) [/big sarcasm] don't let Nvidia fool you into thinking that the RSX is a G70.It's part of their strategy to let you feel comfortable.

 

[London Geezer]

Jaws, your obsession with meaningless peak statistics and your inability to understand predication and the effect of batch-sizing on efficiency all lead me to one simple conclusion.

Jawed

Sorry to sound stupid.. But after all these , i'm still not sure what branch prediction actually does...? And how it makes things better basically...

 

[Shifty Geezer]

The "similar" deductions are from NV's press release (adaptation of G70), transistor count (both at ~300M), and the RSX performance numbers are nearly identical to a G70 @ 550MHz.

Of course there could be changes, significant changes at that! We don't know. Yes, RSX is being made to work in a console environment, but what does that mean?

The only obvious named extra is cache coherency with Cell and two-way local-bus communications. Obviously there's an idea to share data but how and to what effect is up in the air. There's GOT to be more to RSX then just FlexIO though. Unless nVidia engineers have set a new world record for having tea breaks, 50 man-years minimum input on RSX seems to have accomplished very little if all they can come up with is a FlexIO interface

 

[Acert93]

Higher clock, virtually 'free' AA+HDR, comparable transistor count, and unified + tons of goodies. It would seem there's something we're not seeing on the hardware or on the software side of the equation that's holding it back.

Take your pick:

• Beta Kits released in Aug. Less than a month ago and 2 months before going "Gold". Up until this point Devs have been working on X800s and 9800s. These games have not been built with the benefit of actual hardware to test, optimize, and specialize to.
• Beta Kit delays and screwups. They were late, were underclocked to begin with, and even final kits had messups (e.g. 2.8GHz CPUs instead of 3.2GHz)
• Many titles are PC/Xbox ports. Kameo (GCN-Xbox), PDZ (Xbox), Gun (Xbox), Quake 4 (PC), Prey (PC), NFS (PC, Xbox), Call of Duty 2 (PC), FFXI (PS2, PC), Tony Hawk (Xbox), etc.
• Devs are in the chrunch. We have seen with Dead Rising how much can change in 3 months (E3). They don't have time to demo their products AND try to release the game, on time and complete, in November. PS3 devs do have a bit more time.
• We are seeing real gameplay. There is a difference between a cut scene with ingame assets in the engine versus a HUD/gameplay view.
• Devs are struggling with the new hardware. Given the short amount of time, asking them to move to 3 symmetric cores and a totally new GPU--and expect ground breaking games--is a lot to ask. More talented/equiped devs like Bizzare have done well on this, others have not.

As for how much better PS3 demos look, the strategy from Sony is different. They have fewer announced games in development (102 vs 160) and the ones they have shown have been very selective. The number of titles, and the amount of media, is quite different. I cannot remember seeing a single port OR cross platform title, outside Sonic, presented by Sony for the PS3. Those titles DO exist though, as I am sure the PS3 will have its share of UGLY games. There is a very smart and calculated strategy to not demo those products. 1st being PS2 2005 fall sales; 2nd being the "Shock and Awe" factor. Impressions of a product are PRICELSS. It is better to give snapshots--Killzone, MotorStorm, MGS4, etc--that blow people away then to mix in HIGH quality products with LOW quality products.

Sony is playing thigns close to their chest and EFFECTIVELLY swaying public opinion. Another factor is PS3 devs, the lucky ones who have had dev kits for a while, are much closer to the PS3's technical abilities compared to the Xbox360 dev kits in regards to GPU performance.

That said, of the games actually demonstrating actual gameplay on a real engine, the 360 has done ok. PGR3, Oblivion, Gears of War, etc... They all look good. The Ruby Assassin demo demonstrates that technically if something is designed with its performance abilities in mind it can excell.

Another import thing to consider is cross platforming. We already know games like Madden Live, Ridge Racer, Sonic, RE5, etc will be on all the platforms. Obviously different game designs will benefit certain architectures, that said seeing a dev build a game on two platforms can often give an idea of performance. Right now we are not seeing that.

Like some of us have been saying it is only fair to really compare finished products, and even then genre, art style and direction, and technology need to be separated in such comparisons.

Further, we wont be seeing games that exploit the PS3 or Xbox360 until ~2007.

I guess in the end the question will be: How many great looking Xbox 360 titles will be on the market in fall 2006 and how many great looking PS3 titles will be on the market at the same time.

Obviously MS has a lot of ugly games in the pipe and has shown them. They have dropped the ball on the hype a lot. This has hurt them, but it could hurt Sony if they do not live up to the expectations they set.

 

[London Geezer]

The only obvious named extra is cache coherency with Cell and two-way local-bus communications. Obviously there's an idea to share data but how and to what effect is up in the air. There's GOT to be more to RSX then just FlexIO though. Unless nVidia engineers have set a new world record for having tea breaks, 50 man-years minimum input on RSX seems to have accomplished very little if all they can come up with is a FlexIO interface

I agree. They're not british afterall!

 

[one]

[*]Beta Kits released in Aug. Less than a month ago and 2 months before going "Gold". Up until this point Devs have been working on X800s and 9800s. These games have not been built with the benefit of actual hardware to test, optimize, and specialize to.

Beta kits with lower-clocked Xenos and 3-core CPU were shipped at the end of June, while
final kits were shipped a month ago. Also it seems key partners got them prior to E3.
http://www.gamesindustry.biz/news.php?aid=9914

Prior to E3, developers reported that a very small number of these more advanced kits had been manufactured, but it's only in recent weeks that they have begun shipping to Microsoft's development partners on a wide scale.

 

[hugo]

The only obvious named extra is cache coherency with Cell and two-way local-bus communications. Obviously there's an idea to share data but how and to what effect is up in the air. There's GOT to be more to RSX then just FlexIO though. Unless nVidia engineers have set a new world record for having tea breaks, 50 man-years minimum input on RSX seems to have accomplished very little if all they can come up with is a FlexIO interface

Well probably Sony paid them peanuts so they had to take more time offs and consider a 2nd job?[sarcasm]
If Nvidia would announced that the RSX is actually a G70 remodified,it would be considered one of the major blunders/weakness for this PS3 gen.Say byebye to Nvidia for any future collaborations,no more partnership with you cause Ken-san said so.Better not to have a GPU than to have.I would just buy middlewares and development tools from Nvidia if I were Sony.Why continue to pay royalties and extra research funding when I can buy off the shelf products patents to remodel them for my own use?

 

[Alpha_Spartan]

You can't blame Nvidia because RSX being a modified G70 will prove that the RSX was an 11th hour move by Sony when the Cell thing didn't pan out.

 

[mckmas8808]

comparisons.

Further, we wont be seeing games that exploit the PS3 or Xbox360 until ~2007.


but it could hurt Sony if they do not live up to the expectations they set.

I agree with about everything that you said Acert expect these two items. The PS3 will be exploited in 2006. Did you forget that MGS4 is targeted to come out next year? And when that happens Sony would have lived up to their expectations right?

 

[London Geezer]

I agree with about everything that you said Acert expect these two items. The PS3 will be exploited in 2006. Did you forget that MGS4 is targeted to come out next year? And when that happens Sony would have lived up to their expectations right?

Are you actually expecting that developers will be able to get amazing performance out of either Cell or Xenon in the first year? Not gonna happen.

 

[mckmas8808]

You can't blame Nvidia because RSX being a modified G70 will prove that the RSX was an 11th hour move by Sony when the Cell thing didn't pan out.

OMG!! Please don't ever, ever, ever say that again here. B3D is way to smart to actually believe that. We know that when tapped out the RSX would have been in development for 3 years!! Stop with the TXB put down Sony hype. That buck stops here.

 

[hugo]

You can't blame Nvidia because RSX being a modified G70 will prove that the RSX was an 11th hour move by Sony when the Cell thing didn't pan out.

An 11th hour move it is but the idea of buying off the shelf product doens't fit into Ken's vision.He does not believe in outsourcing development/manufacturing cost and would prefer most things done in-house.

 

[Titanio]

Heh my 80 something post not worthy?Perhaps my most favourite phrases in this thread were the RSX is a modified G70 and the Xenos is at least 2 generations ahead.Me like those phrase.

This is why unqualified statements are so dangerous in discussions like these, and should be frowned upon. There are many people reading these discussions simply looking to read what they want to read, and who want things to be nice and simplified, with no regard for the actual subtleties and the "grey" nature of the debate (i.e. things are not black and white). An unfortunately, there seems to be otherwise "respectable" board members who are more than happy to provide.

Unless you're being sarcastic.

Also, Acert, the list of 102 titles in dev for PS3 are for the japanese market only. There are games we know are in dev for PS3 that aren't on that list. Not that it makes much difference, but 102 shouldn't be thrown out there as THE number announced as in development.

Anyway, seems this discussion has brought us nowhere we haven't been before to be honest. We've no new information, really, on Xenos or RSX that we hadn't had before. We're raking over old territory, and its the same people making the same arguments roughly speaking, as far as i can see. Are we going to do this every time someone starts a new thread on RSX or Xenos? Pointing out the search function may be more useful and save everyone a lot more time.

 

[Alpha_Spartan]

I agree with about everything that you said Acert expect these two items. The PS3 will be exploited in 2006. Did you forget that MGS4 is targeted to come out next year? And when that happens Sony would have lived up to their expectations right?

#1. MGS4 won't come out next year since these big games are rarely released on time.

#2. There was nothing revolutionary graphically about MGS4. The textures were weak. You won't see that level of animation in game.

 

[mckmas8808]

Are you actually expecting that developers will be able to get amazing performance out of either Cell or Xenon in the first year? Not gonna happen.

Well it depends if you think MGS4 is amazing performance. If you do then the answer to that question will be yes. If you don't think MGS4 is considered amazing performance then the answer to your question will be no.