Shogmaster
Member
No I hear ya, gofreak. I hear ya.
I can't wait for Dave B's Xenos article either. :lol
I can't wait for Dave B's Xenos article either. :lol
Shogmaster said:I can understand certain claims are a bit much to digest at once, but overall sense I get is that ATI has everything to prove, and nVidia has none to prove. Tell me if I'm off on that.
Anyways, seeing how Xenos has 330M trannies (250M for main shader unit + 20M for the AA, Stencil, Z sort ROP on the daughter and 80M for 10MB of EDRAM minus the ROP on the daughter die), and RSX has 300M (260M for next gen part and 40M for GS perhaps?), It's hard to swallow RSX having 50% more rendering power than Xenos BS Sony and nVidia is throwing around, especially looking at the efficiency built into Xenos.
Shogmaster said:.
Anyways, seeing how Xenos has 330M trannies (250M for main shader unit + 20M for the AA, Stencil, Z sort ROP on the daughter and 80M for 10MB of EDRAM minus the ROP on the daughter die), and RSX has 300M (260M for next gen part and 40M for GS perhaps?), It's hard to swallow RSX having 50% more rendering power than Xenos BS Sony and nVidia is throwing around, especially looking at the efficiency built into Xenos.
HyperionX said:Xenos doesn't have 330M transistors from what I've heard. In fact the GPU unit itself is only 150M not including the eDRAM + AA unit.
It depends on your definition of "better visuals". The xgpu can put out a 4X multisample AA hidef image with little hit on performance. It remains to be seen if the RSX can do the same.Drek said:Hmm, I'm personally expecting more of the reverse, the CPU being used to supplement the GPU, at least from the PS3. With the 7 SPEs and the fast bi-directional bandwidth Sony can effectively cheat to get noticably better visuals than the X360. They'll have to sacrafice general computing and non-graphics in game operations like physics and AI, but they'll still be miles ahead of last generation. Smart move by Sony if that is their plan, since the average consumer equates graphics with overall system power, and also is much less capable of noticing differences in physics, AI, and other general computing functions.
gofreak said:Apparently it's 232m for Xenos itself and 100m for the eDram unit. 20m of the eDram unit's transistors are likely computational logic for AA etc (10MB of eDram likely takes 80m transistors), so non-eDram logic total would be ~250m.
dorio said:It depends on your definition of "better visuals". The xgpu can put out a 4X multisample AA hidef image with little hit on performance.
gofreak said:It's natural that a more critical eye will be cast on ATi's offering, since they're the ones making bolder claims, architecturally anyway.
aaaaa0 said:I would say the same about CELL. But there is a distinct lack of scepticism from certain parties at various message boards (not to point any fingers).
I agreegofreak said:Cell has come under massive analysis etc. for years now. And there's been plenty of nay-saying, alongside the yay-saying. Plenty of questions asked, it came under very close and critical scrutiny. Things have quietened down since debate has gone as far as it can, really, from what we have on it, but some fresh points of argument may re-emerge when IBM opens up the hardware to everyone sometime in the summer. Although we do really have an awful lot of info on it as is, so..
All that said, Xenos has its "groupies" too. How critically or sceptically you think something is being received may be a matter of perception..reminds me of that thread bemoaning GAF's "sony bias". If you're sensitive to something, you'll pick up on it in an unbalanced fashion, and it'll seem like it's everywhere.
midnightguy said:whoa hang on there. I havent heard this before.
official info is: Xenos has 232 million transistors for the GPU core, and 100 million for the eDRAM unit which contains the eDRAM plus some logic, including 192 little "processors" for AA.
RSX: has 300 million or slightly over 300 million transistors, no eDRAM, and no mention of an on-board Graphics Synthesizer.
Shogmaster said:Well, it's my crazy specualtion, but where else would it go? It's gotta go somewhere. And 40ish M trannies is little too big to shove into anything else.....
Shader trannies were not traded for edram trannies. They don't even sit on the same card. It's not like ATI would have doubled the pipes if they excluded the edram.gofreak said:It depends what you mean by hit on performance. That AA isn't free..it may not hit the performance attainable by Xenos much, but that performance might have been higher if the transistors spent on eDram were spent elsewhere (i.e. on shaders). So yeah, little hit on the performance the chip is giving you, but a hit relative to chip that might have been if those trannies were spent on shading power? Yes.
Will the g70 have this 128-bit HDR rendering?I don't think we can really say that when we've no idea what NVidia's next-gen shaders look like, for example. Or how much 128-bit HDR etc. would cost in terms of trannies. It's perfectly plausible than RSX is spending more logic, it's not bound to track the amounts spent in Xenos's shader core.
Pimpwerx said:But I'm also a bit of a dreamer. PEACE.
dorio said:Shader trannies were not traded for edram trannies. They don't even sit on the same card. It's not like ATI would have doubled the pipes if they excluded the edram.
dorio said:Will the g70 have this 128-bit HDR rendering?
That's all subjective. Personally I think AA quality is the biggest difference between realtime graphics and prerendered ie. Toy Story, Shrek etc.Pimpwerx said:RSX has like 30% more trannies dedicated to shaders, and a 10% faster clock speed. If NVidia got it running fp16 or even fp32 HDR at useable resolutions, then I can't see why it won't come out ahead in the GPU race. The AA may end up stuck at 2xMSAA to be free, but that's not a huge deal I don't think. I don't believe the swing between 2x and 4x would be significant enough to make up for potentially big effects advantages. I say this again based on the fact that Sony seems to have placed emphasis on HDR and SSS in their demos. I think a recurring theme was that they had devs shoot for the moon on lighting. The VS power of Cell combined with possibly deep shader pipes on RSX coule make the PS3 graphics advantage significant. But I'm also a bit of a dreamer. PEACE.
dorio said:That's all subjective. Personally I think AA quality is the biggest difference between realtime graphics and prerendered ie. Toy Story, Shrek etc.
Kleegamefan said:Another question...
Did nVidia say for sure RSX would have vertex processing units??
I know there was alot of talk at first of the nVidia GPU having nothing but Pixel Shader units and ROPs, with CELL feeding it verticies...has that been debunked? (just curious)
If, however, RSX has, say, 8 vertex units, they could also recieve vertex assist from SPEs too, no??
Just wanting to know if I am on the right page here....
Kleegamefan said:Another question...
Did nVidia say for sure RSX would have vertex processing units??
I know there was alot of talk at first of the nVidia GPU having nothing but Pixel Shader units and ROPs, with CELL feeding it verticies...has that been debunked? (just curious)
If, however, RSX has, say, 8 vertex units, they could also recieve vertex assist from SPEs too, no??
Just wanting to know if I am on the right page here....
What leads you to believe RSX will not have good AA quality??
Kleegamefan said:Another question...
Did nVidia say for sure RSX would have vertex processing units??
I know there was alot of talk at first of the nVidia GPU having nothing but Pixel Shader units and ROPs, with CELL feeding it verticies...has that been debunked? (just curious)
If, however, RSX has, say, 8 vertex units, they could also recieve vertex assist from SPEs too, no??
Just wanting to know if I am on the right page here....
What leads you to believe RSX will not have good AA quality??
Kleegamefan said:Another question...
Did nVidia say for sure RSX would have vertex processing units??
Kleegamefan said:If, however, RSX has, say, 8 vertex units, they could also recieve vertex assist from SPEs too, no??
David Kirk: SPE and RSX can work together. SPE can preprocess graphics data in the main memory or postprocess rendering results sent from RSX.
Nishikawa's speculation: for example, when you have to create a lake scene by multi-pass rendering with plural render targets, SPE can render a reflection map while RSX does other things. Since a reflection map requires less precision it's not much of overhead even though you have to load related data in both the main RAM and VRAM. It works like SLI by SPE and RSX.
David Kirk: Post-effects such as motion blur, simulation for depth of field, bloom effect in HDR rendering, can be done by SPE processing RSX-rendered results.
Nishikawa's speculation: RSX renders a scene in the main RAM then SPEs add effects to frames in it. Or, you can synthesize SPE-created frames with an RSX-rendered frame.
David Kirk: Let SPEs do vertex-processing then let RSX render it.
Nishikawa's speculation: You can implement a collision-aware tesselator and dynamic LOD by SPE.
David Kirk: SPE and GPU work together, which allows physics simulation to interact with graphics.
Nishikawa's speculation: For expression of water wavelets, a normal map can be generated by pulse physics simulation with a height map texture. This job is done in SPE and RSX in parallel
I don't think its that simple. It's hard to tell since the xephos architecture is so different than current cards but if their claims of 32 pipe equivalance is correct then that performance is in line with what cards coming out at that time frame would have. Their goal though is to achieve the actual peak performance that these pr guys are throwing out there by moving bandwidth consuming functions to another processor and memory store. Now if the xephos had 16 pipes then I'd say yes you're probably right that they traded shading ability to include the edram.gofreak said:If you want to boil it all down to the very basics, money, if MS wasn't spending dollars on a seperate eDram module, they could be spending it on more silicon for the main GPU, Xenos. There certainly is a tradeoff there. It would seem so, yes.
My point was that if the RSX doesn't have edram then that good AA quality will come with a big performance hit if its like current gen cards.Kleegamefan said:What leads you to believe RSX will not have good AA quality??
It's a bit more then that - it's going into the silly realm. People keep talking about how little modification NVidia had time to do and here you'd go and stuff a whole extra chip complete with its external interfaces onto RSX.ShogMaster said:Well, it's my crazy specualtion, but where else would it go?
dorio said:I don't think its that simple. It's hard to tell since the xephos architecture is so different than current cards but if their claims of 32 pipe equivalance is correct then that performance is in line with what cards coming out at that time frame would have. Their goal though is to achieve the actual peak performance that these pr guys are throwing out there by moving bandwidth consuming functions to another processor and memory store. Now if the xephos had 16 pipes then I'd say yes you're probably right that they traded shading ability to include the edram.
dorio said:My point was that if the RSX doesn't have edram then that good AA quality will come with a big performance hit if its like current gen cards.
Fafalada said:It's a bit more then that - it's going into the silly realm. People keep talking about how little modification NVidia had time to do and here you'd go and stuff a whole extra chip complete with its external interfaces onto RSX.
GS on 90nm is just over ~40mm2, it's a pretty tiny chip compared to the rest of the stuff in there. Though personally I don't like the idea of using GS in there - it's a chip that has completely no use in the system other then playing PS2 games.
gofreak said:It should be really quite clear that if they weren't spending money on eDram it could have been spent elsewhere. When you look at the transistor budget of the Xenos total and compare it to PC chips coming out around the same time, or RSX even, that supports the notion of a tradeoff. The total transistor budgets are similar, but MS chose to spend theirs differently.
You can have an impressive level of AA on PC games right now, and it doesn't help them look anywhere near as good as quality CGI looks. As for the specs, 4x AA on R500 is basically free (2-5% hit), but that's not really even close to the level of CGI AA. Also, I think current Nvidia cards have 2xAA for free, and 4x with some hit, but I have no idea how much of a hit.That's all subjective. Personally I think AA quality is the biggest difference between realtime graphics and prerendered ie. Toy Story, Shrek etc.
I'm pretty sure the leaked specs said it will have 64bit HDR and blending, but those are just leaked specs, so who knows.Will the g70 have this 128-bit HDR rendering?
Thanks, hadn't read that from Dave. I think ~250 trannies is in line with the transistor counts of the ATI cards being released this year. Do you know the trannie count for the 520?gofreak said:According to Dave Baumman, the guy who made the "32-pipe" comparison was off base. Such comparison really can't be made without benchmarks on final hardware.
It should be really quite clear that if they weren't spending money on eDram it could have been spent elsewhere. When you look at the transistor budget of the Xenos total and compare it to PC chips coming out around the same time, or RSX even, that supports the notion of a tradeoff. The total transistor budgets are similar, but MS chose to spend theirs differently.
The hit really depends on what you're doing. But yes, relative to its own performance without 4xAA, of course there will be one, but relative to other chips..that's a trickier comparison..
Yeah, I know cgi has an insane amount of samples but my point is the more it has the closer it looks to cgi imo.You can have an impressive level of AA on PC games right now, and it doesn't help them look anywhere near as good as quality CGI looks. As for the specs, 4x AA on R500 is basically free (2-5% hit), but that's not really even close to the level of CGI AA. Also, I think current Nvidia cards have 2xAA for free, and 4x with some hit, but I have no idea how much of a hit.
That would involve redesigning GS for new memory interface - in other words you're no longer getting free compatibility lunch. It makes a lot less sense to me - the place where it's far more likely to be is over that 2.5GB/s north bridge connection (which is more then double the bandwith GS needs anyhow).The only logical place IMO would be in the RSX, where you can reuse the same data pathways that RSX would be using for graphics.
dorio said:Thanks, hadn't read that from Dave. I think ~250 trannies is in line with the transistor counts of the ATI cards being released this year. Do you know the trannie count for the 520?
Shogmaster said:Why limit things to only looking at transistor budget? Surely, the money Sony is putting into BR is money not going into more gaming related features/powers?
Fafalada said:That would involve redesigning GS for new memory interface - in other words you're no longer getting free compatibility lunch. It makes a lot less sense to me - the place where it's far more likely to be is over that 2.5GB/s north bridge connection (which is more then double the bandwith GS needs anyhow).
But if you're going through the trouble of redesigning your chips for backwards compatibility like you suggest, why not just modify RSX for compatibility with GS addressing modes instead.
Kleegamefan said:How soon are we from G70 intro?
Any week now, yes???
dorio said:It would be a shame that they would have to devote 40 million transistors for backwards compatiblity and that's all its able to do. Have they confirmed that bc would be a hardware solution.
Fafalada said:That would involve redesigning GS for new memory interface - in other words you're no longer getting free compatibility lunch. It makes a lot less sense to me - the place where it's far more likely to be is over that 2.5GB/s north bridge connection (which is more then double the bandwith GS needs anyhow).
But if you're going through the trouble of redesigning your chips for backwards compatibility like you suggest, why not just modify RSX for compatibility with GS addressing modes instead.
midnightguy said:having a Graphics Synthesizer in RSX, or even anywhere in PS3, means PS3 has eDRAM
Elios83 said:
he hints that some hardware solutions were required in order to max out compatibility due to the fact that some PS2 games do things with the hardware that are not theoretically possible.
I wonder what that means.Kleegamefan said:he hints that some hardware solutions were required in order to max out compatibility due to the fact that some PS2 games do things with the hardware that are not theoretically possible.
Well, to be fair, during normal rendering GS memory access patterns tend to be cache friendly(all the more as games got more optimized), I have no doubt whatsoever that RSX has more then enough bandwith to sustain such rendering and more.ourumov said:But then...the GS was enough fast to accomplish whatever speed the PSX's GPU achieved. Now there is a difference in bandwith that I don't see so easy to solve...
Fafalada said:The tricky part is when we get to the more exotic render-to-texture shenanigans and stuff
this part 3 is the most interesting part of this interview to me, as wellPimpwerx said:A lot has been made of the Xenos' abilities as a GPGPU. Looks like RSX was designed along the same lines too? This based on this machine translation. I hope the one can get us a proper translation today. This has been the most interesting part of the interview IMO. PEACE.