NVIDIA Fermi: Architecture discussion
- Thread starter Rys
- Start date
Since they most likely will name it GeForce 360 and not JBox360 I don't think Microsoft will have much to object
And how do you figure that ?
A single HD 5870 is on average 40-50% faster than a GTX 285. Even if I don't consider the obvious architectural differences that should improve performance overall, a GT200 with over 50% more Stream Processors (384), more memory bandwidth (though even @ 320 bit, certainly not 50% more), more ROPs and TMUs, should on average be 50% faster than a GTX 285.
Hahaha! .. no.. he asked for proof, not an article fuad wrote because some store-clerk mailed him "This week we sold three GTX295's versus one HD4870X2"!
Funny you're assuming that nVidia's ALUs will scale much, much, much better than ATI's (in your case ideal).
That's like counting architectural improvements twice when you don't even know how Fermi's modifications will impact how it takes on the graphics pipeline at large. Oh well, people are already saying Amen so others assume it's true.
At 4200Mhz (same as Rys' speculation) the 320-bit GTX360 will have 6% more bandwidth than the 285.
And again, if this is a salvage @575Mhz core (reasonable spec wrt GTX260a) or so, you have again a whooping 6% more texel and pixel fillrate.
The only point here is the substantial shader power increase- it's 1.6x base ALUs and say, 1600Mhz (VERY optimistic imo for a firstrun salvage part, I'd put my money on 1400Mhz instead) - 8.5% more shader clocks vs GTX 285.
73% more shader power but nearly flat fills beside Z. My money's on the 5870 being 10+% faster at least (and this is the part of me that's not bullish on future Catalyst releases- the other 99% of me is)
It would had been interesting if you'd heard (there's a barrier between hearsay and real knowledge) what mass production silicon supposedly runs at.
And they have been better. Without doubling G80/G92, GT200 was usually above 50% faster than G80 and G92, especially at higher resolutions.
From HD 3870 to HD 4870 (where ALU count alone increase 2.5x) plus the TMU increase, etc and at best, the HD 4870 doubled the performance of a HD 3870 at higher resolutions. And now we are seeing what happened with the HD 4870 to HD 5870, where the latter is on average 50-60% faster than the HD 4870, even though it doubles almost everything in the HD 4870, in terms of specs.
So yes, I am assuming the ALUs will scale better on NVIDIA's architecture, than on ATI's architecture, since they have been so far.
Tchock said:That's like counting architectural improvements twice when you don't even know how Fermi's modifications will impact how it takes on the graphics pipeline at large. Oh well, people are already saying Amen so others assume it's true.
No, I specifically said I wasn't even counting architectural improvements, since I don't really have data to talk about them...yet.
Tchock said:
I'm actually more inclined to 1500Mhz for the Stream Processors and 600-650Mhz for the core. But that's besides the point. The Stream Processors increase, plus the increase in ROPs (actually ROPs may be the same as GTX 285) a bit more bandwidth and 96 TMUs, should make it quite a bit faster than a GTX 285. How much, we will only know when it's released, but guesstimation on my part, along with Rys speculation of the full Fermi chip's performance, and I really can't see this "GTX 360" not being at the level of the HD 5870 on most occasions. Should be win some, lose some, as was the case with the GTX 260 vs HD 4870.
So nv's perf scales perfectly linearly with alu's,
I didn't say that. I said on average, which means, it may be 60% faster on some occasions and 30-40% in others, depending on the game, resolution, quality settings, etc.
Which would be a financial disaster a la GT200 for nv all over again, with no quick-shrink in sight to save their bacon.
Which would be a financial disaster a la GT200 for nv all over again, with no quick-shrink in sight to save their bacon.
That's assuming that this card, which I called "GeForce 360" is a full Fermi chip, with units disabled. I'm thinking that such card is more like a "GeForce 370" with 480 Stream Processors, one or two ROPs disabled, a 64 bit path from the memory interface disabled, etc.
Which is what GTX260 was to GTX 280. And if you were around at the time of 4870's launch on B3D forums, 260 was said to be DOA on it's then launch price. And how well did it turn out for nv?
Jawed
Legend
GT200's increased ALU:TEX, increased register file and improved texturing hardware all act as multipliers. Z-rate was also substantially increased, not to mention bandwidth.
Performance difference looks fine here, peaking at 97% at 1680x1050 4xAA/16xAF (hardly the toughest setting):
http://www.computerbase.de/artikel/...force_gtx_275/22/#abschnitt_performancerating
1GB of memory makes HD4870 16% faster than the 512MB version of the card in this test. This implies that a comparison of 512MB cards will show lower scaling for HD4870. That's because running out of memory causes additional serialisation during frame rendering. Since any serialisation will consume proportionally more of the frame rendering time on HD4870 than on HD3870, HD4870 will scale less than expected.
I've just shown that a comparison of HD5770 and HD5870 is more useful, and from that you can clearly see ~80% scaling achieved instead of 100% theoretical. The gap is Amdahl's law, basically - perhaps with some driver immaturity thrown in for good measure.
The fact this gap is so large bodes ill for GF100...
That's only because the base for ALU performance on NVidia is so low - much like R6xx's Z-rate base was so low. Whereas ATI ALU performance has always been more than adequate.
Agreed, a GTX360 should come in at least as fast as HD5870 on DX9/10 games.
Jawed
Actually, the overall average I have seen from all reviews has the 5870 in the 30-35% range faster than the GTX285, not 40-50%, unless of course you are referring to the niche market 2560x1600x8xAA.
I remember XMAN saying about no vendor preference a while ago. Now where is that now I wonder...
I have no preference, I'm simply trying expoosing his own biases. I already know he will NEVER admit he was wrong on any of his past predictions. Be it the performance ones or products.
Then why keep up the anti-charlie tirade?
Because I can't stand the to see him post and NOONE calls him out on some of them.
