Originally posted by: Azn
GTX 280 pulling away? Where? :laugh: GTX 275 beats GTX 280 in all of anand's benches.
No it doesn't, there's clearly examples of the 280 winning at 2560. Its also hilarious that you continue to consider benches with < 1 FPS difference as wins.
Treiberversionen
Nvidia GeForce 185.63 (GTX 275)
Nvidia GeForce 182.46 (GTX 260²)
Nvidia GeForce 181.22 (GTX 295, GTX 285)
Nvidia GeForce 180.48
ATi Catalyst 8.592.1 (HD 4890)
ATi Catalyst 9.3 (HD 4870 1GB)
ATi Catalyst 8.11
What's that translate into? Archived benches using 180.48 for the GTX 280 and 185.63 for the 275. Given the very real gains in performance from 180.48 to 185.63 I'd say their results are certainly going to be skewed.
You make me laugh. Did I say bandwidth was more important than SP or TMU? Of course not but it's only in your interest to say ignorant things to derail anyone who challenges your theories. I just said it's part of the GTX 280 equation since you said bandwidth didn't matter on GTX 280. Even in your own benches you've tested your GTX 280 you had minimum frame rates dropping as much 25%. Average frame rates dropping by 5%. To say bandwidth doesn't matter even on a bandwidth saturated card like GTX 280 is quite ignorant when you have tested the theory yourself.
You do realize you're trying to argue both sides of the issue right? Either bandwidth is significant, or it isn't. I've already said it isn't significant on GT200 as I've clearly demonstrated a 27% decrease in bandwidth results in a 3% difference in performance. The fact a GTX 275 can (as the 260 before it) overcome any non-existent bandwidth limitation to come close or exceed a GTX 280, with much greater bandwidth, absolutely proves my point. Once again, you can't have it both ways although you've consistently argued both sides claiming it does or doesn't matter, depending on how wrong you are at any given point.
Clocking it same on the core fine but what about bandwidth or the bandwidth saturation on the ROP itself. You would have to clock the GTX275 1264mhz on the memory to get same bandwidth as GTX 280. Even then High memory clocks has a effect on the ROP since they are tied down to the ROP itself so it would give advantage to GTX 275. This wouldn't be accurate but you would have to know that memory controllers are tied down to ROP to understand this. I've already mentioned this to you numerous times but you never could understand the concept of memory bandwidth effecting how ROP performs. Even then you don't even have the data to prove anything.
Uh, no memory clocks have no impact on the ROP whatsoever, that's why they have independent clockspeeds. The clockspeed of the memory controllers themselves are tied to the core/ROP domain and are not impacted at all by memory frequency.
Clocking it the same reduces the number of variables by eliminating any difference in TMU and SP performance and allows you to isolate performance differences, its that simple. Like I said, you've discovered an overclocked part can overcome hardware deficiencies elsewhere. Congratulations, welcome to 2001.
Since you said ROP makes the biggest performance difference GTX 280 should beat GTX275 with combination of ROP and bandwidth but it doesn't. GTX 275 beats GTX 280 by 5% on average.
LMAO, no, I've never claimed bandwidth mattered, so trying to lump bandwidth in with ROP performance to make the numbers bigger in your favor isn't going to work. If anything the fact bandwidth doesn't hold the GTX 275 back compared to a part like the GTX 280 with more bandwidth further proves my point.
1. GTX280 | 2. GTX 275
ROP fillrate
19264 - 17724 = +8%
GFLOPs
933 - 1010 = -8%
Tex fillrate
48,160 - 50,640 = -5%
memory bandwidth
141.7 - 127 = +11.5%
Which brings us back to my point. Which variable do you think impacts performance the most? I've maintained ROP and core domain, I guess you're claiming "the rest" lol.