trt740I just beat that 2900xt with my 8800 gts 640 in the 3dmark06 bench and my cpu is running about 200 mghz slower than the cpu doing the 2900xts bench. There is also a 8800gtx with a cpu running around 300 mghz slower than the cpu testing the 2900xt in that bench and the 8800 gtx is beating it .

taken from Techpower up Alcapone's 3dmark06 list
ATI list
1, lane - Sapphire HD2900XT @ 880/999 - 13335 - E6700 @ 4120Mhz - 412FSB

Nvidia list
2, ADV4NCED - BFG 8800GTXOC @ 641/980 - 12912 - E6600 @ 3775.1Mhz - 419.4FSB
3, cowie - eVGA 8800GTS @ 726/998 - 12428 - E6600 @ 3746Mhz - 416.2FSB

check it again your fastest run on amd with a gts was 11111 mine is 11025 and your cpu had a 250MHZ clock advantage which would easily make up the difference. and if you took your latest bench on your connie vs my latest you'll see that in the 2.0 tests the gts shines, but not in the 3.0 (or more recently releashed shader model) which shows a more future proof piece of hardware.

my old card was a 320 mb card did you factor that in. New ones was a 640 mb god rest it's soul.

trt740my old card was a 320 mb card did you factor that in.

Doesn't make a difference in 3dMark's default settings. It's optimized for 256MB VRam.

Depends on the resolution , the optimisation is for most usual setting....

HellasVagabondDepends on the resolution , the optimisation is for most usual setting....

I know, that's why I said default. ;)

Pointless really with synthetic benchmarks, if you want to get a more accurate synthetic reading both/all of you run Aquamark 3, ignore the CPU score and just quote the GPU score, that is just about the most GPU biased bench at the moment (of course the CPU plays a part in all benches but AM3 is probably the leaset CPU biased one out there at the moment), there is in fact an Aquamark 3thread here somwhere in any case.

Wile EDoesn't make a difference in 3dMark's default settings. It's optimized for 256MB VRam.

Well it does make a difference I've owned both a 8800gts 320 mb from Foxconn and a white box oem 8800gts 640mb and it made atleast 300 points difference in that bench mark. When both were clocked the same. Don't get me wrong the 2900xt is starting to shape up. Even if it was the fastest thing going there is the loud fan and heat. Still I'm impressed how hard AMD/ ATI are working out it's driver troubles. I saw a 2900xt on sale for 374.00 down from over 400.00 and almost bought it but it's out of my price range. If they could some how get that bad boy down near 339.00 with a rebate they might hook me, but then again it would be a toss up between a 8800 gts 640 mb because they are priced in that range now. Plus the 8800gts needs less power and the fan is better. However, you gotta love the shader potential of the 2900xt. The 2900xt 512 won't beat a gtx unless it is the ddr4 version and has a gig of ram. I wish It would because I really like ATI but I wouldn't bet on it. Who know maybe ATI will prove me wrong and then revise that fan too. They never could make a quiet fan LOL!!!! I have a x800 pro overclocked to a x800xt Wow! Wee! in my system now and that fans loud too. It is in there due to my 8800gt's sudden and un timely death.

trt740Well it does make a difference I've owned both a 8800gts 320 mb from Foxconn and a white box oem 8800gts 640mb and it made atleast 300 points difference in that bench mark. When both were clocked the same. Don't get me wrong the 2900xt is starting to shape up. Even if it was the fastest thing going there is the loud fan and heat. Still I'm impressed how hard AMD/ ATI are working out it's driver troubles. I saw a 2900xt on sale for 374.00 down from over 400.00 and almost bought it but it's out of my price range. If they could some how get that bad boy down near 339.00 with a rebate they might hook me, but then again it would be a toss up between a 8800 gts 640 mb because they are priced in that range now. Plus the 8800gts needs less power and the fan is better. However, you gotta love the shader potential of the 2900xt. The 2900xt 512 won't beat a gtx unless it is the ddr4 version and has a gig of ram. I wish It would because I really like ATI but I wouldn't bet on it. Who know maybe ATI will prove me wrong and then revise that fan too. They never could make a quiet fan LOL!!!! I have a x800 pro overclocked to a x800xt Wow! Wee! in my system now and that fans loud too. It is in there due to my 8800gt's sudden and un timely death.

I agree about the fans, when you crank them, they are noisy. ATI has been bad about that for a while now, but mine never really reaches high speeds, so it remains fairly quiet. Still, watercooling is in my plans for this beast anyway. Eliminates the issue altogether. lol

For the 06 benchmarks, are you sure there wasn't other factors involved? Even futuremark themselves said 320 vs 640mb won't make a difference in 06 at default settings, with all else being equal, as it doesn't even use the full 320MB of the lower model.

Well a bit of difference will always exist between 320 and 640 ram but not much...

its never more than 500 points, usually 1-200. I think its just due to the different ram, like how adding more ram into a desktop PC (assuming timings. the 1T/2T thing etc are all the same) yields a small performance boost.

They could have faster timings, quicker access cause the program fits in less ram modules, things like that.

The bottom line is that 2006 is not so much "otimised" for 256MB GDDR but it requires a "minimum" of 256MB to run. Some of the graphics tests are quite GPU intensive and therefore will demand the full memory bandwidth and will want immediate access, I am willing to bet that at times during the GPU tests it will demand more than 256MB of on board memory even at 1280 x 1024 and that of course requires system RAM "swopping" which slows the process a little making slightly lower real time performance.

Now I couldnt find examples of what impact that could make on 3D Mark 2006 so here is a linkie to some tests run between a 256MB and 512MB card to determine the effect, if you look at Quake4 at 1280 x 1024.....no AA/AF (default 3D Mark 2006 settings) you will see a difference in FPS, what I dont know is the difference in shading/pixel processes between Quake4 and 2006 but as i said, in some of the 2006 tests there is some serious shader/physics going down. In contrast though, in fear that has a completely different architecture there is no difference in speed at these resolutions but as far as fear is concerned it has pretty much a unique architecture.

I suppose what I am saying is that I would expect a little difference, probably as some have indicated, upto a max of 500 points, maybe less, what is clear in my mind though, even with current generation DX9 games that are extremely shader intensive, at resolutions of 16XX x 10XX and higher with everything maxed you could easily see a 25%+ increase in performance between 256 and 512MB.

www.pureoverclock.com/review.php?id=33&page=4