RV770 benchmark results

[MrMilli]

I don't know if it's real or not but here goes:
http://www.computerbase.de/news/hard...rks_rv770-gpu/

[MrMilli]

I thought it would be a bit faster than 3870x2 but it will (possibly) turn out much faster.

[jbunch07]

hmmm i dnt know about that...seems kind of sketchy...

[lemonadesoda]

Anyone here on TPU got some SM2.0 and SM3.0 figures for a QX6600 @3.0, or QX9xxx @3.0, and a HD3870@800 core, (Memory at max).

We can then compare if those SM2.0 and SM3.0 are realistic, based on 480/320 shaders and 32/16TMU and 800/800.

[erocker]

Here, using identical CPU scores:

[jbunch07]

Originally Posted by erocker

Here, using identical CPU scores:

so they didnt have the same cpu?

[erocker]

No, obviously a biased chart. Regardless, the RV770 is still going to stomp the snot out of the GX2. Soon, I'm sure we will see Nvidias next generation beating up on ATi cards. And the great circle of video card life goes on and on and on.....

[jbunch07]

Originally Posted by erocker

No, obviously a biased chart. Regardless, the RV770 is still going to stomp the snot out of the GX2. Soon, I'm sure we will see Nvidias next generation beating up on ATi cards. And the great circle of video card life goes on and on and on.....

haha good point...it definitely is a vicious cycle

[TonyStark]

Regardless of GPU speed, the 3DMark score should be roughly the same for both cards (unless the CPU is an Intel Quad Core overclocked off its tits). The GX2 is heavily bottlenecked at 1280x1024 in 3DMark06.

Also, you can put any values in that 3DMark calculator and it will calculate a result based on a formula. However in real testing, the SM2.0 and SM3.0 are bottlenecked by CPU speed, and will scale with the CPU speed. (ie you cant have 20,000 points on SM3.0 with a CPU that would score 2,000 but you can put those numbers in the calculator).

In other words, that benchmark is bullshit.

[Megasty]

Not this again Everyone knows by now from the previous thread that this circle is 300 degrees of bull followed by 60 degrees of shit. Anyone can makeup nice charts. NO ONE has this card yet so there should be no real benchmarks. Its that simple

[magibeg]

I'm actually pretty sure these exact same charts have already made their rounds in an earlier thread. But yea, can't confirm any of the numbers so.... meh?

[TonyStark]

Originally Posted by Megasty

NO ONE has this card yet so there should be no real benchmarks. Its that simple

There are people who own the card. Ever heard of engineering samples?

[nflesher87]

Originally Posted by TonyStark

Regardless of GPU speed, the 3DMark score should be roughly the same for both cards (unless the CPU is an Intel Quad Core overclocked off its tits). The GX2 is heavily bottlenecked at 1280x1024 in 3DMark06.

Also, you can put any values in that 3DMark calculator and it will calculate a result based on a formula. However in real testing, the SM2.0 and SM3.0 are bottlenecked by CPU speed, and will scale with the CPU speed. (ie you cant have 20,000 points on SM3.0 with a CPU that would score 2,000 but you can put those numbers in the calculator).

In other words, that benchmark is bullshit.

though your theory is good, the SM2.0 and SM3.0 tests aren't reliant enough on CPU speed to swing the overall score more than about 500 points (1K tops if you're talking the difference between CPU at like 2.0 and 4.0 GHz)

I'm not disagreeing that the results are biased, they obviously are, just like most unproven preliminary leaked benchmarks...though I'd say with a completely fair match up the RV770 has at least a 15-20% performance gain on the GX2

[TonyStark]

Originally Posted by nflesher87

though your theory is good, the SM2.0 and SM3.0 tests aren't reliant enough on CPU speed to swing the overall score more than about 500 points (1K tops if you're talking the difference between CPU at like 2.0 and 4.0 GHz)

I'm not disagreeing that the results are biased, they obviously are, just like most unproven preliminary leaked benchmarks...though I'd say with a completely fair match up the RV770 has at least a 15-20% performance gain on the GX2

My "Theory" is actually a fact. Where there is a CPU bottleneck, a more powerful card will still give you the same 3dMark score. Remove the bottleneck and the faster card will score higher.

[nflesher87]

Originally Posted by TonyStark

My "Theory" is actually a fact. Where there is a CPU bottleneck, a more powerful card will still give you the same 3dMark score. Remove the bottleneck and the faster card will score higher.

ok maybe theory wasn't the right word, but irregardless, it's not as big a factor as you made it

[Megasty]

Originally Posted by TonyStark

There are people who own the card. Ever heard of engineering samples?

Of course. I'm not saying that the card don't exist yet. I'm just saying that the ppl that have it now only care about getting the final product out to us, not silly fake benchmarks.

[lemonadesoda]

I have a 3850 that I clock to 800 core, and 1000 memory, on a Q6600 @ 2.7.

So my CPU score is 10% down on their benchmark, clock for clock. My CPU score is 4040. You have to add 25% to get that CPU score they got. I think that's a bit high. That isnt accounted for in the relative clocks. I think the CPU score faked.

Then my SM2.0 is 4500 and my SM3.0 is 5000. They get 8845 and 10612. Thats exactly DOUBLE the performance. But there are only 480 shaders compared to my 320. Thats 50%. The texture units are up from 16 to 32. Thats 100%.

But this is a shader test not a texture fill test. Hence the SM2.0 scores and SM3.0 scores should be up somewhere BETWEEN 50% and 100%, but closer to the 50%. Yet higher than 50% due to faster memory and shader clock. My best guess is a rough average, ie 75%. I dont see how it can be the MAX, ie. 100%, but it should be better than 50%, given the extra RAM speed.

But they are reporting exactly 100% speed increase. I dont think it can be over 75% due to the "upgrade" being averages of 50%-100% improvement. You just cant get the max, max.
>> IMO, these results are BOGUS. F.A. K. E. Theyve taken 3870 figures in 3Dmark06, "doubled them", then used the 3Dmark06 calculator to come up with the scores.

[nflesher87]

Originally Posted by lemonadesoda

I have a 3850 that I clock to 800 core, and 1000 memory, on a Q6600 @ 2.7.

So my CPU score is 10% down on their benchmark, clock for clock. My CPU score is 4040. You have to add 25% to get that CPU score they got. I think that's a bit high. That isnt accounted for in the relative clocks. I think the CPU score faked.

Then my SM2.0 is 4500 and my SM3.0 is 5000. They get 8845 and 10612. Thats exactly DOUBLE the performance. But there are only 480 shaders compared to my 320. Thats 33%. The texture units are up from 16 to 32. Thats 100%.

But this is a shader test not a texture fill test. Hence the SM2.0 scores and SM3.0 scores should be up somewhere BETWEEN 33% and 100%, but much closer to the 33%. Yet higher than 33% due to faster memory. My guess is 50% max, given the extra RAM speed.

But they are reporting 100% speed increase. Not 50%.

>> Their results are BOGUS, IMO. F.A. K. E.

QX9650s get over 7000 CPU score above 4.0GHz

[lemonadesoda]

Benchmark done at: Intel Core 2 Extreme QX9650 @ 3.00GHz

OK, if that is true, that the QX9650 can get 7000 CPU score @4.0, then 7000/4.0*3.0 = 5250, so the CPU score is possible.

[jbunch07]

fake or not im not going to believe anything until the card is for sell and someone here has it and/or even better until i have one!

[Megasty]

Originally Posted by jbunch07

fake or not im not going to believe anything until the card is for sell and someone here has it and/or even better until i have one!

I'm with you on that. I don't believe any benchmarks except for my own anyway

[jbunch07]

Originally Posted by Megasty

I'm with you on that. I don't believe any benchmarks except for my own anyway

you can say that again

[xkm1948]

fake

[trog100]

tony is pretty much right.. both shader tests are so cpu limited in default 2006 a more powerfull card wouldnt boost the score like that.. its fake..

trog

[Morgoth]

not fake
i really want to see how the hd4870x2 and crossfire perfomance in crysis with bloomfield at 3ghz