here is the link to that 7800 scores 7800 thread

[coldpower27]

Well you can't exactly expect the moon on performance here, with Nvidia using only 110nm they had to limit themselves to 24 Pipes for now, like ATI's R520 32 pipes is feasible only on the 90nm node. Judging from the G70 die pic posted a while ago, the G70 die size is around 320mm2, not exactly small is it, a tad larger then the nV40's.

To me 7800 isn't a great score at first glance, for a since I read that the 6800 Ultra gets around 5700 or so.

With the current guess specs on G70 GTX being 10.32 GPixel's and 860 MVertices Fillrates compared to the NV40 Ultra 6.4 GPixel's and 600 MVertices.

I can see the dissappointment with only 5700/7800 = 73% Rate.

With NV40 Ultra having 62% Pixel Fillrate 69% Vertex Rate.

People might have expected a score 8400 or so, but since the score is rather this low, there is a possibility we are seeing the limitation of the memory bandwidth on this card. With the increase of memory bandwidth being rumored to be 100MHZ or so. That would mean 6800 Ultra has 92% of the memory bandwidth of G70. At least that's what I think could be the problem.

[Dazz]

Funny when I say that the G70 is vertex limited everyone is now doing their tests to see what the hit is. I thought it was obvious myself that's why the R480 can out do the NV45 because the R480 has a higher clock speed as such more vertexes despite the NV45 using PS3.0 and Vertex 3.0. Also why ATi Want a unified pipelines so the pipelines can be used either for PS or Vertex if the applications demand it.

[rohit]

I lost all faith in 3dmark when i upgraded from an athlonxp to a 64 and my score went down 200 points, but my games performance took the fu(k off, night and day difference. Then i upgraded to windows x64 and all my games once again responded nicely with less drops in fps, but my 3dmark still lost another 50-100 points....3dmark is worthless, when game performance goes up and scores go down we are in no way getting an accurate score to how well our machines are doing. Sure its the graphics card its measuring, but games dont use just the graphics card. The sooner we put futuremark out of business the sooner we get better gaming video cards
LOLx LOLx LOLx,
When i changed from P4 2.8E @ 3.2GHz to A64 3200+ (no O/C), i lost 180 points in 3dm'05. while the games gained a few more FPS for sure.

Now my question is, How Much are todays Games VERTEX Limited or Vertex dependent? By todays games i mean Doom3, RoE, HL2, BF2 WoW, S.Cell CT..etc etc.

[Dazz]

They make very little use at the moment but may increase as time goes on Unreal3 maybe making heavy use of it for joints on people etc.

[ChrisRay]

Funny when I say that the G70 is vertex limited everyone is now doing their tests to see what the hit is. I thought it was obvious myself that's why the R480 can out do the NV45 because the R480 has a higher clock speed as such more vertexes despite the NV45 using PS3.0 and Vertex 3.0. Also why ATi Want a unified pipelines so the pipelines can be used either for PS or Vertex if the applications demand it.

Everyone is now doing tests?? :eek: I performed those pipeline tests almost a year ago when 3dmark05 was released.


They make very little use at the moment but may increase as time goes on Unreal3 maybe making heavy use of it for joints on people etc.


Vertex processing is always going to take a backseat to pixel fillrate. The reason being is low end cards will never have the vertex crunching capabilities of there high end counterparts. Devs have catered to low end cards. Despite what we high end owners thinks. It's easier to lower pixel fillrate by adjusting resolution, lowering speculiar effects, removing anti aliasing/af, ect. 3dmark pushes over 1 million polygons at any given time. Unreal 3.0 isnt even going to come within shouting distance of that. It is pretty commonly agreed that 3dmark05's emphasis on geometry capabilities is unrealistic with its over emphasis on vertex fetching.