Donnerstag • 30.03.2017 • 18:32 Uhr
 PowerVR FAQ
 SGX544MP angekündigt
 Der aktuelle Stand
 zweiter SGX Artikel ...
 Die Videofähigkeiten...
 neue Treiber für Pou...
 neuer Treiber mit Ha...
 SGX - Aufbau, Techn...
 neuer Serie5XT Grafi...
 Apple und Intel inve...
 Mehrere weitere Lize...
 SGX - PowerVRs Shade...
 Interview mit ImgTec...
 Quack - das andere D...
 Hercules4500 AGP4x-M...
 Optionen des KYRO Tr...
 Nummer 5 lebt!
 Startpartition von W...
 MBX - Mobile Grafik
 Die Grafikfähigkeite...
 Vivid! - PowerVR Gra...
 Videologic Neon 250
 Geforce 6800 - the n...
 Imagination T.
 Planet 3dNow!
 PVR Generations
 Tom's Hardware

3D Center


Imagination Technologies

Geforce 6800 - the next generation

The main purpose of synthetic applications/benchmarks usually is to provide the user with useful information about an accelerators throughput and/or performance under special conditions.

First I used ArchMark from zeckensack: Archmark.html

In general the results are in line with the claimed specifications of the 6800. Results are similar in MDolenc’s Fill-rate Tester:

Display adapter: NVIDIA GeForce 6800
Driver version:
Display mode: 1600x1200 A8R8G8B8 85Hz
Z-Buffer format: D24S8

FFP - Pure fillrate - 3777.976807M pixels/sec
FFP - Z pixel rate - 7359.851074M pixels/sec
FFP - Single texture - 3483.079346M pixels/sec
FFP - Dual texture - 1869.010986M pixels/sec
FFP - Triple texture - 1258.128296M pixels/sec
FFP - Quad texture - 949.485291M pixels/sec
PS 1.1 - Simple - 1277.885742M pixels/sec
PS 1.4 - Simple - 1908.520996M pixels/sec
PS 2.0 - Simple - 1908.838257M pixels/sec
PS 2.0 PP - Simple - 1278.094971M pixels/sec
PS 2.0 - Longer - 961.656250M pixels/sec
PS 2.0 PP - Longer - 958.687744M pixels/sec
PS 2.0 - Longer 4 Registers - 958.385925M pixels/sec
PS 2.0 PP - Longer 4 Registers - 961.671265M pixels/sec
PS 2.0 - Per Pixel Lighting - 245.010010M pixels/sec
PS 2.0 PP - Per Pixel Lighting - 413.563965M pixels/sec

Theoretical throughputs via 3Dmark 01 and 3Dmark 03:

High Poly count (1 light) 104.7 Mtriangles/sec
High Poly count (8 lights) 22,4 Mtriangles/sec
VS 1.1 241,1 fps
VS 2.0 23,3 fps
PS 1.1 310,8 fps
PS 1.4 307,9 fps
PS 2.0 110,2 fps

Judging from the VS 2.0 performance above, the 6800 obviously carries the same 6-way Vertex Shader as it’s bigger brothers. I am somewhat puzzled with the PS1.1 results since they ironically seem a lot higher than what Dave Baumann got in his GF 6800U preview at B3D. If it should be some sort of mistake from me, I’ll correct it later on.

Next comes 3Dmark03. I ran only the 4 game tests since they contribute mostly to the final score. Once with default 1024*768 and once with 1600*1200, yet with both I had 4xAA/8xAF enabled.

The final scores were:

1024*768*32 / 4xAA / 8xAF : 4778 points
1600*1200*32 / 4xAA / 8xAF : 2430 points

Not too bad, considering that past generation high end accelerators used to reach around that default score at 1024 w/o any AA enabled.

Individual scores of all 4 game tests in both resolutions:

One point that deserves attention is Z/stencil performance on NV40. It’s my understanding that they carry 2 Z/stencil ROPs per SIMD channel. Logically 6800Ultra and 6800GT end up with 32 Z/stencil ROPs in total, while the 6800 with 24. When I tested MDolenc’s fill-rate tester though again with 4xMSAA enabled, I got 2283 Mpixels/sec Z/stencil fill-rate, which could mean that albeit one quad of SIMDs is disabled, all 32 ROPs are still active.

It translates into 12.8 Gpixels/s for the 6800Ultra and 10.4 Gpixels/s for the 6800 in stencil fill-rate. PowerVR’s FableMark is a benchmark that measures stencil performance and to that is mainly fill-rate limited. Let’s see how the 6800 fairs with it, with and without AA enabled:

The 6800 behaves quite stable every time 2x more AA samples get added, resulting into a ~15% performance decrease at a time (~15% with 2xAA and ~30% with 4xAA).

NVIDIA has done an excellent job here optimizing stencil performance, especially with MSAA enabled, yet the sample loops (2x at a time) do still cost performance. If stencil performance should remain important especially after the long awaited DoomIII release, it might be a department worth of additional optimizations for the AA sample loops mostly.

Last in line comes another PowerVR demo Deferred Shading. It does what the name implies using dx9.0 Multiple Render Targets. It’s not really a benchmark so I’ve captured only two screenshots with FRAPS enabled in the highest possible resolution (2048*1536). I’ll leave it to anyone’s guess from which accelerator the 2nd screenshot comes from. A hint on performance with all lights enabled you can get from the frame counter in each shot. Yet the differences I’m aiming to point out here, lies in a totally different department. You can see them yourselves; I don’t really need to point them out to you. And before you ask no neither AA nor AF was enabled in either/or case:

Seite 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
 Geforce 6800 - the next genera...
 17 Seiten
 verfasst von Ailuros
 Donnerstag - 26.08.2004 - 18:20 Uhr
[1]  Introduction & Systemsetup
[2]  Antialiasing & Anisotropic Filt...
[3]  2. Anisotropic filtering
[4]  Synthetic Applications
[5]  F1 2002
[6]  Rallisport Challenge
[7]  Serious Sam: The Second Encount...
[8]  Need For Speed: Underground
[9]  Unreal Tournament 2004
[10]  Far Cry
[11]  Hybrid Super-/Multisampling Mod...
[12]  Hybrid Super-/Multisampling Mod...
[13]  2D, DVD / VIDEO Playback
[14]  Summary
[15]  Softmode
[16]  doom3
[17]  DVD