A suggestion

[KONAKONA]

I've thought about this just a smidgen, minus well have it shot down in public.

 

Most (er, um, all) of the 3D benchmarks on the bot are still quite CPU limited. Even vantage still needs a beefy CPU to get good scores. A lot of more casual benchers (Yes, I am biased because I am included) have good GPUs but lack the CPU (and or ram) to back it up in 3D tests. So why not add a bench to the bot that has a very low CPU limitation?

 

I've been told (and I haven't looked this up myself so don't quote me) that the heaven benchmark has little CPU limitation. I haven't checked this myself though. The only other thing I can think of would be running one of the 3dmarks (Probably 03/05 to avoid the CPU tests, but then you are back to the CPU bottleneck) at a higher res and higher AA/AF, but that doesn't seem like it would work well for a number of reasons, mainly that it would force people to buy the full versions.

 

Probably a bad idea, but if you change the points system to give more points to more popular but slower hardware, having some benchmark that properly isolates GPU performance doesn't seem that out of step.

 

Another thought on the same tone would be points for combonations of CPU and GPU in 3D benchmarks. It doesn't seem like it would be that much work, as you can already search for 3D results with CPUs, but it would divide things up a bunch and getting the points to work right with all of that seems like it would be a nightmare.

 

Just throwing stuff out there, please don't eat me. I don't taste very nice.

[Massman]

An important sidenote to make is that a benchmark generally becomes more CPU dependant the older it is. At the release, 3DMark Vantage was pretty GPU bound (eg: even with HD4890 I don't see a difference increasing 1GHz in CPU), but with the 4xGPU configs the CPU bottleneck is just too great.

[KONAKONA]

We've talked to Unigene to have the Heaven benchmark included in the suite, but after initial interest they stopped responding. : /

 

We are very aware of this issue, be sure. Personally I always liked furmark.

 

I forgot furmark had a benching utility. That seems like it would work. They also seem to update it a lot, which helps with the below problem...

 

An important sidenote to make is that a benchmark generally becomes more CPU dependant the older it is. At the release, 3DMark Vantage was pretty GPU bound (eg: even with HD4890 I don't see a difference increasing 1GHz in CPU), but with the 4xGPU configs the CPU bottleneck is just too great.

 

Vantage always had CPU tests in it, though.

 

I really like the idea of using furmark myself.

[Rudster816]

Maybe you just do a GPU score only category with Vantage?

 

I feel your pain. I can "only" do vantage a 4.6ghz on my i7 920, but its getting owned by Gulftowns and 975's. Gulftown really puts a knife in budget orientated benchers. $1100 for a chip is

[K404]

There will be cheaper options as the months pass. It was the same with the FX60, X6800, QX9650 etc etc etc

 

I dont disagree with your idea, but I partly-disagree with your reason for having it

 

@Massman..... to nit-pick, the Vantage situation is unique. The SCORE is bottlenecked, but the GPU sub-score isn't. Not really. If only the GPU score was used, people could skip the top-end CPUs and top-end cooling without completely killing their chances of competing well.

Edited April 10, 2010 by K404

[KONAKONA]

There will be cheaper options as the months pass. It was the same with the FX60, X6800, QX9650 etc etc etc

 

I dont disagree with your idea, but I partly-disagree with your reason for having it

 

By the time a setup gets cheap, the next one is already here. That's part of the reason they get affordable in the first place.

 

If anyone wants I can gather some data on furmark and how it scores related to CPU and GPU speeds. I've done some limited testing on my rig at 1280x1024 16AA and I had maybe a 5 point difference max between CPU@ 3.4 and CPU@1.8. It also scales nicely with GPU speeds. I could probably get some data about how it scales with a variety of hardware, if need be.

 

I'd love to see furmark on the bot.

[BenchZowner]

Along with hundreds of edited screenshots ?

[IanCutress]

I've been developing a bench which is solely this; however it's currently only written for nVidia GPUs in CUDA. I need to port it to OpenCL for ATi cards. Unfortunately OpenCL looks like an arse to program...

 

https://www.benchtec.co.uk/forums/showthread.php?t=4947

 

So far we've seen:

 

Quadro FX580 (in dual E5520) score 164.8

8800GT (in E5520) score 550.5

9800GX2 in Q9650 (@3.6Ghz) score 1173.2 (585.76 for first GPU, 587.48 for second)

9600 (in E6400) score 423.8

260M in i7 920 score 516.8

 

But with any GPU bench, it'll be somewhat effected by the bus speed in any memory copy (RAM to GPU) situation. Mine gets around that by only ever doing it twice and it taking microseconds.

[KONAKONA]

Along with hundreds of edited screenshots ?

 

They already have a submission system, not like it would be any harder to edit an AM3 or 3D01 screen.

 

http://www.ozone3d.net/benchmarks/furmark_scores.php

 

Only problem is all of the top scores there are with the old version (1.8.0), but I don't think that would be a problem when submitting to the bot.