DirectCompute benchmark

gareth170 · 8 Nov 2009 at 11:22

Shocky-FM said:
My results..

HD5870/1GHz vs i7 920/4GHz

5850 @ 1ghz

looks like clock for clock theres not much differents

rippling · 8 Nov 2009 at 11:46

ok so there is def a problem with the bench mark as my 5850 is at 1ghz to and only 28k so this bench mark is poooooo p

Rroff · 8 Nov 2009 at 12:49

Only other PC I have windows 7 on is my laptop - 260m GTX (G92b core) and it seems to really struggle on there... ~7000 GPU score and 600 CPU score.

craig3000 · 8 Nov 2009 at 13:00

Im getting strange scores :/ my 5850 works perfectly fine but for some reason in the benchmark I get just under 10k, I run GPU-z at the same time to check the clocks were up and everything looked normal WTF??

Rroff · 8 Nov 2009 at 13:03

craig3000 · 8 Nov 2009 at 13:04

YEAH BUT LOOK AT THIS

This benchmark is messed up my CPU which is a Q6600 at 2.4ghz is beating a core i7@ 4ghz if you look at the first page of this thread, what the hell is going on hahah

Rroff · 8 Nov 2009 at 13:06

Something deff. not right with your scores...

"M15k for GPU is a 'typical' result for 9800gt" << my laptop which has a slightly faster GPU than the 9800GT gets half the score...

craig3000 · 8 Nov 2009 at 13:09

Well I'm treating it as not reliable on my system

deleting

Rroff · 8 Nov 2009 at 13:11

Oh as far as the benchmarks go (when its giving correct scores) - the benchmark is doing the same workload indiscriminately - using standard DX compute pipeline - with no optimization or bias for any platform - this IS how performance will stack up if the developer isn't using 5.0 specifically and not using any vendor specific optimization - just saying to DirectX - "compute this on the GPU for me".

~15K seems typical for G92 based cards
~30K typical for (single) 200 series cards
~40K typical for 5800 series cards

EDIT: Which means if Fermi does come out in the supposed configuration it will indeed annihilate the 5870 and even 2x 5870 for GPGPU processing... but we knew that was the intention already.

Broken Hope · 8 Nov 2009 at 13:11

craig3000 said:
YEAH BUT LOOK AT THIS

This benchmark is messed up my CPU which is a Q6600 at 2.4ghz is beating a core i7@ 4ghz if you look at the first page of this thread, what the hell is going on hahah

You're running the wrong version, you need 0.25 not 0.15.

The CPU shouldn't be holding anything back in the GPU test, I show as 0% used in task manager when it's running the GPU part of the test.

Mel_P · 8 Nov 2009 at 13:14

I have a 4850 and it says "Directcompute not supported" - do I need to update my drivers or is that a feature of the 4850 - I have Winn 7 DX11?

Rroff · 8 Nov 2009 at 13:17

Nicely noticed Broken Hope.

Mel - ATI haven't exposed directcompute on any older cards than the 5000 series at this point.

Broken Hope · 8 Nov 2009 at 13:18

Rroff said:
Oh as far as the benchmarks go (when its giving correct scores) - the benchmark is doing the same workload indiscriminately - using standard DX compute pipeline - with no optimization or bias for any platform - this IS how performance will stack up if the developer isn't using 5.0 specifically and not using any vendor specific optimization - just saying to DirectX - "compute this on the GPU for me".

~15K seems typical for G92 based cards
~30K typical for (single) 200 series cards
~40K typical for 5800 series cards

EDIT: Which means if Fermi does come out in the supposed configuration it will indeed annihilate the 5870 and even 2x 5870 for GPGPU processing... but we knew that was the intention already.

Care to explain why the 5700 and 5800 series have similar scores, and also care to explain this?

http://futuremark.yougamers.com/forum/showpost.php?p=1247851&postcount=2

Lightnix · 8 Nov 2009 at 13:38

First observation: The scaling's really wonky for the ATi cards. The 5800 series and 5700 series perform within 10-15% of eachother despite the discrepancy (the 5800 series has double everything) in compute resources. Whilst the 5800 series seem to be doing a bit crap, the 5700 series seem to be more or less where you'd expect them.

Second observation:

**** Version 0.xx means the tool is NOT completed, may be buggy or unstable. ****

Third observation: We have no idea what this code is actually running, we only have a very brief description and we have no access to the source code. No I am not saying they're running different code, I'm saying the same code is possibly poorly suited the Radeon cards and CPU's with SSE support.

But yeah, this is totally indicative of every piece directcompute code that will ever be written, sure why not.

Rroff · 8 Nov 2009 at 13:45

Its fairly indicative of anything which is thrown blindly at DX (which will be fairly much the norm for most developers) with the current ATI drivers anyhow...

The 5770 at stock should be scoring around 26-28K tho and the 5870 really should be snucking in just below 60K.

{SAS}TB · 8 Nov 2009 at 14:24

Locky · 8 Nov 2009 at 14:30

Mel_P said:
I have a 4850 and it says "Directcompute not supported" - do I need to update my drivers or is that a feature of the 4850 - I have Winn 7 DX11?

I get the same error msg I have tri sli gtx 280.

Broken Hope · 8 Nov 2009 at 14:36

Locky said:
I get the same error msg I have tri sli gtx 280.

Running Windows 7 or Vista with the platform update? Also I think you need quite new drivers for Direct Compute support on the 200 series.

Rroff · 8 Nov 2009 at 14:59

You need atleast 186.xx drivers + the platform update and the latest DX updates.

hyperst · 8 Nov 2009 at 15:21

heres both for me, dosn't look much diff for gpu

Rroff what scores do you get running this bench?