Call for Benchmarks for CAD and OpenCL

While it's not technically gaming related, it's still GPU related. I got curious, in parts because of the discussion of pro GPUs vs Gaming GPUs for CAD in WNSIB and such as to at what point cuda core (or shader) count tips the scale in favor of the gaming GPUs vs pro GPUs. There is quite a bit of info available on notebookcheck for SPECviewperf, but not for the latest drivers so, if anyone's willing to run it on their gaming GPUs on the current drivers, I'd appreciate it. What I'd like most to see though is benchmarks for luxmark 2.0, the complex benchmark since it's the one for which I got data on the Quadro and Firepros. There's not a lot of data for OpenCL on mobile CPUs as far as I know.

Obviously, a GTX680m won't beat something like a Quadro K5000m or K4000m, but it will at some point beat some quadros and firepros with higher processing power. Same goes for the Radeons.

Here's the link for viewperf: Download GWPG Benchmarks. It's a hefty download at almost 1 GB, so I understand not everyone may want to download it to run it especially since it's not for games.
Here's the one for luxmark: LuxMark - LuxRender Wiki

If you run the benchmarks, please list your system specs if they aren't in your sig. The data i got for viewperf is at 1920x1080, so keep it at that resolution if you do run the benchmarks please.

Latest results here

 

Good stuff, thanks. Will check it out and report back.

Luxmark with 680m @ 950MHz, Room Scene



Luxmark with 680m @ 900MHz + HD 4000, Room Scene.



680m @ 900MHz + HD 4000 + i7-3740QM @ 3.9GHz, Room Scene


Not sure what it all means, but there you go!

Downloading SPECviewperf. Only 128KB/sec from their servers, so gonna be a while.

 

Well, looks like your OC'ed 680m is performing faster in Luxmark than K5000m at stock clocks, wasn't expecting that. Any chance you could run it at stock clocks too?

I know how you feel for specviewperf, I ran into the same slow download speed issue, took over 2 hours to download the darn thing.

EDIT: Did a little more digging, and I think that the score is in frames per sec

Oh and thanks by the way.

 

Downloaded luxmark, but the SPECviewperf will also take a while.
Here are my results:
This is on a 7970m/i7-3610qm with the latest 13.5 beta3 drivers. I didint had the option to select the intel grahic like above, so I just did CPU+GPU and GPU separettly. Hope it helps.

7970m only stock 850/1200


7970m + CPU stock 850/1200


7970m overcclocked 1050/1550


7970m + CPU overcclocked 1050/1550


Weird that the gpu+cpu score is actually a lower than the gpu alone. This looks like another enduro symptome. GPU usage when running with CPU went down to 71% on stock and around 60% when overclock(without CPU 99% all the time). I would be interested to see results of the 7970m on a alinenware ar another system with enduro turned off. Enduro always behave like this- When CPU+GPU is loaded together usage goes down like there is not enough power for both of them.(Battlefield 3 heavy gunfight, or gta 4,...etc)

 

Not sure this helps...

Just 560m's


560m's+cpu

 

7970M @ 1050/1500 (ignore what it says in the screenies - I pushed the clocks further but it didn't show in the benchmark)

4X the perfomance of GTX680M "



CPU limits GPU usage here as well - but no enduro. Maybe I'll need a higher watt PSU?

 

Thanks to everyone so far, I'm pretty busy with university right now so it will be some time before I can put those results in graph format, but I will.

 

I can only go down to 767MHz on the core, stock is 720MHz, but I have a modded BIOS and can only drop -135MHz on the core from my modded stock clock of 900MHz, so not sure if that helps to post that score or not.

 

I'll try running a 680m stock clocks, I only have the 3630qm anyways. Might not be able to disable optimus because of UEFI boot not liking the vbios.

Update:





Stock clocks for the dell 680m (boost 758mhz core)

Everything else on adaptive/stock. not strictly on high performance mode.

 

I took the time to put some of the data you guys gave me in graph format:


AMD dominates in OpenCL, but for the pro GPUs at least, nVidia wins in CAD (no graphs shown, but I could add some).

 

all in all, if you need opencl you dont buy a pro gpu, you buy a 7970m a rejoice, interesting

 

For raw OpenCL performance, yeah, the firepros are only based on the 7700 (M4000) and 7800 (M6000), there is nothing based on the 7970m in the firepro line for some reason.

For lols, I should rerun the complex scene on my 9650m GT, it scored something like a measly 16 if I recall correctly.

 

OpenCL: General-Purpose Computing : AMD Radeon HD 7990: Eight Games And A Beastly Card For $1,000

For scale/reference.

 

hah if you are doing that i will do on my 4670m gddr3

i also found strange that the m6000 is the top that they offer

 

Tell me about it or I would be using the M9000 (or whatever they would have called the M8900's successor).

The benchmark they are running is sala, the medium scene instead of the complex scene for which, my M6000 scores 633, almost at the same level as a desktop GTX680 in OpenCL

EDIT: Oh yeah, now that I checked and double checked, the K5000m results are at 600/1500 clocks since both Dell and HP saw fit to underclock the card on their workstations.

 

is good?

 

SpecViewPerf
680m @ 1100/2400
i7 3740qm @ 3.9GHz

 

Here you go, score of 7970 stock clock 850/1200 with 13.5 beta 3 drivers. The test took for ever to finish :SLEEP:

 

I downloaded the Intel OpenCL driver so I gained CPU support for my GPU




Not sure why my K3000m lists Compute units as 3? Although these changes made no difference on my K3000m result which is still 67

 

Config in signature 7970m @ 920/ 1270 @ 0.975v

 

K5000M stock:

 

Prema, I gotta ask, what's your system, your K5000m score is much higher than the one on the M6700, but the K5000m is underclocked @600 MHz at stock on the M6700. I think that's the case for the Elitebook 8770w, but I could be wrong. Is it one of those few Clevos that offer the K5000m?

 

GPU Only = 906
CPU + GPU = 1133

http://i957.photobucket.com/albums/ae55/FredFlint/LxxGPU_zps35487852.png
http://i957.photobucket.com/albums/ae55/FredFlint/LxxCPUGPU_zps86edf7aa.png

 

Same clocks, but that's on a Clevo P50WMx...don't know how CPU bound that test is...

 

Alright, so after being away for a while, here's some compiled data. So far, I have not included results for the Luxmark normal scene, that's about it.

 

So some conclusions, given Prema's results, it seems like some of those benchmarks are rather CPU bound.
If you have a firepro, you will want to choose the best drive for you given the discrepancies in the scores between version A01 and A03 (those are Dell's internal version numbers, but only Dell offers the M6000 anyways)
The K2000m results for Lightwave are a bit puzzling, they come from notebookcheck by the way.
The K2000m still performs above a 680m in most CAD benchmarks too , to be honest, I wasn't expecting that.
Finally, thanks to everyone who contributed to the data. I have not done anything with the Luxmark normal scene data I got, I'm still deciding whether to include it or not.

Oh and for those wondering about the double post, there's a max of 6 attachments per posts, otherwise, this would only be one post.

 

Are you guys interested in benchmarks for older hardware like the 2000M and GTX 560SE (desktop card)?

 

The Quadro 2000M would be nice.

 

Well, I ran the default benchmark for Lux (GPU only, default scene), and here's what I got:

Quadro 2000M (stock settings):



GTX 560SE (stock settings), just to see how it'd do:

 

M8900 lastest workstation driver from AMD. stock clock.
Luxmark refuse to run for w.e reason ~~. I think fermi result is more interesting.

Viewperf 11.0


Viewset Composite Multisample Performance
catia-03 15.86 no result
ensight-04 21.52 no result
lightwave-01 43.14 no result
maya-03 46.38 no result
proe-05 4.49 no result
sw-02 35.71 no result
tcvis-02 16.75 no result
snx-01 22.17 no result

 

Here's batch two, I noticed one Luxmark result misplaced, added the data I got for the medium scene too and updated the graphs with the firepro's latest driver results which are impressive. AMD did some tweaking, either for the programs or specifically for specview.

Catia:

EnSight:

Lightwave:

Maya:

Pro/Engineer:

Solidworks:

 

Second batch of graphs.

Tcvis:

Siemens NX:

Luxmark Room scene:

Luxmark Sala scene:

Contributors:

Forgot to add Jarhead in the list when I took the screenshot so I'm mentioning it now.

 

luxmark finally run with 12.104
412
192

Viewperf 11 Results Overview (by vendor)

this have result for fermi mobile workstation cards.

 

Those are the scores for medium and complex? I was expecting them to be higher to be honest.

 

Gcn had a major boost in opengl.

 

Dell Precision m6800 with FirePro m6100 and i7 4800qm running on Windows 8.1 x64 with latest drivers from Dell:

SpecViewPort results:

catia-03 38.75
ensight-04 56.29
lightwave-01 68.27
maya-03 82.1
proe-05 8.73
sw-02 63.94
tcvis-02 23.59
snx-01 41.23

LuxMark 2.0
Bonaire [GPU, 12, 1075 Mhz]
gpu, sala 950
gpu room 555
cpu+gpu room 784 [i7 4800qm]

 

I

Wow those are with just a 4800mq? amd driver is pretty sick. I read they are partner with dassault now.

Quadro for cad no longer applies?

Remember The m6100 on paper have less raw pixel performance compare to 680m/k5000m aswell. So, if this is in 7970m version, the result would be blown to...

 

Any data on how the Quadro K1000M performs, by any chance?

 

I have an M4500 in my sig below. Ran this to see how it would do. Nothing to brag about for sure.

Luxmark Room Scene: CPU + GPU = 53

Luxmark Sala Scene: GPU only = 98

It's still running specviewperf. I will check it in the morning.

Sent from my SPH-L720 using Tapatalk

 

I'm still waiting for the bigger one to download, but I just ran LuxMark on my desktop and laptop and these were the results for GPU only. If you want GPU + CPU or CPU data, let me know.

Sala Scene:
Desktop: R9 290 @ 1032/1280 Score: 2165
Laptop: FirePro M6100 Score: 1015

Room Scene
Desktop: R9 290 @ 1032/1280 Score: 1278
Laptop: FirePro M6100 Score: 574

EDIT: I failed to notice that everyone else was only submitting mobile scores. Who would have thought. Only mobile scores on a site with notebook in the name.

Still, it's an interesting point of reference I suppose, and I'm sorry if it was out of place. I think the performance improvement from the M6000 to the M6100 is also pretty impressive.

 

Viewperf 12.0
Viewset Composite Window 1900 x1060 (Looks like I did it wrong?)
catia-04 7.88
creo-01 5.29
energy-01 0.06
maya-04 2.49
medical-01 2.69
showcase-01 0.01
snx-02 7.44
sw-03 10.84

 

I just noticed that what I downloaded was specviewperf 12, and you all are running 11, or so it would seem. Why the older version, and where can I get it if that's what we're using?

 

This thread is a bit old... I didn't notice the version difference.

 

I guess I'll just run the version 12 ones, and if someone wants the 11 and can point me to the download, I'll be glad to run those as well.

 

Alright, on the laptop in my sig, I got the following:

catia-04 39.64
creo-01 33.97
energy-01 0.48
maya-04 34.10
medical-01 11.70
showcase-01 22.36
snx-02 41.49
sw-03 48.89

 

Using the latest driver 337.5 on k5000m, don't have result for previous driver atm ~~
luxmark

room :403
sala :8xx

Clock is same as 680m default, the result seem to be much higher than old 680m results ~~
though I don't think it is a quadro thing. magic driver really?

 

Anyone with the latest generation NVIDIA GeForce 8xxM GPU want to run some of these benchmarks tests? It would be nice to see how 'neutered' these cards actually are compared to the latest pro cards.

 

Post 42 above already has some of these benchmarks for 880M.. Look properly before posting next time

 

Are you sure? I did a Ctrl+F for "880" and couldn't find anything except everyone's laptop configurations in their sigs. The only benchmark tests in this thread for a gaming GPU are for the 680M.

 

680, 780 and 880 are all within about 10-15% in pro apps. they are actually all the same base die, the 770 and 880 just have a few extra cores enabled with a tiny overclock. We are finding the Maxwell 860m is no faster than the Kepler 860 either

my husbands 880m ranges from 5% slower in Catia, MAYA and Lightwave to about 12-18% faster in ProE, Solid and Ensight compared to these old 680M benchmarks and is part of the reason we hanve not updated them ..... they still apply until another architecture change happens or a die shrink possibly.