Jump to content

Luxmark v3.1 (Sept 2015) - OpenCL Benchmark


210 posts in this topic

Recommended Posts

LuxMark has gained, during the past year, a good level of acceptance as OpenCL benchmark. It has been used by AnandTech, Tom's Hardware, Vr-zone and other sites as standard GPU computing benchmark.
AMD has used LuxMark as one of the 5 GPU computing benchmarks to present the new HD7970.



DL V3.1 (NEW, added final version of September 2015)

http://www.luxrender.net/release/luxmar ... 4-v3.1.zip



DL V3.0  (NEW, added end of Feb 2015) - due to OpenCL driver (compiler fails) probs (OS X, Linux) on most gpus only the small scene luxball can benched. The two bigger, complexer scenes fail.




Luxmark 3.0 Results Database



Luxmark 2.0 Results Database - you can look / search for already submitted results (by the app, not posted here)

DL V2.0  


Bildschirmfoto 2012-01-12 um 09.45.12 Kopie .png

Bildschirmfoto 2012-01-12 um 21.09.12.jpg

Bildschirmfoto 2012-01-31 um 09.56.55.jpg

  • Like 3
Link to comment
Share on other sites

Added new render.cfg. No speed changes but not overbright result anymore.

File attatched (+ screenshoot were to put) in first posting.

For me submitting result worked. Perhaps some problem like special characters? in name / pwd ? I would use an simple one if it fails.

Never use your main pwds (email & Co) for such cases. So doenst matter if you let it simple.

Link to comment
Share on other sites

GREAT ! First place with Single GPU for npwski






PS: Woudl be nice if you add the CPU + GPU result.

i7 860 + 3x ATI 5870 around 2700 in the results cpu+gpu db,

maybe you bet them or came close with only one GTX 580 + same cpu ?

background: The more gpus the faster has an limit - the cpu is the bottleneck in that case.

So 4 5870 are only very little faster than 3 5870 with same cpu. cpu + systembus + PCIe Bus is overloaded by huge datatransfers. GPUs must wait to get their results back to cpu.

To fire lots of such highend gpus fast enough you would need an cpu + systembusspeed + PCIe speed

which isnt developed already ;)

Link to comment
Share on other sites

Woudl be nice if you add the CPU + GPU result.

i7 860 + 3x ATI 5870 around 2700 in the results cpu+gpu db,

maybe you bet them or came close with only one GTX 580 + same cpu ?

Thank you, Mitch :)

And no, i can't beat "i7-860 + 3x Cypress", because my quad-core CPU not so fast as 3x top-level GPU. Anyway, it's CPU+GPU score:


i7-860 @4100Mhz + single GTX 580 @915Mhz = 1341

Link to comment
Share on other sites

219 = GPU only or CPUs + GPU. Other 9800 GT got 88 gpu only.

If gpu+cpus please post gpu only result too.



4 * AMD 7970 GPU only got 6153 (win, http://www.luxrender...etail/result/87 )

I must buy 90 of 9600 GT to get close to that ? ;)

Also you can see that in such case, many highend gpus must be fired fast with data from cpu, it happens that

gpus+cpus is much SLOWER than gpu only. Thats because the cpu is on the limit communicating with the gpus already and if the cpu has do do own work beside that the overall speed getting very worse compared to gpu only.

This chart (from that 4* AMD 7970 hero) shows that drasticly:



Also the very small improvement from the OS X hero GTX 580 vs GTX 580 + i7 , 1200 vs 1300 shows that already.

cpu is on the limit - already with one fast gpu.

Link to comment
Share on other sites

Hi friends, soon there will be an new beta which has added an third scene - complex with more than 2 Mio triangles to compute.

Its unclear if gpus 512 MB VRAM it should be no problem.

EDIT: V2.0Beta2 (with new scene room ) now available at same DL link first posting.

i added new results section room in the first post. I got 23 with gpu only GT 9600 512 MB.

The results DB website also updated. complex = room scene.

NEW: The results DB now has an search function , where you can search with OS type, scene type,.... so you will see your and similar

results even you are not listed in the TOP 20.




Link to comment
Share on other sites

Dont forget to DL new Beta2 (same DL Link as Beta1, 1. posting) and test new very complex scene room (> 2 million triangles, huge VRAM consumption)

512 MB VRAM is little to less VRAM for good speed, so it will be much speed diff between same gpu with 512 or more MB (640,....).

640 MB VRAM s already enough for that room scene.

Link to comment
Share on other sites


  • Create New...