LuxMark: OpenCL Benchmark based on LuxRender

OpenCL logo

LuxMark OpenCL Benchmark

LuxMark is an OpenCL benchmark based on LuxRender. LuxRender is a physically based and unbiased rendering engine. Based on state of the art algorithms, LuxRender simulates the flow of light according to physical equations, thus producing realistic images of photographic quality.

You can download LuxMark HERE.

I quickly tested LuxMark on my dev machine. I have two OpenCL platforms: the first is from Intel and the second from AMD. LuxRender selects the first platform so I only benchmarked my quad core X9650 CPU @ 3GHz. Score: 1439 points.

On my test bench, I plugged a Radeon HD 6950 (primary card) and a HD 5770 (secondary card). LuxMark is able to use all available OpenCL GPUs and here is the OpenCL GPUs-only score: 7819 points.

LuxMark OpenCL Benchmark - Radeon HD 6950 + HD 5770

LuxMark OpenCL Benchmark - Radeon HD 6950 + HD 5770

I will use LuxMark in my next graphics cards reviews.

And you my friends, what are your scores?


44 thoughts on “LuxMark: OpenCL Benchmark based on LuxRender”

  1. sirpalee

    MY gtx 480 scored 7483, but with work group size 64? That’s kinda odd…

  2. Promilus

    Only 493 points on HD4670 (OCL GPUs only)
    768 points on HD4670 + A64 X2 2.8GHz as OpenCL devices (OCL CPUs+GPUs)
    405 points OpenCL CPUs only (same as above)
    369 points native CPUs only
    I guess it’s true that OpenCL CPU code is almost as good as hand optimized.
    CPU+GPU does not scale so good.
    Second scene went terribly wrong. Low speed and visual artifacts.

  3. Psolord

    I’m getting 15719 on GTX 570 SLI.

    SlI was disabled but both gpus were at 100% load.

  4. Xeus32

    I have test on my system Q6600 @ 2.6ghz and 9600gt
    with this result:
    *321 gpu + cpu
    *1271 cpu
    *320 gpu

    Or my video card is very very bad or there is some problem 🙁

  5. nou

    you can chage a OpenCL platform. go into scenes/luxball subdirectory, open render.cfg and change opencl.platform.index = 1

  6. mareknr

    I had GTX460 GLH SLI – 7909. It’s really low againnst your HD5770. Maybe my CPU is weak – Phenom X4 940 BE 3GHz.

  7. mareknr

    Second round: GPU(GTX460 GLH SLI) – 8090, only CPU(Phenom II 940 BE 3GHz) – 1250.

  8. mareknr

    Sorry for all these comments. I didn’t noticed your primary card so my result isn’t so bad. 🙂

  9. spiked_mistborn

    My score was 2295 on my Radeon 5770 @ 940 / 1333 core / mem. System is Athlon64 X2 2.3ghz, Win7 Pro x64, Cat 11.2, Stream 2.3. Probably held back slightly by the slow cpu, but cpu usage is low while the benchmark is running.

  10. WacKEDmaN

    GTX470 @ default clocks
    Score: 6102

    GTX470 @ 800/1600/2000/1.075v
    Score: 7848

    GTX470 + 9400GT caused driver to reset multiple times, and didnt finish the benchmark.. prolly due to the miss matched timings

    cant get Intel OpenCL running on my Q6600 or Q8200 (no sse4.1)

  11. akira

    i7 2600k@4.5GHz
    MSI GTX460 Hawk@815MHz
    Sparkle GTX460@700MHz
    Win 7 x64

    LuxMark 1.0 64bit

    OpenCL GPUs: 8278

    Native CPUs: 4036

    Can’t find OpenCL CPU devices.

  12. Hugie

    LuxMark V 1.0, OpenCL GPU Only, HDR, Geforce 240 GTS.

    Thats 498 points… not so much.. but for a 240.. i think its ok 😉

  13. MfA

    OpenCL GPU Only = 5092
    GTX 460 1GB @950/1900/4600MHz

    Native CPU = 2393
    Q9650 @4.00GHz

  14. gopu

    It hangs on my computer.
    In other, a OC CPU is faster than the GPU… so where’s the revolution?

  15. Komar

    Will test this evening back from office …

    Think my OCed 5970 will rule :d against my q9550 c1 @.3.93ghz

  16. Zibri

    CPUs: 3459 (Core i7 950 @ 4.2 ghz)
    GPUs: 8795 (2x GTX 460 SLI (820/1640/2025)
    GPUs: 9020 (2x GTX 460 SLI (840/1680/2025)

  17. mirror

    Hi. could you possibly test an nvidia card as well ? it would be nice to see a comparison of similar budget nvidia-amd cards

  18. Stan Zaske

    I got 4263 With my Palit factory overclocked ( 800 MHz) GTX 460 1 GB card. OpenCL “CPU’s only” didn’t work and “Native CPU’s” works great at loading all cores on my X6 1055T for a score of 1976. Thanks for not requiring me to register to comment. That means a great deal to me.

  19. Miles

    I got 13029 points on the GPU only test using two 6970s (910 MHz core clock on each). 🙂

  20. Promilus

    The results from my previous comment were from WinXP. I did a try on ubuntu x64 (with few libs I had to install)… and
    Luxball hdr
    OpenCL CPU 403
    native CPU 494 (!)
    OpenCL GPU 481
    OCL CPUGPU 524 (!)
    Scene Luxball
    OCLGPU 54
    natCPU 104
    Ubuntu seems to have much higher results on native CPU while much lower on hybrid cpu&gpu (in comparison with WinXP x86). OCL GPU has about the same speed and OCL CPU too.
    Also from what I saw on polish sites GF1xx is the best while older architectures aren’t just as good. G80, G92 & G94 (compute capabilities 1.0 and 1.1?) are even slower than their AMD counterparts. What is weird… OCL GPU workgroup size is half of the recommended (64 against 128 iirc).

  21. Athlonite

    well it gets 4702 on 2x HD5770 with cat 11.2 get even worse if I try it with both GPU and CPU 708 which is weird you’d think with an extra 4 cl cores it’d get better than just two gpu’s

  22. fellix

    Fermi is benefiting here from its coherent L2 cache. The older generations (G80, GT200) doesn’t have this capability, and on top of it the register pressure is much larger on them, i.e. can’t keep enough in-flight threads to hide the memory access latency. Fermi’s doubled register file and read/write L1 cache helps to keep thread stalling more gracious.
    AMD is a bit better here thanks to its much larger register file, but still behind Fermi due to lack of coherent caching.

  23. jod

    This test seems to be very dependent on the PCI express. I’m getting the same score with a GT240 16x than a 6780 at 2x…

  24. SebbesApa

    1265 gpu only (Mobility radeon HD5830)
    1094 cpu only (Intel i7 720QM)
    1106 native cpu
    2377 cpus+gpus

  25. SebbesApa


    Forgot to mention. With Catalyst 11.2 and Intel OpenCL SDK x64 (January 31th)

    Don’t know if i was using the Intel SDK x64 or ATI Stream during cpu though…

    Also got artifacts during cpu only and cpus+gpus on the “LuxBall HDR”.

  26. pablo

    ok, I finally fixed the OpenCl Cpu issue, so:
    OpenCl 1.0 GPU: 868
    OpenCl 1.1 CPU + GPU: 1288
    OpenCl 1.1 CPU: 1295
    Native CPU: 1329

    Core i7-720, GTS 250M

  27. alex_step

    LuxMark v.2.1beta2 Scene: Sala
    CPU Q9300@2.5GHz
    GeForce GT 240 [GPU, 12, 1340MHz]
    Native CPU: 123
    OpenCL1.2 GPU: 112
    OpenCL 1.2 CPU+GPU: 228

    Something wrong?..

  28. Tugrul_512bit

    Luxmark v2.0, Luxball: 14050 points


    HD7870 @ 1200/1375 + FX8150 @ 3900

Comments are closed.