New Cycles Benchmark

Blender Institute prepared six Blender files for testing Cycles rendering with CPU/GPU, using various settings and design styles but based on actual production setups. On the links below you can inspect the spreadsheet with results, and load the .blend file collection.

The goal is to have an overview of systems that are used or tested by developers of Cycles. We aim at updating it regularly, also when new hardware comes in – and especially when render features improve in Cycles.

Most strikingly so-far is that the performance of CPUs is in a similar range as GPUs, especially when compared to costs of hardware. When shots get more complex, CPUs win the performance battle. That confirms our own experience that fast GPU is great for previewing and lighting work, and fast CPU is great for the production rendering. But… who knows what the future brings.

Feel free to post own stats and observations on this blogpost! Maybe other .blend files should be added?

Cycles benchmark zip (530 MB)

Google doc spreadsheet

-Ton-

 

102 comments 53,301 Views
    • GPU: EVGA 980TI
      Processor: DUAL Intel W3520 @ 2.66Ghz
      Mac OS 10.12.5
      Blender 2.78

      GPU Time: 3:01.76
      CPU Time: 9:05:91

      BMW scene

  1. Gpu: Titan Z x6 (3 card)
    Cooling: Bitspower Waterblock with 6 x 120mm radiator(2 loop)
    Processor: i7 3960X @ 3.3Ghz
    Os: Windows 7 Pro 64
    RAM: 16 Gb
    VRAM: 6 Gb
    Blender Ver: 2.76b
    Tiling: Default

    Note: sometime with different tiling setting can achieve much faster time.

    Result:

    bmw27_gpu

    time: 1:03:95
    max temp: 47

    classroom_gpu

    time: 2:28:53
    max temp: 47

    fishy_cat_gpu

    time: 2:26:76
    max temp: 47

    koro_gpu

    time: 14:21:70
    max temp: 52

    pavillon_barcelone_gpu

    midday:crash
    temp: NA

    sunset:
    time: 3:19:87
    temp: 47

    night:
    time: 5:42:41
    temp: 57

    gpu bench:
    time: 0:26:76
    temp: 47

    • For getting optimal tile sizes automatically, use the auto tile size addon, it is part of Blender already, just needs to be switched on.

  2. These numbers for GPU are great. What does “Titan Z x6” mean though? Can you share the exact type nr (+ brand)

    • The titanZ is an extremely expensive card.

    • the titan z is a dual GPU card so the x6 is just saying they were using 3 titan z cards so they actually had 6 GPUs for the render.

      not sure if the titan z had many variations from different brands. the only ones i’ve seen were from ASUS and EVGA but they both had the same specs as buying directly from Nvidia. http://www.nvidia.com/gtx-700-graphics-cards/gtx-titan-z/

  3. Tried this on 2.77beta2, only the BMW one though
    CPU: AMD FX-8320 @4.4GHz
    RAM: 8GB 1866MHz CL10
    GPU: R9 270 @945MHz
    OS: Arch Linux (with catalyst 15.12 GPU drivers)

    -Times (bmw)-
    CPU: 11:33.20
    GPU: 8:14.71 – Render result is wrong: windshield comes out white

  4. Gpu: MSI GTX 970 Gaming 4G
    Processor: Phenom II 1090T @ 3.6Ghz
    Os: Windows 7 Ultimate 64
    RAM: 16 Gb
    VRAM: 4 Gb
    Blender Ver: 2.76 Fastest (Graphicall Build)
    Tiling: Default

    bmw27_gpu

    time: 4:20:29
    max temp: 66°C

    classroom_gpu

    time: 11:06:14
    max temp: 66°C

  5. Intel i7 3770 3.4GHz
    Gigabyte GeForce GTX760 4GB
    16GB DDR3
    Windows 10 Home

    BMW

    CPU: 15:21:18
    GPU: 6:11:50

  6. Why is the Victor scene not being rendered on GPU?
    because of memory limitations?

    Cheers.

  7. Intel Core i5 3.20GHz
    EVGA GTX 970 4GB
    16GB RAM

    BMW
    cpu = 19:35
    gpu = 06:17

    Classroom
    cpu = 01:01:48
    gpu = 13:25

    Fishy Cat
    cpu =26:21
    gpu = 13:36

    Koro
    cpu = 36:03
    gpu = 59:55

    Pabellon Barcelona
    cpu = 47:06
    gpu = 18:53

    Victor
    cpu = 02:11:30

  8. ASUS
    Windows 10 Pro 64-bit
    CPU:i7-4790K
    Ram: 16326MB RAM
    GeForce GTX 980 vram 4 Gb (without overclocking)
    Blender 2.76b

    Fishy Cat
    Tiles size: 501×230 Sampling – 1000
    time: 12:45:58
    temp: 66°C
    Mem peak:458.35M

  9. Hi guys, my results:

    OS: Linux Mint 17.3 Rosa MATE 64bit , kernel 3.16.0-39-generic x86_64

    PC:

    CPU – Intel i7 5820K no OC,
    RAM – 32GBDDR4,
    GPU – Gigabyte GTX 780Ti GHz-edition

    BMW27:

    CPU – 07:13.83
    GPU – 03:09.64

    Classroom:

    CPU – 21:30.08
    GPU – 09:24.33

    Fishy Cat:

    CPU – 10:52.06
    GPU – 06:18.28

    Koro:

    CPU – 15:44.06
    GPU – 28:15.31

    Pabellon Barcelona:

    CPU – 18:17.35
    GPU – 12:25.60

    Victor:

    CPU – 54:27.02

    For my is result that GTX 780Ti is still better than GTX980 in some results (Fisshy Cat and Koro) more than 50%.

  10. i7-4720HQ@2.6GHz 8 core, 8GB RAM, Geforce GTX 960M
    Windows 8.1, Blender 2.76b
    BMW27 cpu 18:31.62 gpu 12:36.54
    Classroom cpu 56:08.45 gpu 41.24.70
    Fishy cat cpu 25:39.68 gpu 23:18.59
    Koro cpu 34:44.29 gpu stopped halfway with error message
    Pavillon Barcelona cpu 43:17.58 gpu 43:10.64
    Victor cpu lost control over my computer, hardware reset

    • i7-4720HQ @ 2.6GHz 8 core, 8GB RAM, Geforce GTX 960M
      Windows 10, Blender 2.77
      BMW27* cpu 19:30.62 gpu 8:53.91
      Classroom*+ cpu 58:04.94 gpu 31:09.79
      Fishy cat cpu 26:44.51 gpu 25:15.63
      Koro* cpu 36:36.38 gpu 1:21:28.74
      Pabellon Barcelona+ cpu 44:42.30 gpu 36:46.65
      Victor* cpu It does render. But I broke it off with difficulty after 2h and a projected further rendering time of something like 16h. (Don’t use this computer for a chore this heavy, rendering used 6.5GB, so the computer had to start swapping RAM pages to the harddrive and visa versa, which takes ages and invalidates the processor benchmark test as results are heavily influenced, if not dominated, by the hard drive writing, and reading, speed).

      + CPU renders using Blender 2.77a.
      *For comparative newbies like me: Open the blend files as trusted source (rendering requires permission to use the Python code defined in the .blend file). Open the file from Blender’s file browser and check the ‘trusted source’ box in the lower LHS in the file browser prior to pressing the ‘open Blender file’ button top right.
      (If you forget, usually after starting the render you will see a warning on the top line with the option to open the file as trusted. After confirming revert, you can restart the render.)

  11. Hello! Here’s my system:
    – i7 4770K 3.5 GHz
    – 16 GB DDR3 1600
    – GPU – I replaced the 770 with the 970 after going through all the tests with the 770. I’ll note which I used in the results
    + Nvidia GTX 770 4GB (04G-P4-3774-KR)
    + Nvidia GTX 970 4GB (04G-P4-2978-KR)

    I have Hackintosh setup with dual booting:
    – Mac 10.11.3
    + GTX 770 driver was built-in with Mac
    + GTX 970 driver was Nvidia Web Driver 346.03.05f02
    – Windows 10 (using latest drivers for both the GTX 770 and GTX 970)

    Everything used Blender 2.77 RC1 and CUDA on the GPU scenes.

    BMW
    – CPU Mac: 09:58.01
    – CPU Win: 13:24.17
    – GPU 770 Mac: 04:55.25
    – GPU 770 Win: 04:58.60
    – GPU 970 Mac: 04:03.16
    – GPU 970 Win: 04:02.87

    (I didn’t do the CPU on Windows on any of the rest; I’ll update this later if I do)

    Classroom
    – CPU Mac: 30:28.81
    – GPU 770 Mac: 16:56.81
    – GPU 770 Win: 17:18.80
    – GPU 970 Mac: 10:28.43
    – GPU 970 Win: 10:33.68

    Fishy Cat
    – CPU Mac: 14:40.26
    – GPU 770 Mac: 10:29:36
    – GPU 770 Win: 10:52.02
    – GPU 970 Mac: 16:13.78
    – GPU 970 Win: 15:08.56

    Koro
    – CPU Mac: 21:05.10
    – GPU 770 Mac: 42:03.55
    – GPU 770 Win: 43:52.45
    – GPU 970 Mac: 01:06:55.80
    – GPU 970 Win: 01:03:19.85

    Pabellon Barcelone
    – CPU Mac: 24:45.99
    – GPU 770 Mac: 20:21.65
    – GPU 770 Win: 21:30.57
    – GPU 970 Mac: 15:40.58
    – GPU 970 Win: 14:46.75

    (I didn’t finish Victor on either OS because I was just tired of waiting haha)

    It’s interesting that the 970 was better at everything compared to the 770 except for Koro where it took almost 20 minutes longer. I wonder why that was the case.

    Anyway, hope that helps!
    -Nathaniel

  12. i7 5960x gtx780ti windows 10
    victor 46.53.65

  13. Something must be wrong, I don’t know if it’s the same double BMW test that I have tried, but my “score” was 2:23min with a GTX 960 4Gb and 2:53 with an AMD R9 380. The guy with the 3 TitanZ should have been something like 10sec max ?!?!

    • Sorry, I was wrong, I thought it was the old one from blenderartists.org Now it makes much more sense:

      for the BMW: GPU R9 380: 6m36sec

  14. i5-4440, 3.10GHz, Linux 4.4.0, Sabayon (Gentoo binary).
    2x MSI GTX 750 Ti, 2Gb

    Latest Blender from builder: Sun Mar 13 04:29:23 2016
    https://builder.blender.org/download/blender-2.77-861616b-linux-glibc219-x86_64.tar.bz2

    I ran only gpu scenes, because we already know what to expect from i5 non-hyperthread.

    GPU-results:

    bmw27: 04:19.39
    fishy_cat: 12:48.49
    classroom: 14:51.70

    barcelona:
    1- time: midday: 33:23.19
    2- time: sunset: 17:19.51
    3- time: night: 39:42.94

    koro: 43:23.63

    barcelona-sunset was interesting, because while rendering it also used 100% of one core to X-process. Fulle vram was not used, so it is strange behaviour. So I think it did not render as optimal.

    I have been very happy with my hardware, for the horsepower/€ it delivers.

  15. 2xGTX980, Core i7 4790 @4Ghz
    BMW 11:27.91 1:55.11
    Classr 37:38.75 04:44.72
    Cat 15:57.58 7:17.15
    Koro 21:32.35 30:17.09
    Pabel 27:19.10 7:05.55
    Victor 1:17:23.37

  16. AMD FX-8320 OctaCore @ 3.5 GHz
    MSI Nvidia GTX 960 4GB RAM

    GPU result for BMW 6:58.44 min
    CPU result for BMW 38:07.01 min

    Only thing that bothers me is that my GTX 960 is rendering at performance level 3, while it has a level 4 that is not used. My older GTX 660 did use it. I would expect that GPU results could be better at a higher performance level, so what can I do?

  17. Hmm strange, I get 7:04 mins on the BMW scene with an EVGA Titan X SC.
    Well in any case, based on the benchmark results posted here it all comes down to the amount of cores. The more ‘workers’ the quicker the result.

    The dual Xeons may have the highest thread count but if you calculate price vs power you’ll see that it’s far better to put that cash into multiple graphics cards. Amazon lists the Xeon 2697 as roughly $2500. Multiply that by 2 for a dual setup and you have $5000. That same amount gets you 5 Asus Titan Xs or 8 EVGA 980ti cards. That means you can build 2 PCs with 4-way graphics in each machine! Hands down GPU computation will win.

    • It’s not that simple. Many times large scenes don’t fit into vram.

      Also used xeons are not that expensive. Go to ebay.com and search for: “2x xeon workstation”. The first hit was titled: “Dell Precision T5500 PC Workstation 2x Intel Xeon 8 Core 2.4GHz 24GB RAM 1TB ” and costs US $449.00. Almost the same price range applies to european ebays.

      • Also a true observation.
        I have however built many scenes with a fair level of complexity and they rendered just fine on my old Gtx 570, and that was a 1gb card. Add to the fact that Blender now pretty much allows keyframing any setting means that subsurf levels can be animated according to the distance of the objects, or decimate modifiers etc.
        Not to criticize the scene but just as an example, the classroom in these benchmarks has so much detail on the desk you could do a full HD close up render of it. Yet in the current chosen camera angle the desk items account for less than 5% of the shot. That geometry is inevitably wasted. Large scenes speaks of production. Production means planning was involved. Planning would let you know in advance what will be near to the camera and what won’t be so you would know where the detail needs to be and if camera angles change have different level versions of the model on other layers. I don’t mean to sound arrogant but I have a lot of experience in scene optimization and I can count how many of my scenes couldn’t fit in the GPU.

        What you said about used Xeons is also a good point.

  18. Ι noticed there is an autorun script on the classroom scene (I haven’t tried the others yet) what does it do? Why is it there? Is the benchmark compromised if I don’t run it? and why is the file saved without the compositor ticked?

  19. Computer:

    Intel 5820K 3.3 GHz
    64 GB DDR4-2133 MHz
    EVGA Titan X (nVidia driver 364.51)

    OS: MS Windows 10

    Blender version used: 2.77 rc2

    CPU GPU

    BMW 9:28.69 7:56.82

    classroom 29:03.20 22:45.90

    fishy cat 13:09.32 49:39.41

    koro 18:16.93 *crash*

    pavillion 1 29:08.23 *crash*
    2 22:41.22 32:07.85
    3 57:19.09 1:08:53.14

    victor 1:05:14.17 *crash*

    * koro crash:
    CUDA error: Launch exceeded timeout in cuMemcpyDtoH(uchar*) mem.data
    – pointer + offset, (CUdeviceptr)(mem.device_pointer + offset), size)

  20. Intel 4790
    16GB DDR3
    GT 720 Graphics (not really up to rendering, only good for for screen display, so CPU result only).
    Dual Boot Win 10 / Linux Mint
    Win 10 with Blender 2.75 – 13:05.37
    Linux Mint 17.2 with Blender 2.77 – 9.29.53

    Normally I run Windows because I’m not so familiar with Linux, but now I see Blender renders almost 30% faster in Linux Mint. I was so surprised I ran the benchmarks again just to make sure. Or is some of this time due to Blender 2.77 vs 2.75 ?

  21. Sorry forgot to say my post was for BMW 27.

  22. OK I downloaded the latest 2.77 version of Blender for Win 10 and did the run again on CPU. Time was 13:04.13, just 1 second faster than last run using Blender 2.75.

    Both O/S are on the same SSD drive on different partitions, so Linux Mint at 9:29.53 is certainly much faster than Win 10 on my machine!
    By the way full spec for processor is i7 4790 3.6 Ghz

  23. Is there a Python script available that runs these benchmarks in sequence, storing the resulting timings (and other stats)? That might make it easier to run the whole set on a varity of our hardware, making sure the settings are the same each time.

  24. iMac (Retina 5K, 27-inch, Late 2014)
    GPU: AMD Radeon R9 M290X 2048 MB
    CPU: 3.5 GHz Intel Core i5

    BMW27
    GPU: 29:00.80

  25. So far the greatest GPUs are (BMW):

    1- Gigabyte GTX 780Ti GHz-edition
    03:09.64

    2- EVGA GTX980 (I don’t know which edition is. 04G-P4-2982-KR or 04G-P4-2983-KR)
    03:30.49

    3- Nvidia GTX 970 4GB (04G-P4-2978-KR)
    04:03.16

  26. Hi everyone. I want to buy an EVGA GTX980 and I don’t know which one is better. Could somebody tell me which one is better?
    The speed of the render, supporting 4k monitor and being under $1000 are important to me.
    https://www.amazon.ca/s/ref=nb_sb_noss?url=search-alias%3Daps&field-keywords=%22EVGA+GTX980%22&rh=i%3Aaps%2Ck%3A%22EVGA+GTX980%22

    Thanks

  27. Hi again
    I recently bought a used HP Z600 Workstation off ebay with a X5650 Xeon (6 core 2.66GHz, 3.06 GHz turbo), then I got a second X5650 Xeon (+heatsink & fan)also off ebay and upgraded the board to dual processor, 12 cores, 24 threads. Also added 24gb 1333Mhz Reg ECC RAM. Total investment around $600 US. Also included is a GTX 465 GPU which is not much good for rendering but fine for display.

    I’m very pleased with the CPU render times considering how little this system cost! These used dual Xeon Workstations seem great value right now, great build quality too. I have made clean installs of Windows 10 and Linux Mint 17.3 as dual boot. Both O/S loaded with Blender 2.77.

    Windows 10 BMW27 CPU 7min 26sec
    Linux Mint 17.3 BMW27 CPU 5min 48sec

    Linux Mint 17.3 Victor CPU 42min 17sec

    Again I see with the BMW27 times that Linux Mint renders over 20% faster than Windows 10, with everything else being equal. Why is Linux Mint so much faster than Windows 10?

    • You cheek your configuration, because with only one xeon x5650 I have 1min:6sec for BMW in win10 with 16gb ram.

    • Correction:
      You cheek your configuration, because with only one Intel xeon x5650 I have for max 6min and 6sec for BMW27 in windows10-pro with 16gb ram.

  28. Hey Ton,
    I’m sorry it took so long to finally run the performance tests on my problem machine. I made my own spreadsheet based on the linked spreadsheet above, and added a column for multi-GPU results, and an extra column for the GPU tests at the “Magical 160×120 tile size” as mentioned in the 980 Ti bug thread. (https://developer.blender.org/T45093)

    What I find most interesting is that any time there is fur or hair in the scene, the 980 and 980 Ti pretty much suck, but the 780 Ti is just stellar. I hope that these results help add some clues to the mystery. All tests were run on 2.77a. My machine specs are posted in field that pops up over the machine name.

    https://docs.google.com/spreadsheets/d/16IMEGEGDy7OwK3yL3ekmIToZKMwe9boTIBpCED_Ul0U/

  29. Here are remaining Blender 2.77, Linux Mint 17.3 CPU times for my HP Z600 12 core 24GB on the renders missed previously:

    Classroom 17 min 14 sec
    Fishy Cat 8 min 49 sec
    Koro 13 min 04 sec
    Barcelona 14 min 09 sec

  30. UPDATE

    Corrected crashes due to using a single graphics card, both for rendering
    and display, by increasing the TDR timeout value as described at:

    https://www.blender.org/manual/render/cycles/gpu_rendering.html
    and
    http://artificialflight.org/blog/2013/cycles-crash-cuda-tdr-error/

    EVGA Titan X (nVidia driver 364.72)

    Blender version used: 2.77 Build date: March 23,2016

    ________________GPU_(DEFAULT)___GPU_(160*120)

    BMW________________8:09.88_________4:16.84

    classroom_________23:15.64_________8:38.98

    fishy cat_________46:02.47________15:49.32

    koro____________1:42:20.84______1:11:31.46

    Pabellon__1_______34:32.87________29:55.67
    __________2_______33:08.97________19:22.49
    __________3_____1:09:49.75________42:11.52

    Blender version used: 2.77 Build date: April 9,2016

    ________________GPU_(DEFAULT)___GPU_(160*120)

    BMW________________8:45.07_________4:30.56

    classroom_________24:22.00_________8:51.88

    fishy cat_________41:00.99________15:55.34

    koro____________1:54:45.04______1:48:36.85

    Pabellon__1_______25:40.38________27:36.35
    __________2_______34:03.57________20:04.41
    __________3_____1:13:04.44________43:19.95

    SUMMARY

    For most (but not all) files, there was a fairly substantial reduction in
    render time using the “magic” tile size of 160*120.

    From the March 23 to April 9, 2016 build, the render times were about the
    same, or slightly longer; with the exception of ‘koro’; where the gains made by setting the tile size to 160*120 seemed to disappear.

  31. Blender 2.77a, Windows 8.1 64-bit
    i7-6700 3.4Ghz
    16GB DDR4
    GTX780

    BMW27:

    CPU – 11:41.84
    GPU – 04:16.72

    Classroom:

    CPU – 35:36.07
    GPU – 12:50.34

    Fishy Cat:

    CPU – 16:01.95
    GPU – 08:03.66

    Koro:

    CPU – 22:08.33
    GPU – 38:24.63

    Pabellon Barcelona:

    CPU – 27:48.53
    GPU – 16:51.77

    Victor:

    CPU – 1:18:17.36

  32. Here are some times for a desktop build I’m still putting together, based mainly around used server parts including two of those used Ebay $70 E5-2670 Xeon bargains that all the enthusiasts are getting excited about.

    2 x E5-2670 Xeon, total 16 cores, 32 threads, base 2.6GHz turbo 3.3GHz. 64GB 1600MHz Reg ECC RAM.
    Linux Mint 17.3.
    Blender 2.77a
    All times CPU only, no GPU yet.

    BMW27 3min 50sec
    Classroom 11min 18sec
    Fishy Cat 5min 41sec
    Koro 9min 31sec
    Pabellon Barcelona 9min 3sec
    Victor 28min 0sec

  33. CPU: AMD 1055T
    GPU: Nvidia GTX 670
    OS: Win7

    GPU-Times:

    BMW: 05:51.54
    Classroom: 19:35.09
    Fishy Cat: 11:33.55
    Koro: 46:29.70
    Barcelona: 23:40.42

  34. UPDATE

    Computer:

    Intel 5820K 3.3 GHz
    64 GB DDR4-2133 MHz
    EVGA Titan X (nVidia driver 368.22)

    OS: Win 10 Pro

    Blender version used: May 26, 2016 build

    GPU Results:

    BMW: 3:16.52
    Classroom: 7:25.42
    Fishy Cat: 12:29.37
    Koro: 1:12:45.87
    Barcelona 1: 22:44.74
    Barcelona 2: 13:22.01
    Barcelona 3: 33:28.11

    SUMMARY

    For most (but not all) files, there was a fairly substantial reduction in
    render time using the latest Win10 64 bit blender build for May 26, 2016.

    Comparing to HOLISTER, BIGSTU80, and The BEAR’s results, which use non-Win10 and non-GM200 based graphic cards, their render times for ‘Fishy cat’ and ‘Koro’ (in most cases) are substantially less.

  35. Incredible! just bought a used GTX 780Ti on stock blender 2.77a got:
    1:02 on bmw27 (so more then 2 minutes less then cpcat report)
    0:34 on fishycat (I just pressed F12 with the default settings what’s going on?)
    9:42 on classroom (slightly more)
    12:42 on pavilon (slightly more)
    so I wonder why bmw and fishycat test went so quick …!?

  36. CPU: AMD FX-9590
    RAM: 32gb
    GPU: Nvidia gtx 650 (384 cuda cores) / Nvidia gtx 970 (1664 cuda cores)
    OS: Windows 10 64-bit

    1.
    DRIVER: Nvidia 347.25

    gtx 650 + gtx 970
    BMW 03:28:29

    gtx 970
    BMW 04:06:16

    2.
    DRIVER: Nvidia 368.39

    gtx 650 + gtx 970
    BMW 03:31:45

    gtx 970
    BMW 04:09:82

  37. New PC
    ———
    CPU: Intel Core i7 5820K (6 core/12 thread) 3.3 GHZ
    RAM: 32 GB
    GPU: Nvidia GeForce GTX 970 4 GB (x2)
    OS: Windows 10 64-bit
    Blender 2.77a

    Fishy Cat
    13:05.15 CPU
    08:03.69 GPU (SLI Off)
    08:12.99 GPU (SLI On)

    SLI difference is negligible, so further stats are all the same (Off).

    BMW
    09:16.20 CPU
    02:09.87 GPU

    Classroom
    28:16.65 CPU Yikes!
    05:34.53 GPU

    Koro
    17:51.80 CPU
    34:40.50 GPU

    Pavilion Barcelone
    22:14.18 CPU
    08:00.91 GPU

    Old PC
    ———
    CPU: AMD Athlon II X4 620 2.6 GHZ
    RAM: 8 GB
    GPU: Geforce 460 GTX 768 MB
    OS: Windows 8.1
    Blender 2.77a

    Fishy Cat
    01:05:05.59 CPU
    Not enough memory for GPU

    BMW
    43:35.59 CPU
    13:57.96 GPU

  38. ***Replaced 650 gtx with 780 ti***

    CPU: AMD FX-9590
    RAM: 32gb
    GPU: Nvidia gtx 780 ti (2880 cuda cores, 3gb) / Nvidia gtx 970 (1664 cuda cores, 4gb)
    OS: Windows 10 64-bit

    gtx 650 + gtx 970
    BMW 02:04:25

  39. ***edit***

    Should be:
    GTX 780 Ti + GTX 970
    BMW 02:04:25

  40. Mine must be the “budget setup” here, but still I think the best value for money (gets hot as hell though):

    CPU: Core2 6600 2.4GHz
    RAM: 8GB
    GPU: 2x MSI gtx580 3GB
    OS: Ubuntu 14.4 LTS

    All on GPU:

    BMW 3:23.39
    Classroom 9:16.95
    Cat 9:34.70
    koro 56:01.66
    pabellon 11:05.55

  41. CPU: i7-3930k (changed bios to disable hyperthreading, as it’s too costly to run all the time)
    RAM: 32 Gbytes
    GPU: GeForce GTX 760
    OS: Windows 10

    I wanted to test my rig, so I chose to render Victor.
    Time Taken: 1 hour 51 mins
    🙂

    Screenie is here:
    http://i.imgur.com/H2srz7g.png

  42. blender 2.77a

    Windows 10 pro 64bit
    32gb ECC Ram
    2x Xeon e5-2670v3 (22cores 48threads) @2.3ghz
    Gtx 550ti (just for display)

    BMW CPU: 03:23

    Classroom CPU: 09:20

    Fishy Cat CPU: 04:32

    koro CPU: 08:22

    Victor CPU: 23:25

    Barcelona CPU: 07:21

    Windows 7 Pro 64bit
    32gb ram
    i7-3930k @4.5ghz
    Gtx TitanX + 2xGtx Titan 6gb (classic)

    BMW GPU: 01:15

    Classroom GPU: 03:31

    Fishy Cat GPU: 03:12

    koro GPU: 16:41 (wtf??)

    Barcelona GPU: 05:56

  43. CPU: Dual E5-2683 v4 32C/64T
    OS: macOS Sierra

    CPU Results:
    BMW 01:45.16
    Classroom: 04:57.05
    Fishy Cat: 02:29.33
    Koro: 03:21.53
    Pavillion: 04:13.07
    Victor: 13:34.56

  44. Windows 10 Pro 64-bit
    CPU: xeon x5650
    Ram: 16Gb RAM
    GPU: GeForce GTX 690
    Blender 2.77a

    BMW GPU : 01:04
    BMW CPU : 06:05

  45. os: windows 10 pro
    blender 2.77a

    ram: 16 Gb
    cpu: intel xeon x5650
    gpu: nvidia gtx 690

    renderig times for BMW27
    opengl: 00:22.78
    gpu: 01:04.55
    cpu: 06:00.63

  46. Win 10pro 64 bit Version 1607
    cpu: i7-6700K
    gpu: nvidia 980ti Driver 372.90
    ram: 32GB

    bmw27
    gpu: 2:56
    cpu: 10:21

    classroom
    gpu: 06:34
    cpu: 33:59

    fishy_cat
    gpu: 08:04
    cpu: 14:36

    pavillion
    gpu: 11:14
    cpu: 26:28

  47. I forgot Blender 2.78

  48. os: windows 10 pro 64 build 14393
    blender 2.78

    ram: 24 Gb (3x8gb)
    cpu: intel xeon x5650
    gpu: nvidia gtx 690

    renderig times for BMW27
    opengl: 00:24.71
    gpu: 01:10.11
    cpu: 05:55.92

  49. BMW
    I7-4790K = 13:04.83
    GTX580 = 5:54.81

  50. SO = win7
    cpu = i5-4590

    BMW
    19:26.06

  51. SO = Win7
    RAM = 16GB
    Benchmark = BMW
    CPU i7 4790k = 13:04.83
    GPU GTX 580 = 5:45.02

  52. Phenom II 1090T
    6 núcleos 3.2 GHz,
    16 GB RAM – 1333 MHz.

    BMW = 22:27.74

  53. laptop i7-3630QM
    ram 16gb 1600mhz
    ——
    BMW CPU = 22:50.86

  54. Intel I7-3770k 3.9GHz
    16 GB RAM – 1333 MHz
    Windows 10
    (2) GTX-580

    BMW GPU: 3.55.90
    CPU: 34.02.08

  55. Intel I7-3770
    16 GB RAM – 1333 Mhz
    Windows 10
    Blender 2.78
    Geforce GTX-770 3 GB

    GPU: BMW 5:37min

  56. Amd fx-8320e
    16gb ram
    LinuxMint
    15:39.02

  57. Hey Chaps & Chapettes,

    I installed Blender 2.78 and thought I’d check if everything is installed correctly. But opening the BMW benchmark and then hitting F12, I’m getting 34+ minutes with GPU rendering time. CPU rendering is like 4 -5 minutes faster.
    I don’t know anything about Blender and 3D so I can’t tell where things go wrong.

    Could you give me a hint?

    My system : Manjaro Linux w/ Gnome Desktop, Linux kernel 4.8.1, Intel Q9550, 8GB DDR3, AMD RX 480 8GB with Open Source driver, Sandisk Ultra II 240GB+480GB SSD.

  58. CPU: AMD FX-8320e
    GPU: Nvidia GTX-970
    OS: LinuxMint 18 (KDE)
    Benchmark: BMW
    ——————–
    CPU: 15:37.76
    GPU: 04:20:18

  59. Gigabyte GA-X99-UD5 v1.0
    Intel(R) Core(TM)i7-5930K CPU @ 3.50GHz (6Cores – 12 Threads)
    RAM 32GB Corsair Dominator Platinum DDR4 2666MHz (4x8GB)
    2x GPU NVidia Geforce Gigabyte GTX980ti Extreme 6GB DDR5 VRAM
    Windows 10 Pro(64x)

    Cosmos Laundromat (Victor):

    Total Time CPU Render: 01:09:57.17
    Total Time CPU Render: 00:59:44.97 (OC 4.30GHz)
    Total Time GPU Render: Cuda error: Out of Memory in CuMemAlloc

    BMW27:

    Total Time CPU Render: 00:08:59.53
    Total Time GPU Render: 00:01:32.33

    Classroom:

    Total Time CPU Render: 00:28:29.88
    Total Time GPU Render: 00:03:32.33

    Fishy Cat:

    Total Time CPU Render: 00:12:47.78
    Total Time GPU Render: 00:03:09.80

    Koro:

    Total Time CPU Render: 00:17:48.81
    Total Time GPU Render: 00:18:11.16

    Pabellon Barcelona:(Sunset)

    Total Time CPU Render: 00:22:46.73
    Total Time GPU Render: 00:05:38.80

  60. Edit:
    I did my benchmarks in Blender 2.78a

  61. I am thinking of buying a medion laptio – i7 6700, 2.6 -35.ghz, Nvidia GTX960M (2gb), 32MB Ram, 5400rpm, ssd 256 – can anybody help me to know whether this si a good option for blender – i do a lot of texture baking and use open gl a lot for rendering and experimentation

  62. Thinkpad W520
    i7-2820QM @2,3GHZ Nvidia Quadro2000M
    8GB Ram Debian Stretch
    BMW:
    CPU- 10:20.52
    GPU- 24:20.59

  63. HP Envy i7 6700K @4GHz CPU, AMD R9 200 GPU
    BMW CPU: 1:57.78 GPU: 3:48.23

    Blender 2.78a threw an error on GPU rendering, blender-2.78.0-git.def365e-windows64 works.

    • This is incorrect, I was using a different version of the blend file. See below for the corrected version

  64. Dell Precision T7500 dual Xeon X5650 @2.66GHz 12 cores, 24 threads. 24GB RAM.
    PNY GTX 780 OC 3GB GDDR5
    Windows 10 64bit
    Blender 2.78a 64bit

    Just upgraded to the dual 6 core processors and tried the BMW benchmark without trying to optimize the system at all.
    CPU 07:32.73
    GPU 04:25.12

    Not bad considering this whole rig cost me about $400 off eBay to put together.

  65. HP i7 920 @2.67GHz Windows 10 Blender 2.78a
    NVidia 570 GPU
    BMW: 6:57

  66. This is a correction to my earlier post which was using a different version of the BMW file. If a moderator could remove my Nov 26 post that would be good, it’s misleading.

    HP i7 6700K @4GHz
    AMD R9 270 GPU
    Windows 10

    Blender 2.78a release version
    CPU 10:38.16
    GPU crashed

    blender-2.78.0-git.def365e-windows64
    CPU 8:11.61
    GPU failed – just fog

  67. Slow GPU rendering issue resolved. Thanks to citizen iuno at the Arch Linux forum.
    In my previous post the slow rendering was not executed by the GPU, but by my ancient CPU.

    CPU: Intel Core 2 Quad Q9550
    GPU: AMD RX 480 8GB (Gigabyte)
    OS: Arch Linux 4.8.11-1-ARCH x86_64 GNU/Linux (Deepin)
    Benchmark: BMW
    ——————–
    CPU: 34:03:44 (Ancient Intel Q9550)
    GPU: 04:22:97 (Open Source AMDGPU drivers with OpenCL ICD from here https://bbs.archlinux.org/viewtopic.php?id=220090 )

  68. I’m getting poor results with a GTX1060. My old GTX560Ti (384 cores) did the new (harder) BMW benchmark in 8:24. The GTX1060 (1280 cores) managed 3:47, which is only 2.2x faster. Should I expect better than that? I was hoping for 3-4x speedup considering the increased core count and four-generations-newer Pascal architecture!

  69. Blender 2.78

    2 x K80 on Azure:
    GPU: 02:25.33 (Saving 00:01.11)(Best out of 3)

    Gonna test 2 x M60 as well later.

  70. Blender 2.78

    2 x M60 on Azure:
    BMW 2.7
    GPU: 02:14:00

  71. Blender 2.78

    4 x M60 on Azure
    BMW 2.7
    GPU: 01:24:00

  72. Blender 2.78

    4 x M60 on Azure

    Classroom: 03:00.41
    Fishy Cat: 02:56.71

  73. GPU: MSI Armour GTX 1060 6GB
    Cooling: Standard Heatsink & CPU Fan
    Processor: AMD Phenom II X6 1095T Black Edition
    OS: Windows 7 Pro 64
    RAM: 1600MHz DDR3 12Gb
    Blender Ver: 2.78a
    Tiling: Default

    bmw27_gpu
    Time: 04:14.68
    Mem: 140.72M
    Peak: 142.97M

    classroom_gpu
    Time: 11:43.48
    Mem: 298.66M
    Peak: 312.91M

    fishy_cat_gpu
    Time: 08:19.85
    Mem: 464.60M
    Peak: 466.85M

    koro_gpu
    Time: 38:20.46
    Mem: 445.23M
    Peak: 469.16M

    pavilon_barcelone_gpu
    Time: 16:01.35
    Mem: 151.29M
    Peak: 154.54M

  74. WINDOWS_10
    BLENDER_2.78a

    CPU: i7-860@3.8GHz
    RAM: 8GB_DDR3@900MHz
    GPU: MSI_nVidia_GTX970OC@Core:1550/Mem:4000MHz
    DRV: nvidia_v.376.33

    BMW: CPU: 17:15:58″ GPU: 03:39:18″

  75. Im getting some nice speed-ups on GPU 980Ti, using newest Buildbot for MacOS. (I am assuming this would be across different platforms as well)

    Most notably is Fishy Cat, I am guessing this is due to the new Hair BVH additions to 2.78.4 Buildbot?

    https://docs.google.com/spreadsheets/d/13KZ8RrN8yAlLVTQhUuHVJ1Z3PAolWskOgSEHJ2emL9Y/edit#gid=0

  76. System:

    CPU: FX-8350 1866Mhz 32GB
    GPGPU: RX-480 8GB XFX Black Edition
    SSD: Plextor PX-128M5Pro
    Debian Sid
    AMDGPU-Pro OpenCL Stack 16.60

    BMW: GPU Results (3 runs 5:48.70)
    Pavilion Barcelona GPU: (3 runs 20:19.25)
    Classroom GPU: (3 runs 14:52.85)
    Fishy Cat GPU: (3 runs 9:18.14)
    Koro GPU: (3 runs 39:02.31)

  77. i7 3770k @4.6Ghz Noctua NH-D15S
    gtx 1070 x2 w

    BMW gpu = 1:34:55

  78. GPU | CPU |
    GTX-970 4GB & GTX-660 2GB | AMD FX-8150 Eight-Core 3.87 ghz |
    (same CPU & RAM) | RAM-16 GB |
    | |
    bmw27 3:03.83 | 19:50.36 |
    | |
    classroom 7:58.00 | 59:45.46 |
    | |
    fishy_cat 5.40.08 | 24:22.34 |
    | |
    koro 24:31.58 | 32:21.42 |
    | |
    pabellon_barcelona 10:32.76 | 49:36.27 |
    | |
    victor -incompleat- | 1:59:36.24 |

  79. What are we testing here? Cycles has many parameters which affect performance without effecting quality https://youtu.be/dFdaoxcnvCc
    On top of that you have 3 OS’s to choose from !
    And two bit depths 32 or 64
    Really I think it is time to drop 32bit
    I did my own tests back in 2013 and the difference was very large (46 secounds on 32bit 21s on x64, call it 50%)

  80. Blender 2.78c (release)
    4770K Stock
    32 GB Ram
    GTX 1080 Ti Founders Edition (driver 378.78 – released 9th March 2017)
    Windows 10

    GPU when idle sits around 56 C (Corsair Link). Rises to 68% to 70% during renders. Room temp is 23 C

    BMW Scene
    F12 : 2 minutes 20 seconds
    4 Tiles: 2 minutes
    1 tile : 2 minutes 11 seconds

    Classroom
    F12: 5 minutes 7 seconds
    4 tiles: 6 minutes 5 seconds

    Fish Cat
    F12: 6 Minutes 28 seconds

    Koro
    F12: 32 minutes 24 seconds

    Pavallion
    F12: 7 Minutes 49 seconds

    =============
    GTX 1080 Ti + GTX 980

    BMW Scene
    1 minute 32 seconds

  81. A few weeks ago I did some cycles benchmarks with the BMW27 file. I recall the following:

    System 1: Two K5000 Quadros and a Titan Z 700 series was about 35 seconds.
    System 2: Three GTX 1080s was about 25 seconds.
    System 3: One GTX 1080 and four more connected via PCIE expansion tower (5 total 1080s) was about 27 to 32 seconds.

    Need to test again with Blender 2.78c.

  82. I’m using Blender 2.78c in Windows 10. My PC is Intel Core i5 750@ 2.67Ghz with 16Gb DDR3 memory.I have a EVGA Geforce GTX 580 graphics card with 3Gb GDDR5 memory. GPU-Z says I have 384bit bus width 192.4GB/s bandwidth and I am running the latest drivers. I have seen all the benchmark results for the GTX580 both here and on Youtube and I am getting much longer render times. I am sure I have set up the CUDA option correctly in user preferences and render panel but I get nearly 40 minutes for the BMW27 benchmark scene! I left all the other settings in the file alone and ran GPU-Z to check what, if anything, was happening to the graphics card. The readings showed about 400mb in memory (which seems to be the norm) but the GPU load stayed on 0-1% for the duration of the render. Surely this can’t be right – am I doing something wrong?

  83. TEST Benchmark BMW by Mike pan on my new System:
    with Blender v2.78c official – Windows 10

    CPU (i7 5820K – 12 threads 3.30 Ghz) :
    – standard 3.30 Ghz:
    06:58.12

    – overclock 4.45 Ghz (Voltage 1.3) + external cooler
    06:02.58 (OC stable only with Blender alone)

    ########################

    GPU (Nvidia Titan X pascal 2016)
    – with 1 titanX (display) overclock :
    02:09.09

    – avec 1 titanX (solo) overclock:
    01:58.26

    ########################

    with 2 titanX :
    – no Overclock :
    01:09.54

    – overclock at 130% + Cooling unbridled:
    00:54.93

    ########################

    Optimisation :
    – 2 titanX OC + div 4 tile than 12 (Bench defaut)
    00:50.12

    – 2 titanX OC + div 4 tile than 12 (Bench defaut) + quit open program + disable windows theme + disable other screen + minimize Blender :
    00:47.98 << Best time 😉

  84. Blender 2.78c
    AMD 1800X on Asus Crosshair Hero VI (beta bios AGESA 1.0.0.6)
    32Go DDR4 @ 2666mHz
    2 X 7870 GHz Edition
    Homemade watercooling (EKWB)

    ##########################
    BMW CPU @ stock : 05:00:92
    BMW CPU @ 3.9GHz : 04:48.45

    Classroom CPU @ stock : 16:05.31

    Not run with the GPU, the time will be really bad.
    Overclocking the CPU help a little, but is not really interesting on single frame render.

  85. Blender 2.78a on Red Hat Linux
    2x AMD EPYC 7601 (32 cores/64 threads each)
    512 GB DDR4 @2400 MHz
    No GPUs, CPU only

    BMW: 01:04.98
    Classroom: 02:42.08
    Fishy Cat: 01:31.43
    Koro: 03:58.52
    Pavillion: 02:30.17
    Victor: 10:17.17

    Nice to see Blender can really keep all the cores busy.

  86. Blender 2.78a on Red Hat Linux
    2x Intel Xeon Gold 6140 @2.3 GHz (18 cores/36 threads each)
    192 GB DDR4 @2666 MHz
    No GPUs, CPU only

    BMW: 01:25.39
    Classroom: 03:50.80
    Fishy Cat: 02:02.06
    Koro: 03:56.29
    Pavillion: 03:19.27
    Victor: 10:46.45

    These CPUs only have about half the number of cores as the AMD EPYCs in the comment above do, but are nearly as fast and only cost about half as much!

  87. Blender 2.78c on Windows 10
    CPU: Ryzen 1600 @3.4GHz (6 cores – 12 threads)
    GPU: EVGA GTX 1060 SSC
    RAM: 16GB DDR4 @2666MHz

    BMW Test
    CPU – 07:56:06
    GPU – 04:31:31

  88. Blender 2.78c on Mint Linux 18.1
    CPU: I7-5820K 3.3Ghz (6 cores – 12 threads)
    GPU: ASUS Strix ROG 1080 Ti – Gaming OC. (11 GB).
    RAM: 32 GBB DDR – 2133 MHz (4 x 8 Quad Config)

    BMW Test:
    GPU: 01:59:21 (factory OC)
    CPU: 06:13:00 (non OC)

  89. Blender 2.78c on Mint Linux 18.1
    CPU: I7-5820K 3.3Ghz (6 cores – 12 threads)
    GPU: ASUS Strix ROG 1080 Ti – Gaming OC. (11 GB).
    RAM: 32 GBB DDR – 2133 MHz (4 x 8 Quad Config)

    Victor:
    GPU: 22:23:83 (factory OC)

    The 1080 Ti is an absolute BEAST. It rendered the victor scene without even bothering to speed up the fans, or crash or anything, but then again – it does have 11 GB onboard.

  1. Leave a Reply

    Your email address will not be published. Required fields are marked *

     

share this story on