Page 51 of 140 FirstFirst ... 41495051525361101 ... LastLast
Results 1,001 to 1,020 of 2793
  1. #1001
    Hi, I just changed the tiles to 256x256 and I get

    GPU CUDA - 00 min 58 sec 16

  2. #1002
    What other times are you comparing? From the beginning of this thread? That was a few versions ago. My times:
    Win 7, GTX 580 with i5 2500K @4326MHz, blender 2.65a - 52 seconds
    Win 7,GTX 580 with i5 2500K @1600MHz, blender 2.65a - 59 seconds
    Win 7, GTX 580 with i5 2500K @4326MHz, blender 2.66a - 55 seconds
    Win 7, GTX 580 with i5 2500K @4326MHz, blender 2.61 - 46 seconds

  3. #1003
    Member Zoltar's Avatar
    Join Date
    Feb 2013
    Originally Posted by knacki View Post
    Almost same config but with a Gainward Titan GTX instead:
    New system nothing done except oc for cpu:

    CPU: 2:17 min
    GPU: 0:34.81 @ 512x512
    Have you had a chance to fiddle about with the GPU Boost settings? It would be interesting to see how much headroom you've got for getting more performance out of it.


  4. #1004
    Member YAFU's Avatar
    Join Date
    Mar 2013
    Originally Posted by Sielan View Post
    What other times are you comparing? From the beginning of this thread? That was a few versions ago.
    You can see dates and which versions have been used in the spreadsheet:
    So it's good that in addition to posting the results here, fill form data (using the default settings in the .blend file for benchmark):

  5. #1005
    Originally Posted by Zoltar View Post
    Have you had a chance to fiddle about with the GPU Boost settings? It would be interesting to see how much headroom you've got for getting more performance out of it.

    Just made a careful (no experiences in overclocking) try with evga's precision tool at 106% power.
    Turned out close to 33 seconds.
    Never came close to the 30 seconds mentioned here before.

  6. #1006
    Hi everyone,

    I just recently started using Blender on my old notebook and will finally use the opportunity to buy something new. So this benchmark is really helpful for me. Or rather: your results of the benchmark -- Thank you!

    Now hopefully this will be helpful for someone as well:

    As already discussed, since there are so many different builds and tile-size settings used, it is kind of hard to compare results. Not to mention most people just stated the name of the graphics chip, w/o memory or frequencies. Therefore, I thought I would try to make some statistics. Unfortunately not for every card there are enough results, so obviously not everything will just average out. But it might still give you a good idea.

    Just the major single-GPU GTX cards (Ti and non-Ti versions counted as the same). No SLI! (too few similar entries)

    ('Ntot' is for the total number of entries for each card/chip. 'std' is for standard deviation.)

    [I have a feeling this is not necessary, but anyway: what you see is a histogram of the GPU render times as taken from the results sheet. For each card the percentage of times in each 'bin' is given by the height of the bar (different color for each card) in that bin. the bins are all 20 s wide (see gray dashed lines). so everything from 20-40 seconds is in the first bin, everything from 40-60 in the secon and so forth. so, for example, for the 580 approx. 28 % of all entries were between 20 and 40 s, ~62% between 40 and 60, ~8% between 60 and 80 and ~2% between 80 and 100 (oO). (Of course, taking 10s bins would be nicer. But the histogram is already ugly enough and for 10s you would need much more data . Either way, the most interesting part is probably the table.]

    I know it's kind of hard to read -- try to focus on one color at a time . Keep in mind that for some chips there are only very few results! (But those 8 entries for the GTX 480 look really interesting! =))

    In general, you might want to compare the median values as they compensate for outliers. But again, they will only make sense with a certain amount of data (so not from 4 values).

    I did not have a chance to read the whole thread but I think there are much more data posted in the replies than in the results sheet. Especially of people who posted new times with optimized settings. That would be really helpful to have in the results table.

    Now for the discussion of those results: Interestingly, it looks like every single 5XX card wins over its supposedly superior 6XX 'sister-model'. Also the results for the 480 are interesting. Though there are only 8, they are very consistently fast.
    The fastest times for 680, 670, 580, 570, and 480 are very close -- what do we learn from that? Assuming the fastest results are from people that used optimal settings/builds, is it that Blender does not use the full potential of the cards? Or, to put it differently, for the type of calculations that Blender needs, you do not need the extra features that the, say, 680 offers over the 480? (This is from someone who has almost no idea how a graphics card works).

    Again, hopefully someone will find this useful & I would be curious to hear your conclusions!


    ps, what a sweet job that was to reformat all those weirdly entered times! Mike -- are you interested in getting the table to update yours? (just GPU times, though) also, there should be a way to force people to enter their data in the right format

  7. #1007
    Member JuanJosé Torres's Avatar
    Join Date
    Nov 2011
    Madrid, Spain
    The fact that 5xx are faster than 6xx for GPU rendering has been known for a very long time. CUDA performance took a step back in the last generation, and it has nothing to do with Blender.

    I wouldn't go as far as recommending a 4xx over a 6xx, no matter how cheaper they might be (if you can find a 4xx nowadays in mint condition, which is probably not an easy task). There are two other major factors to be taken into account when selecting a new GPU for Cycles: video memory and power consumption. In both of those areas 6xx excell over 4xx. In particular, Interactive render preview needs a considerable amount of video memory, and I have already found scenes that could be tile-rendered on a GTX 560 (1250 MB), but wouln't render in interactive viewport mode because of memory limitations (Cycles must be able to fit the whole rendered scene inside the GPU's onboard memory). However, that same scene could be rendered in the viewport with a GTX 580 (3GB) without any problem.
    Last edited by JuanJosé Torres; 26-Mar-13 at 06:20.

  8. #1008
    My little contribute to the thread:

    i7-3930K @4,50 GHz
    32 GB DDR3
    Two gtx 680 SLI
    Blender 2.66.1 x64

    time:36.26 @512x512.

  9. #1009
    I'm on a laptop
    CPU: Intel i7-3630QM @ 2.40GHz
    GPU: NVIDIA GeForce GTX 670MX 3gb
    RAM: 24gb
    OS: Windows 7 64bit

    I left the tile size to 120x67

    CPU render time: 05:00:97
    GPU render time: 02:49:84

    It's kinda slow for a GTX 670MX, isn't it? I mean, compared to other GTX 670 that took between 40sec and 1:20... or the MX version on a laptop is supposed to be that slow?

  10. #1010
    Join Date
    Aug 2012
    Warsaw, Poland
    CPU: Intel i7-2600K @ 3.40GHz
    GPU: NVIDIA GeForce Evga GTX 670 SC 4gb
    OS: Windows 7 64bit
    Blender: 2.66a official release
    Tile size: 240x270

    GPU x1: 01:03:91
    GPU x2: 00:33:86
    GPU x3: 00:26:00

  11. #1011
    i cant render any way...!!!
    when i press f12 nothing render
    even i append car or other object to new file
    Last edited by firewhale; 30-Mar-13 at 23:19. Reason: more detail

  12. #1012
    Join Date
    Jan 2012
    New GTX 650 TI BOOST 2 GB V-Ram just released by nVidia. It is a GK 106 based on the TITAN architecture. It has GPU boost and a 192bit memory bandwidth.

    Here you can have a look at a Gigabyte windforce model ( silent ).

    If someone gets a hold of this please post the results here as this might be a best bang for buck at around 170 $.

  13. #1013
    Member Kelly Bellis's Avatar
    Join Date
    Jan 2013
    Ellsworth, Maine
    @reC - I very seriously considered the 650 Ti Boost, but decided (just last night) on the 660 SC since Tiger had it at $190 after 1) an instant rebate and 2) a mail-in $10 rebate. It and the 500 Corsair PSU probably won't be here for a week..

    FWIW, the test done today :

    CPU: Intel i7 860 @ 2.8GHz
    GPU: AMD Radeon 5770 1gb
    OS: Windows 7 Pro 64bit
    Blender: 2.66.1 official release
    Tile size: unaltered from original test, 120x, 67y
    GPU1: n/a
    CPU: 05:30.09

    Date and time of my entry in the spreadsheet: 01/04/2013 02:44:02

  14. #1014
    Looks like nvidias beast (Titan) is a great invest for cycles rendering.


  15. #1015
    CPU: i7-2600k @ 4.2 GHz
    GPU: EVGA GTX 670 FTW 2gb x2 SLI
    Driver: 314.22
    OS: Windows 7 Pro 64 bit
    Blender 2.66a official release
    Tile size: unaltered
    -GPU x2
    -GPU x1
    -CPU (open shading language disabled)
    -CPU(open shading language enabled)

  16. #1016
    OK this is ridiculous, Blender 2.63.17 is now twice as fast as the latest 2.66.5 just installed.

    CPU i7-3770K
    GPU GTX 560 Ti
    driver 304.84

    GPU rendering time with Blender 2.63.17: 1m02.95
    GPU rendering time with Blender 2.66.5: 2m02.05

    By tweaking everything I get optimal rendering time with 9x 320x180 tiles, but thats only 1m59 with the latest Blender.

    How can the first cycles version be twice as fast as the current. And don't give me any BS about tile settings or whatever: I tried them all.

    Blender 2.63.17 is simply twice as fast. Is it so hard to get it back to that inital performance??

  17. #1017


    My time:CPU: Intel Core 2 Quad 2.9GhzGPU: Nvidia GTX 570OS: Windows 7 64bitTime: 0 min 53 seconds (GPU - CUDA)Time: 0 min 53 seconds (GPU - OpenCL)Time: 8 min 9 sec (CPU)Hello all,I have a GTX 680, why does it show as GTX 570 in the benchmark? And shouldn't my card perform better? Thanks ahead for any help

  18. #1018
    Member scyguo's Avatar
    Join Date
    Jul 2002
    Poland, Bystra
    @edgarishankulov: You talking about text area bottom-right? is not Your's time. It's message text from author - MikePan. Yours is in render window upper-left corner.

    btw my times:

    GPU (tiles 480x540) 00:31.49 scr:
    CPU (tiles 16x16) 01:25.71 scr:

    spec: GTX590, i7 [email protected], 64GB DDR3 @1600
    os Linux Mint 14 (Nadia) Cinnamon Desktop
    cuda toolkit 5.0

  19. #1019
    @scyguo: Oh I see, thank you. In this case my situation is even worse. I have:
    Core i7 3930k [email protected]
    16gb RAM
    Win 7 64bit ultimate
    GTX 680 with about 1500 CUDA cores

    But my time is only 56 seconds GPU.
    Is something wrong? Please Help

  20. #1020
    Join Date
    Jun 2010
    56 seconds for a gtx 680 is a good timing imho
    Blender Cycles: Lighting and Rendering Cookbook -
    ArtStation -

Page 51 of 140 FirstFirst ... 41495051525361101 ... LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts