Blender Cycles Render Engine Benchmarks With NVIDIA CUDA On Linux

[phoronix]

Phoronix: Blender Cycles Render Engine Benchmarks With NVIDIA CUDA On Linux

Here is a look at the performance of the Blender 3D modeling/creation software with its Cycles Engine when making use of NVIDIA's CUDA API for GPU acceleration. Tests for this initial comparison include NVIDIA GeForce GTX 1000 "Pascal" and GTX 900 "Maxwell" graphics cards.

http://www.phoronix.com/vr.php?view=23358

[roboman2444]

Can we get an openCL comparison as well? I would like to see how the opencl code compares to cuda, and how AMD cards perform as well.

[-MacNuke-]

You can't compare AMD and NVidia here because the OpenCL version of Cycles is not feature complete and highly experimental. And nobody at Blender is really working on it. Mostly CUDA fanatics there.

[Michael]

Can we get an openCL comparison as well? I would like to see how the opencl code compares to cuda, and how AMD cards perform as well.

Not sure OpenCL works with NVIDIA GPUs, oddly. When I checked on one of my GPUs the other day with the way I am setting the compute accelerator, it said the supported options were either 'NONE' or 'OPENCL' so not sure if Blender developers are only exposing CL if it sees an AMD GPU string?

[roboman2444]

Not sure OpenCL works with NVIDIA GPUs, oddly. When I checked on one of my GPUs the other day with the way I am setting the compute accelerator, it said the supported options were either 'NONE' or 'OPENCL' so not sure if Blender developers are only exposing CL if it sees an AMD GPU string?

Yeah, i must've remembered wrong.
Anyway, nvidia only supports opencl 1.1 because they really want to push cuda. Latest AMD supports 2.whatever, and cycles requires 1.2 or above.
It would still be neat to see how the AMD cards compare among themselves.

[-MacNuke-]

NVidia supports OpenCL 1.2 since around April 2015. But as far as I know NVidia has no plans in supporting OpenCL 2.x since it would be more powerful that CUDA. Market leadership at its best... and Blender supports that bullshit...

[microcode]

Total power consumed in Joules or KWh would be a great measurement for Blender cycles rendering. If one is building a renderfarm and concerned mainly about power consumption over the course of a render, it would be highly relevant. (it'd basically be average wattage multiplied by time to complete.

An interesting way to display it, would be normalizing the power consumption chart as a ratio of completion time. Then you get the cool power consumption chart showing the variations, but you also get the normalized result.

[juno]

Something seems to be completely wrong here...
In Cycles, tile size is very important. Maybe this setting is really some kind of a worst case here.

I remember doing that BMW test in less than 60 seconds on Hawaii.


Yes, the Cycles guys are really into CUDA. AMD devs themselves made it happen at all on GCN GPUs by splitting up the OpenCL kernel back in the days, which helped Nvidia, too, btw. So yes, NV should also work on OCL, however better w/ CUDA, ofc.

[Michael]

Something seems to be completely wrong here...
In Cycles, tile size is very important. Maybe this setting is really some kind of a worst case here.

I remember doing that BMW test in less than 60 seconds on Hawaii.


Yes, the Cycles guys are really into CUDA. AMD devs themselves made it happen at all on GCN GPUs by splitting up the OpenCL kernel back in the days, which helped Nvidia, too, btw. So yes, NV should also work on OCL, however better w/ CUDA, ofc.

The blend files used are from: https://code.blender.org/2016/02/new-cycles-benchmark/ and from there these people put their benchmark result files in: https://docs.google.com/spreadsheets...56d095bd#gid=0 That spreadsheet shows the best BMW GPU result as being 3 minutes, 36 seconds for an AMD Fire Pro Duo with split kernel. So not sure you got less than 60 seconds or must have been using a different file...

[torsionbar28]

Hmmm test is not working for me I get the following:

Blender 2.77a:
    pts/blender-1.0.0 [Blend File: BMW27 - Compute: CUDA]
    Test 1 of 4
    Estimated Trial Run Count:    1
    Estimated Test Run-Time:      11 Minutes
    Estimated Time To Completion: 43 Minutes
        Started Run 1 @ 21:48:05Traceback (most recent call last):
  File "/home/george2/.phoronix-test-suite/installed-tests/pts/blender-1.0.0/blender-2.77a-linux-glibc211-x86_64/setgpu.py", line 3, in <module>
    bpy.context.user_preferences.system.compute_device_type = 'CUDA'
TypeError: bpy_struct: item.attr = val: enum "CUDA" not found in ('NONE')

        The test run did not produce a result.

    Test Results:

    Average: 0 Seconds
    This test failed to run properly.