As I run the exact same data science GPU code on identical hardware software configuration except for the change from Mathematica 11.2 to Mathematica 11.3 and see my neural network performance go from 295 seconds on 11.2 to 2038 seconds on 11.3.

When training the net, could you set the BatchSize and MaxTrainingRounds options to the same value in both versions, and then report the training time again? (we changed the heuristic for choosing a value of MaxTrainingRounds which sometimes uses more training rounds for extra accuracy. And: the inputs/s is a better indicator of speed than the total training time as it doesn't depend on MaxTrainingRounds). If there is still a speed difference, could you give information about the GPU you are using, the output of $Version and if possible the training script to reproduce this training?

And then I see that Mathematica 11.3 still does not support current XCode LLVM/GCC compiler or NVIDIA for CUDA tools

I am not sure exactly what you mean by this. Could you expand a bit about which functionality you are referring to that doesn't support "current XCode LLVM/GCC compiler or NVIDIA"?

I must have mixed up my versions ... I was certain that I just tried this in 11.3 ("ReleaseID" -> "11.3.0.0 (5944637, 2018030701)") and it did not work. Now I tried it again and TargetDevice -> "GPU" seems to be working fine.

Stefan, this solution worked for me. Also, I see the correct results when evaluating SystemInformation[]

Hi, I'm using macOS 10.13.3 and I have the following results for Nvidia and CUDA:

Sebastian, thank you for the information. Yes, the BatchSize seems to default to a much smaller value in 11.3 from 11.2, raising it from the default of 256 to 700 gets the timing down to 247 seconds in my test example. The MaxTrainingRounds seems to be the same value of 10 between the versions of MMA. The 247 seconds on MMA 11.3 is about 50 seconds faster than the results on 11.2. Raising the BatchSize above 700 (until the kernel crashes around 900) does not yield any improvement in speed. Is there anyway MMA can detect the limits of the GPU and not just crash?

Still evaluating error results.

The new ImageRestyle function continues to fail however. Is this a problem with CUDA or a problem with Neural Network package? Stefan? Sebastian?

Attachments:

I'm not sure where SystemInformation is getting its "Tookit version" info but it's complete bogus. Please use the information returned by CUDAResourcesInformation[] instead.

The "core count" shown in SystemInformation is also misleading, but it doesn't have any effect on computations. I'll see if I can get this fixed.

After doing the update you describe Stefan, I am still not sure I have a configuration that works. I have attached 4 pictures that I hope give a view of what I am seeing. The mandelbulbGPU example from your documentation still fails. It still seems to be reporting: "Name" -> "CUDAResources", "Version" -> "10.5.0" This is on macOS 10.13.2 with MMA 11.3.0

Attachments:

I meant: "What about support for AMD cards". Sorry for the confusion.