5/11/2023 0 Comments Rocm vs opencl benchmark![]() VRAM allocation and GPU usage flip between 0% and 30%. PlaidML does allow me to see how much VRAM im using where as TensorFlow pre allocates 96% and you cant see any deeper than that. When i saw this happening i just quit and Google will explain HBCC far better than I. This seem to much more of a thing on Windows than Linux and im not sure if its because of Vega's Memory too and swapping between this and VRAM. TensorFlow will very happy to throw a OOM error and quit. I have yet to find out exactly what the problem is. With SAE its immediate, with H128 its hit and miss. Same error as i do with DFL-H128 at batch size 24. ![]() I am also unable to test DFL-SAE as get the This has only been ran by me on my (poor) dataset. Interesting right? There are some other points id like to make but first id like to remind you that Please allow for a slight margin of error, these things are never perfect.īatch size: ROCm EGs: PlaidML EGs: Windows EGs: I had to change the Dlight output to 128 in order to get any meaningful data. The same TensorFlow code that it uses for Nvidia and ran some tests.Īll of these tests were ran on the same data using the FaceSwap default configs except for Dlight. I then made the changes required so that FaceSwap would use I booted up a fresh Linux enviroment, installed the ROCm stack with the modified TensorFlow andĪ recent git copy of FaceSwap. If one is written with OpenCL and the other with low-level primitives there must be a difference. What would be the performace difference between PlaidML and Tensorflow on AMD? This is one of the ways FaceSwap developers has managed to allow you to use any hardwareįor ML and that is something we should thank them for. PlaidML is a OpenCL v1.2 ML libary but does not conform to my previous 'v1.2 and CUDA' rant. This has allowed (i think this is the reason anyway) TensorFlow to be built ontop of MIOpen soįor those of you that aren't aware FaceSwap uses PlaidML for AMD and TensorFlow for Nvidia. In short HIP is cuda clone designed to allow easy migration from CUDA, perform a find+replaceįrom 'cudaFunction()' to 'hipFuntion()' and you're 90% done. Part of this software stack is MIOpen, a low-level ML libary and a new runtime API "HIP". In the attempt to catch up they have built the Radeon Open Compute stack What i did, whether is was based on false pretenses or not.ĪMD has been trying to catch up in the ML game, how well they are doing is something you canĭecide for yourself. It doesn't matter though, i'm mearly explaining why i belived I was a very ignorant observation and I'm sure there's someone reading this than knows farīetter than I and can say why this is. Now i do not know what the benefits of migrating from OpenCL v1.2 to v2.0+ are if thier are any atĪll. With Nvidia OpenCL, but would then also have a seperate CUDA path. That a program would be written in OpenCL v1.2, what looked to me as a way of staying compatible My belief came from looking at software requirements, CUDA or OpenCL.ĭue to Nvidia pushing CUDA for the past 10+ years they stopped thier OpenCL support at v1.2, whileĪMD and Intel continued to support up to v2.2 and v2.0 respectively. Jen-Hsun Huang stated 10 years ago that Nvidia is a To CUDA, which shouldn't surprise anyone. What i do mean is that the entire software stack is biased The code i looke at was very easy to follow. I do not mean that the developers have done a 'bad' job of coding, i think FaceSwap is excellent and RTX GPUs aside, i was very much in the belief that this was a software rather than a hardware issue. To some people who cannot / do not wish to buy another GPU. What i already knew, the comparison in performace as terrible. Looking at the forums i saw multipule posts that essentially summed I have just started looking at DeepFakes and ML and have an AMD GPU, a position im sure people The code i've written is not the best and would require some cleanup before anybody programming This was done purely as an 'I wonder if'. Nvidia is the king of ML, this isn't me attempting to deny that. This was a small project by myself, a non-expert in either Python or Machine Learing (ML).
0 Comments
Leave a Reply. |