|Regarding ROCm, AMD's competing framework to CUDA, here is a 9 minute slide-cast by Ben Sander, AMD Fellow and Lead Software Architect for ROCm and HSA, published last year:|
Here he demos HIP, the porting tool for CUDA code, at SC16, the ACM/IEEE Supercomputing Conference last year.
By the way, SC17 runs next week, and AMD will attend with lots of presentations. Hopefully, we'll get some exciting new information about EPYC and AMD in the server space. Here are some of the demos planned, including a couple on machine learning:
- Latest AMD EPYC™ Processor based solutions from our OEMs, Channel ODMs, and System Integrators.
- TensorFlow Deep Learning Acceleration using Radeon Instinct MI25 GPUs on ROCm Software Platform with MIOpen Libraries.
- LS-DYNA® simulations on AMD EPYC processor based Supermicro A+ BigTwin Solution.
- Deep Learning Model Inference (I.E. extraction) from Trained Caffe model using AMD ANNIE model compiler accelerated using Radeon Instinct™ MI25 GPUs on the ROCm Software Platform with MIOpen Libraries on Inventec P47 platform.
- Reverse Time Migration (RTM) Computations on ultra-dense AMD EPYC processor based HPE Cloudline CL3150 server.