Has anyone built an AI-optimized computer around AMD? Stuff like this? If so, why did you go with AMD given Intel’s AI optimizations like DL-boost? More cores? Lower cost?
exxactcorp.com/AMD-Ryzen-Deep
Conversation
Replying to
Here’s an interesting comparison. You rely on NVidia for the heavy-lift part in either case, but Intel vs. AMD likely shapes the overall system price/performance. Trying to think about the key variables in the design space here.
1
3
I mean AMD CPUs for the general compute part... I’m assuming Nvidia is the only game in town for the AI part.
Quote Tweet
Replying to @vgr
Sadly no one does deep learning with AMD GPUs. ROCm is kind of a joke. I'm hoping vulkan API for compute will pick up with time but right now it's research quality code at best.
Ryzen are great CPUs for CPU heavy tasks though.
2
1
3
I think I last speced and bought a desktop in 2006. Never built one myself. Considering it as a project this year, and since I’m not a gamer, trying to think of other kinds of workload to solve for. Running deep learning apps at home seems to be an interesting thing to try.
3
6
Replying to
Sadly no one does deep learning with AMD GPUs. ROCm is kind of a joke. I'm hoping vulkan API for compute will pick up with time but right now it's research quality code at best.
Ryzen are great CPUs for CPU heavy tasks though.
1
2
Replying to
My understanding is that DL-boost is mostly for inference on the CPU (which can still be cost effective for certain workloads). Users of these sort of machines will be doing training on the GPU, and in that case the raw extra core count is more useful.
Replying to
If you mean on the CPU side, sure, plenty of people use AMD. It's better.
Ryzen > intel for most code. The intel wins are on stuff no one uses. On-CPU DL stuff is for inference, not training, so who cares, and almost nothing uses AVX-512 because like 7 cpus have it.



