5 ms·
There is a newly announced Chinese alternative to those GPUs - the Biren BR100. Fabbed at TSMC. (I guess we are about to see if SMIC can do their 7nm at scale.
by throwaway4good 4y ago
There is a newly announced Chinese alternative to those GPUs - the Biren BR100.
Fabbed at TSMC. (I guess we are about to see if SMIC can do their 7nm at scale.)
https://www.nextplatform.com/2022/08/25/china-launches-the-inevitable-indigenous-gpu/ https://www.nextplatform.com/2022/08/25/china-launches-the-i...
CHINA LAUNCHES THE INEVITABLE INDIGENOUS GPU
…
Here is how Biren says it stacks up on various machine learning workloads, pitting the BR100 against the Nvidia A100:
…
We presume this is for AI training workloads, not inference. The average speedup over the A100 is around 2.6X. It is not clear if the Nvidia machines were using sparse matrix features, which doubles the throughput, or not. Our guess is they were not.
- curiousgal 4y agoDoesn't really matter if they're cheaper and more easily available.
- throwaway4good 4y agoThis article has more technical details: https://www.servethehome.com/biren-br100-gpu-for-datacenter-compute-and-ai-workloads/ https://www.servethehome.com/biren-br100-gpu-for-datacenter-...
- LeifCarrotson 4y agoI'm optimistic that this diversity will unseat the CUDA framework's dominance in high-performance compute. OpenCL, AMD ROCm, and now "Birensupa" will compete with CUDA; hopefully leading to reduced dependence on any individual manufacturer's proprietary APIs.
- mrguyorama 4y agoDoes anyone actually use ROCm? It seems barely supported by AMD, with minimal development, and difficult to run in nearly any scenario. It's been a massive disappointment for someone who generally prefers AMD gpus.