Is there any information on how this thing performs on actual AI workloads? It is priced similarly to a dozen of DGX A100s, but is it faster on something like t

Is there any information on how this thing performs on actual AI workloads? It is priced similarly to a dozen of DGX A100s, but is it faster on something like training big transformer models(such as CLIP or GPT-3)?