Any comparison with Nvidia ?
I think earlier in this thread. It was said that H200 with 8 card cluster can do 3800 tps.
So, maybe Ascend-910C can do around 2000 tps, possibly more with optimization (910C should be > 2x of 910B in compute). That matches the 910C being 60%+ inference performance of H100 theory. H200 should be better than H100 in even inference since it holds more memory & has much higher memory bandwidth. These things don't matter as much for inference as training, but I'd imagine still matters a little bit.
Baidu is in trouble imo.
Tencent has now fully integrated DeepSeek into search and assistant across all its platforms. Why would anyone still use Baidu to do the searching if Tencent search uses the same technology?
