Watch Desk posted an update
Huawei’s Ascend 910B chips reportedly reached 93% of an Nvidia A800 cluster’s performance on mixture-of-experts training, based on customer disclosures discussed by HelloChinaTech.
Why it mattersThe same report says the chips were about five times slower than Nvidia’s H200 in long-context sampling. That is a useful distinction: chip performance can look competitive on one workload and lag sharply on another. These are figures reported by HelloChinaTech, not a broad independent benchmark. For people weighing AI hardware, the practical question is which workloads the numbers actually represent.
Discuss: When comparing AI chips, should buyers prioritise strong results on a particular workload or performance that holds across a wider range of tasks?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.