3 ms·
Seems like for an AI data center you’d really want a dedicated AI chip, no? The current M chips have way more CPU and GPU than inference workloads need or want,
by forrestthewoods 2y ago
Seems like for an AI data center you’d really want a dedicated AI chip, no? The current M chips have way more CPU and GPU than inference workloads need or want, right?
- wmf 2y agoWhat if they are using the GPU for inference.
- forrestthewoods 2y agoStill a waste of CPU. Seems hard to believe that workloads would saturate both GPU and NPU. It’s got to be sub-optimal. I’m just not sure by how much. My ill informed spidey sense says by quite a bit. But I don’t know which is why I asked!