3 ms·
I don’t think they are very comparable. This is still very much a conventional supercomputer despite the marketing. Here’s a fun comparison of total silicon wa
by jagger27 5y ago
I don’t think they are very comparable. This is still very much a conventional supercomputer despite the marketing.
Here’s a fun comparison of total silicon wafer space used.
A Cerebras die is 46,255 mm^2.
1,500 Milan 64-core CPUs * 8 compute chiplets each is 1,004,832 mm^2. (Not including the I/O chiplet).
6,159 NVidia A100 dies is 5,087,334 mm^2.
These are all made on TSMC 7nm, funnily enough.
- ganzuul 5y agoAI needs bandwidth. Those 1500 chips might as well be orbiting the Earth.
- mirker 5y agoThere’s more to performance than peak theoretical performance. The architectures that stick around have tended to be ones with more software support (e.g., x86). If you mean network bandwidth, the level of batching controls the bandwidth bound and is configurable. If you mean chip bandwidth, that relies on advanced compilation that is pretty darn hard to get right. For a chip like cerebras’s to win, it’d have to have a bandwidth bound workload and deliver on the software to eliminate the bottlenecks.