3 ms·
nice summary. are you working in HPC? edit: In case you work for Cerebras, I gotta say, 18GB is indeed low for such a compute dense beast. I'd expect it to hav
by m_mueller 6y ago
nice summary. are you working in HPC?
edit: In case you work for Cerebras, I gotta say, 18GB is indeed low for such a compute dense beast. I'd expect it to have access to at least 200GB off-chip high bandwidth memory.
- copy-pastries 6y agoThanks! Yes, I work as an HPC researcher, but not for Cerebras. I just watched the presentation of this paper at the SC20 Supercomputing conference last week. I probably should have said, according to the specs, a CS-1 does have loads (TBs?) of off-chip DRAM to spill over into if the 18GiB isn't enough. That 18GiB is on-chip SRAM which you could think of as a giant 18GiB cache, which is astounding. Once you spill over that though, you're limited to how fast you can get data to the chip (probably multiple PCIe connections to a host CPU on the system with loads of RAM - I think the details are in the paper), so the results won't be nearly as impressive as this 10000x claim.
- m_mueller 6y agosounds quite capable actually, for anything that fits to its memory system. HPC is a small-ish community where you can tell quite immediately who's part of it from the way of thinking. I've switched to financial industry in a position where (other engineers think) performance doesn't matter much, but it has always stayed an important factor for the design decisions I make.