NEC Is Looking To Contribute SX-Aurora VE Accelerator Support To LLVM

[coder]

Then read the _real_ doks by NEC (Dr. Erich Focht).
Starting page 29

https://sx-aurora.github.io/posts/Vector-Computer-Evolution/

_Current_ VE (selling since Feb 2018, 16 nm FinFET) is nearly Tesla V100 but with better/lower power draw.
Price is significantly lower then V100.

What I see is that it has more bandwidth and BW/W (we knew this). As for compute, they're very close on Himeno, but V100 stomps it on DGEMM and HPCG. And they conspicuously excluded the V100 from their deep learning performance comparison.

Since the V100 is almost 2 years old, I'm guessing we'll soon see a successor at 7nm, sometime this year. I'd love to know how AMD's MI60 compares.

Also, I have to ask which price we're talking about. The sticker price of V100's is indeed high, but I'd imagine big customers often pay much less.

Finally, I finding it worrying that Slide 39 makes no mention of OpenCL (or HSA).