What do you do with 5,280 V100 GPU's?

[Phynaz]

This

660 petaFLOPS

 

[ElFenix]

if they're exactly 12" long you can have a mile of GPUs. like beer mile, but dorkier.

 

[Krteq]

660 petaFLOPS

*(AI) / 80 PFLOPS (FP32)

Not so impressed

Anyone knows how they get that - ehm - "AI" number and what is that number represents? It's not for INT8/FP16 peak performance, so what is it?

 

[Phynaz]

*(AI) / 80 PFLOPS (FP32)

Not so impressed

Anyone knows how they get that - ehm - "AI" number and what is that number represents? It's not for INT8/FP16 peak performance, so what is it?

Matrix math.

A single Tensor Core performs the equivalent of 64 FMA operations per clock (for 128 FLOPS total), and with 8 such cores per SM, 1024 FLOPS per clock per SM.

https://www.anandtech.com/show/1136...v100-gpu-and-tesla-v100-accelerator-announced

 

[crisium]

I'd let it turn loose on itself - use them FLOPs with machine learning to design faster, more efficient GPUs at the same node. GPU singularity.

 

[Phynaz]

*(AI) / 80 PFLOPS (FP32)

Not so impressed

Why?

 

[Elixer]

This

660 petaFLOPS

Thought for sure you were going to say...
Play Crysis!

 

[Phynaz]

^ I was tempted!

 

[Arachnotronic]

Sell them on eBay and take a $ bath.

 

[EXCellR8]

...complain that SLI still doesn't work as well as it should.

 

[Qwertilot]

I'd let it turn loose on itself - use them FLOPs with machine learning to design faster, more efficient GPUs at the same node. GPU singularity.

Good plan up to the point where it works out to do molecular self assembly to make them and we've got a homogonising swarm on our hands