Tesla T4 benchmarks
#4
(12-06-2018, 10:36 AM)s3in!c Wrote: Thanks for posting the interesting benchmarks!
Did you measure the effective power usage or did you just wrote the value given by nvidia in the specs?

I've monitored it during a longer SHA1-run and it is stable around 70W, the Max-powerlimit is also 70W.
I noticed it jumping once to 71.5 watts or so, the official PCIe slot limit is 75W so it should be tight.
Idle it did about 23W, probably due the large amount of memory and it being a Turing chip that is more power hungry in idle as Pascal. The clock dropped a bit from a turbo frequency of 1225 to 11-something when the server fans spun down to a lower rate and the temperature climbed to ~80 degrees C.
Some server fan profiling is necessary as servers nowadays try to be silent and green ;-|
I use a R640, though it's officially an unsupported platform for this card, but who cares.

I'm happy to do more test if anyone is interested.

Concerning price, yes you can buy two 2080ti's for just one of these. On the other hand you can fit three of these in a regular 1U server. If you are also planning to do a lot of Deep Learning, it might be an interesting product but you need deep pockets.

Cheers


Messages In This Thread
Tesla T4 benchmarks - by lapwing - 12-05-2018, 11:36 PM
RE: Tesla T4 benchmarks - by Mem5 - 12-06-2018, 09:49 AM
RE: Tesla T4 benchmarks - by s3in!c - 12-06-2018, 10:36 AM
RE: Tesla T4 benchmarks - by lapwing - 12-06-2018, 05:35 PM
RE: Tesla T4 benchmarks - by Flomac - 12-06-2018, 07:11 PM