Advanced search

Message boards : GPUGRID CAFE : 170 TFLOPS AI in a box?

Post to thread Subscribe Sort
Author Message
Profile Francois Normandin
Send message
Joined: 8 Mar 11
Posts: 71
Credit: 654,432,613
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 44996 - Posted: 29 Oct 2016 | 3:35:54 UTC

NVIDIA POWER! :)

http://www.nvidia.com/object/deep-learning-system.html

ID: 44996 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
kcharuso
Avatar
Send message
Joined: 7 Oct 13
Posts: 2
Credit: 671,965,120
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 45095 - Posted: 1 Nov 2016 | 14:24:52 UTC - in response to Message 44996.

mann!! that's so so dangerous. I felt inferior for fraction of a second there...

ID: 45095 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
Greger
Send message
Joined: 6 Jan 15
Posts: 76
Credit: 23,801,427,249
RAC: 51,731,724
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 45097 - Posted: 1 Nov 2016 | 19:57:23 UTC

GPUs 8x Tesla GP100
TFLOPS (GPU FP16 /CPU FP32) 170/3
GPU Memory 16 GB per GPU
CPU Dual 20-core Intel® Xeon®
E5-2698 v4 2.2 GHz
NVIDIA CUDA® Cores 28672
Maximum Power Requirements 3200W

Price $129,000

Yes please

ID: 45097 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
[CSF] Aleksey Belkov
Send message
Joined: 26 Dec 13
Posts: 86
Credit: 1,258,731,270
RAC: 178,506
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45105 - Posted: 2 Nov 2016 | 0:27:45 UTC - in response to Message 45097.

This hardware for a very different kind of tasks/computations.
Performance is specified for a half or single precision, while in our case we used double precision.
So it's not so cool.

ID: 45105 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45119 - Posted: 2 Nov 2016 | 17:26:01 UTC - in response to Message 45105.
Last modified: 2 Nov 2016 | 17:57:33 UTC

GPUGRID uses single precision and the GeForce range is financially realistic, unlike the Tesla's and Quadro's. The Titan's are a cross between the high end GeForce and Quadro's/Tesla's and are really intended for flag wavers.

8 GTX1080's dropped into 2 high end gaming/crunching systems would cost just under $10K and would get the same credit here as the above $129K Tesla system; the GTX1080's actually boost to over 2GHz making their actual boost single precision GFlops 10TF each - the same as a GP102 (slightly better than P100's).
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

ID: 45119 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
[CSF] Aleksey Belkov
Send message
Joined: 26 Dec 13
Posts: 86
Credit: 1,258,731,270
RAC: 178,506
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45171 - Posted: 4 Nov 2016 | 14:14:54 UTC

I apologize for my misconception.
For some reason, thought that it uses double precision.

ID: 45171 | Rating: 0 | rate: Rate + / Rate - Report as offensive    Reply Quote
Post to thread

Message boards : GPUGRID CAFE : 170 TFLOPS AI in a box?

//