It appears this limit to which you refer is an artificial one imposed by the NVIDIA drivers (and people get mad when I mention that NVIDIA can be capricious in its implementations 😊 ).
Here's an article that discusses the enablement of multi-instance/ virtualization of GeForce cards:
https://videocardz.com/newz/modders-find-a-way-to-enable-gpu-virtualization-on-geforce-graphics-cards
Jim
James Kress Ph.D., President
The KressWorks® Institute
An IRS Approved 501 (c)(3) Charitable, Nonprofit Corporation
“ENGINEERING THE CURE” ©
(248) 573-5499
Learn More and Donate At:
Website: http://www.kressworks.org
Confidentiality Notice | This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential or proprietary information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, immediately contact the sender by reply e-mail and destroy all copies of the original message.
-----Original Message-----
From: Jason Hogrefe <jason.hogrefe.exxactcorp.com>
Sent: Friday, October 22, 2021 1:07 PM
To: AMBER Mailing List <amber.ambermd.org>; jimkress 58 <jimkress_58.kressworks.org>
Subject: RE: [AMBER] Maximizing GROMACS Throughput with Multiple Simulations per GPU Using MPS and MIG
If you are looking to test this out on an A100 system you can apply for our remote Test Drive program:
https://www.exxactcorp.com/Test-Drive
Jason Hogrefe
Marketing Operations Specialist
______
P 402.250.9203
Exxact Corporation | exxactcorp.com
-----Original Message-----
From: Matias Machado <mmachado.pasteur.edu.uy>
Sent: Friday, October 22, 2021 8:36 AM
To: jimkress 58 <jimkress_58.kressworks.org>; AMBER Mailing List <amber.ambermd.org>
Subject: Re: [AMBER] Maximizing GROMACS Throughput with Multiple Simulations per GPU Using MPS and MIG
[EXTERNAL]
Dear James,
I'm not an expert in here... but they used A100 cards (~USD 20k) which allows for splitting the GPU into 7 independent instances... no other card on the market can do this as far as I know...
I remember Adrian Roitberg saying that AMBER should benefit from that feature 'out-of-the-box', but I guess developers require to get their hands into A100 cards to implement/test the code... :-/
All the best,
Matías
----- Mensaje original -----
De: "James Kress" <jimkress_58.kressworks.org>
Para: "AMBER Mailing List" <amber.ambermd.org>
Enviados: Jueves, 21 de Octubre 2021 17:29:16
Asunto: [AMBER] Maximizing GROMACS Throughput with Multiple Simulations per GPU Using MPS and MIG
How would we replicate this simulation set with Amber?:
https://developer.nvidia.com/blog/maximizing-gromacs-throughput-with-multiple-simulations-per-gpu-using-mps-and-mig/?mkt_tok=MTU2LU9GTi03NDIAAAGAQ2HX780vGeCR3yEDCYePFc_rZOlFwC0ZafOGfVQCZgaYCdtP1sRih7Hv4b-rPHNx7dcm2ot8E7-6-rCX9GunutbjqRUakD7ydwauLEYgBnSqAs4
James Kress Ph.D., President
The KressWorks® Institute
An IRS Approved 501 (c)(3) Charitable, Nonprofit Corporation “ENGINEERING THE CURE” ©
(248) 573-5499
Learn More and Donate At:
Website: http://www.kressworks.org
Confidentiality Notice | This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential or proprietary information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, immediately contact the sender by reply e-mail and destroy all copies of the original message.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Oct 22 2021 - 12:00:02 PDT
