Re: [AMBER] 1x PCI-E Riser card for MD/TI?

From: Ross Walker <>
Date: Tue, 22 May 2018 17:08:48 -0400

Hi Adam,

Wow, that looks like some serious Frankenstein kit right there. I'd be shocked if that even worked for GPUs let alone for AMBER. If it does actually work then in principal for single GPU simulations where you do minimal I/O it should actually run fairly well with less than PCI-E x16. How much less? I haven't tested below 4x. 1x is probably too low but would at least work in theory. If however you wanted to do anything where there was more than minimal I/O - say ntwx<5000 or you were doing something that required the CPU as well, e.g. NFE simulations then performance would be very bad.

My advice would be that for $6 it might be worth a gamble just for the coolness of having the GPU hanging outside of the box like some half severed limb but I'd do that assuming that my $6 investment was a writeoff for what would be an interesting experiment. The gamble is if it catches fire it could take your $600+ GPU with it.

If you do try it I'd be interested to hear the results.

All the best

> On May 22, 2018, at 10:20 AM, Hadházi Ádám <> wrote:
> Dear Amber Users,
> - Do you know if it is possible to use a PCI-E riser card with "1x lane
> speed" in GPU accelerated Amber MD/TI simulations?
> e.g.:
> (We are not interested in combining more GPUs to one MD task... but a riser
> would provide flexible spacing to avoid overheating GPUs)
> Best,
> Adam
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Tue May 22 2018 - 14:30:02 PDT
Custom Search