The important thing to realize about the Tesla T4 is that it is a
server-oriented device designed for low power consumption. One can install
and engage four T4s for the same electrical consumption as one V100, and I
expect that the aggregate throughput of the four cards would somewhat
exceed the V100, but at substantially reduced output per card.
Another thing to consider is price of the card. I hate to say it, but if
you don't pay the power bill (if your university does), then you have
little incentive to buy such energy-efficient things when a more powerful
RTX-2080Ti is available for about half the price.
The T4 is a great option for someone with deep pockets and a high demand
for scientific computing or other distributed, arithmetic-intensive or DL
problems. NVIDIA did a really good job balancing the computational
capacity and power consumption there. For an individual researcher,
though, the price point remains RTX-2080Ti.
Dave
On Wed, May 15, 2019 at 11:52 AM Shenglong Wang <sw77.nyu.edu> wrote:
> There are some numbers from NVIDIA and Dell
>
> https://developer.nvidia.com/nvidia-hpc-application-performance-v100-t4 <
> https://developer.nvidia.com/nvidia-hpc-application-performance-v100-t4>
>
>
> https://www.dell.com/support/article/us/en/04/sln316570/hpc-applications-performance-with-turing?lang=en
> <
> https://www.dell.com/support/article/us/en/04/sln316570/hpc-applications-performance-with-turing?lang=en
> >
>
> Best,
> Shenglong
>
> > On May 15, 2019, at 11:46 AM, Luca Codutti <
> luca.codutti.oci.uni-hannover.de> wrote:
> >
> > Dear all,
> >
> > I would like to know if there are any amber benchmarks made for the
> > nvidia tesla T4. More importantly, do you have any indication on how it
> > does compare with the V100? It apparently does not have a really good
> > double precision benchmark according to this link
> > <
> https://urldefense.proofpoint.com/v2/url?u=https-3A__www.microway.com_hpc-2Dtech-2Dtips_nvidia-2Dturing-2Dtesla-2Dt4-2Dhpc-2Dperformance-2Dbenchmarks_&d=DwICAg&c=slrrB7dE8n7gBJbeO0g-IQ&r=gI7YPDDMIksUN29Y65_UCg&m=bt-3yhOxLakoG5Nbj2khd3L4risxJj6fxCQJGDOONio&s=tWvVnTNWKhCCvQ3JMif4LmVX105QU4sAf3e-DHjoG3s&e=>so
>
> > I was wondering if and how the relatively poor performance compared with
> > V100 and P100 is going to influence MD calculations speed.
> >
> > Kind regards
> >
> > Luca
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> >
> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ambermd.org_mailman_listinfo_amber&d=DwICAg&c=slrrB7dE8n7gBJbeO0g-IQ&r=gI7YPDDMIksUN29Y65_UCg&m=bt-3yhOxLakoG5Nbj2khd3L4risxJj6fxCQJGDOONio&s=Ar4tOwZxqdyhSmcg0o2z5RrHjThnU0y3aqqf7haABTQ&e=
>
> _______________________________________________
> 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 Wed May 15 2019 - 09:00:04 PDT
