Someone did a benchmark for the cuda cores.

https://www.phoronix.com/scan.php?page=article&item=nvidia-rtx2080ti-compute&num=1

It looks like that the new RTX are about 50-60% ish stronger than 1080. Unlike games these results won't change much with drivers, so I think that we might expect such improvements.

And if someone will make the calculations to work with the RX cores we could get even better results.

Yehuda Halfon



-------- Original Message --------
From: Yehuda Halfon <[log in to unmask]>
Sent: Thursday, September 20, 2018 01:01 AM
To: [log in to unmask]
Subject: Re: [ccpem] RTX 2080 Benchmarks?

I think will see it in the long iterations. The short ones will be the same, but when the calculations will be on the GPU the bottleneck will be between the iterations.



-------- Original Message --------
From: Dari Kimanius <[log in to unmask]>
Sent: Thursday, September 20, 2018 12:21 AM
To: [log in to unmask]
Subject: Re: [ccpem] RTX 2080 Benchmarks?

A ~30% improvement in the overall performance of general-purpose calculations is what we've seen in every new generation of NVIDIA cards the last couple of years. 

But as Shintaro also mentioned RELION currently has both an IO and CPU bottleneck and unless you also get a powerful CPU and better disk access, you're not very likely to see those ~30%.

Best,
Dari

On Wed, Sep 19, 2018, 19:38 Marin van Heel <[log in to unmask]> wrote:

My take-home message from a recent NVIDEA presentation in our lab was that number of floating-point processors within the GPUs is shifting towards 16bit FPUs at the cost of reducing (or maintaining) the number of 64bit FPUs .  As a general trend, this means a shift towards faster graphic processing rather than towards faster scientific processing. Indeed in the quick search on benchmarks I just did, I only saw ~30% improvement mentioned with respect games that I have never even heard of. Thus keep your scientific processing expectations under control!

My two cents,

Marin

On 19/09/2018 13:38, Shintaro Aibara wrote:
We'll wait and see, it'll largely depend on your CPU's ability to keep your GPUs fed with data.
I suspect the typical workstation will not see much speed gain unless you have a very powerful CPU.

I suspect the 1080Ti will still be the better buy for its larger memory capacity.

On Wed, Sep 19, 2018 at 6:26 PM Yehuda Halfon <[log in to unmask]> wrote:

Hi Douglas,

I don't think they aren't out for the public yet. I think the "founder" edition will be out at the 21.9 and I think the relion users will wait for the "blower" edition so our workstations won't die of overheating.

Yehuda Halfon



-------- Original Message --------
From: Douglas Duckworth <[log in to unmask]>
Sent: Wednesday, September 19, 2018 07:04 PM
To: [log in to unmask]
Subject: [ccpem] RTX 2080 Benchmarks?

Hi

Has anyone had the opportunity to benchmark Relion on Nvidia's newest card the RTX 2080 verses GTX 1080?  Non-Relion benchmarks seem to show 20-30% performance increase vs 1080.


Thanks,

Douglas Duckworth, MSc, LFCS
HPC System Administrator
Scientific Computing Unit
Weill Cornell Medicine
1300 York Avenue
New York, NY 10065
E: [log in to unmask]
O: 212-746-6305
F: 212-746-8690


To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1


-- 
==============================================================

    Prof Dr Ir Marin van Heel

    Laboratório Nacional de Nanotecnologia - LNNano
    CNPEM/LNNano, Campinas, Brazil

    email:  marin.vanheel(A_T)gmail.com
            marin.vanheel(A_T)lnnano.cnpem.br
    and:    mvh.office(A_T)gmail.com  

--------------------------------------------------
    Emeritus Professor of Cryo-EM Data Processing
    Leiden University
--------------------------------------------------
    Emeritus Professor of Structural Biology
    Imperial College London
    Faculty of Natural Sciences
    email: m.vanheel(A_T)imperial.ac.uk
--------------------------------------------------

I receive many emails per day and, although I try, 
there is no guarantee that I will actually read each incoming email. 


To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1



To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1



To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1



To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1