Hi Alessio,
You won't gain a lot (other than GPU memory you may not need depending on workload) from the 'professional' parts - as long as the vendor is warrantying the cards you are fine. If you want to save money I'd cut back on the RAM - 384GB is overkill and you are unlikely to use it. 192GB or even 96GB is likely plenty. You might also want to switch to the new Cascade Lake CPUs instead of Skylake. 6240 is cheaper than 6148. It is 18 core vs 6148 being 20 core but it has higher clock speed 2.6GHz vs 2.4GHz so it is likely a wash performance wise for a reduced price.
2 x 3.8TB SSD or 6.4TB NVME is also probably overkill for cache. You might find a single 1.6TB NVME (or a couple of raid 0 1TB NVMEs) to give better performance for lower cost (could also save on not needing hardware raid card). Then include some 12TB HDD's if you need appreciable local storage.
For the 4 x RTX2080TI option I'd also consider changing that to a 2U rack mount system - the ducted cooling gives better performance - this assumes you aren't looking for a workstation format.
Hope that helps.
All the best
Ross
> On Jun 11, 2019, at 10:26 AM, Alessio Accardi <[log in to unmask]> wrote:
>
> Dear all,
>
> At Weill Cornell Medicine we are in the process of expanding our computational resources to process cryoEM data. For technical reasons we will adopt a model where multiple nodes operate as virtual workstations that run independently with a queue system for users. In this expansion, we are looking to purchase multiple nodes, with exact numbers depending on the configuration/price of the nodes.
>
> We are discussing with our Scientific Computing Unit on what is the optimal configuration for these nodes and we have 3 possibilities listed below. For reference, most of us work on membrane channels/transporters, often reconstituted in nanodiscs.
>
> We would like to have suggestions on whether these specs are appropriate for our purposes or if they are overkill and current software would not take advantage of the specs of the hardware. Additionally, one important aspect to consider is the choice of using professional grade (options #1 and #2) or consumer grade (option #3) hardware.
>
> #1 Dell server
> CPU: Intel Xeon Gold 6148
> Memory: 384GB
> SSD: 6.4TB Samsung HLLL
> GPU: 4 x v100 32GB NVLINK
> Chassis size: 1U
> Power: 2 x 2400W
> Support: 5 years, NBD Onsite
> Price: $33,326
>
> #2 Penguin server
> CPU: Intel Xeon Gold 6148
> Memory: 384GB
> SSD: 6.4TB Micron NVMe
> GPU: 4 x v100 32GB PCIe
> Chassis size: 2U
> Power: 2 x 1600W
> Support: 5 years, parts mailed
> Price: $35,009
>
> #3 Exxact workstation
> CPU: Intel Xeon Gold 6148
> Memory: 384GB
> SSD: 2 x 3.8TB RAID 0
> GPU: 4 x RTX 2080 TI 11GB
> Chassis size: 4U
> Power: 2 x 2000W
> Support: 5 years, parts mailed
> Price: $20,079
>
>
> If anyone prefers, please email me directly at [log in to unmask]
>
> Thank you very much for your help and suggestions.
>
> Alessio Accardi
>
> ########################################################################
>
> 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
|