Question Monster server for 3TB RAM application/workstation or terrible idea?

[J-dubya]

I currently use a machine that I built a year ago. It has the 32 core AMD Threadripper, 256GB of RAM, 60 something TB of HDD (mostly M.2), and Windows 10 Pro OS. This is too slow for the timeline I'm trying to work with and not nearly enough RAM to have the accuracy I would like. It currently takes about 20 days to get through a set of solves for evaluation and I am limited on the size of the game tree due to RAM.

I am considering buying a Dell Poweredge R930 with 4 CPUs: E7-8867v4 2.4GHz 72-Cores total, 3TB RAM. I'm not interested in buying the Windows Server license for all 72 cores so I would be using Linux Ubuntu. The solver application will work well on that OS I hear. I've never used Linux OS, but am somewhat savvy with tech stuff (built multiple PCs, did some coding in college).

Many people that run this application rent servers like this for $1200 a month and just use it 1 month or 2. I will need to use it for several months and much more in the future. It seems like it makes more sense to just buy the server. I know it will be loud. I know it will use TONS of power.

Has anyone messed with something like this? What kind of issues will I be running into? How hard is it for a Linux newbie to set up Ubuntu with a decent GUI (dont need much in the way of graphics, just a GUI to interact with the solver app)? Is it possible to change out the fans in a big server like this to quieter fans? The fan plugs are not standard 4 pin, could possibly modify plugs..? Is there a better OS for this server to do what I'm trying to do? Is it a completely insane idea to start with? Are there builds that would work better, considering I want 3TB of RAM and at least 60 CPU cores (preferably more)?

Thanks!

 

[J-dubya]

The reason I ask is because when you have a hardware problem, it doesn't stop everything. The rest of the cluster continues to process and will pick up the failed jobs. Much more resillient than a single server. If it was me, I would investigate if that is an option.

Thank you. I don't really have a hardware problem. I currently have no issues with my set up. It is operating as expected. I am just considering an upgrade that involves using a server. I am not in question of what hardware I need. Just trying to figure out how hard it will be to get a server like the one I listed to operate like a workstation and run mainly just the 1 application. I was hoping there might be a shortcut of some sort since I am not really using it as a server. It's just a cheap way to get the hardware set up I need.

 

[USAFRet]

Thank you. I don't really have a hardware problem. I currently have no issues with my set up. It is operating as expected. I am just considering an upgrade that involves using a server. I am not in question of what hardware I need. Just trying to figure out how hard it will be to get a server like the one I listed to operate like a workstation and run mainly just the 1 application. I was hoping there might be a shortcut of some sort since I am not really using it as a server. It's just a cheap way to get the hardware set up I need.

This level of hardware, the only "shortcut" is to do it correctly from the start.
Both hardware and software.

 

[sonofjesse]

If you can get this for 1200/month rent, it looks like you would have to use this for 5 months before break even (6k server your looking at not counting server racks, UPS, room, power usage, installing/maintenance of drives etc.. In my mind you would need to run this at least 12-18 months consistently before I would even consider this route. I'm not for sure you said this not commercial, but if your not making money is it worth doing this and spending this money?


https://www.solverglobal.com/ (is this it)

This must be uber niche science/university stuff maybe. Is their a local college or place with a super computer you could rent?

Another option is to hire a server admin for X amount of hours to assist you with this, but that will just add to your cost. Do a spreadsheet with full costs and how many months you plan on running this.

Good luck sounds like an interesting journey with a lot of variables.

 

[kanewolf]

Thank you. I don't really have a hardware problem. I currently have no issues with my set up. It is operating as expected. I am just considering an upgrade that involves using a server. I am not in question of what hardware I need. Just trying to figure out how hard it will be to get a server like the one I listed to operate like a workstation and run mainly just the 1 application. I was hoping there might be a shortcut of some sort since I am not really using it as a server. It's just a cheap way to get the hardware set up I need.

You missed my point. I am saying when your 930 goes down, it takes 100% of your resources. With a cluster, on part of your resources go fown.

 

[J-dubya]

You missed my point. I am saying when your 930 goes down, it takes 100% of your resources. With a cluster, on part of your resources go fown.

I see. No, if it goes down, I lose the current solve that it is working on.

 

[USAFRet]

I see. No, if it goes down, I lose the current solve that it is working on.

And that is yet another reason why people pay someone else to host this....guaranteed uptime and failover.

 

[J-dubya]

If you can get this for 1200/month rent, it looks like you would have to use this for 5 months before break even (6k server your looking at not counting server racks, UPS, room, power usage, installing/maintenance of drives etc.. In my mind you would need to run this at least 12-18 months consistently before I would even consider this route. I'm not for sure you said this not commercial, but if your not making money is it worth doing this and spending this money?


https://www.solverglobal.com/ (is this it)

This must be uber niche science/university stuff maybe. Is their a local college or place with a super computer you could rent?

Another option is to hire a server admin for X amount of hours to assist you with this, but that will just add to your cost. Do a spreadsheet with full costs and how many months you plan on running this.

Good luck sounds like an interesting journey with a lot of variables.

Thank you! I will likely be using it much more than 12-18 months. Maybe hiring someone to assist is the way to go.

 

[J-dubya]

And that is yet another reason why people pay someone else to host this....guaranteed uptime and failover.

That is not really an issue tho. It would only cost me 1 day if it goes down. My current set up takes 2-3 days per solve and is a much less reliable system and that seldom comes up.

 

[kanewolf]

That is not really an issue tho. It would only cost me 1 day if it goes down. My current set up takes 2-3 days per solve and is a much less reliable system and that seldom comes up.

This new box might take 2-3 days, since you are increasing the problem size (assuming you are limited because of RAM). These 930 cores will be slower than your current AMD.

 

[J-dubya]

This new box might take 2-3 days, since you are increasing the problem size (assuming you are limited because of RAM). These 930 cores will be slower than your current AMD.

I see. Yes, the problem would def be amplified in that way. I do have a UPS. I would likely need a larger one for that server. Not a deal breaker tho.