Jump to content

Wikipedia:Reference desk/Archives/Computing/2020 February 14

From Wikipedia, the free encyclopedia
Computing desk
< February 13 << Jan | February | Mar >> Current desk >
Welcome to the Wikipedia Computing Reference Desk Archives
The page you are currently viewing is a transcluded archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.


February 14

[edit]

A gaming computer for scientific work?

[edit]

Hi there. I need to increase the speed of my computing and my machine now is a 2 year old HP Pavilion Windows 10 Pro. I looked around and found that gaming machines appear to be the fastest and they are advertised as such. I will describe my typical software: Microsoft SQL Server, Visual Studio 2017 with C++ and C#, perhaps also Python in the near future. Is it a good idea for me to fork two grand on a Dell or even Talon gaming computer?

Thanks, AboutFace 22 (talk) 22:04, 14 February 2020 (UTC)[reply]

No, a big chunk of the cost will be on the graphics card in the machine. You're not running anything that needs that kind of power. - X201 (talk) 22:20, 14 February 2020 (UTC)[reply]
Agreed, though I suppose he could go with a supplier like Alienware that lets him custom-design the rig and just "cheap out" on the graphics card. Matt Deres (talk) 20:21, 15 February 2020 (UTC)[reply]

Thank you. Very useful. AboutFace 22 (talk) 22:37, 15 February 2020 (UTC)[reply]

Don't buy hardware until there is specific application code you want to run on it. But there are many Python-hosted machine learning libraries like PyTorch that use GPU's, and there are also some NumPy gpu integrations and numpy-like gpu-using libraries. So whether the gpu helps depends on what you are doing. Try some web searches like "numpy gpu". 2601:648:8202:96B0:0:0:0:7AC0 (talk) 06:05, 16 February 2020 (UTC)[reply]

You would be better served by a workstation-class machine. One that has Xeon processors, ECC DRAM, and a near-server-class motherboard. The gaming computers are going to be needless flashy woo that meet benchmarks you don't need. There should be plenty of workstation machines aimed at business users who have your kind of workflow, although I have to say that more development work is going to be done in the cloud, and that points toward thin clients, going forward. Enjoy the fat clients while they still last. Elizium23 (talk) 06:10, 16 February 2020 (UTC)[reply]
Elizium23, imo a Threadripper would be better for a workstation system, partially because it's vastly cheaper. 32 cores at $2000 vs 28 cores at $9000. They still support ECC and everything else. However remember that this is only good advice if they do any form of machine learning. —moonythedwarf (Braden N.) 15:43, 19 February 2020 (UTC)[reply]
As mentioned below, *all* Ryzen products support ECC. So even a standard consumer grade 3950X (16C @ $700) or 3900X (12C @ $500) would be suitable —moonythedwarf (Braden N.) 15:47, 19 February 2020 (UTC)[reply]
Gaming computers are not for getting work done. The software you mention may or may not need a lot of power, depends on what you are doing. You could go for a high-end CPU with onboard graphics. For getting work done and not costing an arm and a leg, I have seven used (off lease) Sandy Bridge Xeon workstations. Now, Xeons vary a lot in speed. Some are slow and mainly useful for serving files - faster one are for doing work. Bubba73 You talkin' to me? 06:26, 16 February 2020 (UTC)[reply]
Really we need to more about your application to make any hardware recommendations. It's true that if you don't need a lot of local bandwidth and visualization, you may be better off with remote servers. Gaming GPU's (e.g. NVidia GTX/RTX series) cost a heck of a lot less than workstation ones (Quadro), mostly because of marketing scams, and they do about the same thing, so tons of professionals use them. 2601:648:8202:96B0:0:0:0:7AC0 (talk) 07:06, 16 February 2020 (UTC)[reply]
Also keep in mind that older Xeons (especially pre-Sandy-Bridge) while still reliable, use a heck of a lot more electric power than newer chips. You can get them cheap because data center users threw them out because of the newer stuff's savings on power bills. 2601:648:8202:96B0:0:0:0:7AC0 (talk) 19:08, 16 February 2020 (UTC)[reply]

I design applications that are computationally very intensive. Numerical integrations are a big part of them, integration of spherical harmonics multiplied by some 2-D functions defined on 2-sphere. I don't know how to characterize that in more detail, though.

Thank you very much for your posts. AboutFace 22 (talk) 16:12, 17 February 2020 (UTC)[reply]

One question is what numerical libraries you are using, and whether they use GPU's. Another is what you are doing with the end results. Typical computations end up getting repeated and tweaked lots of times, so if you're worried about reliability from lack of ECC, you could use non-ECC computers for initial development and ECC servers to check the results before final publication or deployment. Note that current AMD Ryzen cpus used in gaming computers support ECC, though not all motherboards do. You haven't even put any numbers on "computationally very intensive" which would give us a clue about whether you need a PC or a supercomputer. What are you running your stuff on now, and how long is it taking? 2601:648:8202:96B0:0:0:0:7AC0 (talk) 19:35, 17 February 2020 (UTC)[reply]