Building a Computer for Computational Chemistry

[spaceshipjammer]

Budget (including currency): 3000

Country: USA

Games, programs or workloads that it will be used for: Computational chemistry research, specifically Maple, GAMESS, Gaussian, TensorFlow, Python, PyTorch

Other details (existing parts lists, whether any peripherals are needed, what you're upgrading from, when you're going to buy, what resolution and refresh rate you want to play at, etc): 

Hello everyone, my wife was recently hired to become a professor of Chemistry and is looking to put together a workstation. While she will have some access to supercomputers, she would like to be able to do as much of her work locally as possible. While she has been able to talk to some advisors, none are super knowledgeable about PC hardware specifically, most just order what Dell tells them to. Her work uses the above programs and has an emphasis on machine learning and electronic structure theory. While she does have a budget up to $3,000, she also doesn't want to overspend on anything, so feel free to let us know if something seems overkill to you. The PCPartPicker list is below, thanks in advance for any help!

 

PCPartPicker Part List: https://pcpartpicker.com/list/6KCWVw

CPU: Intel Core i9-13900K 3 GHz 24-Core Processor  ($549.97 @ Newegg)
CPU Cooler: Noctua NH-D15 chromax.black 82.52 CFM CPU Cooler  ($119.95 @ Amazon)
Motherboard: Gigabyte Z790 AORUS ELITE AX ATX LGA1700 Motherboard  ($254.99 @ Amazon)
Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-5200 CL40 Memory  ($164.99 @ Amazon)
Storage: Samsung 980 Pro 2 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive  ($129.99 @ Adorama)
Video Card: Asus TUF GAMING GeForce RTX 4080 16 GB Video Card  ($1199.00 @ Amazon)
Case: Fractal Design Pop Air ATX Mid Tower Case  ($105.98 @ Newegg)
Power Supply: SeaSonic PRIME Ultra Platinum 1000 1000 W 80+ Platinum Certified Fully Modular ATX Power Supply  ($408.97 @ Amazon)
Total: $2933.84
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2023-06-12 18:36 EDT-0400

[Slottr]

Maybe wait until she's actually started and comfortable before locking into and anything

Its more than likely that the school's supercomputer network will be far more efficient for large model work than anything she can do locally.

 

If computation work can be strictly done on there, then the excess cost of local computing becomes fairly stale 

[igormp]

17 hours ago, spaceshipjammer said:

CPU: Intel Core i9-13900K 3 GHz 24-Core Processor  ($549.97 @ Newegg)

I'd say to go with a 7950x or even the X3D version, both are easier to cool and for some of her software the X3D version might net a nice performance bonus (ideally one should perform benchmarks to see if it's worth it or not, also depends on the datasets she's working with).

17 hours ago, spaceshipjammer said:

Memory: Corsair Vengeance 64 GB (2 x 32 GB) DDR5-5200 CL40 Memory  ($164.99 @ Amazon)

Wouldn't 2x48GB be a better option? More RAM and more stable frequencies.

17 hours ago, spaceshipjammer said:

Video Card: Asus TUF GAMING GeForce RTX 4080 16 GB Video Card  ($1199.00 @ Amazon)

Sadly I don't see a way to make a 4090 fit within this budget

Is getting used an option? If so, an used 3090 may be a better pick due to the extra vram.

 

17 hours ago, Slottr said:

Maybe wait until she's actually started and comfortable before locking into and anything

Its more than likely that the school's supercomputer network will be far more efficient for large model work than anything she can do locally.

 

If computation work can be strictly done on there, then the excess cost of local computing becomes fairly stale 

As someone who also has access to a pretty heft cluster, and while what you said is true, being able to iterate locally and try out stuff before submitting a job to a cluster is way more comfortable. Having to rely solely on submitting jobs all the time and checking their outputs in a non-interactive way is really annoying, not to mention the quotas and limitations imposed.

 

So it's not much about efficiency but rather convenience.