Advice for new MD Simulation Research/Gaming Workstation (Work from Home)

Hey everybody I am an experienced PC builder but I’m looking for advice on workstations since in prior years I built PCs mostly from an Enthusiast perspective with other considerations second. However, I have recently started a graduate school program in physics and I will be supplementing my Theoretical Physics Research. I will probably have access to a Cluster for more intense calculations but I will also be working a lot from home due to COVID. Tons of new parts have been released (or will be released) and I’m looking to get a great bang for the buck workstation that can both game and run short(er) (MD simulations always take a ton of time) simulations from home before I deploy calculations onto a Cluster.

Here are my current specs

  • Ryzen 7 1700X
  • GTX 1080 Ti
  • EVGA 850W PSU
  • 32 GB DDR4 3200Mhz
  • Gsync 1440p UW monitor (Least Concern part I just want fast performance in my applications)

Here is the Software Suite I am planning to use:

  • Mathematica
  • Quantum Espresso
  • Lecture & Office Hours Streaming/Video
  • Python/C++ Programming

Here are my Gaming use Cases

  • 1440p Ultrawide
  • Valve Index VR
  • 60 fps minimum w/ no tearing
  • 90 fps preferred due to VR gaming
  • Monster Hunter World
  • Dark Souls 1-3
  • Don’t Starve Together
  • Crusader Kings III
  • Space Pirate Simulator
  • Half-Life Alyx

The Bulk of my research will likely be done with Pen and Paper Mathematics as well as Mathematica Notebooks, but for the times I need to run simulations and test code at home I would like to have something fast enough to perform the computations quickly enough to make testing models less of a headache. Additionally, I do some streaming and make videos on the side when I teach students and I want my streams to be as high of quality as they possibly can be. When teaching heavy multi-tasking is a must.

I will mention that I don’t need anything like a Quadro, Xeon, Threadripper, or EPIC due to the fact I will have access to a cluster but Ryzen, Gforce, Radeon, and Intel Rocket Lake are all probably fine. It just needs to be fast enough so I can do testing and iteration quickly. I also live in a tiny apartment so I will likely be in the same room as this thing most of the time. However, I do know that my build may be memory-constrained and I may need some RAM and VRAM upgrades to keep up with the MD simulations (would intel OPTANE be a good thing to consider?)

The build is 3 years old and is starting to show its age but I could probably live 1 more year with it if I needed to since my research will not hit full tilt until the summer of 2021. I know how to build gaming PCs and simple servers that make sense and fit within reasonable budgets but this is my first time building a true workstation and I’m looking for advice from anyone that either has similar usage cases as me or who has used the programs I want to use to recommend hardware to me. The build will likely be an AMD CPU unless one of the programs I use heavily favors intel but I’m just unsure about what performance is like with these programs.

Thanks for all the Advice!

I say go with a Ryzen 9 5900X+ an RTX3080+64GB ram (tuned well).
The only reason to go with the Nvidia GPU is because of the software stack. If you have the budget I’d also consider a 5950X+3090.

Thanks for the Reply!

Is the 3090 recommendation purely for the extra VRAM? Benchmarks indicate that the 3090 vs 3080 are very similar but VRAM may be a bottleneck for MD simulations.

What benefit would that 16 core Ryzen have over the 12 core for my use case? I am certainly aware that I will want more than 8 cores in an ideal scenario.

What size of data sets are you working with? Do you actually need the VRAM?

How much can you parallelize your workload?

Simple MD simulations can easily start getting into the 10s of Gigabytes for even simple Computations. Larger More representative Datasets can easily Exceed 100 GB. On the Cluster, we can easily exceed 1 TB. Mind you this isn’t me giving it a data set this is a data set it creates when it creates the particles for the simulation. However the Workload is very parallelizable. I think for GROMACS GPU Acceleration is the standard. I’m just not sure which GPU would work best for my use case.

Best you can afford for your budget if its cuda required Nvidia is your only option. You best performance / $$ will probably be 3080, if you need the vram prob 3090. If you are just doing small tests you could get away with lower for sure, just really depends on the cost of your time. (if not cuda new amd cards might be quite nice)

As for CPU do your calculation rely on cpu a lot or mostly GPU?

For GROMACS GPU is the standard use case and the workload is very very parallelizable. However, for Mathematica notebooks, it can be a bit tricker since it depends on the type of computation being performed. Some computations can’t be parallelized, some can be lightly parallelized. However, for some computations, single-core performance is all that matters. Mathematica notebooks and Compiling code will be the bulk of my computation research needs. The MD simulations I would test would be toy models to then deploy on the cluster. However, all the results need to be rendered locally after post-processing. Additionally to display the results of simulations usually requires real-time rendering capabilities of some type.

So literally just going to any 5xxx cpu will be huge as the gains were crazy from 1st gen Ryzen. I would probably target 8 or 12 core, you can do 16 core if you want but since you have a mix of needs (single and multi, it still would probably be better but idk about the value)

Do you have a budget to go form what you have to the upgrade?

1 Like

Yeah the 3090 recommendation is for the vram+ software stack.

A budget figure would be nice.


I mean top priority imo is getting anything Ryzen 5xxx, gpu if cuda required = wait till stock if it ever appears.

1 Like

My budget is preferably no more than $2000 -$2500. I could go a little higher if I save up for a little longer. I already have many of the components and peripherals so likely this would be a CPU, MOBO, Memory, and GPU upgrade. Perhaps an extra SSD if I can fit it into my budget.

I will mention that Mathematica’s Parallelization is a bit strange in the fact that it can’t utilize multi-threading very well. If you are going to parallelize the computation it needs to be full cores only.

Posted a reply to mutation666 including budget and component details.

Replace the 3900xt with the 5900xt. Also, try to get the 3080 at MSRP. The founders edition seems like the best one out of the bunch imo.

Edit: a used 2080Ti is probably a good deal too for around 450usd

1 Like

Thanks to both of you for your recommendations and assistance. I’m inexperienced with how the software stack affects the workstation experience. This discussion was very helpful to me!

An optane drive is a good choice if you need high iops drive with high endurance. But you can get away with cheaper options as well for sure.

Good places to look for benchmarks phoronix or pugetsystems for certain work loads.

Well, I was thinking that I could use optane as a scratch disk for my simulations that way when I"m running simulations that take up 100 GB of memory it wouldn’t always need to go out to the hard disk to accelerate. Either way, I think an SSD scratch disk will be in my future as well, but I’ll worry about that when I get to it. However, it would need to be high endurance since it will constantly keep getting overwritten with raw data.

Yeah optane would be good for that kind of use for sure, there are other intel drives that have quite high endurance as well that can be had for pretty cheap used on ebay (probably a better option then optane due to the pricing is so high on those still)

1 Like

What might you recommend for a scratch disk that his high endurance other than optane? I have never needed a high endurance drive so this is new to me.

Good example is P3700

Even used it should have a ton of life on it
1 Like

Thanks! I’ll look into that option as well!