New build for scientific simulation.

[Antony Leung]

So I have a laptop that I use for Chemistry simulations, and is so slow. The programs I use is gaussian and chemcraft. Take days to do a molecular simulation, and I always wonder if I could speed up by building a new PC for that. What you guys recommend me for those programs.

 

Says a chemistry professor who doesn't know about computer hardware.

 

Ideal budget. 1000$ USD, can overbudget but needs a good reason for it.

 

No case is needed he can recicle a old one that he have from his father.

 

Even myself don't know if this build needs much ram. like 16 or 32.

[Kiani]

Which rig are you using already? For a reference point.

[Antony Leung]

Just now, Kiani said:

Which rig are you using already? For a reference point.

It doesn't matter, he says is a old laptop, if you haven't notice yet this is a build recomendation for my professor of chemestry, just for the sake of helping him, lets assume he doesn't have a computer and need one for molecular simulation. 

[argyle]

I happen to work in higher ed and have a little experience with these types of programs.

 

Gaussian can work on GPUs if needed, so you could build a rig similar to a mining rig for maximum power with a few GPUs. It's not a simple setup but it may be the best option. I think that would need to run in Linux, though. We have an HPC that runs things like this.

 

Chemcraft isn't optimized for GPUs, and unfortunately they don't provide a lot of information on how it runs or even what the minimum specs are. It is optimized for Windows though.

 

If you want to run both of these on one system I personally would recommend finding a used Workstation class system with a higher core count Xeon (6+ with hyper threading). I would expect that you'd need to upgrade the RAM and add an SSD for boot (most used workstations do not come with SSDs, though YMMV).

 

It may be possible to dual boot Windows and Linux, put a higher end video card in and run Gaussian on the GPU through Linux and Chemcraft on windows as needed.

[argyle]

This might be a good starting point: https://www.newegg.com/Product/Product.aspx?Item=9SIA68F5256685&cm_re=hp_z420-_-1VK-001E-09ED2-_-Product

[Antony Leung]

1 minute ago, argyle said:

I happen to work in higher ed and have a little experience with these types of programs.

 

Gaussian can work on GPUs if needed, so you could build a rig similar to a mining rig for maximum power with a few GPUs. It's not a simple setup but it may be the best option. I think that would need to run in Linux, though. We have an HPC that runs things like this.

 

Chemcraft isn't optimized for GPUs, and unfortunately they don't provide a lot of information on how it runs or even what the minimum specs are. It is optimized for Windows though.

 

If you want to run both of these on one system I personally would recommend finding a used Workstation class system with a higher core count Xeon (6+ with hyper threading). I would expect that you'd need to upgrade the RAM and add an SSD for boot (most used workstations do not come with SSDs, though YMMV).

 

It may be possible to dual boot Windows and Linux, put a higher end video card in and run Gaussian on the GPU through Linux and Chemcraft on windows as needed.

Thanks, very helpful. I came across with this slideshow from nvidia, and they mentioned "Active GPU acceleration projects" and gamess, and if i remember correctly, chemcraft use that right?

 

I was thinking about this build https://pcpartpicker.com/list/2ZkKvV, heavily gpu bounded but not a mediocre and good multi-core tasking processor. and If those programs dont use a lot of ram. like photoshop or sony vegas.

[Antony Leung]

Hey @argyle and @Kiani, What do you guys think? I also call for help @STRMfrmXMN, @deXxterlab97, @Dackzy, @nerdslayer1, and @herman mcpootis.

[Antony Leung]

Last resource, I call @DocSwag for a hand. What do you think about this?

[DocSwag]

7 minutes ago, Antony Leung said:

Last resource, I call @DocSwag for a hand. What do you think about this?

Do you know much about what these programs like? 

 

From some quick googling gaussian seems to be able to take advantage of lots of cores and seems to be able to be gpu accelerated (considering they mainly mention Nvidia GPUs though I'm guessing amd isn't a great option). However, I'm still unsure if Gaussian likes Ryzen and whether or not it would prefer a quadro.

 

Unfortunately though I can't find much info for chemcraft.

 

[Antony Leung]

7 minutes ago, DocSwag said:

Do you know much about what these programs like? 

 

Unfortunately, no. He did told me some clues, like gauss, take the calculations of the molecules, and chem renders or makes a 3D representation of it.  So my guess is gauss likes both but is gpu bounded and chem gpu.

 

So with that build of 1000$ is still viable?

 

Last time he talk to me, he says he can 1500$. So this one should be ideal right?

[DocSwag]

4 minutes ago, Antony Leung said:

Unfortunately, no. He did told me some clues, like gauss, take the calculations of the molecules, and chem renders or makes a 3D representation of it.  So my guess is gauss likes both but is gpu bounded and chem gpu.

 

So with that build of 1000$ is still viable?

 

Last time he talk to me, he says he can 1500$. So this one should be ideal right?

Those two look ok to me. Of course, I can't be 100% certain as it's possible they like quadros more but assuming they don't hate Ryzen and GeForce is better for them both look good

[Antony Leung]

5 minutes ago, DocSwag said:

Those two look ok to me. Of course, I can't be 100% certain as it's possible they like quadros more but assuming they don't hate Ryzen and GeForce is better for them both look good

Sure, but again the problem is money. 

 

Indeed would be perfect a P100 for his build and xeon cores, but again money. Something that I forgot, to ask him is ram. I dont know if 16 would be enough, Im thinking like those programs of rendering videos, they tend to like more 32 GB.

[Moress]

On 9/28/2017 at 2:55 PM, Antony Leung said:

 @deXxterlab97

When you're out of the loop..

[DocSwag]

1 hour ago, Antony Leung said:

Sure, but again the problem is money. 

 

Indeed would be perfect a P100 for his build and xeon cores, but again money. Something that I forgot, to ask him is ram. I dont know if 16 would be enough, Im thinking like those programs of rendering videos, they tend to like more 32 GB.

All I can tell you is I saw that Gaussian said that you should have 4gb ram per processor. However, I'm not sure if by processor it's referring to individual cores or actual individual CPUs  

13 minutes ago, Moress said:

When you're out of the loop..

Press F to pay respects

[Moress]

2 minutes ago, DocSwag said:

Press F to pay respects

F

[argyle]

I would still suggest looking at an off-lease Z series workstation and adding a higher powered GPU if you think you can run it that way. A lot depends on the optimization of the software and the OS settings. We generally run simulation software like those on higher core count Xeon with hyper threading enabled and a mid-range workstation GPU.

 

Truthfully while building might be fun it's probably a lot easier to use a workstation class machine with a longer expected lifespan and much easier hardware support (anyplace can get HP replacement parts easy and there are lots of them, plus they make troubleshooting easy with their own hardware testing). Plus since workstation systems usually get cycled out after a 3 year lease you can get really good machines for very little money.

[The Sloth]

On 2017-09-28 at 12:55 PM, Antony Leung said:

Hey @argyle and @Kiani, What do you guys think? I also call for help @STRMfrmXMN, @deXxterlab97, @Dackzy, @nerdslayer1, and @herman mcpootis.

i would suggest looking at old high core count XEON servers and a used Quadro GPUs. 

[Antony Leung]

On 10/2/2017 at 9:25 PM, nerdslayer1 said:

i would suggest looking at old high core count XEON servers and a used Quadro GPUs. 

Which quadro gpus? I'm lost on that.

[argyle]

The wikipedia page for Quadro shows a table that gives you an approximate consumer card to compare:

https://en.wikipedia.org/wiki/Nvidia_Quadro

 

For instance, a Quadro P6000 is closest to a Titan XP in the consumer space. It will give you an idea of the rough power to compare to a consumer card and you'll be able to look at pricing (and even looking at used Xeon workstations) and compare it.