Current best hardware for Nastran?

[will_roe]

Hi,

I'm currently specing a workstation with a near open end budget. 

I'm hoping for recommendations for hardware/specs specific for Nastran. I've done a fair bit of research in the past revoling around Inventor modeling and drawings specs and have watched the full TFI/Tech3D buying the right workstation video. But can't seem to find much on what will boost Nastrans preformance.

Would there be any difference between a Ryzen 7950x and a intel 13900k specifically for Nastran?

Does Nastran use vram the same way inventor?

Is there a limit on how much ram is useful?

Which analyses can utilise parallel processing?

Are we better off running linear analysis in Inventor stress analysis if possible?

Any help appreciated 👍

[delaroca · Accepted solution]

Hi, @will_roe 

Inventor Nastran makes the most use of the CPU and the Hard Drive (HD or SSD). So those two should be your hardware priority (CPU: high clock speed*core count; Hard Drive: Fast read/write speed).

Comparing CPUs without testing is mostly a guessing game. My guess is that the Ryzen would perform slightly better due to the overall higher base clock speed throughout the cores, and higher overclock speed. But either of them would be a good choice. The intel Xeon-W series is usually recommended for heavy workloads across long periods of time, for personal workstations. But in terms of "finishing an Inventor Nastran simulation as fast as possible", I think it should lose to the new generation i9 or Ryzen 7950x.

Keep in mind that Nastran is not an optimal software for parallel computing, after a certain number of cores (such as 8 or so), the performance boost of additional cores doesn't add significant improvements such as from 1 to 2, 2 to 4, etc. At that point, a higher clock speed should be more beneficial. As far as I know, any Inventor Nastran simulation can benefit from parallel computing, by altering the NPROCESSORS parameter to match your number of CPU physical cores.

The hard drive (SSD) also plays a big role, as the solver is constantly writing/reading information during the simulation. That should be easy to solve, the faster, the better. The AORUS Gen4 AIC of 8TB can reach about 15000 MB/s of both read and write speed, that's the fastest I found.

Inventor Nastran doesn't use GPU power. So you can ignore that aspect. And as for RAM/VRAM, it should be only enough to handle the size of your assembly files with ease. But regardless of how powerful your computer is, you'll always want to optimize your simulation, and simplifying the model/reducing interacting parts is usually the best route. Even the fastest supercomputers on earth have a hard time solving big element count/project parts simulations (that's when simulating by simplified analytical mathematical models comes in).

For the linear analysis, I usually prefer the Nastran one, since it has more control options. But they both behave the same.

Best regards,

Leonardo de la Roca

[cfagerst · Accepted solution]

Hello,

I can only speak of experience in inventor and playing around with invmark. There, the 13900K having some more IPC takes the crown in the mostly single threaded inventor. In the multithread benchmark (rendering) of invmark and in CB23 it's quite close between the 13900K and 7950X. If Nastran is only partly multithreaded, or does not so well make use of lots of thread, my guesstimate is Intel will perform slightly better of the two. With both, one would like to go for fast RAM, I did a better invmarkscore on both Intel and AMD with going from 4800 to 5600 and 6000MHz DDR5. Even faster could be good if money not an issue. Seems to be a bottleneck with these fast CPUs of today. Another very important factor is cooling. I wouldn't go for anything less than a 280mm AIO, something in the likes of an Arctic Freezer with a fat radiator. This ofc means custom build is the way to go. Prebuilt manufacturers don't seem to understand (or care) how much these new chips need power and put a cooler on them that looks like a freaking joke (Like the 120mm AIO on a 12900K Dell Precision, haven't seen any 13900K or 7950X prebuilts yet, but I can imagine what it will be.). Bad cooling/not supplying enough power is the most efficient way of killing the otherwise great performance of these new CPU:s What comes to the SSD, from what I tested in invmark, this doesn't make much impact at all, ofc again not familiar with nastran's use of storage. If it's very storage intensive, one would think that some raidcard with 4 NVMEs in a PCIE x16 slots could kick some butt, but in invmark that would hardly make any difference. Most if not all Mobos you like to pair with a 13900K (or 7950X) will have PCIE5 for NVME, and those drives should be able shortly from what I've read so if you can wait, get one of those as OS drive (Or 2 in raid0 if feeling like it) and speed should be similar to 4x PCIE4 NVMEs in Raid0.

[will_roe]

Awesome! Thanks for your input.

Many Thanks.