Great video, next time if you can give a little more details about the computer build with some why you choose some components, it will be very helpful i am mechanical engineering myself and I'm trying to get a little deeper in the world of simulation.

You are missing one crucial aspect of a good FEA/CFD workstation when running on a "gaming" CPU. While the single core performance of these CPUs is great, these CPUs have only 2 memory channels while server CPUs have 4 and upwards memory channels (todays CPUs can have 8 or even 12 memory channels). Having a high clock speed core is moot when you don't have enough memory channels to transfer the data across the cores, that's the main reason FEA workstations work with server grade CPUs, you most likely bottlenecked the CPU with the constrain on the memory distribution between cores. Although if you only have a license for 8 cores you are probably fine (the rule of thumb is 4 cores per channel), but then you could have chosen much more "economically efficient" CPU since most of your cores are un-usable during a simulation run.

Hello, just wondering why you focus on raw CPU throughput instead of GPU acceleration? Ansys has a GPU override env var, and SOLIDWORKS can be overridden as well (all within ToS of course), so I don't quite understand why you wouldn't simply utilize multiple RTX 4090s or several V100 SXM2s (using the Supermicro X10DGO-SXMV and RSC-G-6)? I've built a quad-V100 SXM2 server for US$2000, and it performs quite while especially at that price. Please feel free to correct me, I feel like I've missed some important details.

@enriquelopez5624

2 ай бұрын

hello, my friend, regards from Arequipa, Peru. Southamerica. I am begginer worker as a mechanical engineer. I totally trusth and agree with your comment. Can you share to me the parameters and characteristics of your computer of 2000 dolars, thas is the money i have to buy a desckop computer. Many thanks

copy of copy of copy lim tends to infinity 🤣 4:35

hey mate nice video btw, can you put the specs on it? i want to build one like yours, i mainly use ansys and solidworks. do ansys really benefit from a gpu that have double precision one or you can get away with gaming gpu like rtx 4090? thanks in advance

@elementengineering

Жыл бұрын

Hi Manray8513. Sorry, i've just seen this comment. Have you built a machine yet? To be honest, we haven't tested double precision yet. This new machine has a Quadro RTX A4000. We previously used a GeForce RTX 3090. Using the 10,000odd CUDA cores does speed certain simulations up, but I believe we have found it to be marginal. But we should probably test it out. Let us know if you have a specific use case and we might be able to organize running a simulation on either machine with different parameters and it could help you make a decision if you haven't made one already?

@manray8513

Жыл бұрын

@@elementengineering all cool man! keep rippin, i finally bought refurbished dell t7910 with dual xeon e5-2699 v4, 512 gb ram @ 2600 mhz, rtx 3090 turbo and 3 tesla v100 16gb watercooled them all, what i can say is, this machine is really fast compared to my old setup dell t5610 from 2013 is like 3 times faster for solidworks simulation and 15 times faster on ansys mechanical

@klemen4607

7 ай бұрын

@@elementengineering I understand the GPU choice but i'm curious about CPU. Wouldn't the use of HEDT CPUs like xeon or threadripper have benefits in combination with 4 channel / 8 channel ECC ram?

@elementengineering

7 ай бұрын

@@klemen4607 Good question! I believe the answer is in single core computational performance. I would have to check some synthetic workload performance tests, but from memory, I believe we did look at the HEDT machines and the single core performance and multicore when only using 4-8cores was much better from the gaming CPUs. Also availability of parts. Ayrton

Are you using or considered water cooling?

@elementengineering

Жыл бұрын

Yes, it's an off-the-shelf All In One (AIO) cooler.