[EternalSeekerX]

Hello,

So as a design challenge, we had to design a prosthetic arm for a pro cyclist. One of the aspects was exploring possible improvements, which I did by looking at aero of different arm shells for cyclist performance on races likethe Tour de France. The geometry of the cyclist was a 3D scan (no wheel spoke) and the arm shells were overlapped with the 3D scan geometry and united. This was done in RANS using Fluent. Mesh sensitivity study showed 15 million cells as the sweet spot and for the same flow conditions, I validated my cd to 1 thousand. You can see the before and after as a picture attachment below:

Baseline
Baseline.jpg

Airfoil Arm
AirfoilArm.jpg


Basically, for a 15 million cell case, a 16GB Intel i7 7700HQ was enough, it took approximately 9835.885 sec. I compared this to my i7 5960X (overclocked to 4.0Ghz), it took 5778.035 sec. (For reference my 5960X runs motorbike using 6 cores in 181 seconds).

This got me thinking: If my old laptop is only approximately two times slower (which makes sense due to dual vs. quad channel DDR4), newer laptops and desktops should see huge improvements and become more viable. Is this true?

Can decently sized problems now be solved with consumer laptops? Or are HEDT and server chips still the de facto? How does newer hybrid architecture affect CFD? IIRC, dual-channel DDR5 has similar memory bandwidth to quad-channel DDR4. Is this true?

If so, does that mean any laptop or consumer desktop with the highest clock speed, memory speed, and lots of cores will perform CFD very well?

Would love hear from the experts!

[Simbelmynë]

Originally Posted by EternalSeekerX

Hello,

So as a design challenge, we had to design a prosthetic arm for a pro cyclist. One of the aspects was exploring possible improvements, which I did by looking at aero of different arm shells for cyclist performance on races likethe Tour de France. The geometry of the cyclist was a 3D scan (no wheel spoke) and the arm shells were overlapped with the 3D scan geometry and united. This was done in RANS using Fluent. Mesh sensitivity study showed 15 million cells as the sweet spot and for the same flow conditions, I validated my cd to 1 thousand. You can see the before and after as a picture attachment below:

Baseline
Attachment 102247

Airfoil Arm
Attachment 102248


Basically, for a 15 million cell case, a 16GB Intel i7 7700HQ was enough, it took approximately 9835.885 sec. I compared this to my i7 5960X (overclocked to 4.0Ghz), it took 5778.035 sec. (For reference my 5960X runs motorbike using 6 cores in 181 seconds).

This got me thinking: If my old laptop is only approximately two times slower (which makes sense due to dual vs. quad channel DDR4), newer laptops and desktops should see huge improvements and become more viable. Is this true?

Can decently sized problems now be solved with consumer laptops? Or are HEDT and server chips still the de facto? How does newer hybrid architecture affect CFD? IIRC, dual-channel DDR5 has similar memory bandwidth to quad-channel DDR4. Is this true?

If so, does that mean any laptop or consumer desktop with the highest clock speed, memory speed, and lots of cores will perform CFD very well?

Would love hear from the experts!

If you talk about a laptop where battery life is prioritized highly, also provided that you have compatible software, then I would easily go with a Macbook Pro. The M3 Max did the benchmark in about 54 seconds, even while on battery (!). I have not seen any M4 Max benchmarks yet, but it should be a significant improvement.

OpenFOAM and Comsol works natively on Apple silicon, but there is a huge amount of software that does not work natively yet. I have a colleague that runs Fluent in a virtual machine (parallels) on his Macbook and it works really well for pre- and post-processing, and the occasional solution of a smaller system.

[andy_]

Originally Posted by EternalSeekerX

If so, does that mean any laptop or consumer desktop with the highest clock speed, memory speed, and lots of cores will perform CFD very well?

Generally consumer hardware can run CFD OK but not as well as equivalent server hardware. Consumer hardware tends to have corners cut particularly w.r.t. to memory handling in order to get the price down which shows up in CFD benchmark as poorer scaling to realistic problem sizes and poorer scaling with numbers of cores/processors compared to more expensive equivalent server/workstation hardware. You can see this in the last few posts in the OpenFOAM benchmark thread pinned at the top of the forum for both Apple and Intel consumer hardware in comparison with earlier results for server/workstation hardware.

Whether consumer hardware or server hardware is a more appropriate choice is going to depend on the CFD problem sizes, the number of CFD problems that need to be run, what else one wants to do with the hardware, budget, etc... For occasional smallish CFD runs consumer hardware is likely to be fine but for sustained running of largish CFD problems server hardware is likely to be more cost effective and reliable.

[EternalSeekerX]

Originally Posted by Simbelmynë

If you talk about a laptop where battery life is prioritized highly, also provided that you have compatible software, then I would easily go with a Macbook Pro. The M3 Max did the benchmark in about 54 seconds, even while on battery (!). I have not seen any M4 Max benchmarks yet, but it should be a significant improvement.

OpenFOAM and Comsol works natively on Apple silicon, but there is a huge amount of software that does not work natively yet. I have a colleague that runs Fluent in a virtual machine (parallels) on his Macbook and it works really well for pre- and post-processing, and the occasional solution of a smaller system.

Awesome! The numbers you quoted is approx twice as fast as my overclocked i7 5960x which released in 2013. I can see why though, unified memory goes brrrt. I also saw how well it compares to new x86_64 systems too. Pretty nice. I still feel x86_64 > arm64 though. However I'd love to see more love for ampere systems, maybe an ampere laptop.

P.S. checked out the max mini m4, super cute ��

Originally Posted by andy_

Generally consumer hardware can run CFD OK but not as well as equivalent server hardware. Consumer hardware tends to have corners cut particularly w.r.t. to memory handling in order to get the price down which shows up in CFD benchmark as poorer scaling to realistic problem sizes and poorer scaling with numbers of cores/processors compared to more expensive equivalent server/workstation hardware. You can see this in the last few posts in the OpenFOAM benchmark thread pinned at the top of the forum for both Apple and Intel consumer hardware in comparison with earlier results for server/workstation hardware.

Whether consumer hardware or server hardware is a more appropriate choice is going to depend on the CFD problem sizes, the number of CFD problems that need to be run, what else one wants to do with the hardware, budget, etc... For occasional smallish CFD runs consumer hardware is likely to be fine but for sustained running of largish CFD problems server hardware is likely to be more cost effective and reliable.

I do wonder, with laptops capable of 128gb ram, they could handle some sizable problems. It got me.wondering if memory speed, core speed and core count are the only key factors since all laptops are still locked in memory channel. In my opinion, I think runs that take 1 or even 2 days on a laptop is appropriate if you don't need the laptop in a mean time.

I mean we would all want dual eypc and a few a100's. But as a student it's expensive for sure. I just wanted to get a graps on the type of specs you would target for a laptop (even I guess regular desktops) for cfd loads. Or would gaming laptops/desktops or computer for other workload be better since cfd will perform the same across the range of cpus (I.e. cfd performance is the same between low end and high end ryzen for example)? Would intel be off the table with its e/p cores? How about zen 5/zen 5c?

To bad used server parts suck to buy in canada. I saw a few v100 for sale, but I wonder if they would accelerate my old i7 desktop if I use gpgpu now.