[crush56]

I’m currently solving some personnal simulations on my 5900x / 32 gb ddr4
Most of these simulations take weeks to a month to complete (around 5 millions mesh element) with 8 cores used (i have no licence limit when it comes to cores)
Do you think upgrading to a 7950x (and by the way adding 32 extra gb of ddr4) would change much ?
In a other hand i could wait for next Xeon, i was thinking about W-2445 which should be around 850$ and brings back avx 512 support.

What do you think ?
Thank you

[evcelica]

With analyses this large (5M mesh elements) and only 32 GB RAM, you are probably solving out of core? If so, adding enough RAM to solve in core would make the largest difference. The fastest CPU and RAM in the world will still be slow if you don't have enough RAM space to fit the model, as you are limited completely by drive speed. Increasing drive speed would help if solving out-of-core is unavoidable.

[crush56]

Originally Posted by evcelica

With analyses this large (5M mesh elements) and only 32 GB RAM, you are probably solving out of core? If so, adding enough RAM to solve in core would make the largest difference. The fastest CPU and RAM in the world will still be slow if you don't have enough RAM space to fit the model, as you are limited completely by drive speed. Increasing drive speed would help if solving out-of-core is unavoidable.

Thank you for your answer, so basically should I go for 64/128 gb ?

[flotus1]

Check if you are currently running out-of-core. Very likely. And check how much disk space is used. Then you can decide how much memory you need, and if 128GB are even enough.

[crush56]

How do I know I’m running out of core ? (Dumb question sorry)
And it takes some times 40/60 gb on my SSD

[flotus1]

I'm not familiar with mechanical, but there should be some log files where this information is kept.
Anyway, with 60GB used on your drive, it is safe to assume that you need an upgrade to 128GB to avoid out-of-core.

[crush56]

Originally Posted by flotus1

I'm not familiar with mechanical, but there should be some log files where this information is kept.
Anyway, with 60GB used on your drive, it is safe to assume that you need an upgrade to 128GB to avoid out-of-core.

I can do that (and I will), but aside the ram upgrade and assuming I’m no longer memory «*bottlenecked*», do you think upgrading the cpu to 7950x is worth the money or it’s a waste ?

[flotus1]

Total waste of money in my opinion.
On top of the CPU, you would need to pay for an expensive motherboard, and DDR5. Considering you need 128GB, doing that with DDR4 is much cheaper. Decent DDR4 memory is less than 350€: https://geizhals.eu/g-skill-ripjaws-...c=uk#offerlist
And I don't think this would run significantly faster with a 7950x.

[crush56]

Originally Posted by flotus1

Total waste of money in my opinion.
On top of the CPU, you would need to pay for an expensive motherboard, and DDR5. Considering you need 128GB, doing that with DDR4 is much cheaper. Decent DDR4 memory is less than 350€: https://geizhals.eu/g-skill-ripjaws-...c=uk#offerlist
And I don't think this would run significantly faster with a 7950x.

Okay thank you very much !

[evcelica]

Originally Posted by crush56

How do I know I’m running out of core ? (Dumb question sorry)
And it takes some times 40/60 gb on my SSD

While or after it is solving, check the "solution information" tab in workbench (just below the solution on the tree on the left). This is the output file (text file) which will also be in your scratch directory, called "solve.out".

Somewhere in there it tells you the solver type, RAM requirements and whether it is solving In-core or out-of-core. Below is one of mine, where it is solving on 12 processors using the Direct solver, and In-core:

This Analysis had 600k nodes, used the direct solver, and required:
26GB RAM for in-core.
7GB RAM for out-of-core
The iterative solver requires far less RAM than the direct solver.

************************************************** ************
DISTRIBUTED SPARSE MATRIX DIRECT SOLVER.
Number of equations = 1752991, Maximum wavefront = 1242

Local memory allocated for solver = 3094.120 MB
Local memory required for in-core solution = 2705.939 MB
Local memory required for out-of-core solution = 562.808 MB

Total memory allocated for solver = 29739.451 MB
Total memory required for in-core solution = 25998.543 MB
Total memory required for out-of-core solution = 6880.073 MB

*** NOTE *** CP = 117.250 TIME= 13:03:59
The Distributed Sparse Matrix Solver is currently running in the
in-core memory mode. This memory mode uses the most amount of memory
in order to avoid using the hard drive as much as possible, which most
often results in the fastest solution time. This mode is recommended
if enough physical memory is present to accommodate all of the solver
data.
************************************************** ***********