[coffee_break]

Hello,
I am planning to build a new PC for OpenFOAM simulations. I have two options for the CPU: the AMD Ryzen™ 9 9900X and the AMD Ryzen™ 7 9800X3D. Which one should I prefer? The 9900X has 12 cores, while the 9800X3D features 3D V-Cache technology.


Additional question: Is it worth waiting for RAMs with speeds over 8000 MHz?

[wkernkamp]

Originally Posted by coffee_break

Hello,
I am planning to build a new PC for OpenFOAM simulations. I have two options for the CPU: the AMD Ryzen™ 9 9900X and the AMD Ryzen™ 7 9800X3D. Which one should I prefer? The 9900X has 12 cores, while the 9800X3D features 3D V-Cache technology.


Additional question: Is it worth waiting for RAMs with speeds over 8000 MHz?

I have a Ryzen 7 7800X3D:


1. The V-Cache is worth a lot for CFD. More cache is less memory access needed.

2. I get my best openfoam (stable) results with memory at 7200, fclk at 2167 and uclock at 1800 (this is uclk=mclk/2)
3. I get a slightly faster result with 6400 and uclk=mclk, but not quite stable.
4. My DDR5 is good for 8000, but motherboard not. It is a cheap $88 Gigabyte A620M S2H.
5. It should be possible to get faster solutions with 7950X3D if you deactivate 5 cores on the none 3D chiplet. You should get about 11/8 speedup because you make use of the additional infinity fabric from the non-X3D chiplet while the fabric to the X3D chiplet is already maxed out. Nobody on openbanchmarking.org has done that. My computer is the fastest Ryzen 7000 on the drivaer medium solution at just under 1800 seconds.
6. You should go to openbenchmarking.org and compare 7900X and 7800X3D. The performance ratio should be similar for 9900X versus 9800X3D. I recommend the 9800X3D.
7. It is possible to run the 7800X3D with DDR5-8000 and for openfoam that should help. Same probably holds for 9800X3D. Make sure your motherboard can do that speed (probably just with one DIMM per Channel.) Since the motherboards are the limiting factor for now, focus on the motherboard capability.

[coffee_break]

Originally Posted by wkernkamp

I have a Ryzen 7 7800X3D:


1. The V-Cache is worth a lot for CFD. More cache is less memory access needed.

2. I get my best openfoam (stable) results with memory at 7200, fclk at 2167 and uclock at 1800 (this is uclk=mclk/2)
3. I get a slightly faster result with 6400 and uclk=mclk, but not quite stable.
4. My DDR5 is good for 8000, but motherboard not. It is a cheap $88 Gigabyte A620M S2H.
5. It should be possible to get faster solutions with 7950X3D if you deactivate 5 cores on the none 3D chiplet. You should get about 11/8 speedup because you make use of the additional infinity fabric from the non-X3D chiplet while the fabric to the X3D chiplet is already maxed out. Nobody on openbanchmarking.org has done that. My computer is the fastest Ryzen 7000 on the drivaer medium solution at just under 1800 seconds.
6. You should go to openbenchmarking.org and compare 7900X and 7800X3D. The performance ratio should be similar for 9900X versus 9800X3D. I recommend the 9800X3D.
7. It is possible to run the 7800X3D with DDR5-8000 and for openfoam that should help. Same probably holds for 9800X3D. Make sure your motherboard can do that speed (probably just with one DIMM per Channel.) Since the motherboards are the limiting factor for now, focus on the motherboard capability.

Thank you it is very detailed answer.