I'll be looking to purchase a new PC soon to upgrade to Windows 11. Just thought to poll Forum members about their current hardware setups they are happy with, particularly with FEA performance.
I have also been thinking but about the similarities between FEM and AI. The issue with larger models seems to be bandwidth. The new unified systems might be good for both, though the only economical X86 one at the moment is the Ryzen AI MAX+ 395, the 8060S is not an Nvidia card so Callculix could not use it to help with 64 bit ops (no Cuda). If the followon to the Ryzen AI MAX+ 395 if it had a fast pci bus an Nvidia Blackwell card could be installed on it. Maybe a riser from the M2 slot for the Nvidia GPU. A single Epyc Server might get around this with a GTX5090, though the bandwidth to the GPU by pci would be relatively slow. The MAC systems are pricey and not Cuda or X86. A rewrite of Callculix for RocM might do the trick, but that is a big task. Nvidia's price per 64 bit flop is rather high. I think there will be something worthwhile in a year or so but there would be a lot of reprogramming. I think we are a year too early.
I am just thinking about Hanging with 10 for another year and getting a GTX 5070 TI Super when they come out in the spring to transfer to my next computer. Been thinking about my grandson who is a computer wiz. Taking an AI graduate course so I investigated how to get usable local AI from his old Ryzen 1700 16GB linux rig. The consumer blackwells are still crippled in double precision, but that part is not used in AI.
Comments