AMD Ryzen Threadripper 3990x - Parallel GA

20 visualizaciones (últimos 30 días)
Antonio Intini
Antonio Intini el 20 de Nov. de 2020
Comentada: Antonio Intini el 20 de Nov. de 2020
Hi there,
Is Matlab going to make use of all 64 cores of a AMD Ryzen Threadripper 3990x in parallel processing, such as generic algorithm?
Best regards

Respuestas (1)

Jason Ross
Jason Ross el 20 de Nov. de 2020
The "local" cluster supports up to 512 cores, so you can set the cluster size to 64 and it should use the cores. The ga function also has built-in parallel support.
Some other concerns might come up:
  • Each parallel worker is essentially another MATLAB session, so you need to have enough RAM to support 64 MATLAB sessions running simultaneously. The mimimum RAM requirement is 4 GB, recommended is 8 GB, and of course if your problem is bigger, you need more. So for supporting those workers you need 256 or 512 GB of RAM, maybe more.
  • Disk I/O : if you have 64 processes accessing disk simultaneously doing heavy I/O you can become bottlenecked on read or write. So you may need to look at technologies like RAID to spread the reads and writes over several drives. A single drive might become your bottleneck (especially if it's also doing OS stuff). I only recommend SSDs nowadays.
  • OS: Windows has different core limits for different versions. Make sure that you select a version that's allows the cores you have. Note also that it may be beneficial to disable hyperthreading, so you keep all your precessors in a processor group. On Linux this isn't an issue. For heavy math operations, hyperthreading doesn't really do anything for you anyway.

Categorías

Más información sobre MATLAB Parallel Server en Help Center y File Exchange.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by