Outlining my GPU cluster


  • Cluster will have 8-16 disk-less nodes with GPU-s and 1 server;
  • I like NVIDIA GTX 295 cause of 55 nm technology, less heat, less current per PPD.  WE have 4 PCI-s to spare on 1 host, density is very important. Each new host/node adds overhead cost. I believe, that next core will support more streaming processors.
  • Most probably next GTX 300 family is still 55nm, not 40 nm (as ATI), and there is no point to wait GTX 300; 
  • I’m wondering, why ATI is so poor in computing and good in games…. NVIDIA is my choice for GPU;
  • 19″ rack mount is excellent basis. I would put at least 2 ATX side by side, both equipped with 4 GPU double cards. All air-cooled. Total thermal output is 9-20 KW;
  • Diskless nodes, PXE-booting up via Ethernet from server HDD;
  • Custom-made central power supply ( One for all 8 to 16 nodes, ca 20 KW output);
  • I would leave CPU / SMP folding out of picture. I’d rather use MSI K9A2 Platinum 4 x PCIx16 MB’s and cheap Phenom X3 CPU-s with 2 GB RAM (enough);
  • Each node will gave 4 x double-GPU cards — GTX 295 for example);
  • Very important aspect is management in such system. I dont have solution Yet, but VMWARE, Windows HPC will do it most probably. 
  • I would prefer Win platform, cause of better drivers for GPU-s.

Leave a comment

No comments yet.

Comments RSS TrackBack Identifier URI

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s