In the last of my GPGPU/PowerEdge C410x trilogy I offer up a whiteboard session with the system’s architect, Joe Sekel.
Some of the topics Joe walks through:
- How does having remote GPGPUs connected via cable back to a server compare in performance to having the GPGPUs embedded in the server?
- The topology of the PCI express x16 (16 lanes per link) plumbing: from the chipset in the host sever through to the GPGPU.
- The data transfer bandwidth that x16 Gen 2 gives you.
Extra-credit reading:
- Blog: Say hello to my little friend — packing up to 16 GPGPUs
- Blog: Deep dive tour(s) of the PowerEdge C410x
- Blog from our GM about today’s announcement: Paving the way for powerful thinkers
Pau for now…