Archive: GPU Accelerated Servers

IBM originally envisioned the new HC-10 Workstation Blade as a workstation replacement, say IBM’s Dave Laux and George Dolbier. It‘s a high density server with embedded GPU acceleration. “We found it also worked incredibly well as a headless render node,” Laux says a little wryly. Isn’t interesting when hardware reveals its own mission?


First let’s just say that a workstation blade is an interesting idea for those who can afford the infrastructure. It means that every seat in the house can share a common pool of workstations easily (in theory).


But think about the GPU accelerated renderfarm, Laux says. while things like 2D rendering, lighting, shading texturing move to the GPU (asssuminig you have Gelato, similar, or proprietary code to split the instructions), your same 1/2 RU hardware does more work faster for the same electric bill, heat load, liscensing costs, etc.


Time for a clarification. The name “workstation blade” can be confusing. It’s clear when you’re thinking about using them as workstations. But if you move to the renderfarm idea, understand that these blades are not workstations working as servers (as is sometimes the case in renderfarms). they are real servers with the floating point peformance, durability, networking, power supply, redundancy, and security advantages of servers as well as the server architecture that connects the CPU to the memory, network, disc drive.


Not for the faint of budget, but for the few–worth a look. Don’t try to penetrate the IBM maze. Just email Laux at dlaux@us.ibm.com.

Digg Syndication Del.icio.us Syndication Google Syndication MyYahoo Syndication Reddit Syndication

Related Topics: Workflow, Rendering, 2D, Hardware

Comments are closed.