Another New Computer Thread - Dual 6 Core Processors

[mkennedy]

If more cores are better, what about a phi co-processor? adds a TON of cores for relatively (per core) low cost.

[jcaffee]

Matt, Xeon Phi has looked interesting for some time. However,... From an article on Xcelerit, "Note that currently the Xeon Phi Co-processor is only supported with the Linux operating system. Given that most financial institutions use Windows for software development and production grids, we believe that this is a major limitation to its adoption."

Perhaps as the platform matures? Otherwise the new Tesla K40 sort of rocks the off-the-shelf graphics world. But then Chief Architect would need to be CUDA aware, and it's not. (Actually, only Phoebe would need to take advantage of CUDA to see an improvement within the NVidia tech stack)

jon

[HumbleChief]

Never heard of Xeon Phi and best I know less than more as it looks like more money down the computer rabbit hole.

Quick follow up on my new system. Motherboard can be finicky which is something i don't like but something that has to be sacrificed in this case in order to OC these Xeon processors.

And I'm learning that Fast is such a relative term. The GTX 780 is fast but doesn't seem that much faster than my old clunker 560. The prices have come down and that makes it a bit more compelling but I'm not sure a lesser 770 or 760 wouldn't be just fine.

The Dual Xeons do indeed RayTrace more than 3 times faster than my old machine but it still seems slow. Not sure what it will take to ever feel like computers are keeping up other than a complete change in CA's architecture engine and the Cuda Hybrid type graphics cards. Haven't really noticed any drastic speed ups in any other operations but those speed ups can be subtle and can lend themselves to a better 'feeling' system which I think is how I'd describe this new system.

Bottom line for me is still the same. Get as much processor and Graphics card as you can afford but be prepared to be undewhelmed at the performance increases.

[mkennedy]

Yeah, if phi is only supported in linux, the only use would be to run a virtual render farm, boot multiple instances of CA and run renders in parallel instead of qued. The overhead of dealing with booting CA in Linux seems pretty steep and the price of the phi, while reasonable per core, is still pretty high. Anyway, it all sounds WAY too complicated for me.
Larry, how many cores was the old machine? Is the speed\cores relationship fairly linear? i7\4 cores 1/3 the speed of xeon\12 cores?

[HumbleChief]

Larry, how many cores was the old machine? Is the speed\cores relationship fairly linear? i7\4 cores 1/3 the speed of xeon\12 cores?

Old machine quad core i7 - see sig. Impossible for me to say Matt as the clock speeds are different but one of the Chiefers said that the amount of cores will increase the speed in a fairly linear fashion. Twice the cores, not quite twice the speed increase..

[Doug Park]

Xeon Phi and GPU support is something that needs to be specifically programmed into the application. The Tesla HPC card is really no different (other than having many more cores) than what is already available in your GPU. Moving forward we are likely to support using OpenCL to access the GPU which would make it possible to leverage the Tesla cards. As I understand it the Xeon Phi can be accessed from a Windows machine, but also requires special programming to use.

[mkennedy]

Doug, thnx, LArry, likewise.
It sounds like cores are as, or more, important than clock speed.
I'll try pricing clock speed against cores, see which is cheaper, but it sure seems that parallel processing should be faster\cheaper than serial?