Need Help With Ultimate Workstation & Gaming Build

DrGreer

Reputable
Apr 16, 2016
136
0
4,690
I need some advice for a new build I'm doing. I'd like to satisfy my use-cases for at least a few years. Those are...


  • Running deep learning libraries against CPUs & GPUs (Very important)
    OS virtualization with VGA passthrough (Would be nice)
    Audio & video production.
    VR gaming.

Because a couple of the machine learning libraries I use don't support GPUs, I want to do resource intensive virtualization, and adding more CPU power will obviously keep my build relevant to me longer, I've been thinking about using a dual socket board. In particular the ASUS Z10PE-D8 WS (http://www.newegg.com/Product/Product.aspx?Item=9SIA2F84D69056&cm_re=Z10PE-D8_WS-_-13-132-268-_-Product) in which I would place two i7-6850K (http://www.newegg.com/Product/Product.aspx?Item=N82E16819117647&cm_re=i7-6850K-_-19-117-647-_-Product). Plenty of speed and PCIe lanes (40 per chip). Asus lists Xeons as being compatible with that board but I believe Broadwell-E will work in it. Or should I just use some comparable Xeon model? Of course then overclocking is largely missed out on. Plenty of RAM slots for virtualization and I was also thinking about trying RAM disk for various applications including games. The 4-way SLI of the board is important to me though I'll be starting out with two GTX 1080s.

As for the dual-cpu issue, I read a post from April saying the BIOS for the board would support Broadwell-E shortly. I've seen no posts since. Maybe it's in the changelog if I can find it. This thread is mostly to weigh in on the board and CPU(s) and what other options (If any) I might entertain. I know some might say that dual CPU sockets aren't necessary but I believe that the use-case of my programming libs and running several OS simultaneously on a hypervisor warrants it.

Some case, cooling, and PSU recommendations would be appreciated as well.

thanks!
 
4-way SLI is less usefull with 1080s, as it is limited to only 2 cards in SLI.

Alsi gaming software is moving AWAY from SLI, especially for VR, because microstutter causes nausea in a lot of users.

Instead of 2 1080s, you might be better off getting a Titan Pascal.
Most games can handle multi-threading single CPUs, but not-so-much multiple CPUs.
 


There's some SLI problem in VR? Even with NVidia having a new SLI bridge just for the 10 series? The multiple CPUs aren't for gaming. They're for virtualization.
 
microstutter (getting frames out of sequence, or even just slightly delayed) can cause headaches and nausea in VR, instead of just being annoying.

In addition, support for multi-GPU setups has always been flaky—and it’s getting worse. Numerous big-name games over the past couple of years released with late, buggy, or no SLI or Crossfire support. What’s more, Nvidia and AMD seem ready to shrug even more of the multi-GPU support work off the shoulders of their drivers and onto game developers now that next-gen graphics APIs like DirectX 12 and Vulkan enable deeper hardware access for game makers. Game developers are a notoriously overworked and under-resourced bunch, which doesn’t bode well for SLI and Crossfire support getting more attention in the near future.

So single card solutions are best. for the cost of a pair of 1080s (about $1300) you can get a Titan X (Pascal) for $1200, which is the first single GPU able to push 60 fps at 4K.
 


I read in an article that not being able to SLI more than one 1000 series GPU is a myth. If you go on the NVIDIA site where they sell the new bridge for the 1000 series you'll also see that they do support 3 and 4way. Now, I did read that SLI is no good for machine learning but that multiple graphics cards on PCIe can be utilized. I'm leaning towards a Titan X and in the future possibly installing a second. We'll see. Of course, if I could be guaranteed that installing two 1080s (SLI or not) would double my machine learning crunching tasks for a single task I would rethink wanting a Titan X Pascal. That's the big question because if I have two 1080s for machine learning I can always just use one of them for VR gaming and game dev.