There are reasons for each of your observations. My experiences have lulled me into making these estimations (some old, others new):
1. CURRENTLY, Linux (free) is the best/stable OS for systems built to house eight or more GPU processors. If I had been asked "which distro" two weeks ago, I would have then responded - "Ubuntu Studio and Linux Mint." However, I'm studying what appears to be a real challenger for the crown - whimsically called "PuppyLinux." [See
http://puppylinux.com &
http://puppylinux.org/main/Overview%20a ... tarted.htm ] Windows (not free) is best for systems built to house less than eight GPU processors if one wants to overclock them. Windows systems are more easily overclocked by the free overclocking utilities that are available. Notice that I use the term GPU processors because, like you, I also have dual processor cards such as GTX Titan Zs, GTX 590s and a GTX 690 [such cards can be designed to handle IO space issues more efficiently because the two GPU processors share almost all other card components]. Also, different OS versions can make/break the or/deal. In the situation you describe, is it ordeal (without the slash)? Luckily, Linux has been supporting the top GPU maxed out systems long before Windows 7-10 were announced.
2. Newer GPUs tend to use more input/output (IO) space than their predecessors. This trend I project will continue. New/additional/improved features lures us to buy them. Those features can come at a hidden cost to those in need of running more many GPU processors in a single system - they require more & more IO space. The upside is that "Newer GPU Processor" generally means "Faster Rendering GPU Processor."
3. Nvidia erects IO caverns in its products a lot differently than AMD does in its products. That difference can be a good thing - that means than even when a system lets one know that it's IO space for GTXs has been tapped out, usually there's still a lot of empty IO space to support some AMD GPUs. So, if one also has a need for OCL rendering, there's room enough to make CUDA and OCL rendering occur simultaneously with a bunch of AMD and Nvidia GPU processors. I've been mixing them for years. But I do have many, many more GTXs than I have AMDs.
4. Different motherboards support different amounts of IO space. When comparing motherboards, note the difference in the number of visible PCI-e slots. Generally a good indicator of which motherboard will support the the greater number of GPU processors is the one with the most PCI-e slots. The number of visible slots isn't necessarily the max, but I suggest most often that the motherboard with more such slots is more likely to support even more GPUs than the comparator
*1/].
5. Generally, the more CPUs a system has, the greater the number of GPU processors that can be supported. Why? A two CPU system generally has the potential to handle more IO activity than a single CPU system has. Moreover, a quad CPU system generally has the potential to handle more IO activity than a dual CPU system has, etc. etc. I still recommended
*2/ ],that we completely digest the graphical depiction of the target system's motherboard layout. How many "words" can a picture be worth?
6. Regarding your last post, try to give me each and every change that you made to each component of the system, when the change occurred and what OS was used at what period in time. Are (or were) you using Amfeltec components and/or other splitters/other special connectors, and if so, which ones and has there been a change involving any of them, the nature of the change and when did the change occur? Frankly, I'm not surprised that you cannot now run more GPUs in the Windows version that you now have. What would be surprising is that you hadn't made any changes to the system and could run six Titan Z cards in the past under Windows pre-version 10 and not Server.
*1/ I stand on the shoulders of the greatest pioneer. Back on a day in late 2013 - that great pioneer -
zzz1000 posted a topic entitled - "18 GPUs in a single rig and it works." [
https://devtalk.nvidia.com/default/topi ... -it-works/ ]. I've read that topic (and others of his posts elsewhere) many times since I was blessed to find them and on each occasion I've said, "How did I miss that next point that was revealed so subtilely by his recorded observations?" Note what motherboard he says that he used and what motherboard model that I currently have seven of. Also, note who zzz1000 was inspired by, then study the posts regarding FASTRA.
*2/ [
viewtopic.php?f=40&t=43597&start=200#p241271 ]
Because I have 180+ GPU processers in 16 tweaked/multiOS systems - Character limit prevents detailed stats.
