Known Socket 1150 Gigabyte boards with working vt-d.

[Robstarusa]

Has anyone confirmed any gigabyte boards function properly with vt-d (and hopefully vga passthrough)?

I'd like to pass through either a 6850 or 6870 to a windows guest.

I'm looking for a board recommendation.

Thanks,

Robert

[RootWyrm]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

Has anyone confirmed any gigabyte boards function properly with vt-d (and hopefully vga passthrough)?

I'd like to pass through either a 6850 or 6870 to a windows guest.

I'm looking for a board recommendation.

Thanks,

Robert

*puts on the virtualization hat*
Hi Robert. I'm RootWyrm. I'm kind of an expert on this stuff.

And you, uh, kinda neglected to mention what hypervisor you're intending to run. Which matters a lot. Though honestly the TL;DR version as far as 1150 desktop boards and VT-d for video is almost always "nope." Reason being that the BIOS is not capable of having it's video output snatched away, and snatches all outputs by default. And I presume what you actually mean is doing PCIe access via VT-d so the guest can access the GPU directly for offload or render.

Not saying impossible, but honestly I've yet to encounter a desktop board that can do VT-d for video in anything resembling a sane fashion. Even with multiple distinct models (e.g. mixing non-crossfire capable cards like 6970 and 4870.) It's a common thread across all manufacturers. The desktop chipsets and BIOSes are just not meant for it by design. There are several non-desktop boards I can recommend depending on hypervisor, but need more information on how exactly you're planning to virtualize and configure the hardware.

[Robstarusa]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

*puts on the virtualization hat*
Hi Robert. I'm RootWyrm. I'm kind of an expert on this stuff.

And you, uh, kinda neglected to mention what hypervisor you're intending to run. Which matters a lot. Though honestly the TL;DR version as far as 1150 desktop boards and VT-d for video is almost always "nope." Reason being that the BIOS is not capable of having it's video output snatched away, and snatches all outputs by default. And I presume what you actually mean is doing PCIe access via VT-d so the guest can access the GPU directly for offload or render.

Not saying impossible, but honestly I've yet to encounter a desktop board that can do VT-d for video in anything resembling a sane fashion. Even with multiple distinct models (e.g. mixing non-crossfire capable cards like 6970 and 4870.) It's a common thread across all manufacturers. The desktop chipsets and BIOSes are just not meant for it by design. There are several non-desktop boards I can recommend depending on hypervisor, but need more information on how exactly you're planning to virtualize and configure the hardware.

Hypervisor: First choice is KVM, second Xen -- either OSS 4.3 or XenServer/XCP. Not interested in vmware in any fashion as it's too limited (last I checked).
Yes, I'd like the guest to have exclusive access to pci-e video as well as other pci-e cards. What do you mean by "in a sane fashion" ? I currently have several pci-e x1 sata boards (ASM1061 chipset, SIL3132, iirc), network cards (intel based) as well as a radeon 6850 & 6870.

What boards do you know actually worK? I'm only interested in single socket stuff, and I need at least 2 pci-e x16 slots.

[RootWyrm]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

Hypervisor: First choice is KVM, second Xen -- either OSS 4.3 or XenServer/XCP. Not interested in vmware in any fashion as it's too limited (last I checked).
Yes, I'd like the guest to have exclusive access to pci-e video as well as other pci-e cards. What do you mean by "in a sane fashion" ? I currently have several pci-e x1 sata boards (ASM1061 chipset, SIL3132, iirc), network cards (intel based) as well as a radeon 6850 & 6870.

What boards do you know actually worK? I'm only interested in single socket stuff, and I need at least 2 pci-e x16 slots.

Ouch. Well. Here's the super short answer:
What you want is not possible with 1150. End of story.

Longer answer is that it is only possible with devices which have at minimum a fully interposed switch or bridge, period. One that does dual 16x mechanical even at 2.0 is going to cost you more than board, CPU and memory combined. And you'll only get single 16x 2.0 bandwidth since that's how it connects - and splits to 2x8x electrical. Again, not a Gigabyte thing - an EVERYONE thing. Chipset's not designed for it, BIOS isn't designed for it. And with all those low quality SATA HBAs in the mix, you'd just have a study in suffering.
On top of that, you're talking about an 1150. Even the top end desktop 1150's don't do 2x16; it's 1x16+1x8. (See G1.Sniper M5, Z87X-UD5H, etc.) I searched high and low for one with a true 2x16 and it doesn't exist from any manufacturer in an ATX form factor. Found one board that did 2x16 in uATX with zero other slots. (And it was discontinued due to being unbearably buggy.)
The absolute only chipset where you can use a on-CPU graphics and VT-d is the C226 and exactly one, and I do mean absolutely only one to date all other claims are false Z87 board - Supermicro C7Z87-OCE, only with BIOS R1.1 and later, only with explicit VT-d CPUs (as confirmed by ARK.) And that still only gets you 8/8 or 8/4/4 in the 16x mechanical. Maybe Gigabyte GA-6LX family (C224/C226) but those are pure unobtanium unless you're building a hundred of these boxes and you're STILL at a 16x+8x maximum. It's a CPU limitation.

2x16 means you need a 2011. Wanting to do VT-d with GPU means server board, pretty much. Probably if you could get your hands on one, the GA-6PXSV2 (second cousin to the X79S-UP5-WIFI) would be your best bet as it has dual 16x spaced for dual slot, 6x SATA2 + Mini-SAS which is expander capable. Which means yes, throwing away all your SATA cards. But for a much better SAS controller so yeah. Other than that? There really aren't any good options out there that I've found to date, not in a desktop form factor.

[Andres Gonzalez Biront]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

I have a Gigabyte GA-X79-UD3 and PCIe passthrough works perfectly fine with USB2, USB3, GPU (HD7950) and SATA devices (I don't remember If I tested with G-SATA or Intel SATA). I also have an old BIOSTAR S1156 with a Core i5 650 on it working "fine" with VT-d, with a FXO card and an Intel NIC (PCI and PCIe) but not video, I didn't test it.

Everything on VMware ESXi 5.0 (5.1 and 5.5 have problems with the USB devices, and VMware states that it's not supported).

The only thing I couldn't get to work was the Creative X-Fi PCI card. It's on PCI bus, so, that's not very good for Passthrough. The Guest OS sees it, but the moment I install the drivers the sound gets corrupted. But it's Creative, I can't really blame the motherboard, nor the Hypervisor. If I'm not mistaken, I think the onboard sound worked just fine.

So, if VT-d feature is listed as supported, you could give it a try. It will not work as a server grade board, but it will get things done with almost no performance impact. I had some benchmarks but I can't find them...

[RootWyrm]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

Andres;

Yeah, not too surprised on the X79-UD3. Problem being: that's an X79 and we're looking for 1150's. ;( It's probably Intel SATA with SCU enabled. That's the nominal configuration on C200/C600 too. Same deal with the i5 650 - that's a VT-d CPU plus a chipset pre-VT-d limiting. Hell, for maximum fun? 3420 chipset plus X3400 CPU.

Also don't confuse passthrough for VT-d. Too many people do that. Totally different technologies. PCI pass-through is actually far easier than PCIe or VT-d, as a matter of fact, so I don't know what's wrong there. Creative is flat out crap, no question, but there are literally dozens of people using PCI pass-through on X8SIL-F's. (Even though the bus routing is screwed up so the Matrox gets pulled in too.) I could go on at great length about the right and wrong way to do PCI and PCI-X but all I need to say is AMD8132, PLX or GTFO. ;)

But as I said: the key problem I have seen on 1150 is that A) you don't get 2x16x period B) VT-d is prohibited/misconfigured so communications get screwed up C) BIOS tends to grab exclusive. So yeah, there's significant performance impacts, and so on. Even if you get around the BIOS impacts, you can't get around the physical limitations of the CPU and wiring. Same as those boards that brag about 'flexible 4x SLI!!!' and then when I check the traces it's actually fixed 16x/16x/8x/4x.

[moben]

Re: Known Socket 1150 Gigabyte boards with working vt-d.

Hi,

sorry to hijack your thread a bit, but my question is actually related.

Does anyone know if the H87N-WIFI (or Z87N-WIFI) supports GPU passthrough (with Linux/KVM as the hypervisor)?
It does have Vt-d in the Manual, but I've hesitated to buy it since I haven't found a confirmation yet that this actually works.

And if it doesn't has anyone used a different LGA1150 m-ITX board successfully with GPU passthrough?
(I read that most Asrock have it working, but then google doesn't yield useful results for those either :( )