Intel Announces First Release Of KVMGT

[kobblestown]

Does this work without VT-d?

My understanding is that this is a KVM port of the, currently under development, XenGT solution for mediated pass-through. AFAIKT, it doesn't require VT-d. They shadow some privileged resources (like Graphics Translation Tables) but give direct access for the performance-critical resources, i.e. once the memory is mapped correctly by the GPU it can be accessed directly by it.

It's a brilliant idea and I cannot wait for it to get into the upstream projects and trickle down to the simple users like me who don't want to compile from source.

BTW, concerning VT-d, a lot of current (most Core i5's and i7's) processors support it. There is, of course, the problem with the MB support. For LGA 1150 I think Q87 is the only desktop chipset that officially supports VT-d. However, for the other chipsets you can get mainboards that do support it. I think what happens there is that if the chipset doesn't support it but the MB does (whether it works is, of course, a different thing), then it's only in effect for the PCI-e slot that is directly connected to the CPU. It won't work for the integrated devices or the other PCI-e slots - that would require support from the chipset. But you're most likely to want it for a graphics card anyway.

[miskol]

vmware workstation 11 and AVX/AVX2

vmware release workstation 11 on 1.12
And I did some benchmarks with x265 encoder in guest ubuntu
I wasn't really happy with speed if I run x265 on host machine or guest machine
x265 encoding on guest machine is 1/5 slower then if I run it on host machine directly

Anybody can benchmark compare KVM host/guest speed with x265 ?

[dragonn]

My understanding is that this is a KVM port of the, currently under development, XenGT solution for mediated pass-through. AFAIKT, it doesn't require VT-d. They shadow some privileged resources (like Graphics Translation Tables) but give direct access for the performance-critical resources, i.e. once the memory is mapped correctly by the GPU it can be accessed directly by it.

It's a brilliant idea and I cannot wait for it to get into the upstream projects and trickle down to the simple users like me who don't want to compile from source.

BTW, concerning VT-d, a lot of current (most Core i5's and i7's) processors support it. There is, of course, the problem with the MB support. For LGA 1150 I think Q87 is the only desktop chipset that officially supports VT-d. However, for the other chipsets you can get mainboards that do support it. I think what happens there is that if the chipset doesn't support it but the MB does (whether it works is, of course, a different thing), then it's only in effect for the PCI-e slot that is directly connected to the CPU. It won't work for the integrated devices or the other PCI-e slots - that would require support from the chipset. But you're most likely to want it for a graphics card anyway.

But the k-series cpu doesn't have VT-d :/. My idea is using the my Nvidia gpu for host system and the Intel HD for VM's, the question is, can by the video passed from integrade graphic to my nvidia graphic so I don't need to connect monitor to integrate graphic.

[kobblestown]

But the k-series cpu doesn't have VT-d.

But the i5-4690K and i7-4790K do! :P

Well, they are rather the exception (apart from the LGA 2011 ones) so you are right. Intel have segmented their CPU lines like a minefield.

BTW, I specifically bought an i7-4770 (non-K) because of VT-d and TSX. It turns out TSX is broken in Haswell and early Broadwell, and I don't really have any use for VT-d (and I paired it with a Q87 board so I get the full treatment). I kinda like the idea behind VT-d though.

BTW, Haswell is also missing the SMAP feature which was supposed to be there. It's a pity that AMD cannot put any pressure on Intel any longer...

[dragonn]

But the i5-4690K and i7-4790K do! :P

Well, they are rather the exception (apart from the LGA 2011 ones) so you are right. Intel have segmented their CPU lines like a minefield.

BTW, I specifically bought an i7-4770 (non-K) because of VT-d and TSX. It turns out TSX is broken in Haswell and early Broadwell, and I don't really have any use for VT-d (and I paired it with a Q87 board so I get the full treatment). I kinda like the idea behind VT-d though.

BTW, Haswell is also missing the SMAP feature which was supposed to be there. It's a pity that AMD cannot put any pressure on Intel any longer...

Oooo I have think this effects all k-series, I have 4790k wow . I need to try VT-d out .