not sure which kernel parameter or libvirt option that would be either unfortunately.
If you or someone else can point to some clear directions I am still happy to help.
On 06/04/17 20:06, Joshua Lee wrote:
If you get that running, run FIrestrike both "before and after" using posted interrupts for me... and show what scores you get, maybe some other benchmark like one of the ones from Unigine if you have the time.
On Thu, Apr 6, 2017 at 1:00 PM, Joshua Lee <joshua613 gmail com> wrote:
I'm not sure, I know there's a kernel command line argument that activates support for it, then there might be something also within your libvirt XML and QEMU command-line?
On Thu, Apr 6, 2017 at 8:00 AM, P. Pronk <vfio pronk nl> wrote:
With regards to your first question, I am using an E5-1620v4 with X99 chipset but I am not sure how I can check for you if the 16xx supports this. Let me know if you would like me to check something specifically.
Kind regards, Pim
On 06/04/17 13:18, Joshua Lee wrote:
I'm not clear if posted interrupts, a feature of Broadwell-EP, is available on both E5 1xxx V4 and E5 2xxx V4, or just E5 2xxx V4? Intel's marketing literature seems to stress the latter family of processors as having it, but reviews on sources such as Anandtech do not mention anything but Broadwell-EP E5 Xeons having it in general.
Also, a second question, what amount of performance increase in GPU performance could one expect for a KVM/QEMU virtual machine with PCIE VGA passthrough using the posted interrupt feature of Broadwell-EP Xeons, using say a 1620 V4, 1630 V4, or 1650 V4, assuming those have the feature (if not, insert a frequency-optimized E5 2xxx instead) versus say a i7-5820k?
_________________ vfio-users mailing list vfio-users redhat com https://www.redhat.com/mailman /listinfo/vfio-users _________________ vfio-users mailing list vfio-users redhat com https://www.redhat.com/mailman /listinfo/vfio-users