Hello! Good day!
I have Nvidia Tesla k40 and SuperMicro x9drg-qf. My server is having 2 CPU and 64GB RAM, PCIe 3.0.
I will try power up my Virtual Machines in passthroug mode. But I receive error:
PCIPassthru: 02:00.0 : Device BAR 0 requested 64-bit memory address that exceeded MPN type
PCIPassthru: 02:00.0 : Device BAR 4 requested 64-bit memory address that exceeded MPN type
BIOS SuperMicro is having option: Memory Mapped I/O above 4GB == Enabled. I can`t change this option to Disable, Tesla will not work (no POST message in BIOS, Insufficient PCI Resources).
How I can use vDGA mode in my VM in this configurations.
I thank you for answers.
Hello, Iforbes! I am using qemu-kvm. Kvm is working with k40m! I have two cards(k40c and k40m). Please, look at this specification: http://international.download.nvidia.com/tesla/pdf/tesla-k40-passive-board-spec.pdf (6 page) http://www.nvidia.ru/content/PDF/kepler/Tesla-K40-Active-Board-Spec-BD-06949-001_v03.pdf (7 page) This information was listed in nvidia specification and I can see in output by command "lspci": Tesla k40c - BAR1 256M Tesla k40m - BAR1 16gb. BAR1 is very big for Xen and VMware... It can not maping the memory... May be Xen fixed this problem... I don't know. I wish you an easy fix of your problem! You can write me on my email.
Check that you have the latest available bios from SuperMicro and the latest patches for vSphere.
If there is still a problem start with posting the vmx-file from the vm and the vmware.log.
// Linjo
UP! 🙂
UP!
Surprised nobody has answered this for you yet:
VMware KB: Systems with MMIO regions larger than 4GB are not supported
Hi mgordeev. How did you get this to work? I'm in the same boat with Supermicro and Tesla K40 card.
Thanks
Hello, Iforbes! I am using qemu-kvm. Kvm is working with k40m! I have two cards(k40c and k40m). Please, look at this specification: http://international.download.nvidia.com/tesla/pdf/tesla-k40-passive-board-spec.pdf (6 page) http://www.nvidia.ru/content/PDF/kepler/Tesla-K40-Active-Board-Spec-BD-06949-001_v03.pdf (7 page) This information was listed in nvidia specification and I can see in output by command "lspci": Tesla k40c - BAR1 256M Tesla k40m - BAR1 16gb. BAR1 is very big for Xen and VMware... It can not maping the memory... May be Xen fixed this problem... I don't know. I wish you an easy fix of your problem! You can write me on my email.
Hi mgordeev. Yah it seems that vSphere cannot handle MMIO above 4GB (VMware kb 2050443). I'm working with SuperMicro to see if there is a workaround.