1812

Also, maybe look into "iommu_inclusive_mapping=1" I've seen a few instances online where it has been used to help with rmrr issues but really know very little about it.

This is their fix for the gen 8/9 servers: https://support.hpe.com/hpsc/doc/public/display?docId=emr_na-c04781229&sp4ts.oid=5249566 Not sure how applicable it is for micro servers. Here is also a redhat white paper on it : https://access.redhat.com/sites/default/files/attachments/rmrr-wp1.pdf Additionally, here is a proxmox post (in German) about passthrough problems on a gen 8 microserver. https://forum.proxmox.com/threads/hp-proliant-microserver-gen8-raidcontroller-hp-p410-passthrough-probleme.30547/ They also describe changing the boot to include "iommu=on" or "intel_iommu=on" but I have not tried those. (there are also other options intended for AMD processors which include "iommu=pt" and "iommu=1" . I also remember reading somewhere that in some newer bios for newer servers that HP had an option (don't remember what it was called) that enables the exclusion of cards for passthrough (but this was in passing about 6 months ago)

It put the bridge back in with the GPU...... fun. is there a different slot you can try? my guess is no since it's a micro server.... to be clear, this is an HP issue, not unRaid issue. HP has been a pain about actually fixing this. And their "fix" is a workaround that is not a simple as "click this to make it work."

WELL, at least we got them divided into groups now. This is a new error one for me. Try unstubbing the bridge id and reboot. If same error occurs, try unstubbing the card id and reboot (confirming iommu groupings along the way each time.) If that doesn't work, then it may be time for research.

If you read the section above regarding RMRR problems, you'll see it relates to your audio portion of the gpu. Different problem than previous posters. Try removing 8086:0151 from the override line and see if that allows the video and audio to be split into different IOUMMU groups, and if so, don't pass through the audio to the vm.

The bridge is causing the problem on passthrough. Try this to see if you can pry it away from the ethernet ports: append pcie_acs_override=id:8086:10bc vfio_iommu_type1.allow_unsafe_interrupts=1 vfio-pci.ids=,8086:10bc initrd=/bzroot or append pcie_acs_override=id:111d:8018,8086:10bc vfio_iommu_type1.allow_unsafe_interrupts=1 vfio-pci.ids=111d:8018,8086:10bc initrd=/bzroot if that doesn't work and the bridge is still with the 2 ports in the groupings, then you can try manually adding the bridge for the group. The xml would look like this (per your posted grouping for 26 ): <hostdev mode='subsystem' type='pci' managed='yes'> <driver name='vfio'/> <source> <address domain='0x0000' bus='0x09' slot='0x04' function='0x0'/> </source> </hostdev> <hostdev mode='subsystem' type='pci' managed='yes'> <driver name='vfio'/> <source> <address domain='0x0000' bus='0x0b' slot='0x00' function='0x0'/> </source> </hostdev> <hostdev mode='subsystem' type='pci' managed='yes'> <driver name='vfio'/> <source> <address domain='0x0000' bus='0x0b' slot='0x00' function='0x1'/> </source> </hostdev> I removed the vm address lines as they will be populated on vm start. I'm not sure if that will work, but it could. On my machines, I use a PCIe card out of a semi-irrational belief that it's better to keep the firewall traffic slightly more isolated off my server that way, than using the internal For reference, I use append pcie_acs_override=downstream vfio_iommu_type1.allow_unsafe_interrupts=1 vfio-pci.ids=1b73:1100,8086:10e8 initrd=/bzroot and my card splits to: IOMMU group 14: [111d:8018] 02:02.0 PCI bridge: Integrated Device Technology, Inc. [IDT] PES12N3A PCI Express Switch (rev 0e) IOMMU group 15: [111d:8018] 02:04.0 PCI bridge: Integrated Device Technology, Inc. [IDT] PES12N3A PCI Express Switch (rev 0e) IOMMU group 16: [8086:10e8] 03:00.0 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01) IOMMU group 17: [8086:10e8] 03:00.1 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01) IOMMU group 18: [8086:10e8] 04:00.0 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01) IOMMU group 19: [8086:10e8] 04:00.1 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01)

I have tested this fairly extensively at the link below. Long story short: virtualized windows doesn't care about hyper threaded processor topology and will run each cpu at 100%. OSX doesn't know better and throttles what it perceives has hyper threaded cores to (presumably) not choke the processor. Want the best osx performance? Don't specify a topology.

unRaid is fine on x1 1070 is supported once you install the Nvidia web drivers. I own a 1050. it won't do 4k. I own a 1060. it will do limited 4k. 1070 is a bit more powerful than both and will handle 4k. GPU prices are high because of miners. this is life. If you give up on 4k, then you can get by on spending less. I use the 1050 for photoshop editing and it handles it just fine.

any GPU will be bottlenecked in a x1 slot. I run several osx vm's on gt 710's. they do fine for basic use/tasks. if you want great 4k then go gtx 1070/80. A gtx 1060 can do some lower bitrate 4k as well (I use one for 4k video editing using proxy sources but it can't handle 150mbps 4k raw files)

No. But my first installer image was somehow not good and wouldn’t install a working os, so I started over from scratch and everything went fine from there.

I have 2 osx vm's running on 6.4 stable, so it works.

just a follow up, i could't boot my gtx 1060 with 2 monitors attached. IIRC (from this morning) it showed the apple logo, progress bar, then everything went black (using hdmi and dvi) and if i also remember correctly, about 6 months ago on 10.12.x I had issues using displayport on a gtx 1060 (would only show black screen) but if you let it boot up on it own then plug it in, everything was fine. I read someone on a forum that it was an issue either with the driver or clover.

Question for @gridrunner, do yo know why GPU's would run slower? I have both of mine (gtx 1050 and a gt 730) in x16 slots, but cinebench shows them at 36 and 33 fps respectively.

Hats off to @gridrunner! I finally got around to doing 10.13.2 and got it working (after screwing up the first installer image....) GPU passthrough worked ootb with the gt 730 (expected) on native drivers and updated to web drivers without issue.

I am not opposed to this.

+1

in for no other reason than it could be fun to play a vGPU..

I didn't read this all, but it looks fairly comprehensive and I'm 100% sure that I will need it in the future. Thanks!

Grab the window and adjust the size. Vnc should then auto readjust the size based on window size.

I get this on my dl580 g7... actually booted halfway and then appeared o hang.. so i forced it off. (unrelated to my previous post/issue)

any idea why when doing passthrough, it pegs all the cores for a few minutes before booting normally? This behavior is not exhibited using vnc viewer, only with gpu/usb card pass through. I've used 2 different GPU's and same result.this is on 6.4.0_rc15e, using q35-2.10, on a working sierra img (not hight sierra). --seems related to memory allocation. It takes longer the more I assign the vm.

this is odd. post Tools>Diagnostics the only time i've had issues with web GUI not being available is when I was running pfsense as my router with no backup and had multiple dockers trying to check for updates at boot time. They can't reach the Internet because they check before vm's start in 6.2.4, and then have to each time out. like 300 seconds each I think? What is your normal idle power usage? for my main server running 2 E7- 4870's (40 threads and 2 psu's) I idle at 300ish depending on how many desktop i'm running. But my other running 4 E7- 4870's (used for massive video editing projects, 80 threads, 3-4 psu's) idles at 550 watts.... so just curious. not me!

one of my unraid servers has 80 cores... the other , my main server has 40. They are both sufficient in their duties.

read this and test as outlined: Following the test procedure above confirms that its working on mine (currently dl580 g7, previous dl380 g6). In the bios, you need to put power to os control, or at least I had to. My tips and tweaks show the following: Unraid won't show you turbo frequencies because of some reason I don't remember. And vm's wont report it correctly either, nor show you boosted frequencies even when it is boosting.

Post -- System>Tools> Diagnostics Are you trying to pass through half the card or the whole card? Did you isolate it from unRaid ( as in, unRaid isn't trying to use it for server communication)? do you have an older bios you can try?