belliash Posted July 4, 2019 Share Posted July 4, 2019 Hello, I plan to buy z Ryzen7 2700 CPU and some B450-based mATX mainboard. I woud like to passthrough GPU to Windows 10 running in VM, thus I will need to buy 2 graphics cards. Actually I am thinking of the following configuration: Motherboard: Gigabyte B450 Aorus M CPU: Ryzen7 2700 GPU1: GeForce GT 710 GPU2: GeForce 1660Ti I read about problems with IOMMU groups, but I found information in manual that this mainboard allows to choose which GPU to use as primary card. I am not sure if this is going to work, but according to motherboard's manual, it should be possible to use GT710 in bottom PCI-E slot in Linux and to pass GTX1660Ti from top PCI-E slot to VM. This seems to be best scenario, because: * first of all, top PCI-E slot seems to come from CPU PCI-E lanes and seems to be in separate IOMMU group. It is also full 16x slot. * the bottom PCI-E slot is slower and share IOMMU group I think. At least users reports that in other B450 based motherboards. There would be no need to use ACS if this card could be used as primary card. Can any owner of B450 based mainboard post IOMMU groups here including mainboard model and BIOS version? Maybe someone own the B450 Aorus mainboard from Gigabyte? From what I checked, all models (not only M) should be capable of useing bottom GPU as primary card. Can anyone confirm this is really working? Thanks in advance! Regards! Quote Link to comment
belliash Posted July 6, 2019 Author Share Posted July 6, 2019 I exactly need the output from the following command: for d in /sys/kernel/iommu_groups/*/devices/*; do n=${d#*/iommu_groups/*}; n=${n%%/*}; printf 'IOMMU Group %s ' "$n"; lspci -nns "${d##*/}"; done; Quote Link to comment
Squid Posted July 7, 2019 Share Posted July 7, 2019 (edited) On 7/6/2019 at 7:02 AM, belliash said: I exactly need the output from the following command: for d in /sys/kernel/iommu_groups/*/devices/*; do n=${d#*/iommu_groups/*}; n=${n%%/*}; printf 'IOMMU Group %s ' "$n"; lspci -nns "${d##*/}"; done; FWIW, instead of entering in the command that might scare people off, you can also get the info via Tools - System Devices. And also via the various ACS override options, you can also potentially separate devices that are natively in a single IOMMU group to their own. Edited July 7, 2019 by Squid 2 Quote Link to comment
Caduceus Posted July 8, 2019 Share Posted July 8, 2019 (edited) I recently built an unRaid server with almost this exact setup. Same motherboard, same cpu, I have the 1660ti (Gigabyte 1660ti mini-itx) setup in the primary pcix16 slot) but I have an old Radon 5770 in the second slot. The only caveat is, I need to run the F5 version of the bios. I had upgraded to F32 and F40. Both gave me serious issues that would not allow Windows VMs to start with those bios revisions. I did a full post in the VM Troubleshooting section. I hope this helps. You can see my IOMMU groups here:Post with IOMMU groups Edited July 8, 2019 by Caduceus added link to IOMMU groups 2 Quote Link to comment
belliash Posted July 8, 2019 Author Share Posted July 8, 2019 3 hours ago, Caduceus said: I recently built an unRaid server with almost this exact setup. Same motherboard, same cpu, I have the 1660ti (Gigabyte 1660ti mini-itx) setup in the primary pcix16 slot) but I have an old Radon 5770 in the second slot. The only caveat is, I need to run the F5 version of the bios. I had upgraded to F32 and F40. Both gave me serious issues that would not allow Windows VMs to start with those bios revisions. I did a full post in the VM Troubleshooting section. I hope this helps. You can see my IOMMU groups here:Post with IOMMU groups And were you able to use Radeon 5770 as primary display for host, and to pass 1660Ti to VM? Your problem with BIOS doesn't look good, especially I am considering a different CPU to buy - Ryzen7 3700X and it needs F40. Quote Link to comment
Caduceus Posted July 8, 2019 Share Posted July 8, 2019 4 hours ago, belliash said: And were you able to use Radeon 5770 as primary display for host, and to pass 1660Ti to VM? Your problem with BIOS doesn't look good, especially I am considering a different CPU to buy - Ryzen7 3700X and it needs F40. I was able to use the Radeon 5770 as the primary display by configuring the bios to use card two as primary. I was able to pass 1660Ti to VM, but only with the F5 bios. I got the 2700 because I was able to get the motherboard and chip for about $220 total. I planned on upgrading to 3700x in a year or two when the prices come down. Yes, if you get that chip, unless Gigabyte fixes their bios, you won't be able to pass through the 1660Ti. Also, I was able to pass through both videocards, 1 to each vm by running unRaid headless. It's a little more complicated, but you can use a vbios rom with a vm to let it take control of the video card in use by the host. Quote Link to comment
belliash Posted July 8, 2019 Author Share Posted July 8, 2019 (edited) Yeah, that's how I passed optimus-based nvidia card in my laptop. I wonder if this problem is really related to BIOS, not some combination of BIOS, BIOS settings and Linux kernel. I found several threads over the net, some contained patches for kernel, as well as mostly UEFI was used. I wonder if problem would still exist when booted in legacy mode. I finally came across this patch: https://clbin.com/VCiYJ that most probably fixes problems introduced by AGESA 0.0.7.2 (BIOS version F30). According to the information found over the net it works at least with Linux Kernel 5.1.3. Maybe you could try it? There is also a chance, AGESA 1.0.0.3AB will bring some bugfixes. Edited July 8, 2019 by belliash Quote Link to comment
Caduceus Posted July 9, 2019 Share Posted July 9, 2019 11 hours ago, belliash said: Yeah, that's how I passed optimus-based nvidia card in my laptop. I wonder if this problem is really related to BIOS, not some combination of BIOS, BIOS settings and Linux kernel. I found several threads over the net, some contained patches for kernel, as well as mostly UEFI was used. I wonder if problem would still exist when booted in legacy mode. I finally came across this patch: https://clbin.com/VCiYJ that most probably fixes problems introduced by AGESA 0.0.7.2 (BIOS version F30). According to the information found over the net it works at least with Linux Kernel 5.1.3. Maybe you could try it? There is also a chance, AGESA 1.0.0.3AB will bring some bugfixes. unraid is currently on kernel 4.19.56 . I've read about the patches but haven't attempted anything. I was going to wait it out for the next AGESA. I did try both UEFI and legacy when I had F32 and F40. It didn't matter. Plus I read there could be problems with Windows using Nvidia cards if you booted unraid with the UEFI mode. I have no idea. I'm really happy with the way everything is working for the moment. I picked up some g.skill dual channel 3200mhz 32gb kit for $109 and a bunch of IronWolf 4TB hdds for $95. Quote Link to comment
belliash Posted July 9, 2019 Author Share Posted July 9, 2019 (edited) Yep, read that too, that's why asking about legacy mode. I think I would get rid of UEFI and secure boot. Shame unraid does not provide some experimental kernel with above patch applied. BTW: What is the performance of your GPU in VM? And one more question: Could you post your VM XML config file or command line? Edited July 9, 2019 by belliash Quote Link to comment
chris_netsmart Posted July 12, 2019 Share Posted July 12, 2019 (edited) I am looking at the same board but with R7 2700x CPU with a M.2 SSD for Cache, but I just read something that has stopped in my tracks. did I read this right that if I use a M.2 SSD, then I will loss 4 SATA ports ? so going from a 6 SATA to 2 SATA. and for myself I am looking at a older graphics card the first is MSI G4MX440 or 1050 TI to allow me to passthrough to a VM guest Edited July 12, 2019 by chris_netsmart Quote Link to comment
belliash Posted July 13, 2019 Author Share Posted July 13, 2019 Thats right. Its better to buy M.2 PCI-E. You will lose just 2 SATA ports. When you choose another mainboard, you will lose PCI-E port. There are no as many PCI-E lanes, so when you decide to use M.2 you lose either SATA or PCI-E port. Quote Link to comment
chris_netsmart Posted July 13, 2019 Share Posted July 13, 2019 @belliash thanks. I will look in to the pci-e board. And i will also look into PCI SATA extension board to see how i can get the MAX sata ports when i plug in my M.2 SSD card Quote Link to comment
belliash Posted July 14, 2019 Author Share Posted July 14, 2019 Dont use M.2 .... buy PCI-E NVMe... Quote Link to comment
belliash Posted July 16, 2019 Author Share Posted July 16, 2019 @Caduceus: Have you tried F41a? Quote Link to comment
belliash Posted July 30, 2019 Author Share Posted July 30, 2019 I have finally bought a New computer and almost everything works as expected. I will write Some short info about it as well as how to launch virtualization well on Zen2. Quote Link to comment
belliash Posted August 1, 2019 Author Share Posted August 1, 2019 https://www.reddit.com/r/Amd/comments/ckr5f4/amd_ryzen_3000_series_linux_support_and/ Quote Link to comment
helin0x Posted August 1, 2019 Share Posted August 1, 2019 How’be you applied those fixes to unraid? Quote Link to comment
belliash Posted August 2, 2019 Author Share Posted August 2, 2019 7 hours ago, helin0x said: How’be you applied those fixes to unraid? I didn't. Anyway the PCI quirks patch should be no longer necessary since AGESA 1.0.0.3ABB, as AMD claims they fixed this problem. Yet, this is something I still have not checked. The 2 other patches gives you only hardware monitoring and you can safely omit them. Quote Link to comment
belliash Posted August 2, 2019 Author Share Posted August 2, 2019 10 hours ago, belliash said: I didn't. Anyway the PCI quirks patch should be no longer necessary since AGESA 1.0.0.3ABB, as AMD claims they fixed this problem. Yet, this is something I still have not checked. The 2 other patches gives you only hardware monitoring and you can safely omit them. I confirm, AMD fixes PCI quirks problem with AGESA 1.0.0.3ABB! No patching required any longer! Quote Link to comment
helin0x Posted August 2, 2019 Share Posted August 2, 2019 You are running amd gpu or nvidia? Quirk being the gpu Passthrough unknown device header issue? Quote Link to comment
belliash Posted August 6, 2019 Author Share Posted August 6, 2019 On 8/3/2019 at 12:31 AM, helin0x said: You are running amd gpu or nvidia? Quirk being the gpu Passthrough unknown device header issue? You got specification on reddit. Quote Link to comment
sturmstar Posted August 6, 2019 Share Posted August 6, 2019 On 8/2/2019 at 6:20 PM, belliash said: I confirm, AMD fixes PCI quirks problem with AGESA 1.0.0.3ABB! No patching required any longer! I can't confirm - at least with my setup - it's a Gigabyte X470 Board and I have two Nvidia GPUs, neither is working with 1.0.0.3ABB - I had to go back to F6 Quote Link to comment
belliash Posted August 6, 2019 Author Share Posted August 6, 2019 2 hours ago, sturmstar said: I can't confirm - at least with my setup - it's a Gigabyte X470 Board and I have two Nvidia GPUs, neither is working with 1.0.0.3ABB - I had to go back to F6 Do you have the "Unknown PCI header type '127'" error? Quote Link to comment
sturmstar Posted August 6, 2019 Share Posted August 6, 2019 52 minutes ago, belliash said: Do you have the "Unknown PCI header type '127'" error? Yes, after the second attempt to start the VMs - first attempt is stuck in D3 Quote Link to comment
belliash Posted August 6, 2019 Author Share Posted August 6, 2019 This is strange. I also had this problem,but bios f42a with agesa 1.0.0.3ab resolved it for me. Quote Link to comment
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.