Processors

[skyzos]

I just bought a 1950X for a build I will do in the next month or so, and my plan is to do lots of virtualization and passthrough. Rreferably with KVM. I hope this (and the temperature sensors on Linux) get working soon. It would be a real shame to manufacture CPU with specifications that are so perfect for this purpose, and then fail to make the necessary collaboration with the KVM team happen such that this gets fixed once and for all. If I had known about this apparently long-standing issue, It might not have bought a Threadripper. I could do Xen or ESXI, or maybe even Bhyve, but KVM + ZFS in the form of Proxmox is my first choice.

Please don't make me regret my first AMD build.

[starlightknight]

That was my plan as well, but I can't even get far enough for this bug to be an issue with the 1950x. PCI bus is a mess regardless of motherboard, pass-through is totally broke and some common PCIe cards cause the system not to even post.

See:

• https://www.reddit.com/r/Amd/comments/6vbe6w/threadripper_broken_on_linux_for_pci_passthrough/
• PCI Passthrough does not work (at all) on X399 Threadripper
• https://rog.asus.com/forum/showthread.php?93634-ROG-Zenith-Extreme-(X399-socket-TR4)-info-experience...​

[filthyscum]

I have an idea: if AMD can't allocate any personell to resolve this issue, then just figure out how many lost TR sales they're looking at due to this and/or how many engineering hours and put up a bug bounty equal to that.

And Bob's your uncle!

[starlightknight]

I think that would be sending the wrong message to them - basically that its ok to leave things like this broken. It would be far more effective for every person interested to write in personally to support and let them know that they either are not buying the product or are buying intel instead because of this.

[skyzos]

I don't care if they do it in-house or if they throw somebody money to fix it.. but if the PCI is a mess, that should obviously be the priority. As an early adopter, I expect it to take a while for the bugs to get ironed out. But if six months from now PCI and/or IOMMU isn't working well, I'll be pretty irritated. Here's to hoping that it can all be fixed without changes to hardware.

[starlightknight]

I think we interpreted this differently. I'm not against *AMD* funding a bug bounty for it and/or hiring someone external. I'm against *us* (the customers) putting up a bug bounty - especially for current generation hardware, where the competition works.

[filthyscum]

If it was unclear, that was what I meant. We've already paid for it to work  

[skyzos]

Indeed.

[efeu]

It seems community started to debug, if interested in more details see here:

https://www.reddit.com/r/VFIO/comments/6ubboq/possible_fix_for_kvmamd_npt_bug/

[sjengstah]

I'll guess i wait for another month or 2 if still people complain due to the lack IOMMU support I'll unfortunately will be buying intel then.

Rather buy AMD but only if IOMMU will be working properly, don't really want to spend 1.1k on a proccessor that is not capable of this.