lionceau

Ubuntu 18.04 LTS with all powertop settings on "good" idles at 16W on my Dell T20, Unraid with the exact same hardware idles at 31W for no good reason. CPUs were idle and disks were spun down in both tests. Even with this quite power efficient hardware and 0,28 cent per kWh that's ~37 euros per year or the price of a Tier 2 "Plus" license in two years. I'd rather pay Lime Technology than my power company, it would help save the environment too.

It looks like somebody has recently reverse engineered throttlestop and figured out how to undervolt Haswell and newer CPUs in Linux. Is there any chance this can get ported to unraid, perhaps even as a plugin with a GUI? https://github.com/mihic/linux-intel-undervolt

That leaves Facebook, SuSe and Fujitsu as the major BTRFS contributors. This acquisition seems to be related: https://www.redhat.com/en/about/press-releases/red-hat-acquires-permabit-assets-eases-barriers-cloud-portability-data-deduplication-technology They intend to open source Permabit's technology. Perhaps XFS will get deduplication and compression in the near future? So far I do not regret my decision of using XFS for all my array and cache drives.

I had problems with something "thrashing" my disk with reads causing one disk to always stay spun up. After some troubleshooting I found that it was cache_dirs plugin. It seems that one of my backup folders contains a large amount of directories, apparently too many to fit into RAM, so cache_dirs read the structure from disk every 10 seconds because the cached structure got evicted in the mean time. This resulted in endless directories reads from disk to memory. Not great. Excluding that folder worked for the most part but some reads were still not cached; now I doubled my RAM, included the folder again and had zero read/write from any disk in the past few hours. Is there any way to figure out how much memory the cached directory structure currently uses?

It looks like somebody found a way to patch the Nvidia driver to remove the virtualization check that causes Error 43. Perhaps this is helpful for some edge cases where people can't dump their BIOS for whatever reason. https://github.com/sk1080/nvidia-kvm-patcher I have not tried this, attempt at your own risk.

I have updated the original posting. If people affected by the AMD/KVM/NPT bug could take a look at https://bugzilla.kernel.org/show_bug.cgi?id=196409 and perhaps comment on it then that'd be great because it would help raise awareness.

Agreed but that's in the KVM developer's hands, there's really nothing the unRAID team can do about it.

I am in the same position, looking to replace my quadcore Haswell which idles at only 35W, about 40% of what an overclocked Ryzen system seems to consume. Since Pauven says it's an unRAID specific problem I may just keep my unRAID Haswell system as a strict NAS and build a second Ryzen-based system with a different Linux distribution as my workstation which I'll shut down when it isn't in use.

http://www.insanelymac.com/forum/topic/309087-insanely-fast-virtual-mac-qemu-ovmf-clover-and-native-graphics/page-25 Looks like this requries QEMU 2.9.0 to work. The parameters are simple enough to add to the unraid XML, but I haven't updated to 6.4-rc yet. Once 6.4 is stable and those parameters are confirmed working we should be able to use macOS on unraid with unpatched clover versions.

Are Global C-States stable with the new 6.4-rc? It brings a new Kernel that should fix some Ryzen issues.

somebody (helgrind over at insanelymac) seems to have found a way around the clover patch, does anybody know how to convert this to unraid compatible XML? I think I've found a way to avoid the ugly Clover patch: -cpu Penryn,vendor=GenuineIntel,kvm=on,+invtsc,vmware-cpuid-freq=on vmware-cpuid-freq requires kvm to be exposed and invtsc enabled, hence the kvm=on,+invtsc Tested on qemu 2.9.0.

I'm curious has anybody managed to get Filevault working in a unRAID Sierra VM? It seems like it's working on regular Hackintoshes that support UEFI. I think it would be really nice to have the disk image encrypted since unRAID itself does not focus on security and might be prone to getting hacked.

Does anybody happen to know if this works in unRAID? https://github.com/r4m0n/ZenStates-Linux If so, I wonder if some of these options may fix the Global C-State bug at a finer granularity than completely disabling the feature in the BIOS.

I guess lime tech is well aware...

I'm curious, do these drivers work inside a Windows VM? http://wccftech.com/amds-new-ryzen-...iency/?utm_source=wccftech&utm_medium=related

I feel like fixing the Global C-State and NPT bugs, both issues that look like they could be software related, would go a long way towards making Ryzen attractive for Unraid. A 10-20W idle increase over my quadcore haswell Xeon would be acceptable, but an additional 40-50W of needless idle power consumption and mysteriously crippled multicore VM performance is not. A shame really, looks like I'll have to wait for Skylake-X/Kabylake-X and lose the ECC support or wait a year or so for the refresh of Ryzen.

Hi, I use three "desktop" VMs (Win,Mac,Linux) with unRAID and each of them has the same GPU passed through, so I can't use them at the same time. Is there a way to "queue" VM actions so I can for example stop Linux and immediately start Windows from within Linux using the webgui? Right now I have to use a second device to shut down one VM and then start the other, because once the first shut down is done I cannot interact with UnRAID to start the next VM.

I also wanted to chime in and thank the early adopters and developers for their testing. Ryzen seems great on paper but I value stability over anything, so I'll keep using my Xeon until we are sure that unRAID on Ryzen has reached parity in terms of reliability.

Yes, GPU PCIe bandwidth has surprisingly little effect on gaming performance in my experience. Many OSD programs show PCIe bus saturation and the most I see is 5-7% on loading screens where the VRAM is being filled with textures from RAM or SSD. Typical bus utilisation during gaming is around 1-2% even when with remote streaming. Of course this could be wildly different with other workloads running CUDA. The GPU PCIe bandwidth topic often comes up with people running external GPUs over Thunderbolt. What they don’t realise that it's the significant latency from protocol encapsulation/translation that robs performance, not the raw throughput.

Digital Foundry reports 39W idle in their Ryzen review. Specs mentioned: R7 1800X, MSI X370 "Gaming Titanium", 32GB RAM, Titan X Pascal, Windows 10

Yeah, I think you are on the right track regarding power consumption. Based on other reviews Ryzen should be fairly economic in idle, perhaps 5-10W higher than Kaby Lake. 5 Gbit/10 Gbit LAN controllers are known to be very power hungry. It'd be quite disappointing if it turns out that the BIOS only disables access but not power to the chipset. Gaming mainboards in particular don't seem to pay much attention to these details because what's another 10W here and there when you have 8 case fans and six LED strips. I suspect a basic B350 board without all the extra features would idle much lower but unfortunately these usually have very questionable 4+2 phase power layouts. Idle power consumption between a 1700 and 1800X should be almost identical when all cores are clocked down. TDP really only suggests power consumption at maximum load and in the case of Ryzen that's maximum load on one core at boost. Actual power consumption at 100% load on all cores is usually much higher.

Perhaps it is the GPU? Unlike Windows 10, unRAID doesn't have graphics drivers with power managmentby default. My power consumption drops by ~12W when I start my Win10 VM with Nvidia WHQL drivers and my GTX1060 passed through. Thanks for the numbers by the way. Quite a bit higher than I hoped they would be. I wonder if the full featured mainboard with a 5-Gbit NIC, Soundblaster and other fancy features is party to blame. As for EDAC I don't think Ryzen supports it. I remember reading that ECC and single error bit correction is silently functional but not validated and unsupported. Two-bit error correction with reporting and kernel halt is not implemented and won't work with standard desktop Ryzen. I guess that's where they draw the line for Naples or some other dedicated workstation SKU (Ryzen Pro?)

Yes, thanks for the data point. I'm very curious what a "bare" system would idle at. 10W for the pump, 7W per idle graphics card, 1W per fan would bring it down to 40W OC before any minor "tweaks" like deactivating the soundcard or any unused controllers, etc. That's not as bad as I thought it would be.

Has anybody measured idle power consumption at the wall with Ryzen? I read that typical Ryzen systems idle at 60W (without harddisks or with HDDs spun down). My current Haswell only sips 30W with a 1060-6GB. Perhaps this has to do with the fact that most benchmarks were done using Windows and AMD strongly recommends "High Performance" power saving setting? In any case it looks promising, my ideal setup would be an R5 1600X with 3200 Mhz ECC RAM but neither product exists at the moment.

Have you tried this in unraid? CPU tools inside VMs generally show strange frequencies, at least the VMs on my Haswell Xeon do. grep MHz /proc/cpuinfo