Thanks to NordPass for sponsoring today's video. Take control of your passwords! Visit nordpass.com/Craft to take advantage of their best offers, risk free for 30-days!
Wow, great update, and so soon after the first part! Big kudos @DotBowder for pinpointing this to the microcode. I did not really think about that, because we generally update our BIOS on our retail boards and have the microcode installed (Not only for stability, but also for security reasons). BTW, we already have a section in our reference documentation about the microcode if you want to link to that ( Section 3.3 Firmware Updates) It is also on our wiki. Thanks for the good work!
Thanks a bunch for following up on this. This kind of actual hands-on testing and reporting of functionality - and especially your follow-up, when you got the microcode updated and stable and redid all the benchmarks, and your new findings - is incredibly valuable, harder to come by presented in such a comprehensive manner, and happens to line up with stuff I'm tinkering with right now, so, cheers!
The new intro is cool and all but I really liked the old intro. Smooth jazz playing over the pint filling up, while looking like it's recharging a battery, captures your channel perfectly. WIth the new intro I'm not sure what the matrix has to do with anything, doesn't seem like a good fit. Also it's a bit too long.
This was PERFECT! I support a high end gaming PC with a 13700K and two RX7600xt GPUs with Proxmox 8.1 and two VMs for my kids to split the system. I applied the code and ran Passmark CPU bench, and noticed how greatly improved the results were, and how it all balanced well. Much Kudos to Jeff for this one!
Two things: 1) Nice to see that Proxmox is able to do what Windows has been able to do, and then do it better than Windows. Always a plus. 2) re: 1x VM x 4 P-core threads -- you can test that explicitly by specifying the CPU affinity for that VM to 0,2,4,6 and you should get pretty similar results where it is only just using the P-core, without HTT. That's also assuming that you don't want to use taskset for the VM process that's running inside Proxmox 8.1 (which you can also do, if you don't specify the CPU affinity field in the Hardware -> CPU section of your VM.
The interesting part to cover in part 3 is how these boards perform in idle mode. I mean power consumption, falling into c6+ states, etc. Everyone measures performance, but my home server is idling most of time, and though I need to squeeze every GHz out of it sometimes , 80% of time its just idle and costs me $$$ And many thanks for such niche videos, it's insaly hard to find how virtualization works on these Chinese mobos
Slightly off topic .... I recently switched from VMware Workstation to Proxmox VE. I should have made this switch years ago! I'm still a newbie on Proxmox, but I'm looking forward to learning more about Proxmox VE!
I did the same some years ago. Was a VMware enthusiast for years, and even taught VMware Cert courses for the US Navy. But with all the features that are worth a damn behind a paywall, Proxmox is the way to go. That said, Proxmox is no panacea. To my knowledge, Proxmox offers nothing to compare to VSphere Distributed Resource Scheduler for balancing load in clusters. And while this might seem like a minor item, I really miss how ESXi graphically presents vSwitch configuration (I wish Proxmox had something to compare). Now that Broadcom has come in and pretty much destroyed VMware, I hope more Enterprises and SMB's jump ship and come to some of the more open altenatives (such as Proxmox and XCP-NG). I guess we'll see how that all plays out (probably most who consider leaving VMware would likely go to Microsoft Hyper-V or Nutanix).
FWIW, I've been running a 13900k in Proxmox 8.0 for nearly a year now with no instability issues. I'm not pushing the system to its limits by any means, but it runs a dozen or so VMs without issue. I'll update and apply the microcode patches soon though, won't say no to improvements.
@@allandresner My understanding is that's just how it's reported to the VM, but the host will boost higher. Try running cat /proc/cpuinfo | grep "MHz" on your host to see current clocks.
I was running an i5-13500 with a w680 motherboard for quite some time with zero issues. I had the p-cores pinned to a VM that I use for gaming, and it worked well, too.
This jogged my memory on virtualization and right-sizing VMs compared to bare metal, and why a mid-way loaded host (with noisy neighbor VMs) is faster with a diverse workload. It's because the scheduler should be able to extract more performance during the delay intervals of other threads when they don't overlap too much. Oversubscription can still net performance to an extent when many guests are nearly idle.
This lines up with my own testing of Proxmox vs. bare metal, and it explains why I some VM results outperformed bare metal (I never investigated the why because my better results were right around margin-of-error levels at single digit percentages, though consistently so).
I have been running the Erying i5-1250 SRLCY for almost a year now. I even started with Proxmox 7. I have multiple container VMs running 24/7 with a Tesla P4 passthrough to one of the VMs, a SATA expansion card passthrough to a second VM and a plex container with quick sync. The current runtime ist about 40 days and I never experienced any stability issues. So maby I got lucky with my board
@@jdpdata 3?!? Wow I just bought one and I am waiting to see how good it is. I may replace my 2 others Proxmox nodes with 2 more ms-01 later. We’ll see.
The biggest issue I've had trying to run VMs on my erying board (non ES, 11800H) is with PCIE passthrough. I actually have an RTX A2000 12GB that fails to passthrough. Granted, I was trying to passthrough via unraid, rather than proxmox. Perhaps its time to give proxmox a go. On the previous note, however; are there any bios settings that need to be changed to enable proper passthrough on these boards?
Thank you for the thorough and timely part 2! I was salivating over the cheap 13th gen engineering sample boards from Erying in the past few days, and trying to convince myself that I have a need for one as a low power VM host. Seeing that they have potential stability issues definitely dampens my interest.
If you want to actually run it, just avoid ES parts and pay a little extra. The ES is fun in that it can be overclocked and at a low price, but the VRAM of erying motherboard doesn't seem to be able to handle too much power
@@harrytsang1501 im rolling with a 13th gen 13900 ES from "10729", same layout, only thing different is they still have the older mobo rear connector (like steve from GN), whereas the erying 13th gen has usb c (and i think theres also a clear cmos instead of the older dvi(or vga) port. overclocking im able to hit 5.4 all core @ 1.24 or so volts, stil tuning it now but i think im inching myself closer to 1.2v all core, chip still is mostly in the 75w range but it can still go +/- 25w
0:44 here, Kyiv Craft Computing connects people (:
9 месяцев назад+1
I bought the Frankenstein 12900H (no ES) immediately after the first set of videos on Erying. With their RAM and SSD, to avoid any lengthy troubleshooting steps... And I've been using it as a daily driver light "workstation" (with regular RAM and SSDs), I don't do as complex 3D as I used to, rendering is done in the GPU, since the lack of PCIE lanes isn't so much of an issue for rendering, because (I think) it loads all the textures once and doesn't shuffle them in and out like in games...? And it's been rock solid! Even bought an PCIe E-key 2.5Gb for the wifi slot (thx Wendell for the hack) and it just works under Windows! But the previous video got me worried, since my next step would be to buy an actual workstation and turn this into a low power high speed server. Now I'm so glad it works! Thank you for the info!
Great video!! I was thinking about this kind of set up inspired by your previous videos with Erying products. Please do more this kind of tests also show us part 3 with installation of this boards in your garage server :D
Yeah the ES Boards can be a bit weird, I mean after all they are not rated for stability (or anything really, there's a reason they are ES and not retail and you pay less). You can get away with good stability with just a bit less clockspeed (if any) but I run into this issue as well, when I tested my 11980HK ES Erying board on an unraid server. It run very well and stable unless the system was really doing something because every time after about 5mins of full load the system would reboot. The new bios and the better VRAM coolers did help a bit so I wouldn't see crashes under normal use but I was able to crash it repeatedly when stress testing. For a homelab it's probably fine for me but I wouldn't use it as a NAS or anything where it could corrupt data even. I might play around a but with lowering the multiplier and trying to reduce strain on the VRAMs and see if that may help so there's that.
I dont see the use of this for virtualisation, when you are limited on the ram side. CPU is one step but then you need Ram as well. and 96GB Ram are used up pretty fast, when you are hosting game servers or windows vms. And buying several of them is more expensive then just one professional server.
Can you add testing with containers as well? I thought the situation with containers vs VMs and hybrid architecture was a bit funky too. But perhaps that has been addressed as well?
Sorry if this is a dumb question. But, basically let the kernel do the work? As long as you are running Proxmox with the latest kernel-it will allocate the best cores / threads to use? Or do you still need to install the Intel Microcode for this to work at all????
Having newest microcode is good idea either way, but depending on how you installed proxmox you might already have microcode installed - I always install clean debian first and proxmox on top of that (due to partitioning limitations with proxmox installer) and when you select expert mode in the debian installer and select non-free repo you should get the microcode package automatically I think.
Thank you for making this update video! It's wonderful to see someone making timely updates to keep the community informed when new information comes to their attention! Keep up the amazing work!!!
Good sir, thank you! I’ve had my eye on this exact motherboard for my first homelab and am so grateful to have stumbled across your channel! Would you by any chance mind sharing what ram and speed you got working on these and if your made any bios changes? Edit to add: would you feel this board would be reliable enough to be used in a NAS build and would there be any compatibility issues?
Glad you took another look into this 😊 I wish you would take another look at the 3070m GPU that got BIOS bricked (if you were unable to unbrick it, you could always send the GPU to another RUclips content creator for the unbrick video - then use the Frankendriver on the fixed GPU)
Hey Jeff, have you tried to get vgpu pass through working with the es motherboards? I’m using the 12900hx es lga 1700 cpu, I can get the vgpu to show up in Proxmox but nothing can actually use them when passed through, windows sits on error 43, and jellyfin in a Ubuntu lxc container can’t use it either, I’m wondering if a microcode update would fix this or maybe it’s just ES weirdness?
Probably kinda niche but when you get a free moment could you do a segment on setting up different windows vms on proxmox to be able to use RDP across vlans? In my case i just setup an xp vm on proxmox vlan 2 and enabled RDP, but my main pc on vlan 1 won't connect to it, im using ubiquiti gear if it matters. I have yet to set up another vm (win 10 or 11) to test it. Maybe a bonus you could include is how to make sure the VM vlan can talk to the local network(s) but not to the actual internet if it's so desired.
Are you talking about setting up a VM specifically for RDPing into other VMs? Check out my video on Guacamole. ruclips.net/video/Sq-irDBauvo/видео.html
@CraftComputing nope, trying to make it so I can rdp into the vm (on a designated vm vlan) from my main pc (on my home vlan). At the moment I've only setup the one win xp vm, but my win 7 desktop is claiming the vm isn't on the network when I tell the rdp connection program to connect. In my udmpro I have to where the home vlan can talk to any other vlan, established and related is allowed, but vm vlan is locked down like my iot one is. Edit: I just finished setting up both a win 10 and 11 vm, both work if I type in the ip for RDP, not the pc name though.... Edit 2: I've since setup "dns name" (I think it was) and using computername.networkDNSname.local and it worked on all 3 (10, 11, and xp). Still have to figure out how to block that vm from actual internet access and have it only be intranet.
@blondeguy08 no, it really isn't too hard to set up the different vlans to talk back and forth, I just did the same rules as when I made my iot vlan from watching crosstalk solutions and mactelecoms channels set up videos. Eventually I want to find out how to block all internet access to a specific vlan (vm one) since I'm only using the vms for specific programs that don't need internet access and xp can't do anything in terms of updating anymore so it'd be a major security hole to let it access the internet where nefarious people could find it, not that I'm too concerned but better safe than sorry.
I've had a lot of issues with the 12000 series CPUs, but after applied the microcode, you linked to, it started working as it should :D Usually I am sticking with AMD 5000 series CPUs....
Very interesting and useful, thank you. A grumpy observation: 'Equivalent' Intel desktop and Xeon CPUs - One would expect the more expensive Xeon would be a better-bined part, if not outright a better design. But what does Intel do? Rob the Xeon customer of E-cores for a higher price and excuse that as a VM issue! Worse, it wouldn't surprise me at all if the supposed 'better' and more expensive Xeon is a desktop equivalent otherwise destined for the trash because it's E-cores don't work!
I'm running Intel NUC with 1340p since november without any issues. Usually there is 5 VMs on at all times and I did not notice any stability issues. Even go iGPU passtrough for jellyfin HW acceleration working in a VM I went trough few of tteck proxmox helper scripts after install which includes microcode update, that was most likely what saved me from issues
Nice work, great content. Would be even better if you considered power usage statistics when performing these kinds of discovery test. For most, that seems to be one of the biggest concerns with home lab equipment.
I watched both the videos in this two part series with great interest when they came out. Since then, I've ordered a Minisforum MS-01 i9-12900H that I'm now waiting to be delivered. Thanks so much for all the very useful information. Especially regarding the processor microcode obviously. Interesting times ahead! 👍🙂
I built a 12900k system for low latency audio processing in windows and found that disabling hyperthreading improved performance considerably in that work load. Not really the same thing obviously but thought it was interesting.
Interesting stuff. Won't be able to test it myself though as I only have a Epyc 7543 and a Xeon E-2314. Neither have E cores, and the Xeon doesn't even have SMT :p
I should take a look at some of those micro code updates because I'm running KVM on my 13th gen Intel laptop so I can have a Windows machine for the couple of things I have to do that use that. Might solve some of the small performance issues I'm seeing
Thank you. It's interesting to see the testing you went through. Also, I am curious to see what stores you buy from on AliExpress as I have never bought from there and want to choose a reputable place.
I still like going full manual if I have a performance target and wants it to be predictable and load independent. Have a 13900k on proxmox with isolcpus set so that host only uses last 8 E cores, and manually pin each vCpu to the unscheduled P and E cores. The E cores are great for integer workloads such as hosting VM, virtio, and I do keep smaller VMs sharinf those 8 E cores. But when I pin the mix of P and E cores to vCpu, I get the same cinebench results every time, -10-15% if all cores are loaded and power draw hits 253w. I will won't trust that proxmox will handle it well, since even on bare metal, disabling E cores sometimes does make the machine more responsive.
could disabling e cores allow a much higher power and thermal throttling ceiling for the P Cores be a factor for a more responsive machine? it could just be something Proxmox has no control over.
@@tanmaypanadi1414 I don't think it is power and thermal related. That would only slow down the CPU so much. I just find that if I want predictable and consistent performance under load, manually setting CPU affinity and CPU pinning helps so much. Especially when your goal is to make it behave like a real machine. The E cores are great at doing integer tasks, such as handling network stack, Host OS and other qemu/KVM management tasks. But they make slowdowns more often and the performance inconsistency is immediately remedied when I taskset the vCpu onto pinned P and E cores. So that when the guest OS scheduler expects P core performance from core 0, it always gets a P core on core 0 and not whatever proxmox decides.
I was looking through the ERYING store today and come across 10 and 11th gen Xeon integrated motherboards. Would there be any benefit of W 11955M 8C/16th vs the i5 13420H 8C/12T ? both at same Price point of $400 ish AUD. * A bit of background looking to upgrade my Dell Optiplex 7070 Mini running i5-9500t and separate box running a few HDD's in RAID to a new all in one solution with a bit more grunt for some Extra VM's. Currently running Poxmox 7 with container for all my docker Arr's + Ubuntu VM with iGPU passthrough for Plex.
Hey, is there a plan for beer glasses that can hold 0.5L (about 17oz)? The 16oz pint glass can only hold 473ml and our beer bottles hold 0.5L. I know it's a strange request. But if I order from the USA to Germany, it must be worth it ;)
I wonder what Joyent Illumos would do. The thread/core thing was "solved" by Sun Microsystems and Solaris about 20 years ago in their Niagara architecture. Illumos is the current Free-ish version of Solaris and I'm sure some of that scheduling mojo was carried over. Unix and later Linux always had the ability to manually pin tasks and processes to CPUs/CPU sets. I'm wondering if. you could run a full Cinebench session pinned to the P-cores and then pinned to the E-cores (alone) and see the results. Also, you gotta try a proper Belgian Witbier or even a Geuze.
@@CraftComputing I was suggesting the inverse; Illumos is the descendant of Solaris for Intel (and Sparc) and could be considered an alternative to Proxmox (although it is a full OS and not really a hypervisor.) I have no idea how it would handle Ecores v. Pcores at this point, though.
I am surprise because I’m Running a 13900k for Two Years now on proxmox With 16 vms running. ECore allways enabeld and i had never Problems with crashes Without applaying some microcode.
Been wondering if It would be worth it to use my spare, 12900K for my main proxmox server. These videos definitely help with making that decision. Was also toying with the idea of possibly turning this 13900KS daily into a proxmox desktop that still runs windows daily on it but can also run other vm's and containers while being able to shut down windows. Have only seen one other person do this on RUclips, and probably isn't a choice many would want to make for their daily. But I need room to start building more than just the networking portion of my home lab. Buying old Xeon E5v4's and trying to get my MSI GS66 to work with proxmox ve without throwing errors, seems so far like it's not going to work.
I wonder if the NICE level of the host OS's needed threads get's bogged down when the VM's are running full bore is causing the stability crash issues? If all the VM's are using ALL system resources, I wonder if the host OS has enough overhead to not crash...
This is an extremely interesting outcome. The gain in the vm due to scheduling especially. Part 3 might male it worth using one of the high end chips on a lab machine if it comes out consistent.
For what it's worth, I've seen similar results on my 13900T server at home. It's the same silicon as the rest of the desktop 13900 chips, just with VPro and a lower power limit. Mine was an ebay salvage find from a lot of water-damaged machines.
Personally on my Alder lake system i only give a VM either Performance cores (threads included) or E cores NEVER both! aka i generally keep the E cores for the base Linux OS and give Widnows Games VM the P cores
Such a shame about the engineering samples, ive been storking them in AliExpress for months now looking to replace one of my servers running a E5-2690 for a while. And it looks like im going to have to hold off for now 😔. Lets hope in another gen or two they will fix the stability issues. They always do and i can wait!
Once again, another niche aspect of a broader, very amazing community has come forward to share their knowledge, potentially saving the day! Gotta love that. 😎😎
Querry: Big little architecture.... Didn't "Big-little" architecture used to refer to Big-endian/little-endian architecture?? That is the direction that they like to chase the bits through the processor?? Where am I slightly confused??
"big.LITTLE" is ARM's name for the power/efficiency core stuff. Big Endian and Little Endian is an entirely different thing that only sounds vaguely similar.
This is very interesting. I recently built a pc with a 7950x3d . I have been thinking of running proxmox on it, with a gaming VM but have always stopped short given the CPU design. Has anyone tried using a 7950x3D for a gaming VM. Would you need to lock the VM to the gaming cores or would proxmox actually be smart enough to figure it out?
The installation instructions in the video description do not mention downloading anything from GitHub at all, just to install the microcode update package from the non-free-firmware Debian repositories. I guess the microcode in Debian's official repositories is new enough.
I would love to see this running Hyper-V. I'm a Windows SysAdmin so my homelab is also Windows currently (at least until i can get a good deal on an Erying board lol).
How are certain types of cores allocated to machines? There is no such setting in the interface, only the number of processors and cores, but not their types.
Results presentation and description is a bit hard to follow. I never touched 12gen+ Intel CPUs in my experiments and probably missing some details on the way they function. But watched this video couple of times and still find the way "tables and graphs" are layed out is not helping but making it more confusing.
@@christopheoberrauch784 to be fair, i've found its very easy to patch the microcode using the Proxmox VE Helper-Scripts utility... that is, unless its not already updated by the mobo bios version 😉
Jeff, I'll send you some IPAs if you can figure this out: 13900K on Supermicro (latest BIOS), Proxmox 8.1.11, pegged CPUs 0-11 on a Windows 11 VM. Cannot get past 3Ghz in the VM. Even I'm locking down the performance cores and microcode is good to go.
How are you measuring/recording CPU speed? If you're checking inside the VM, most performance info isn't actually sent to the VM from the host. Check stats on the host.
@@CraftComputing Good point Jeff, in Hyper-V I have found a bug that if I move a VM from host to host, the VM does not correctly report the turbo boost, but if I create a VM on that host, it will. To your point, in both Proxmox and Hyper-V the host is reflecting the maxed out frequency and if your saying that I should rely on that and not worry about what I see in the VM, then I'm done trying to see that within the VM. Appreciate your feedback, beer on its way. Have you ever tried Knotted Root Brewery out of Colorado?
Well, this makes running a new hybrid architecture for proxmox or really any VM host that can take advantage of the microcode, (Im lookin at you TrueNas Scale, please integrate) really REALLY promising for home labs only running a few VMs or VM that sit Idle a good bit.
100%, no notes. They've never had a single issue, and have been running for almost a year now. Currently one is running a Minecraft and a Palworld server 🤘
with all due respect sir, proxmox itself not a hypervisor. Proxmox creates a nice tool sets and addons around the kvm hypervisor as a whole virtulization suit, which including management interface (webgui+api) clustering, sdn, storage etc...
So no ES in THIS family of CPUs for virtualization experiments? Sure. I think I'll build out my spare 13400F to tinker. On a different note, I suspect a QS sample of Xeon scalable would most likely be fine given that's more of a "native" workload.
Thanks to NordPass for sponsoring today's video. Take control of your passwords! Visit nordpass.com/Craft to take advantage of their best offers, risk free for 30-days!
Is it plain text if I write all my passwords down in Klingon?
@Yuriel1981 If they're all ASCII characters, then yes it is.
cool... but bitwarden for me
Wow, great update, and so soon after the first part! Big kudos @DotBowder for pinpointing this to the microcode. I did not really think about that, because we generally update our BIOS on our retail boards and have the microcode installed (Not only for stability, but also for security reasons). BTW, we already have a section in our reference documentation about the microcode if you want to link to that ( Section 3.3 Firmware Updates) It is also on our wiki. Thanks for the good work!
Thanks a bunch for following up on this. This kind of actual hands-on testing and reporting of functionality - and especially your follow-up, when you got the microcode updated and stable and redid all the benchmarks, and your new findings - is incredibly valuable, harder to come by presented in such a comprehensive manner, and happens to line up with stuff I'm tinkering with right now, so, cheers!
Jeff is great at this stuff.
The new intro is cool and all but I really liked the old intro. Smooth jazz playing over the pint filling up, while looking like it's recharging a battery, captures your channel perfectly. WIth the new intro I'm not sure what the matrix has to do with anything, doesn't seem like a good fit. Also it's a bit too long.
I agree as well. I like the old intro a bit more.
I'll be switching back and forth. The Matrix intro isn't completely replacing the old.
@@CraftComputinghate to say it. Old was better. Sorry. Sorry.
I like the new one actually 😎
Results are about 50/50. Cool, I have two good ones.
This was PERFECT! I support a high end gaming PC with a 13700K and two RX7600xt GPUs with Proxmox 8.1 and two VMs for my kids to split the system. I applied the code and ran Passmark CPU bench, and noticed how greatly improved the results were, and how it all balanced well. Much Kudos to Jeff for this one!
Two things:
1) Nice to see that Proxmox is able to do what Windows has been able to do, and then do it better than Windows. Always a plus.
2) re: 1x VM x 4 P-core threads -- you can test that explicitly by specifying the CPU affinity for that VM to 0,2,4,6 and you should get pretty similar results where it is only just using the P-core, without HTT.
That's also assuming that you don't want to use taskset for the VM process that's running inside Proxmox 8.1 (which you can also do, if you don't specify the CPU affinity field in the Hardware -> CPU section of your VM.
The interesting part to cover in part 3 is how these boards perform in idle mode. I mean power consumption, falling into c6+ states, etc. Everyone measures performance, but my home server is idling most of time, and though I need to squeeze every GHz out of it sometimes , 80% of time its just idle and costs me $$$
And many thanks for such niche videos, it's insaly hard to find how virtualization works on these Chinese mobos
That's my primary concern
Yes please!
then turn it off when you dont use it 😂😂😂
@@sunnycloudy1337 Have a solution for low latency response when you want to use it?
@@indignasmr7379 wake on lan or physical remote bluetooth/wifi switches you can use if you cant push the button yourself
Slightly off topic .... I recently switched from VMware Workstation to Proxmox VE. I should have made this switch years ago! I'm still a newbie on Proxmox, but I'm looking forward to learning more about Proxmox VE!
I did the same some years ago. Was a VMware enthusiast for years, and even taught VMware Cert courses for the US Navy. But with all the features that are worth a damn behind a paywall, Proxmox is the way to go. That said, Proxmox is no panacea. To my knowledge, Proxmox offers nothing to compare to VSphere Distributed Resource Scheduler for balancing load in clusters. And while this might seem like a minor item, I really miss how ESXi graphically presents vSwitch configuration (I wish Proxmox had something to compare).
Now that Broadcom has come in and pretty much destroyed VMware, I hope more Enterprises and SMB's jump ship and come to some of the more open altenatives (such as Proxmox and XCP-NG). I guess we'll see how that all plays out (probably most who consider leaving VMware would likely go to Microsoft Hyper-V or Nutanix).
FWIW, I've been running a 13900k in Proxmox 8.0 for nearly a year now with no instability issues. I'm not pushing the system to its limits by any means, but it runs a dozen or so VMs without issue. I'll update and apply the microcode patches soon though, won't say no to improvements.
Does your VM go past 3Ghz? I have the same processor.. can't get my VM to see past 3Ghz, using host CPU.
@@allandresner My understanding is that's just how it's reported to the VM, but the host will boost higher. Try running cat /proc/cpuinfo | grep "MHz" on your host to see current clocks.
I have been super happy with using 13900 (non-K) in proxmox cluster for a while now, and been super happy without modifying the scheduling.
I was running an i5-13500 with a w680 motherboard for quite some time with zero issues. I had the p-cores pinned to a VM that I use for gaming, and it worked well, too.
This jogged my memory on virtualization and right-sizing VMs compared to bare metal, and why a mid-way loaded host (with noisy neighbor VMs) is faster with a diverse workload. It's because the scheduler should be able to extract more performance during the delay intervals of other threads when they don't overlap too much. Oversubscription can still net performance to an extent when many guests are nearly idle.
This lines up with my own testing of Proxmox vs. bare metal, and it explains why I some VM results outperformed bare metal (I never investigated the why because my better results were right around margin-of-error levels at single digit percentages, though consistently so).
I have been running the Erying i5-1250 SRLCY for almost a year now. I even started with Proxmox 7. I have multiple container VMs running 24/7 with a Tesla P4 passthrough to one of the VMs, a SATA expansion card passthrough to a second VM and a plex container with quick sync. The current runtime ist about 40 days and I never experienced any stability issues. So maby I got lucky with my board
Just in time, waiting on my minisforum ms-01s to arrive. Thank you for taking the time on this
Me too! Got three MS-01 coming
@@jdpdata 3?!? Wow I just bought one and I am waiting to see how good it is. I may replace my 2 others Proxmox nodes with 2 more ms-01 later. We’ll see.
likewise, this clears my remaining Qs on the device.
The biggest issue I've had trying to run VMs on my erying board (non ES, 11800H) is with PCIE passthrough. I actually have an RTX A2000 12GB that fails to passthrough. Granted, I was trying to passthrough via unraid, rather than proxmox. Perhaps its time to give proxmox a go. On the previous note, however; are there any bios settings that need to be changed to enable proper passthrough on these boards?
Thank you for the thorough and timely part 2!
I was salivating over the cheap 13th gen engineering sample boards from Erying in the past few days, and trying to convince myself that I have a need for one as a low power VM host. Seeing that they have potential stability issues definitely dampens my interest.
I have more boards arriving next week, and will be testing out more 12th and 13th Gen ES parts.
If you want to actually run it, just avoid ES parts and pay a little extra.
The ES is fun in that it can be overclocked and at a low price, but the VRAM of erying motherboard doesn't seem to be able to handle too much power
@@harrytsang1501*VRM
@@harrytsang1501 im rolling with a 13th gen 13900 ES from "10729", same layout, only thing different is they still have the older mobo rear connector (like steve from GN), whereas the erying 13th gen has usb c (and i think theres also a clear cmos instead of the older dvi(or vga) port.
overclocking im able to hit 5.4 all core @ 1.24 or so volts, stil tuning it now but i think im inching myself closer to 1.2v all core, chip still is mostly in the 75w range but it can still go +/- 25w
Nice, Part 2 time to Watch it directly. Even its 22:51(10:51 pm) where i live (Germany). But by the way, i love your Videos.
12:38am here lol, live in South Africa and also love CC's vids, great for information as I'm getting into the homelab stuff
0:44 here, Kyiv
Craft Computing connects people (:
I bought the Frankenstein 12900H (no ES) immediately after the first set of videos on Erying. With their RAM and SSD, to avoid any lengthy troubleshooting steps... And I've been using it as a daily driver light "workstation" (with regular RAM and SSDs), I don't do as complex 3D as I used to, rendering is done in the GPU, since the lack of PCIE lanes isn't so much of an issue for rendering, because (I think) it loads all the textures once and doesn't shuffle them in and out like in games...? And it's been rock solid! Even bought an PCIe E-key 2.5Gb for the wifi slot (thx Wendell for the hack) and it just works under Windows! But the previous video got me worried, since my next step would be to buy an actual workstation and turn this into a low power high speed server. Now I'm so glad it works! Thank you for the info!
Great video!! I was thinking about this kind of set up inspired by your previous videos with Erying products. Please do more this kind of tests also show us part 3 with installation of this boards in your garage server :D
Yeah the ES Boards can be a bit weird, I mean after all they are not rated for stability (or anything really, there's a reason they are ES and not retail and you pay less). You can get away with good stability with just a bit less clockspeed (if any) but I run into this issue as well, when I tested my 11980HK ES Erying board on an unraid server. It run very well and stable unless the system was really doing something because every time after about 5mins of full load the system would reboot. The new bios and the better VRAM coolers did help a bit so I wouldn't see crashes under normal use but I was able to crash it repeatedly when stress testing. For a homelab it's probably fine for me but I wouldn't use it as a NAS or anything where it could corrupt data even. I might play around a but with lowering the multiplier and trying to reduce strain on the VRAMs and see if that may help so there's that.
lol i just finished part one exit back to my homepage and see part 2 released 10 mins ago.... nice.
Cool, now forget everything I said in Part 1 :-D
@@CraftComputingThese are the comments that make me wish for a laugh emoji :)
@CraftComputing wait I'm on part 7 of this series.....
I dont see the use of this for virtualisation, when you are limited on the ram side. CPU is one step but then you need Ram as well. and 96GB Ram are used up pretty fast, when you are hosting game servers or windows vms. And buying several of them is more expensive then just one professional server.
Can you add testing with containers as well? I thought the situation with containers vs VMs and hybrid architecture was a bit funky too. But perhaps that has been addressed as well?
Sorry if this is a dumb question. But, basically let the kernel do the work? As long as you are running Proxmox with the latest kernel-it will allocate the best cores / threads to use? Or do you still need to install the Intel Microcode for this to work at all????
Having newest microcode is good idea either way, but depending on how you installed proxmox you might already have microcode installed - I always install clean debian first and proxmox on top of that (due to partitioning limitations with proxmox installer) and when you select expert mode in the debian installer and select non-free repo you should get the microcode package automatically I think.
Thank you for making this update video! It's wonderful to see someone making timely updates to keep the community informed when new information comes to their attention! Keep up the amazing work!!!
Good sir, thank you! I’ve had my eye on this exact motherboard for my first homelab and am so grateful to have stumbled across your channel! Would you by any chance mind sharing what ram and speed you got working on these and if your made any bios changes?
Edit to add: would you feel this board would be reliable enough to be used in a NAS build and would there be any compatibility issues?
Do i need the intel microcode patch with a 2024 version of the kernel?
I'm running an i5-12600K with currently two VMs and I haven't experienced a single crash in about six months. It's been a great system.
so awesome thanks to all the devs and yourself for the support/testing. excited to try a new erying 13900h itx build for homelab server!
Setting up a mini pc home lab with a minisforum ms-01 and will use your learnings here for that set up. Thank you!
I'm jealous you've got yours already! ;_; I ordered 5 of them on the day they went out and still haven't got a shipment notification!
Fab testing found this really interesting! Looking forward to part 3.
It's wonderful to see a RUclipsr that actually has the stones to admit fault. Great job man.
Glad you took another look into this 😊 I wish you would take another look at the 3070m GPU that got BIOS bricked (if you were unable to unbrick it, you could always send the GPU to another RUclips content creator for the unbrick video - then use the Frankendriver on the fixed GPU)
Yay! more multipart continuation of long form content im actually interested in.
Hey Jeff, have you tried to get vgpu pass through working with the es motherboards? I’m using the 12900hx es lga 1700 cpu, I can get the vgpu to show up in Proxmox but nothing can actually use them when passed through, windows sits on error 43, and jellyfin in a Ubuntu lxc container can’t use it either, I’m wondering if a microcode update would fix this or maybe it’s just ES weirdness?
Probably kinda niche but when you get a free moment could you do a segment on setting up different windows vms on proxmox to be able to use RDP across vlans?
In my case i just setup an xp vm on proxmox vlan 2 and enabled RDP, but my main pc on vlan 1 won't connect to it, im using ubiquiti gear if it matters. I have yet to set up another vm (win 10 or 11) to test it.
Maybe a bonus you could include is how to make sure the VM vlan can talk to the local network(s) but not to the actual internet if it's so desired.
Are you talking about setting up a VM specifically for RDPing into other VMs? Check out my video on Guacamole. ruclips.net/video/Sq-irDBauvo/видео.html
@CraftComputing nope, trying to make it so I can rdp into the vm (on a designated vm vlan) from my main pc (on my home vlan). At the moment I've only setup the one win xp vm, but my win 7 desktop is claiming the vm isn't on the network when I tell the rdp connection program to connect. In my udmpro I have to where the home vlan can talk to any other vlan, established and related is allowed, but vm vlan is locked down like my iot one is.
Edit: I just finished setting up both a win 10 and 11 vm, both work if I type in the ip for RDP, not the pc name though....
Edit 2: I've since setup "dns name" (I think it was) and using computername.networkDNSname.local and it worked on all 3 (10, 11, and xp). Still have to figure out how to block that vm from actual internet access and have it only be intranet.
@@AceBoy2099why do you have them setup on a different vlan? Was it hard to allow the vlans to connect just some traffic?
@blondeguy08 no, it really isn't too hard to set up the different vlans to talk back and forth, I just did the same rules as when I made my iot vlan from watching crosstalk solutions and mactelecoms channels set up videos. Eventually I want to find out how to block all internet access to a specific vlan (vm one) since I'm only using the vms for specific programs that don't need internet access and xp can't do anything in terms of updating anymore so it'd be a major security hole to let it access the internet where nefarious people could find it, not that I'm too concerned but better safe than sorry.
I've had a lot of issues with the 12000 series CPUs, but after applied the microcode, you linked to, it started working as it should :D Usually I am sticking with AMD 5000 series CPUs....
Great Video & Beer..!
so which you will recommended for build a proxmox server core i9 13900hx OR ryzen 9 7945hx?
Very interesting and useful, thank you. A grumpy observation:
'Equivalent' Intel desktop and Xeon CPUs - One would expect the more expensive Xeon would be a better-bined part, if not outright a better design. But what does Intel do? Rob the Xeon customer of E-cores for a higher price and excuse that as a VM issue! Worse, it wouldn't surprise me at all if the supposed 'better' and more expensive Xeon is a desktop equivalent otherwise destined for the trash because it's E-cores don't work!
I'm running Intel NUC with 1340p since november without any issues. Usually there is 5 VMs on at all times and I did not notice any stability issues. Even go iGPU passtrough for jellyfin HW acceleration working in a VM
I went trough few of tteck proxmox helper scripts after install which includes microcode update, that was most likely what saved me from issues
This is really interesting. What this also says is that the N100/N300 process have a real place in the marketplace for multi-threaded systems.
Nice work, great content. Would be even better if you considered power usage statistics when performing these kinds of discovery test. For most, that seems to be one of the biggest concerns with home lab equipment.
I have been running proxmox on a non-ES 12th gen Erying board with no stability issues for a month now.
Better than bare metal. Linux ftw. Love the video. Great content as usual. Thanks Jeff!
Hmm… this makes me think the Minisforum MS-01 is a viable Proxmox cluster candidate. Exciting!
I watched both the videos in this two part series with great interest when they came out. Since then, I've ordered a Minisforum MS-01 i9-12900H that I'm now waiting to be delivered. Thanks so much for all the very useful information. Especially regarding the processor microcode obviously. Interesting times ahead! 👍🙂
I built a 12900k system for low latency audio processing in windows and found that disabling hyperthreading improved performance considerably in that work load. Not really the same thing obviously but thought it was interesting.
you know how active SR-IOV on erying motherboard ? i try to use iris xe but im stuck
Hi Jeff, awesome content as always! Did you try i7-13620H 1x VM, pinned 4T from P cores with HT disabled? I'm curious af
Can you please test the following usecase: passthrough the entire iGPU and audio to a VM and have that VM to output video and audio through HDMi.
Interesting stuff. Won't be able to test it myself though as I only have a Epyc 7543 and a Xeon E-2314. Neither have E cores, and the Xeon doesn't even have SMT :p
I should take a look at some of those micro code updates because I'm running KVM on my 13th gen Intel laptop so I can have a Windows machine for the couple of things I have to do that use that. Might solve some of the small performance issues I'm seeing
So glad to see this update!
Welcome to bleeding edge. I’m glad you were able to find the needed help and succeed in your attempts.
Thank you. It's interesting to see the testing you went through. Also, I am curious to see what stores you buy from on AliExpress as I have never bought from there and want to choose a reputable place.
I still like going full manual if I have a performance target and wants it to be predictable and load independent.
Have a 13900k on proxmox with isolcpus set so that host only uses last 8 E cores, and manually pin each vCpu to the unscheduled P and E cores. The E cores are great for integer workloads such as hosting VM, virtio, and I do keep smaller VMs sharinf those 8 E cores. But when I pin the mix of P and E cores to vCpu, I get the same cinebench results every time, -10-15% if all cores are loaded and power draw hits 253w.
I will won't trust that proxmox will handle it well, since even on bare metal, disabling E cores sometimes does make the machine more responsive.
could disabling e cores allow a much higher power and thermal throttling ceiling for the P Cores be a factor for a more responsive machine? it could just be something Proxmox has no control over.
@@tanmaypanadi1414 I don't think it is power and thermal related. That would only slow down the CPU so much.
I just find that if I want predictable and consistent performance under load, manually setting CPU affinity and CPU pinning helps so much. Especially when your goal is to make it behave like a real machine.
The E cores are great at doing integer tasks, such as handling network stack, Host OS and other qemu/KVM management tasks. But they make slowdowns more often and the performance inconsistency is immediately remedied when I taskset the vCpu onto pinned P and E cores. So that when the guest OS scheduler expects P core performance from core 0, it always gets a P core on core 0 and not whatever proxmox decides.
I was looking through the ERYING store today and come across 10 and 11th gen Xeon integrated motherboards. Would there be any benefit of W 11955M 8C/16th vs the i5 13420H 8C/12T ? both at same Price point of $400 ish AUD.
* A bit of background looking to upgrade my Dell Optiplex 7070 Mini running i5-9500t and separate box running a few HDD's in RAID to a new all in one solution with a bit more grunt for some Extra VM's. Currently running Poxmox 7 with container for all my docker Arr's + Ubuntu VM with iGPU passthrough for Plex.
Hey, is there a plan for beer glasses that can hold 0.5L (about 17oz)? The 16oz pint glass can only hold 473ml and our beer bottles hold 0.5L. I know it's a strange request. But if I order from the USA to Germany, it must be worth it ;)
You and your 16.9oz standards :-D I'll definitely look into it.
@@CraftComputing sorryyyyyyyy ;-P but the best, more beer per glass
This is interesting! What you are demonstrating is essentially a worst case scenario operating with near 100% uptime! Thank you sir.
I wonder what Joyent Illumos would do. The thread/core thing was "solved" by Sun Microsystems and Solaris about 20 years ago in their Niagara architecture. Illumos is the current Free-ish version of Solaris and I'm sure some of that scheduling mojo was carried over. Unix and later Linux always had the ability to manually pin tasks and processes to CPUs/CPU sets. I'm wondering if. you could run a full Cinebench session pinned to the P-cores and then pinned to the E-cores (alone) and see the results.
Also, you gotta try a proper Belgian Witbier or even a Geuze.
I am NOT testing Proxmox on my Sun Ultra 45!
@@CraftComputing I was suggesting the inverse; Illumos is the descendant of Solaris for Intel (and Sparc) and could be considered an alternative to Proxmox (although it is a full OS and not really a hypervisor.) I have no idea how it would handle Ecores v. Pcores at this point, though.
I am surprise because I’m Running a 13900k for Two Years now on proxmox With 16 vms running. ECore allways enabeld and i had never Problems with crashes Without applaying some microcode.
Been wondering if It would be worth it to use my spare, 12900K for my main proxmox server. These videos definitely help with making that decision. Was also toying with the idea of possibly turning this 13900KS daily into a proxmox desktop that still runs windows daily on it but can also run other vm's and containers while being able to shut down windows. Have only seen one other person do this on RUclips, and probably isn't a choice many would want to make for their daily. But I need room to start building more than just the networking portion of my home lab. Buying old Xeon E5v4's and trying to get my MSI GS66 to work with proxmox ve without throwing errors, seems so far like it's not going to work.
I wonder if the NICE level of the host OS's needed threads get's bogged down when the VM's are running full bore is causing the stability crash issues? If all the VM's are using ALL system resources, I wonder if the host OS has enough overhead to not crash...
This was a great set of videos and I learned a lot - looking forward to the next.
Now i Just have to Google if the PCI passthrough works on those Chinese mobo. Thanks
This is an extremely interesting outcome. The gain in the vm due to scheduling especially. Part 3 might male it worth using one of the high end chips on a lab machine if it comes out consistent.
For what it's worth, I've seen similar results on my 13900T server at home. It's the same silicon as the rest of the desktop 13900 chips, just with VPro and a lower power limit. Mine was an ebay salvage find from a lot of water-damaged machines.
I have an ES CPU i9-12900hk and i have problems with it. Proxmox just reboots when the host has uptime 1 or 2 days and i start a virtual machine
Personally on my Alder lake system i only give a VM either Performance cores (threads included) or E cores NEVER both! aka i generally keep the E cores for the base Linux OS and give Widnows Games VM the P cores
Waiting for part 3 🎉
Such a shame about the engineering samples, ive been storking them in AliExpress for months now looking to replace one of my servers running a E5-2690 for a while. And it looks like im going to have to hold off for now 😔.
Lets hope in another gen or two they will fix the stability issues. They always do and i can wait!
Once again, another niche aspect of a broader, very amazing community has come forward to share their knowledge, potentially saving the day!
Gotta love that. 😎😎
This is incredible, i guess i can get off the shelf micro PC(s) for proxmox when they are dirt cheap!
Querry:
Big little architecture....
Didn't "Big-little" architecture used to refer to Big-endian/little-endian architecture?? That is the direction that they like to chase the bits through the processor?? Where am I slightly confused??
"big.LITTLE" is ARM's name for the power/efficiency core stuff. Big Endian and Little Endian is an entirely different thing that only sounds vaguely similar.
@@marekfiferna
I was asking if the second one was called the same thing a few years back....
@@montecorbit8280 And I explained it to you...
@@marekfiferna
Mostly....thank you for that.
@@montecorbit8280 What did I miss?
Can't wait for part three
Do the non mobile chips with hybrid architecture require the microcode applied?
Funny that I just impulse brought the $449 Microcenter 12600KF, z790 mb & arc a770 16gb bundle and got recommended this video!
This is very interesting. I recently built a pc with a 7950x3d . I have been thinking of running proxmox on it, with a gaming VM but have always stopped short given the CPU design. Has anyone tried using a 7950x3D for a gaming VM. Would you need to lock the VM to the gaming cores or would proxmox actually be smart enough to figure it out?
Wait, Proxmox isn’t applying microcode updates by default? Or is the version you got from GitHub just newer?
The installation instructions in the video description do not mention downloading anything from GitHub at all, just to install the microcode update package from the non-free-firmware Debian repositories. I guess the microcode in Debian's official repositories is new enough.
I would love to see this running Hyper-V. I'm a Windows SysAdmin so my homelab is also Windows currently (at least until i can get a good deal on an Erying board lol).
Love the intro, interested to know what will happen with ES board, could it be used for something else like a router or something
How are certain types of cores allocated to machines? There is no such setting in the interface, only the number of processors and cores, but not their types.
Now to see how it works in TrueNAS Scale!
I’d love a video on a cpu core / threads and a crash course of what it all means..
Then it can be more performative to run a VM than bare metal? I'm gonna try it for sure.
Where is link to part 1?
Results presentation and description is a bit hard to follow. I never touched 12gen+ Intel CPUs in my experiments and probably missing some details on the way they function. But watched this video couple of times and still find the way "tables and graphs" are layed out is not helping but making it more confusing.
Gotta love when validating results, and you discover a whole new world 🌎!
Pulling the trigger on a MS 01 after seeing this 🎉
Great that it works .. with the retail board
So mixed cores, actually good if you virtualise. 🤯
the real question is, why isnt proxmox supporting this microcode update from the get go? what are they waiting for? 🤨🤨🤨
I asked myself the same question. In particular, this architecture is not entirely new and many people are struggling with this problem.
@@christopheoberrauch784 to be fair, i've found its very easy to patch the microcode using the Proxmox VE Helper-Scripts utility... that is, unless its not already updated by the mobo bios version 😉
Jeff, I'll send you some IPAs if you can figure this out: 13900K on Supermicro (latest BIOS), Proxmox 8.1.11, pegged CPUs 0-11 on a Windows 11 VM. Cannot get past 3Ghz in the VM. Even I'm locking down the performance cores and microcode is good to go.
How are you measuring/recording CPU speed? If you're checking inside the VM, most performance info isn't actually sent to the VM from the host.
Check stats on the host.
@@CraftComputing Good point Jeff, in Hyper-V I have found a bug that if I move a VM from host to host, the VM does not correctly report the turbo boost, but if I create a VM on that host, it will. To your point, in both Proxmox and Hyper-V the host is reflecting the maxed out frequency and if your saying that I should rely on that and not worry about what I see in the VM, then I'm done trying to see that within the VM. Appreciate your feedback, beer on its way. Have you ever tried Knotted Root Brewery out of Colorado?
I've not had that. Sounds delicious 😋
I think you have stumbled upon why Intel is getting rid of hyperthreading on CPUs in the future and going to "rental units".
Well, this makes running a new hybrid architecture for proxmox or really any VM host that can take advantage of the microcode, (Im lookin at you TrueNas Scale, please integrate) really REALLY promising for home labs only running a few VMs or VM that sit Idle a good bit.
i5-13500 is back on the menu, yay!
How stable are your 11th gen Erying motherboards in Proxmox?
100%, no notes. They've never had a single issue, and have been running for almost a year now. Currently one is running a Minecraft and a Palworld server 🤘
with all due respect sir, proxmox itself not a hypervisor. Proxmox creates a nice tool sets and addons around the kvm hypervisor as a whole virtulization suit, which including management interface (webgui+api) clustering, sdn, storage etc...
So no ES in THIS family of CPUs for virtualization experiments? Sure. I think I'll build out my spare 13400F to tinker. On a different note, I suspect a QS sample of Xeon scalable would most likely be fine given that's more of a "native" workload.
Wow. What a change!