Windows7ge

It looks like I'm on v7.4.4. I'm still new to this. I have no idea where get a version eight.

Currently trying to diagnose my F@H client. Since the move to Linux I've gotten mixed results. CPU now bounces anywhere between 270~400K PPD from 200K, considerable gain over Windows. GPU however holds really steady around 350K PPD. On Windows it was 900K. I don't know why. I'm gonna try creating the VM again. Maybe disabling Secure Boot will fix loading a kernel driver but if this is how the GPU is going to perform under Linux I have to go back to Windows. I have to use a version of Windows though that won't reboot itself whenever there's an update. I think I have an .ISO for Enterprise LTSC. Maybe that will do better.

This is starting to make sense now. Some WU's are heavier than others. In some instances multiple WU's on a GPU would make for better overall utilization. In my case with this cute little thing it looks like most WU's already keep it in the 90 percentile and the current WU has it pinned at 100% so trying to divide it will probably hurt more than it helps. I got a LXC container running so the CPU is folding native. I got the QEMU/KVM machine reset with a P4-8Q vGPU. The Windows VM has been repurposed to just run FaH Control. I gave the LXC 60 of 64 threads. So a core for the GPU, a core for the guest, and hopefully that's enough for the hypervisor too. xD I'll drop it down to 58 if I think it might help later.

Well, word through the grapevine says there's a second P4 headed my way for no reason in particular... :3

Nope, it showed up in the mail and I IMMEDIATELY started setting up the Linux KVM host driver and deploying the guest drivers for Windows and Linux. I never passed through the card itself. Does vGPU in itself have noticeable overhead? If you've tried bare metal you seem like a good person to ask. If loss is negligible for 1 vGPU I'd prefer to keep playing with the feature cause this is cool as hell.

I can agree with just handing out one vGPU to a single VM probably being better overall but I don't know if these GPU's can be used bare metal or if they'd only work in this application as a vGPU. I'll probably move the CPU to bare metal in an LXC container. It needs all the support it can get and I know QEMU/KVM does have overhead when you stress the processor. It's measurable.

Now unless I'm completely mis-understanding how folding works (which is very possible) so far using vGPU in a 50/50 arrangement looks to be a detriment... We were over 900K yesterday with just Windows but now that the card is pinned at 100% utilization across two VM's we've dropped off a cliff. FAH Control does show some good information though. It's telling me Linux is a promising investment of time instead of continuing with Windows. Windows VM: Linux VM: At least for compute if splitting the GPU is hurting me here I'll just move the whole card over to the Linux VM. @leadeater Thoughts? For CPU compute I'm either going to disable it entirely or move it to a dedicated LXC container so it can run on the host kernel. This should yield better CPU compute performance over full virtualization which was my experience with BOINC. I hope FAH Control can allow disabling the CPU all together in the slots settings.

@leadeater We are exploring territories I genuinely never once thought imaginable. Windows 10 22H2 virtual machine: vGPU Software Licensed Product License Status : Licensed (Expiry: 2024-1-24 2:41:39 GMT) Ubuntu Server 22.04 virtual machine: vGPU Software Licensed Product License Status : Licensed (Expiry: 2024-1-24 2:52:8 GMT) Both of them using one GPU: Wed Oct 25 23:18:26 2023 +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.104.06 Driver Version: 535.104.06 CUDA Version: N/A | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 Tesla P4 On | 00000000:22:00.0 Off | 0 | | N/A 68C P0 53W / 75W | 7519MiB / 7680MiB | 98% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 3124239 C+G vgpu 3744MiB | | 0 N/A N/A 3157034 C+G vgpu 3744MiB | +---------------------------------------------------------------------------------------+ I'm gonna keep an eye on this new Linux VM. Make sure jobs finish and I hope the average credit for both CPU & GPU go above that of it's Windows counterpart. If it does I'll ditch Windows and spin-up another Linux Server VM. I will keep this in mind. Thanks.

You might be able to use FreeDOS here if you don't already have a means to boot from MS-DOS. What are the contents of the F.bat file? It looks like AFUDOS.EXE is the firmware update program. C99Q3B23.ROM is the firmware image itself. I'm assumeing F.bat is the script that runs the update. So assuming all of these are in the same directory you'd just run the command: f.bat My guess without more context is it will run a command like: afudos.exe -h c99q3b23.rom

That's where I was but now I'm seeing Configure -> Folding slots, default is -1. I'll tweak that as necessary. Just got a vGPU passed to Ubuntu Server: 06:10.0 VGA compatible controller: NVIDIA Corporation GP104GL [Tesla P4] (rev a1) It only has 4GB of VRAM I left the other 4GB on the Windows VM. Now I have to figure out the drivers.

OK I'm starting to learn how this software works. FAH Control is apparently indefinitely dead on Linux because of python-gnome2 no longer being available in Ubuntu and the FAH devs not updating the client to python-gnome3. I have it working in a Windows VM and just got a second (linux server) VM to connect after editing it's config file. The hypervisor CPU is now seeing utilization on most of the 64 threads. Since I'm only seeing Low, Medium, High as options I might artificially lower the cores per VM just so it doesn't choke on itself.

Since I have 64 threads to work with on one of my servers I think I'm going to opt to create two virtual machines and give each of them 32 threads w/ a GRID P4-4Q. Should cut the Tesla right down the middle and let me play with multiple platforms while utilizing more of the CPU.

What do you mean by slots?

CPU utilization is being weird. It seems like either Windows or the folding client can't recognize over 32 threads. May be more reason to use multiple virtual machines.

Ah, so when I looked up your name from that link in the first post that really was you. You have 1 billion points. Ironic. All I know is when they say GO I'm gonna get smoked by everybody.

The number fluctuates wildly day to day but evidence suggests ~900k is what I'm averaging. I want to split it into two GRID P4-4Q vGPU's as I'm seeing a swaying back and forth of % utilization. I think running two project tasks at the same time would be better and on Linux at that if I can get it working. @leadeater Based on the numbers I've been seeing I'm getting the feeling my ~1.2mill PPD is "cute".

I did create a passkey and I used it when I setup my folding client so it should be good to go. I have no context as to what constitutes a lot of PPD so what are some people here averaging?

Oh it hasn't started yet. Alright cool. So I read the first post. Just submit my forum name, folding name, and create a passkey? I already joined team LTT and I'm folding.