New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Comments
sounds good
MB model?
Ancient X9DRi-LN4F+, just got SR-IOV Capability showing up a few minutes ago, when I run lspci -vvv , I flipped a bunch of BIOS params but I think the fix was disabling ASPM. Because this is a PCIe 5 card on a PCIe 3 MB, there were negotiation problems and it degraded to x1 from x16. Resizable bar is second problem but will try to launch a couple of VM's with what I have...
Decided to take a chance. Accepting pre-orders.
@Shakib - I know how exactly the GPU will be shared is a bit up in the air yet, but what would fair share for the CPU be for this offier?
Wait, what? I have several RTX cards, and I haven't entered into any contracts with nVidia when purchasing them. I could install them wherever the hell I wait.
If they were passed through to a VM, I guess you could argue that the end user installing the libraries might have to agree not to use them in a datacenter, but I don't think I've ever noticed such a thing when installing drivers myself. But in any case, that'd be a problem for the user not the provider.
As always, you can use 100% CPU whenever needed, just not 24/7. I don't think you will ever see anyone complaining about CPU fair share/limits with their HostCram VPS service.
I don't like to limit things. Everything is uncapped. I don't even block port 25.
I'm not sure how an IOMMU would even be able to separate slices into areas protected from each other unless the device is able to present itself as multiple PCI devices.
If you just take a random consumer card, any user could read/write any other user's memory.
Watch this >
Can you save me the time of watching it and explain how it does the impossible?
An IOMMU can restrict VMs to accessing a subset of physical memory and a subset of IO devices. But each device has its own permissions to access memory for DMA. A single device would be granted permission to access physical memory for all VMs that were shared that device.
If you craft a malicious root signature, you can easily access memory you're not supposed to be accessing from your VM, but it'd be OK because the GPU was permitted to access it. And modifying a root signature is easy because it lives in memory as well.
The cards that do support sharing the GPU across VMs implement their own memory protections in software rather than relying on the IOMMU.
Unless of course your video has some new revelation. But I don't have time to watch it now, but I'll assume it doesn't.
B50 GPU can be split on 6.17 kernel of Proxmox by default.
The stable 9.x release uses currently the 6.17 kernel since Proxmox VE 9.1. This kernel version is derived from Ubuntu 25.10.
Basically, I don't have to modify anything; it should just work fine on my platform.
I googled a bit more. Seems to be something that's not so new, but newer than what I knew about VMs: https://en.wikipedia.org/wiki/Single-root_input/output_virtualization
A bit more info here: https://www.reddit.com/r/explainlikeimfive/comments/lqrikc/eli5_single_root_io_virtualization/
Ordered all the hardware I needed.
Will update you in a week.
Next question is who do you think is going to want this?
Just seems to me that the B50 isn't that powerful (the internet seems to suggest it's got less processing power than my 4 year old 3080) and its main advantage is 16GB of RAM, which is already quite small for any AI tasks, but seems pointless once it's shared between 4 or 8 other people. So apart from light video encoding tasks (and isn't the market already sown up for that with older and now cheap Intels?) I'm not sure who would want this.
BTW don't mean to sound negative. I'm all about interesting VPS propositions, but I just can't figure out who would want this specifically.
Every Windows RDP users will want this. I have dozens of Windows RDP users as clients. I think I can sell this to them if I can make it work the way I am expecting it to work.
I want the GPU acceleration myself to make my own RDP process 3D and play videos smoothly.
I mean who doesn't want a RDP that feels like a PC?
Ah my mistake! I assumed you were chasing the AI market with this...
AI was never the goal. Windows is.
We are fully automated for Windows OS now.
What's the status now? Figured out the solution?
Waiting for parts.
So we go here:
https://my.hostcram.com/order/main/packages/special-offers/
Select the 8GB option, and just order as usual, then wait?
No. Make a custom deal with me.
Edit: Most of the hardware was received, but the motherboard order was cancelled by the seller for some reason and now I am thinking to buy ASUS Pro WS B850M-ACE SE instead of the regular server motherboard.
We are doing it.
Rosewill FBM-X3 Micro ATX Mid Tower Case
ASUS Pro WS B850M-ACE SE Motherboard
Ryzen 7900, 192GB Corsair DDR5, 2x PNY 2TB NVMe 5.0
Antec 240mm ARGB Liquid CPU Cooler
Intel Arc Pro B50 Workstation GPU
Done. Test in progress.
Its part of the driver flow, https://www.nvidia.com/content/DriverDownloads/licence.php?lang=us&type=GeForce 2.1.3, they expect you to only use enterprise cards and not allow gtx/rtx.
Offer violates rules:
Be informative in your offers. List basic specifications (memory, storage, bandwidth, port speed, virtualisation method) and pricing on a per-plan basis.
Pricing is unlisted.
so how to order?
We are doing 5600 MHz stable for all RAM kits.
Further testing will be done.
You can pre-order.
Just shoot me a message.
Added X550 10G NIC.
The plan is to use both X550 and onboard 10G NIC for I don't know what.