.NET Runtime VS vgpu_unlock

StopTheHostException should be made public to be dealt with gracefully #60600 - GitHub

I ended up filing an issue on GitHub. They then filed a new issue which is still open today.

How .NET Framework (or latest .NET) implemented a certain function / feature, how it varies different design / implementation decisions in reality, etc. (remember 1. the latest .NET is open-source under MIT license. 2. .NET Framework 4.8 Reference Source is not; it is under a thing called MS-RSL. But still you can look at it)

Most of the article basically describes a workaround that Microsoft themselves needed to implement because .NET Core and .NET Framework don’t support in-process side-by-side hosting: https://github.com/dotnet/runtime/issues/53729

runtime/src/libraries/System.Security.Cryptography.Cng/src/System/Security/Cryptography at main · dotnet/runtime (github.com)

There are much better APIs now on types such as Aes, HMAC and IncrementalHash, with support for one-shot operations and Span interop. Your best friends are the API docs, .NET API browser, and just looking at issues and PRs in the runtime repo such as this, this or this 🙂

which could be due to NTLM incompatibility: https://github.com/dotnet/runtime/issues/887

https://github.com/dotnet/runtime/issues/49397 and https://github.com/dotnet/runtime/issues/63331 go into more detail about the APIs that are being added, with just Shuffle left to implement.

Another option is to check out the vgpu unlock project. It allows the resources of a GPU to be split between host and VM. However, it only works on newer Nvidia GPUs. (You can look up valid hardware IDs in the project's python script and a site like this.) There has been some discussion about vgpu here, but I've never done it myself.

That does exist and work, but for NVIDIA cards only. The feature stems from GRID vGPU deployed by NVIDIA on mostly Tesla and some Quattro cards. It is only a software lock on that function however, thus there is some smart folk, who made it usable on most consumer cards with limitation. If you are interested, take a look at this: https://github.com/DualCoder/vgpu_unlock

Yep.

With the money I'd spare by not grabbing a separate NVR I'd also grab a Quadro P400 2GB (roughly 140 euros) for Plex transcoding and Blue Iris hardware acceleration. As far as I saw it could be split between two VMs using vGPU unlocking so I could split this between Plex and Blue Iris so both could use the same card. Since the P400 seems to be able to transcode 2 4K h265 streams to 1080p and possibly can even be unlocked, and most of the time I don't actually need that much power (actually I don't really have 4K content yet due to lack of 4K TV) so I guess half the card should also be enough, especially if I'd use direct streaming.

What you want is called SR IOV. Check this out: https://github.com/DualCoder/vgpu_unlock

#1: Help people help you: put some effort in #2: Vgpu_unlock: Unlock vGPU functionality for consumer grade GPUs | 83 comments #3: Nvidia Resizable BAR drivers released. BIG NEWS: Officially no more Error Code 43!

If your card is supported by https://github.com/DualCoder/vgpu_unlock you can, or if it has vgpu support out of the box, otherwise I don't think there is anything if your card is not supported by vgpu.

> Is that done at the driver level?

No, the FP64 units aren't physically present on silicon in high numbers on FP16 dies.

However, limitations enforced just by the driver and its FW set:

- GPU virtualisation (see: https://github.com/DualCoder/vgpu_unlock)

- NVENC video encoder limitations to 2 simultaneous streams on customer cards

- Lite Hash Rate limitation enforcement to make GPUs less attractive for miners

vgpu_unlock might allow you to use the 2070 simultaneously with the host and guest, but it might cut GPU performance. Using the iGPU for the host seems the best solution.