OpenCL

I've had about three lockups, and I'm averaging about one a day, or actually I was averaging about two when I first started out, but after some updates, I'm in a pretty good area right now.

would highly recommend using like a time shift and also keeping backups of at least three kernels because you just never know when a bad update may come through because you're literally getting bleeding edge not necessarily testing or staging or development but um very very new stable releases so uh overall i'll kind of update this on my next one and touch into it and see how this is going if i'm still getting those freezes because it's been about a day i have about 24 hours of uptime right now on my machine without a freeze so you know hey it might freeze in this video but uh as of right now it's i'm in a really good spot so i'm really liking it as far as the updates is concerned So the last thing I wanted to go over is certain driver installations or assistance such as Vulkan, Mesa, OpenCL, all these things that help accelerate graphics or utilizing your GPO better in Linux.

uh because a lot of times your gpu just doesn't get utilized in many games i think a lot of linux users don't realize this so they try out uh one of the games and then they get in there and go oh well performance is crap it's done this is just horrible so out of the box the gaming performance wasn't great of course because heck nothing set up an arch out of the box for the most part um i got on warframe and i made a video which i'll link up uh up above here it was gaming and linux and i went over warframe when i first launched warframe without doing any modifications whatsoever it worked But it was very, very laggy.

So it was stutter.

And then also, um, GLS Lang for GLX.

Just checking out that information and what it was giving.

But on top of that, the other two I would mention is Mesa and Mesa Demos and OpenCL dash headers and OpenCL Mesa.

And then there's some other miscellaneous lib32 packages that I downloaded for specific games just to make them run a little bit better.

So it's a little bit of a puzzling bug.

I think it's in relation to how I kind of hacked the open source Vega drivers to use the old school OpenCL from the closed source or proprietary drivers that AMD uses.

So I think my little... issue there may remain it's really kind of annoying me but I'm going to see how that goes and I'm only going to be releasing two more videos in this series one at 25 days or around there and then at the very conclusion i think we're getting to the end here uh you kind of have seen the progression of the switching from windows to linux i've given a lot of pros and cons in the earlier videos and kind of transitioned into just an update format because i'm not really booting into windows anymore so um i'm gonna do one more update video probably here uh mid next week and then at the end of uh next week i should be approaching 30 days and i will conclude this challenge with a overview of my thoughts and what i will be using

going forward Thanks for watching my video.

The libraries from your distro that you're using are probably more up to date than the stuff being used in rocky or rel based systems.

The other thing is setting up the GPU, which has OpenCL or CUDA if you're an NVIDIA user.

That's also a problem.

And then codec support is the next thing.

I can still game with plenty of my older games.

But this is how I install this.

So you have OpenCL and you get Resolve actually seeing that that GPU.

So that's the library launch problem.

But the really interesting work involves graphics and AI acceleration.

Qualcomm is working with Mesa's graphics drivers, including Friderino for OpenGL, Turnip for Vulkan, and Rusticle for OpenCL Compute.

They're also upstreaming fast RPC support to give Linux applications access to Hexagon NPU for local AI workloads.

And they're working on power management, thermals, and system reliability.

And they identified that perfect hashing and the compact hashing.

So they implemented in OpenCL and we wrote this portable GPU implementation, the perfect hashing and the compact hashing implementation for the computational meshes in the different implementation including CUDA, HIP, Cocos, Raja, OpenACC, OpenMP, SQL, around eight implementations and we compare the perfect hashing with the compact hashing.

So this is the computational mesh and this is the refined mesh, adaptive mesh and this is the refined mesh and this is the sparse matrix corresponding to this fine mesh and then means this can be stored in the hash table and perfect hash table to space to save the space this can be stored into the compact hash table.

So, and this means for these computational structures, we need the, for computation, we need to look for the neighbors, top right-left neighbors.

Right.

I mean, there's always been OpenCL, but now we're starting to see Vulkan as a direct abstraction layer for AI processing.

But we're also seeing a couple other basically universal execution layers that are able to run AI code on any hardware that you have.

And so not only performance-wise, but you can also run the software on basically whatever you want.

This is a, this is a treasure.

Even, even Apple has recognized, I think, that after the OpenCL moment of January where they sold out all their Mac Minis, there's something happening here, and I think Apple has, has kind of said, "You know, that creator thing that we were doing?"

[laughs]

"There's this whole new group of people who are willing to spend absurd amounts of money on Mac hardware.