WebGPU :: Compute Shaders Crash Course For Beginners

[@xade8381]

plz continue, youtube lacks structured and comprehensive content on webGPU

[@visionary_3_d]

Sure! Thanks for watching

[@MaxScottTheCoder]

very true

[@SuboptimalEng]

As a small Web Graphics/WebGPU KZbinr, it’s great see more folks talking about this! Especially showing how it can be used for AI workflows. Great work 🤓

[@RickBeacham]

Please more WebGPU compute shaders! not enough of this content. Cheers.

[@vokusz]

Thanks for helping bridge enthusiasts to start creating amazing experiences with WebGPU, all thanks to your passion for creating invaluable videos like this. Much love!

[@visionary_3_d]

@@vokusz I’m hoping to see some amazing stuff 😃. Thanks for the kind words, I appreciate it ❤️

[@HamzaQayyum]

Been trying to get into WebGPU for compute with no prior GPU prog. experience (e.g. CUDA). As such, this video was very helpful. Thank you!

[@MrCOPYPASTE]

Congratulations for the very nice and informative tutorial. Love the way you used markdown to write into the video. Hope you get a lot of subs and likes you deserve it. Cheers 🙃

[@visionary_3_d]

Thank You! Yeah that markdown trick was very exciting when I discovered it 😄 Thanks for giving me lots of positive energy! I appreciate it ♥

[@MrCOPYPASTE]

@@visionary_3_d No problem my friend, I know you are awesome!!! Keep up the good work!!!!

[@ozzy-fr7vj]

please, more vedios on compute shaders, this was amazing

[@visionary_3_d]

@@ozzy-fr7vj alright 🤝

[@peoplebehindindustry]

Fk yeah!! Been waiting for this sooo long

[@visionary_3_d]

@@peoplebehindindustry Let’s Go 🔥

[@minecraftermad]

i can do 92% in 6.5 seconds (1epoch) on the cpu without CNN so definitely some optimization needed there :D

[@visionary_3_d]

Yes. Exactly! There's a lot of room for optimization. Can you also share if your algorithm uses SIMD, or other optimization techniques ?

[@dude2542]

Wow i just started it then there's a video.

[@RickBeacham]

This is exactly what i was thinking. Thanks

[@jizhang2407]

Thanks for sharing. Part 6 is...not explained?

[@visionary_3_d]

That's right! I will explain that part in a dev log (soon).

[@int-64]

Thank you

[@visionary_3_d]

yw.

[@ErwinMotor0]

love your conetnt ncie job

[@visionary_3_d]

@@ErwinMotor0 Thanks 🙏 appreciate the kind words 🤝

[@mnajjar85]

Can GPU be utilized to run a Genetic Algorithm run!

[@horizon0514]

thanks bro

[@visionary_3_d]

yw.

[@romanl.4293]

Hello, what's the color schema do you use for coding and code examples in this video?

[@visionary_3_d]

@@romanl.4293 it’s a custom one. but I use Monokai Pro ( Octagon ) as the base.

[@hamzarashid7579]

What font you're using for coding?

[@visionary_3_d]

Glad you liked it! I'm using Consolas ligaturized v3.

[@danser_theplayer01]

Idk man, I wasn't sure if it's faster to bounce off a pointer to the value, or to perform one entire multiplication every time I want a specific index. I went with an array of arrays approcach to access indeces simply by pointers instead of having extra calculations (multiplication is expensive).

[@visionary_3_d]

@@danser_theplayer01 Fair. It could be faster as I haven’t done any performance testing. You brought up a good point 👍 However for building a tensor math system you need a buffer and a shape for flexibility. Also WebGPU only supports CS shaders up to 3 dimensions. For higher dimensional tensors we’re forced to flatten the array anyways.

[@danser_theplayer01]

Lesgo, we gonna lure all the math and ML people over from Python to JavaScript🔥🔥 (which is faster than python from the get go).

[@minecraftermad]

eh, mojo has a higher chance for that to happen, since it's more familiar and it's a sane environment unlike js, TS idk enough to say.

[@novantha1]

I mean, I’m planning on taking everything in this video to C and compiling to WASM / WebGPU, so I’m not sure how many people you’re converting, lol.

[@danser_theplayer01]

Darn shame it's not supported in Node. I'm making some math package and I could really benefit from like a hundred distinct cores on the GPU instead of an average of 12 logical processors in a mid range computer CPU.

[@visionary_3_d]

@@danser_theplayer01 it should be available in node : developer.chrome.com/blog/webgpu-ecosystem Although I haven’t tried it myself.

[@jinil9002]

so, from now on, don't we have to use torch?

[@visionary_3_d]

for serious machine learning stuff ? sure!

[@jinil9002]

@ wow, please make a video for the huggingface transformer alternative.