CPU vs GPU | GPU Programming | Episode 1

[@silience4095]

Great video! Excellent introduction. Keep up, I just want to add that one of the main differences is that CPUs are absolutely incredible at branching code, while GPUs are terrible at that. CPUs have decades worth of optimizations specifically to handle branching really well, and keep the pipeline going as uninterrupted as possible.

[@lqx7]

This is such quality content. Surely it’s going to blow up.

[@dhananjayansudhakar3811]

Thanks for explaining it so well, yet in simple terms.

[@Stefan-td1pw]

Can't wait for this to pick up the traction it deserves, investing at 100 views

[@xeqqail3546]

real XD

[@vastabyss6496]

Commenting for the algorithm!

[@rickpala_]

Bro the visual explanations are amazing! Wish the audio was a little better!

[@Humble_Electronic_Musician]

Excellent!

[@Coder.tahsin]

Nice video , I am learning OpenCL

[@Anjalisharma-dk6tk]

i feel like this channel ,especially this series, need a overwork. wherein, sir Simon has to make sure we know what the pre requites are. i am currently just found this channel as i was looking for good computer architecture animation channel(don't mind me there). and the content went in full speed without any seat belt. on a serious note, what is the pre required for this series?

[@Anjalisharma-dk6tk]

just in case you are like me. there is a zero episode with all these given. i will study everything needed before starting this series. anyone that can guide me for more sources to go before starting this beautiful series?

[@houski4242564]

@@Anjalisharma-dk6tk I am trying to understand computer architecture too. I looked into it and I started learning from coursera course called “Building a Modern computer for first principles: NAND to Tetris” they provide everything to learn computer architecture from absolute beginners.

[@Anjalisharma-dk6tk]

@@houski4242564 thanks for the guidance ski, i might look into that series as well!

[@TheCollectiveHexagon]

@@houski4242564 would you say thats good for a beginner? or is some background needed?

[@SAhellenLily]

🤔🤷

[@louisjx8009]

Dude, wtf. You can't compare CPU core and GPU core. They are not the same. GPU core count take into account the line width of the data and CPU just don't. One CPU core can process 8 to 64 additions in one cycle, but you don't multiply the core count by 8. But that's exactly what GPU manufacturers do...

[@szymonozog7862]

You are right, I just wanted to show a rough outline of the scale but the difference should have been made more clear in the video. Thank you for pointing that out

[@vitalyl1327]

Wrong. GPU core is not just a wide SIMD, there's a lot more there. A CPU equivalent would have been each SIMD instruction sandwiched in scatter-gather instructions.

[@StarsManny]

You can compare any 2 things that you like. After the comparison, you will find that the 2 things are similar, or fairly similar, or different, or very different etc. The fact that "They are not the same" is actually a very good reason to compare them.

[@frenchmarty7446]

"One CPU core can process 8 to 64 additions in one cycle." Yes, but that is still a single instruction on a single contiguous array. We think of a CPU as a set of hardware threads executing independent sequential streams of instructions. This isn't actually what's happening but that's the abstraction we use. This way of thinking just doesn't translate to GPUs. GPU cores *can* execute separate instructions (i.e. the GPU core is the smallest unit of abstraction in terms of instructions), they are not just sub-units of big SIMD instructions. But they are not as flexible as separate CPU threads; for example, GPU threads in the same group suffer a serious performance penalty if they hit a conditional statement and branch off in separate directions (though they *can* do it). It's not fair to say that GPU core counts are arbitrarily inflated. It makes sense within the GPU paradigm.

[@SkegAudio]

your sensitive reaction is not warranted, my guy

[@dovos8572]

what language are you using in the examples? oh it is cuda. welp more to learn than i thought