LeetGPU

89 posts

LeetGPU banner
LeetGPU

LeetGPU

@LeetGPU

The GPU Programming Platform.

参加日 Ocak 2025
3 フォロー中2K フォロワー
alexine 🏴‍☠️
alexine 🏴‍☠️@alexinexxx·
just came across this cool blog, CuteDSL rabbit hole begins
alexine 🏴‍☠️ tweet media
English
6
5
209
9K
LeetGPU
LeetGPU@LeetGPU·
@ens_pyrz Awesome! Do you have any feedback for us? How can we make it more useful?
English
0
0
0
29
Enes Poyraz
Enes Poyraz@ens_pyrz·
Solved all Challenges on #LeetGPU ! It was teamwork of Human + AI! I am the only one that solved all Problems in CUDA :)
Enes Poyraz tweet media
English
2
0
2
97
DM
DM@Sudden_Sudo·
@vikhyatk @sasuke___420 @JohnThilen @willccbb Leetgpu is disappointing. Tensara is actually quite good but could use far more problems. How they score Leaderboards is a bit iffy though.
English
2
0
2
68
will brown
will brown@willccbb·
fake software engineering jobs are dying rapidly. real software engineering jobs, however,
English
47
27
1.4K
146.9K
LeetGPU
LeetGPU@LeetGPU·
We just released a new challenge today ❗ GPT-2 (120M) Transformer Block Compose multiple kernels into a full transformer block. The first of many upcoming challenges focused on real-world inference optimization. Write your solution in CUDA, Triton, PyTorch, JAX, Mojo, or CuTe DSL and benchmark it on state-of-the-art GPUs like the H100, H200, and B200, and more.
LeetGPU tweet media
English
8
53
586
29.9K
LeetGPU
LeetGPU@LeetGPU·
PTX + SASS are now viewable on LeetGPU! See exactly what your CUDA turns into, instruction by instruction
English
0
5
13
741
LeetGPU
LeetGPU@LeetGPU·
We just launched a couple of new features on LeetGPU - PyTorch Profiler Traces for every submission - AI Chat to help explain, debug, and optimize your code Go try them out!
English
0
0
14
1.1K
vøxium
vøxium@v0xium·
@therealkmans @sadernoheart @LeetGPU Thanks, it is specially required for the CLI as a lot of books have their own util headers, and examples cannot be coded in just a single file to learn. I hope it comes out soon.
English
1
0
0
43
sadernoheart
sadernoheart@sadernoheart·
day 76/100 of GPU Programming - wasted no time either, solved it and currently the fastest FP16 batch matmul across all available GPU's on @LeetGPU - notably with a peak runtime speed of 0.0567 ms on the H100 here's how i did it🧵:
sadernoheart tweet mediasadernoheart tweet media
sadernoheart@sadernoheart

spoke to one of the builders at leetgpu.com, @therealkmans, about having an FP16 batched mat mul so i can try programming with tensor cores and he got it up in no time. i'll say it again, this is the best place to learn and build muscle memory on writing cuda

English
3
7
137
11.1K
LeetGPU
LeetGPU@LeetGPU·
How can we make LeetGPU better for you? Let us know 👇
English
2
0
1
491
LeetGPU
LeetGPU@LeetGPU·
We just shipped JAX support on all challenges 🚀 Try it out!
English
0
1
8
539
LeetGPU
LeetGPU@LeetGPU·
Congrats!! We're super glad we could help you grow as a GPU programmer ❤️
sadernoheart@sadernoheart

day 100/100 of GPU Programming Didn't write a kernel today. I spent the day reflecting. 100 days writing kernels and I didn't miss a single day, not one. On some days, I learnt to write new ones, some days I practiced kernels I've written before. I took on something my younger self would never have imagined taking on. In learning how to write kernels, I learnt how to learn. When I first heard about this challenge from @hkproj, I had zero knowledge about GPU Programming, I barely knew how a GPU looked or worked. I had no experience whatsoever with C/C++. GPU resources were scattered and scarce. And aside from that, my biggest challenge was being GPU poor. But that didn't stop me, I still decided to take on the challenge. I had no definitive path. If there's something I've witnessed is that If you solely decide to take something head on, resources will come your way. You don't have to have a very clear path, just get started. You'll get to see the road ahead as you go. I went from writing and profiling CUDA kernels on Google Colab to finding out about @LeetGPU which accelerated my progress exponentially because I now had some kind of definitive path writing and learning specific types of kernels. These past 100 days have been my best in terms of Programming. Through this journey, I made friends, got haters, got job offers and learnt so much. X is the best place to be, if you consistently put yourself out there, alike people will show up. 100 days ago I couldn't even write a vector add, fast forward 99 days later, I'm writing FP16 Matrix Multiplication Kernels via WMMA APIs. There are those who felt going on an 100 day challenge was obnoxious and performative. I didn't find that so, I pushed myself to the limit. I wrote kernels on days I felt burnt-out and depressed. I wrote kernels when I was busy with school and life, some days I didn't play football just so I can get a kernel done. In the process, I learnt to be consistent. I didn't do this for anybody by myself. Before @ludwigABAP and @0xmer_ noticed me and put me out there, I was basically showing my progress to myself(I had about 200 followers). Having more people inspired by my work made it even more fulfilling for me because I had people looking up to me. So for anyone who want to take on the challenge on study something for 100 days straight, I encourage you and for the people already in the process, keep going. Put your head down, ignore the voices and accelerate. You really can just do things!

English
5
7
579
58.5K
LeetGPU
LeetGPU@LeetGPU·
It's so nice
LeetGPU tweet media
English
0
0
5
2.1K