GPU MODE أُعيد تغريده
@marksaroufim @GPU_MODE @NVIDIAGTC
Award ceremony for the nvfp4 kernels.
Come hang out at GuildHouse!
English
GPU MODE
226 posts
GPU MODE أُعيد تغريده
What’s the best model you can train in a day if someone hands you a pile of Blackwell GPUs?
You can try out yourself
On April 9 in Paris, @GPU_MODE + @verdacloud + @sestercegroup are hosting a GPU hackathon with a bunch of GPUs to run on and even more of them for the winners.
English
@EthanHe_42 This needs a 5-lecture sequence at @GPU_MODE
English
My last open-source project before joining xAI is just out today. Megatron Core MoE is probably the best open framework out there to seriously train mixture of experts at scale. It achieves 1233 TFLOPS/GPU for DeepSeek-V3-685B. arxiv.org/abs/2603.07685
English
GPU MODE أُعيد تغريده
Join the GPU MODE Hackathon, sponsored by AMD, and push the boundaries of LLM inference performance on leading open models, optimized for AMD Instinct MI355X GPUs.
Finalists will compete for the $1.1M total cash prize pool across two independent tracks, each focused on a specific model and inference stack.
Learn more and get registered here: luma.com/cqq4mojz
English
GPU MODE أُعيد تغريده
LLMs are now superhuman at reward hacking our kernel competitions
Natalia Kokoromyti, was #1 on last problem of the NVFP4 competition for around 10 min before we scrubbed the reward hack
I know of very few humans who can write such a hack gpumode.com/news/reward-ha…
English
GPU MODE أُعيد تغريده
A bit impromptu but we're having a small GTC pregame in SF on March 11 with a goated speaker lineup
@tedzadouri the first author of Flash Attention 4, @bfspector flapping efficiency expert at @flappyairplanes and Orian Leitersdorf at Decart
English
@amaratatva It’s free via the kernelbot platform github.com/gpu-mode/popco…
English
Our next kernel competition is now open for submissions! A $1.1M cash prize competition sponsored by AMD on optimizing DeepSeek-R1-0528, GPT-OSS-120B on MI355X
Registration: luma.com/cqq4mojz
English
GPU MODE أُعيد تغريده
Asymmetric hardware scaling is here. Blackwell tensor cores are now so fast, exp2 and shared memory are the wall. FlashAttention-4 changes the algorithm & pipeline so that softmax & SMEM bandwidth no longer dictate speed. Attn reaches ~1600 TFLOPs, pretty much at matmul speed!
joint work w/ Markus Hoehnerbach, Jay Shah(@ultraproduct), Timmy Liu, Vijay Thakkar (@__tensorcore__ ), Tri Dao (@tri_dao)
1/
English
黄老板所说的 “买得越多，省得越多” 真的靠谱么？
3月7日，GPU Mode 第100场激情拆解。黄老板亲传子弟 Dylan Patel，基于 @SemiAnalysis 全新 InferenceX 基准，直播拆解 GB300 NVL72（假设你能买到）和 H100 的真实推理性能天花板，扒一扒核弹厂的底裤。
顺便，InferenceX 实锤：老黄的刀法，确实比苏妈的芯片更锋利 🔪
算力军火商的私房课，来了
中文
We're celebrating our 100th lecture with @SemiAnalysis_ tomorrow at 9am PST and we'll be talking about arguably the most important OSS benchmark suite out today InferenceX
Time flies, thank you to everyone who stuck around for this long 🥲
English
Breaking: # of cuda RL agents set to outpace # of ML DSLs by 2030
GIF
DailyPapers@HuggingPapers
CUDA Agent by ByteDance A large-scale agentic RL system for high-performance CUDA kernel generation. Achieves SOTA on KernelBench—100% faster than torch.compile on Level-1/2 splits and 92% on Level-3, beating Claude Opus 4.5 and Gemini 3 Pro by ~40% on the hardest tasks.
English
GPU MODE أُعيد تغريده
I hired someone at this hackathon and he’s doing great. Honestly this is top recruiting grounds
And he stood out cause he just happened to have a printed resume with him haha
GPU MODE@GPU_MODE
We did this at the Jane Street hackathon last year, the room. It was intense af, this was the room a few minutes before the winner was declared on the public leaderboard
English
We did this at the Jane Street hackathon last year, the room. It was intense af, this was the room a few minutes before the winner was declared on the public leaderboard
0xm℮r@0xmer_
AI hackathons should be more like gpu mode kernel competitions and not "who can put together agentslop the quickest"
English
There's more fun stuff you can see here gpumode.com/news/jane-stre…
We treat all our events with the same seriousness it'd take to ship a small indie game
English
GPU MODE أُعيد تغريده
It was a pleasure to be able to speak at @GPU_MODE today
I spoke about some of my work at @exolabs, discussing ML on consumer devices, and how to do distributed inference and training on Apple Silicon with MLX
youtube.com/watch?v=sV0PJC…
YouTube
English