Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Two days ago, Deepseek surprised everyone with an "undefined-behavior" PTX optimization speeding up particular ML workloads on a Hopper NVIDIA GPU Kernel. Let's reverse engineer the hack, implement it ourselves, and benchmark the speedup on an H100.

LaurieWired

155,104 subscribers

228,591 views • 1 year ago •via X (Twitter)

Science & Technology News & Politics Education

Anya Rossi• Live Now

Private livecam show

11 Comments

LaurieWired1 year ago

Full Video:

LaurieWired1 year ago

My test code:

NetMind.AI2 years ago

Get access to a wide range of GPUs like H100, A100, 4090, 3090 and save over 90% at NetMind Power. Rent Now!

numanumabruh1 year ago

You'd never know she's 6'5"

Jason Ho1 year ago

laurie supremacy

Bob (Moderna #7) Kerns1 year ago

Until recently, I'd only seen your tweets; the first video I encountered was the 2025 prediction ones. Assumptions violated: higher voice, younger. Always good to have one's assumptions flagged, but especially the age. I was struck by the maturity of your analysis!

Dave 🚀1 year ago

LFG!

KnowledgeisMostValuable1 year ago

I'd fight off a bear for you

nisten - e/acc1 year ago

lfg

🥀shiVam🥀1 year ago

wait, did you film this at Google HQ? (must appreciate the audio recording and editing)

Calcs1 year ago

Fantastic video, more please, lol 😂

Related Videos

Just built an MCP for Ghidra. Now basically any LLM (Claude, Gemini, local...) can Reverse Engineer malware for you. With the right prompting, it automates a *ton* of tedious tasks. One-shot markups of entire binaries with just a click. Open source, on Github now.

Just built an MCP for Ghidra. Now basically any LLM (Claude, Gemini, local...) can Reverse Engineer malware for you. With the right prompting, it automates a ton of tedious tasks. One-shot markups of entire binaries with just a click. Open source, on Github now.

LaurieWired

285,649 views • 1 year ago

a Tsinghua University lab just put a project on GitHub that replaces a $400,000 H100 rack with a single 24GB graphics card. it's called ktransformers, and the trick is almost stupidly simple: the experts you actually use stay on the gpu, the ones you don't sit on the cpu until they're called / deepseek-v3 and r1 with 139K context in 24gb of vram / up to 28x speedup over the standard setup / fine-tune deepseek-v3 across four rtx 4090s instead of a datacenter / built by tsinghua university's madsys lab, not a startup with a landing page apache 2.0, and already past 17,000 stars. -> bookmark it.

a Tsinghua University lab just put a project on GitHub that replaces a $400,000 H100 rack with a single 24GB graphics card. it's called ktransformers, and the trick is almost stupidly simple: the experts you actually use stay on the gpu, the ones you don't sit on the cpu until they're called / deepseek-v3 and r1 with 139K context in 24gb of vram / up to 28x speedup over the standard setup / fine-tune deepseek-v3 across four rtx 4090s instead of a datacenter / built by tsinghua university's madsys lab, not a startup with a landing page apache 2.0, and already past 17,000 stars. -> bookmark it.

dunik

312,503 views • 1 month ago

Woow Nvidia has just released a 2.6B open-source world model 🔥 You can turn a single image, text prompt and trajectory into controllable worlds... And on a single GPU! - Code available on GitHub - Paper as well on arxiv You can use it for many things like embodied AI and robotics research, simulations, etc. Because it can run on a single GPU (like an RTX 5090 or H100) it makes world models accessible to basically everyone!

Woow Nvidia has just released a 2.6B open-source world model 🔥 You can turn a single image, text prompt and trajectory into controllable worlds... And on a single GPU! - Code available on GitHub - Paper as well on arxiv You can use it for many things like embodied AI and robotics research, simulations, etc. Because it can run on a single GPU (like an RTX 5090 or H100) it makes world models accessible to basically everyone!

Paul Couvert

174,008 views • 3 months ago

You can run CUDA, on a Mac ARM GPU, in the browser. It sounds ridiculous but it actually works. HipScript chains CUDA, to OpenCL, to Vulkan, to Tint (Google’s shader translator), to a WASM WebGPU. I got a plasma simulation in running in just a few minutes, no NVIDIA GPU!

You can run CUDA, on a Mac ARM GPU, in the browser. It sounds ridiculous but it actually works. HipScript chains CUDA, to OpenCL, to Vulkan, to Tint (Google’s shader translator), to a WASM WebGPU. I got a plasma simulation in running in just a few minutes, no NVIDIA GPU!

LaurieWired

160,427 views • 1 year ago

Timelapse #177 (14 hrs) - Picked up an RTX Pro 6000 Blackwell and wired in another 1000W PSU - running two of them now - DeepSeek v4 Flash running locally on both GPUs - Distilling DeepSeek v4 Flash down into Nanbeige (still cooking) - Kernel bench runs for DeepSeek v4 Flash and Qwen 3.8 Max - Got it into chat but couldn't reliably get it into the OMP harness yet - Private contract work - days like this are stuff I actually want to build. and when it's the high-intensity grind, i still have to show up and get it done

Timelapse #177 (14 hrs) - Picked up an RTX Pro 6000 Blackwell and wired in another 1000W PSU - running two of them now - DeepSeek v4 Flash running locally on both GPUs - Distilling DeepSeek v4 Flash down into Nanbeige (still cooking) - Kernel bench runs for DeepSeek v4 Flash and Qwen 3.8 Max - Got it into chat but couldn't reliably get it into the OMP harness yet - Private contract work - days like this are stuff I actually want to build. and when it's the high-intensity grind, i still have to show up and get it done

Elliot Arledge

17,401 views • 14 days ago

YOUR PARENTS PAID FOR THE CUDA MOAT! The #1 contributor to the CUDA MOAT isn't the the developers at NVIDIA, but it is the millions of developers outside of NVIDIA that invent new algorithms for CUDA like Flash Attention. For most of them, it started with an GeForce gaming GPU. NVIDIA is the only companies that has an reasonable good developer stack on consumer grade GPUs. As people grow up beyond playing CSGO & League of Legends & Minecraft, they either become anime weeaboos or they start programming on their existing computer with has an GeForce GPU

YOUR PARENTS PAID FOR THE CUDA MOAT! The #1 contributor to the CUDA MOAT isn't the the developers at NVIDIA, but it is the millions of developers outside of NVIDIA that invent new algorithms for CUDA like Flash Attention. For most of them, it started with an GeForce gaming GPU. NVIDIA is the only companies that has an reasonable good developer stack on consumer grade GPUs. As people grow up beyond playing CSGO & League of Legends & Minecraft, they either become anime weeaboos or they start programming on their existing computer with has an GeForce GPU

SemiAnalysis

25,230 views • 4 months ago

Unboxing the ROG Zephyrus G14 🔥 Featuring an Intel® Core™ Ultra 9 386H processor and up to an NVIDIA® GeForce RTX™ 5080 Laptop GPU, it delivers stunning visuals on a 14-inch 3K ROG Nebula HDR OLED display with up to 1100 nits peak brightness.

Unboxing the ROG Zephyrus G14 🔥 Featuring an Intel® Core™ Ultra 9 386H processor and up to an NVIDIA® GeForce RTX™ 5080 Laptop GPU, it delivers stunning visuals on a 14-inch 3K ROG Nebula HDR OLED display with up to 1100 nits peak brightness.

ROG Global

20,645 views • 4 months ago

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Daniel Romero

38,220 views • 1 month ago

🧠 Chat with Reasoning A few days ago the DeepSeek team released a LLM model with reasoning in various sizes. This we show is an example of 1bl that can run on machines with low GPU power like a mobile, but have enough power to answer complex questions. With these advanced models it is possible to link it with #IoT equipment to control information and use it in advanced control environments. All this under Open Source models and decentralized networks such as #Neurai #XNA $XNA #DeepSeek #Reasoning #AIchat

🧠 Chat with Reasoning A few days ago the DeepSeek team released a LLM model with reasoning in various sizes. This we show is an example of 1bl that can run on machines with low GPU power like a mobile, but have enough power to answer complex questions. With these advanced models it is possible to link it with #IoT equipment to control information and use it in advanced control environments. All this under Open Source models and decentralized networks such as #Neurai #XNA $XNA #DeepSeek #Reasoning #AIchat

NeurAI Project / XNA

17,691 views • 1 year ago

"If Wemby is an A+ on the defensive end, and let's say he's an A- on the offensive end – does anybody match up with those grades? And the answer is nobody." – DP on Victor Wembanyama's chances to win MVP.

"If Wemby is an A+ on the defensive end, and let's say he's an A- on the offensive end – does anybody match up with those grades? And the answer is nobody." – DP on Victor Wembanyama's chances to win MVP.

Dan Patrick Show

177,071 views • 4 months ago

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

elvis

261,622 views • 2 years ago

If you have compute hungry AI ML tasks, you can spin up an Intel Gaudi2 instance and watch it just rip. I wrote a quick Python script to use the facebook/detr-resnet-50 model (🙏🏻 Hugging Face) to detect and label objects in a video. It's interesting that no one is talking about Intel Gaudi vs NVIDIA A100 or H100. Gaudi2 isn't cheap at $10.42/hour but it feels like a viable alternative to fighting someone for an A100 or H100. Here's my code if you want to try it out:

If you have compute hungry AI ML tasks, you can spin up an Intel Gaudi2 instance and watch it just rip. I wrote a quick Python script to use the facebook/detr-resnet-50 model (🙏🏻 Hugging Face) to detect and label objects in a video. It's interesting that no one is talking about Intel Gaudi vs NVIDIA A100 or H100. Gaudi2 isn't cheap at $10.42/hour but it feels like a viable alternative to fighting someone for an A100 or H100. Here's my code if you want to try it out:

Ryan Carson

41,328 views • 2 years ago

Gave Gemini 3 Deep Think an image of a 3D spider web and asked for an interactive design tool. It generated a full design suite (procedural control, simulation, optimization) with STL export capability. I used it to engineer new metamaterials and a spider-web inspired bridge design, 3D printed it, then validated the structural integrity with a NVIDIA DGX Spark load test. A mind-blowing example of the future of material and architecture design - image in, fabrication-ready design out! Google Gemini

Gave Gemini 3 Deep Think an image of a 3D spider web and asked for an interactive design tool. It generated a full design suite (procedural control, simulation, optimization) with STL export capability. I used it to engineer new metamaterials and a spider-web inspired bridge design, 3D printed it, then validated the structural integrity with a NVIDIA DGX Spark load test. A mind-blowing example of the future of material and architecture design - image in, fabrication-ready design out! Google Gemini

Markus J. Buehler

98,719 views • 6 months ago

A couple of days ago at an Avenged Sevenfold show, the band invited me up on stage to play rhythm guitar on 'Beast and the Harlot' with them. An absolute dream come true — a high that I'll be riding on forever. I compiled a few short clips but full video ➡️

A couple of days ago at an Avenged Sevenfold show, the band invited me up on stage to play rhythm guitar on 'Beast and the Harlot' with them. An absolute dream come true — a high that I'll be riding on forever. I compiled a few short clips but full video ➡️

Adam (Speedy)

14,892 views • 1 year ago

NVIDIA’s new humanoid robot brain is officially on sale for $3,499. NVIDIA has announced the general availability of the Jetson AGX Thor developer kit and production modules. Powered by an NVIDIA Blackwell GPU and featuring 128GB of memory, it delivers up to 2,070 FP4 teraflops within a 130-watt power envelope. Early adopters include Agility, Boston Dynamics, and Figure.

NVIDIA’s new humanoid robot brain is officially on sale for $3,499. NVIDIA has announced the general availability of the Jetson AGX Thor developer kit and production modules. Powered by an NVIDIA Blackwell GPU and featuring 128GB of memory, it delivers up to 2,070 FP4 teraflops within a 130-watt power envelope. Early adopters include Agility, Boston Dynamics, and Figure.

The Humanoid Hub

369,047 views • 11 months ago

Until a few days ago, everyone on social media became an expert on why gold and silver are bad investments. The favourite argument? “Gold doesn’t produce cash flows.” Really? Since when? #Gold #StockMarket

Until a few days ago, everyone on social media became an expert on why gold and silver are bad investments. The favourite argument? “Gold doesn’t produce cash flows.” Really? Since when? #Gold #StockMarket

CA Neha Niharika

12,566 views • 1 month ago

Quick video on the stages of reverse engineering🙂

Quick video on the stages of reverse engineering🙂

LaurieWired

35,048 views • 2 years ago

A female software engineer, Shampa Pathak (Pandey), working with IT major Infosys in #MadhyaPradesh’s #Indore, was crushed to death by a speeding car driven by the young son of an electrical engineer in a posh residential society late on Wednesday night. Hours after the incident, the father-son duo, Kuldeep Chaudhary and his son Mohit Chaudhary (18), were arrested by police and booked for murder. 🔗

A female software engineer, Shampa Pathak (Pandey), working with IT major Infosys in #MadhyaPradesh’s #Indore, was crushed to death by a speeding car driven by the young son of an electrical engineer in a posh residential society late on Wednesday night. Hours after the incident, the father-son duo, Kuldeep Chaudhary and his son Mohit Chaudhary (18), were arrested by police and booked for murder. 🔗

The New Indian Express

120,201 views • 4 months ago

I'm sharing a #RARP after a #HOLEP. The patient I operated on two years ago was completely asymptomatic, however, he presented with an elevated PSA of 2.4 ng/ml, when he had previously managed below 1 ng/ml. I ordered an MRI and PET scan PSMA, followed the protocol, and performed the radical prostatectomy. The enucleation appears to have been perfect, and it's interesting to see the completely clear prostatic bed inside. Do you routinely follow up your Holep patients even if they are asymptomatic?

I'm sharing a #RARP after a #HOLEP. The patient I operated on two years ago was completely asymptomatic, however, he presented with an elevated PSA of 2.4 ng/ml, when he had previously managed below 1 ng/ml. I ordered an MRI and PET scan PSMA, followed the protocol, and performed the radical prostatectomy. The enucleation appears to have been perfect, and it's interesting to see the completely clear prostatic bed inside. Do you routinely follow up your Holep patients even if they are asymptomatic?

Bruce Rubí HoLEP 🇲🇽

13,032 views • 8 months ago