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.

We built Talos - a full CNN inference engine running directly on silicon. Every multiply, buffer, and data path lives as real digital logic on the FPGA. This is what deep learning looks like when the model becomes hardware👇

luthira

2,712 subscribers

92,371 views • 5 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

What happens when physical props meet real-time VFX? This deep dive looks at Muzzle Report, custom hardware that connects directly to Unreal Engine—giving actors something real to perform with and creates instant feedback on set. Watch the full deep dive:

What happens when physical props meet real-time VFX? This deep dive looks at Muzzle Report, custom hardware that connects directly to Unreal Engine—giving actors something real to perform with and creates instant feedback on set. Watch the full deep dive:

Unreal Engine

31,855 views • 6 months ago

First light! 🔥 ao486 boots DOS on the #TangConsole 138K. HDD, VGA, BIOS, and VGA BIOS are all up and running — watching it hit the DOS prompt on real hardware is surreal. #FPGA #RetroComputing 🎥👇

First light! 🔥 ao486 boots DOS on the #TangConsole 138K. HDD, VGA, BIOS, and VGA BIOS are all up and running — watching it hit the DOS prompt on real hardware is surreal. #FPGA #RetroComputing 🎥👇

nand2mario

15,437 views • 11 months ago

.Reiner Pope's new blackboard lecture goes all the way down: how AI training and inference are built up from logic gates on silicon. He walks me through a 4-bit multiply-accumulate by hand, and shows how that primitive is the foundation for the matrix multiplies in training runs.

.Reiner Pope's new blackboard lecture goes all the way down: how AI training and inference are built up from logic gates on silicon. He walks me through a 4-bit multiply-accumulate by hand, and shows how that primitive is the foundation for the matrix multiplies in training runs.

Dwarkesh Patel

38,959 views • 2 months ago

Georgia Tech built a free tool that animates every data structure and algorithm in real time this is what DSA study looks like when it actually makes sense

Georgia Tech built a free tool that animates every data structure and algorithm in real time this is what DSA study looks like when it actually makes sense

Vaishnavi

49,150 views • 4 months ago

Every man lives a similar life path. The difference in outcome is determined by what we do on the path.

Every man lives a similar life path. The difference in outcome is determined by what we do on the path.

Andrew Tate

262,163 views • 1 year ago

56,000+ tokens/sec at just 80 MHz. 🤯 I burned a full Transformer with KV cache into a custom chip. Designed gate by gate as a 100% digital integrated circuit. Prototyped on a FPGA. (No GPU. No CPU) Just pure digital silicon running Andrej Karpathy microGPT, spelling out names on a tiny LCD. This is GateGPT 👇

56,000+ tokens/sec at just 80 MHz. 🤯 I burned a full Transformer with KV cache into a custom chip. Designed gate by gate as a 100% digital integrated circuit. Prototyped on a FPGA. (No GPU. No CPU) Just pure digital silicon running Andrej Karpathy microGPT, spelling out names on a tiny LCD. This is GateGPT 👇

Fabio Guzman

756,578 views • 2 months ago

How many of the big ideas of the past 15 years of AI are downstream of hardware constraints? The big hardware story over that period is that logic has become way cheaper than data transfer. Stacking huge numbers of matrix multiplies was perfect for this hardware regime, because matrix multiplication is logic-intensive but requires less data transfer. And so we got matmul-heavy deep learning. It's interesting to think about what AI would look like in a world where these costs didn't diverge so much.

How many of the big ideas of the past 15 years of AI are downstream of hardware constraints? The big hardware story over that period is that logic has become way cheaper than data transfer. Stacking huge numbers of matrix multiplies was perfect for this hardware regime, because matrix multiplication is logic-intensive but requires less data transfer. And so we got matmul-heavy deep learning. It's interesting to think about what AI would look like in a world where these costs didn't diverge so much.

Dwarkesh Patel

61,147 views • 1 month ago

We deployed a fully private AI agent on NuNet in under 5 minutes 🚀 OpenClaw🦞 running Qwen through ollama , one of the hottest open source model families right now, entirely on decentralized compute. No cloud. No API keys. No data leaving the machine. This is what private AI looks like when you actually build it instead of just talking about it. Your model. Your hardware. Your rules. Full walkthrough showing exactly how it works: What should we deploy next?

We deployed a fully private AI agent on NuNet in under 5 minutes 🚀 OpenClaw🦞 running Qwen through ollama , one of the hottest open source model families right now, entirely on decentralized compute. No cloud. No API keys. No data leaving the machine. This is what private AI looks like when you actually build it instead of just talking about it. Your model. Your hardware. Your rules. Full walkthrough showing exactly how it works: What should we deploy next?

NuNet 🌐

87,604 views • 4 months ago

If the engine is strong enough, you should be able to build real products on top of it. That's the whole point of LTX-2.3. Introducing LTX Desktop. A fully local, open-source video editor running directly on the LTX engine, optimized for NVIDIA GPUs and compatible hardware.

If the engine is strong enough, you should be able to build real products on top of it. That's the whole point of LTX-2.3. Introducing LTX Desktop. A fully local, open-source video editor running directly on the LTX engine, optimized for NVIDIA GPUs and compatible hardware.

LTX

954,160 views • 5 months ago

When the PS Vita was first revealed as the NGP, Kojima showed up on stage to present a demo of Metal Gear Solid 4 running on the handheld “This game used the model data and environments from PS3, and it was exported directly to NGP. On NGP, we can enjoy the same quality as PS3"

When the PS Vita was first revealed as the NGP, Kojima showed up on stage to present a demo of Metal Gear Solid 4 running on the handheld “This game used the model data and environments from PS3, and it was exported directly to NGP. On NGP, we can enjoy the same quality as PS3"

Radec

98,167 views • 5 months ago

🕹️ Is this the smallest language model in the world? I just managed to squeeze JAM, real artificial intelligence into 30 kilobytes, running on a 1979 Atari 800. Just A Model. Fully generative, deterministic language model, powered by a neural network and built to run on 8‑bit hardware. Atari Forever.

🕹️ Is this the smallest language model in the world? I just managed to squeeze JAM, real artificial intelligence into 30 kilobytes, running on a 1979 Atari 800. Just A Model. Fully generative, deterministic language model, powered by a neural network and built to run on 8‑bit hardware. Atari Forever.

Marek Spanel

113,732 views • 4 months ago

Super Smash Bros. 64, running natively on iOS and iPadOS for the first time. Not an emulator. BrawlerPad is built on BattleShip, which already supplied the SSB64 source-port core, Apple Metal path, and SDL controller/audio stack. BattleShip ran on Apple Silicon macOS and Android. There had never been a public iOS or iPadOS port. Now there is. The game logic is compiled into a native ARM64 app and runs through BattleShip/libultraship’s Metal/Fast3D path. No N64 hardware is simulated, and no ROM is interpreted at runtime. Smash deserves a controller, so BrawlerPad supports physical controllers through the same input path as the game. Touch works too. I used Codex 5.6 Sol on the iOS integration. Full build guide and source below.

Super Smash Bros. 64, running natively on iOS and iPadOS for the first time. Not an emulator. BrawlerPad is built on BattleShip, which already supplied the SSB64 source-port core, Apple Metal path, and SDL controller/audio stack. BattleShip ran on Apple Silicon macOS and Android. There had never been a public iOS or iPadOS port. Now there is. The game logic is compiled into a native ARM64 app and runs through BattleShip/libultraship’s Metal/Fast3D path. No N64 hardware is simulated, and no ROM is interpreted at runtime. Smash deserves a controller, so BrawlerPad supports physical controllers through the same input path as the game. Touch works too. I used Codex 5.6 Sol on the iOS integration. Full build guide and source below.

Kahris

14,612 views • 9 days ago

The Chat Room isn’t just a concept. It runs real AIR tasks, in real time, and we’re showing it. Watch an Agent process public data, structure it, and return actionable answers in seconds. This is what it looks like when data becomes usable. This is the Chat Room in action. 🎬 🔗

The Chat Room isn’t just a concept. It runs real AIR tasks, in real time, and we’re showing it. Watch an Agent process public data, structure it, and return actionable answers in seconds. This is what it looks like when data becomes usable. This is the Chat Room in action. 🎬 🔗

Teneo Protocol

68,489 views • 9 months ago

Dreamcast running on FPGA isn't about what the DE10-Nano can do today, it's about what FPGA gaming could become tomorrow. The Dreamcast's SH-4 CPU, PowerVR graphics, and NAOMI heritage push far beyond the limits of current MiSTer hardware, but the work being done now is laying the foundations for future FPGA platforms. This isn't a playable core story. It's a glimpse into the future of hardware preservation. Dreamcast. NAOMI. Beyond. Thanks ElectronAsh, Stefanos Kornilios Mitsis Poiitidis, Corn and Co #Dreamcast #SEGA #FPGA #MiSTerFPGA #RetroGaming #GamePreservation #NAOMI #FPGAGaming

Dreamcast running on FPGA isn't about what the DE10-Nano can do today, it's about what FPGA gaming could become tomorrow. The Dreamcast's SH-4 CPU, PowerVR graphics, and NAOMI heritage push far beyond the limits of current MiSTer hardware, but the work being done now is laying the foundations for future FPGA platforms. This isn't a playable core story. It's a glimpse into the future of hardware preservation. Dreamcast. NAOMI. Beyond. Thanks ElectronAsh, Stefanos Kornilios Mitsis Poiitidis, Corn and Co #Dreamcast #SEGA #FPGA #MiSTerFPGA #RetroGaming #GamePreservation #NAOMI #FPGAGaming

Pixel Cherry Ninja

13,156 views • 1 month ago

🚨 This is Atlas Brief — my Iran War Cost Tracker. $1,213,532,473 spent so far. Live. Updated in real time from public data. $2,546 per second. $220 million per day. Built on USASpending[.]gov, Congress[.]gov, and DoD contract data. Every number is auditable. Every source is cited. This is what institutional intelligence looks like — Data. Atlas Brief tracks the war they don’t want you to put a price tag on.

🚨 This is Atlas Brief — my Iran War Cost Tracker. $1,213,532,473 spent so far. Live. Updated in real time from public data. $2,546 per second. $220 million per day. Built on USASpending[.]gov, Congress[.]gov, and DoD contract data. Every number is auditable. Every source is cited. This is what institutional intelligence looks like — Data. Atlas Brief tracks the war they don’t want you to put a price tag on.

Brian Allen

713,993 views • 5 months ago

The Machine That Learns The Law Behind The Data A very very interesting US Patent US10963540B2 - Physics Informed Learning Machine describes a learning system that does not begin with data alone. It begins with a physical model, usually written as a differential equation (or PDE) dx/dt = f(x,t) A normal Machine Learning model sees scattered data and tries to fit it. A physics-informed learning machine starts with a law. Then it treats the data as evidence that updates what the model believes about the physical system. For this application, I use the patent idea on NASA C-MAPSS Turbofan engine data. The machine watches multivariate telemetry from a degrading engine and infers a hidden health state that is not measured directly. From that posterior belief, it estimates the engine’s remaining useful life. In the main 3D scene, the engine lifetime is turned into a tunnel. The spiral ribbons are real sensor channels evolving over cycle-time. The glowing core is the inferred health state. The surrounding cloud is uncertainty. The orange wall ahead is the predicted failure horizon. So the big picture is: sensor evidence comes in, posterior belief tightens, and the machine moves from uncertainty toward a concrete failure prediction. The inset posteriors make that explicit. The health posterior shows where the model believes the hidden engine condition sits at the current moment, and how sharply it believes it. The RUL posterior shows the same idea for remaining life... early on it is broad, later it shifts left and narrows as the machine becomes more certain about how close failure is. This idea is not limited to engines. The same idea can apply to data centers, CPUs, GPUs, cooling systems, power grids, robotics, batteries, and any machine that produces telemetry while obeying physical constraints. In an age where machine learning runs on massive hardware infrastructure, this kind of model matters: it can turn noisy sensor streams into early warnings before expensive systems fail.

The Machine That Learns The Law Behind The Data A very very interesting US Patent US10963540B2 - Physics Informed Learning Machine describes a learning system that does not begin with data alone. It begins with a physical model, usually written as a differential equation (or PDE) dx/dt = f(x,t) A normal Machine Learning model sees scattered data and tries to fit it. A physics-informed learning machine starts with a law. Then it treats the data as evidence that updates what the model believes about the physical system. For this application, I use the patent idea on NASA C-MAPSS Turbofan engine data. The machine watches multivariate telemetry from a degrading engine and infers a hidden health state that is not measured directly. From that posterior belief, it estimates the engine’s remaining useful life. In the main 3D scene, the engine lifetime is turned into a tunnel. The spiral ribbons are real sensor channels evolving over cycle-time. The glowing core is the inferred health state. The surrounding cloud is uncertainty. The orange wall ahead is the predicted failure horizon. So the big picture is: sensor evidence comes in, posterior belief tightens, and the machine moves from uncertainty toward a concrete failure prediction. The inset posteriors make that explicit. The health posterior shows where the model believes the hidden engine condition sits at the current moment, and how sharply it believes it. The RUL posterior shows the same idea for remaining life... early on it is broad, later it shifts left and narrows as the machine becomes more certain about how close failure is. This idea is not limited to engines. The same idea can apply to data centers, CPUs, GPUs, cooling systems, power grids, robotics, batteries, and any machine that produces telemetry while obeying physical constraints. In an age where machine learning runs on massive hardware infrastructure, this kind of model matters: it can turn noisy sensor streams into early warnings before expensive systems fail.

Mathelirium

17,843 views • 3 months ago

After 8+ years on the Tesla Autopilot team and 3 years at Intel, I started Apex Compute to design a new architecture for efficient AI inference. For the past 9 months, we’ve been building our custom inference accelerator. Today we’re releasing Unified Engine v1. Last June we raised our seed round with Maxitech , DeepFin Research, Soma Capital and an incredible group of angel investors. In less than 9 months, we completed our RTL architecture and brought our first pre-silicon prototype to life on FPGA. Our architecture combines systolic array and vector processing in a single compute engine with multiple architectural optimizations, achieving very high FLOPs utilization. A single engine is super lean and it uses less than 90K LUTs and 1 MB Block RAM. It may also be one of the smallest logic-footprint compute engines developed so far. Our Unified Engine v1 supports: -matrix-matrix multiplication (~95% FLOPs utilization) -softmax (~90% FLOPs utilization) -broadcast and element-wise operations -RMSNorm / LayerNorm -block quantization/dequantization (fp4, int4) -multi-engine synchronization and many other operations. We even implemented memory-efficient attention similar to FlashAttention, reaching ~90% FLOP utilization. Full benchmarks and the software stack are available on our GitHub: We have basic compiler written in Python and it supports PyTorch tensors directly to easily test and transfer tensors between the accelerator and host using bf16, fp4 and int4 formats. Our FPGA prototype can already run LLM inference and outperform NVIDIA Jetson Orin Nano, even on a mid-tier FPGA setup (6.4x lower memory bandwidth, 18% slower clock speed at 4.5 Watts). Check the side-by-side comparison video below. Our GitHub includes low-level operator implementations, examples for tiled matrix multiplication, operation chaining, tensor parallelism, attention kernel and a full Gemma 3 1B model implementation. Many more models(Vision Transformers and VLA) are coming soon. Our accelerator IP is AXI-ready for deployment on any AMD(Xilinx) FPGA platform today. Even better, our two-engine prototype runs on an entry-level AMD(Xilinx) FPGA as a PCIe accelerator card. You can purchase it here for $50 to experiment our pre-silicon prototype on your desktop PC or Raspberry Pi 5. We will be releasing hardware bitstream updates as the architecture gets new features. More to come soon! We are expanding our team and looking for compiler engineers and floating-point hardware design engineers. If you're interested, please send me a DM.

After 8+ years on the Tesla Autopilot team and 3 years at Intel, I started Apex Compute to design a new architecture for efficient AI inference. For the past 9 months, we’ve been building our custom inference accelerator. Today we’re releasing Unified Engine v1. Last June we raised our seed round with Maxitech , DeepFin Research, Soma Capital and an incredible group of angel investors. In less than 9 months, we completed our RTL architecture and brought our first pre-silicon prototype to life on FPGA. Our architecture combines systolic array and vector processing in a single compute engine with multiple architectural optimizations, achieving very high FLOPs utilization. A single engine is super lean and it uses less than 90K LUTs and 1 MB Block RAM. It may also be one of the smallest logic-footprint compute engines developed so far. Our Unified Engine v1 supports: -matrix-matrix multiplication (~95% FLOPs utilization) -softmax (~90% FLOPs utilization) -broadcast and element-wise operations -RMSNorm / LayerNorm -block quantization/dequantization (fp4, int4) -multi-engine synchronization and many other operations. We even implemented memory-efficient attention similar to FlashAttention, reaching ~90% FLOP utilization. Full benchmarks and the software stack are available on our GitHub: We have basic compiler written in Python and it supports PyTorch tensors directly to easily test and transfer tensors between the accelerator and host using bf16, fp4 and int4 formats. Our FPGA prototype can already run LLM inference and outperform NVIDIA Jetson Orin Nano, even on a mid-tier FPGA setup (6.4x lower memory bandwidth, 18% slower clock speed at 4.5 Watts). Check the side-by-side comparison video below. Our GitHub includes low-level operator implementations, examples for tiled matrix multiplication, operation chaining, tensor parallelism, attention kernel and a full Gemma 3 1B model implementation. Many more models(Vision Transformers and VLA) are coming soon. Our accelerator IP is AXI-ready for deployment on any AMD(Xilinx) FPGA platform today. Even better, our two-engine prototype runs on an entry-level AMD(Xilinx) FPGA as a PCIe accelerator card. You can purchase it here for $50 to experiment our pre-silicon prototype on your desktop PC or Raspberry Pi 5. We will be releasing hardware bitstream updates as the architecture gets new features. More to come soon! We are expanding our team and looking for compiler engineers and floating-point hardware design engineers. If you're interested, please send me a DM.

Hasan

37,648 views • 5 months ago

Congrats to the Kimi.ai team! This is awesome. Great to see this level of research coming from open-source frontier model labs. I liked the paper so much I built a Rust implementation of it ;) Full AttnRes + Block AttnRes with two-phase inference, built using Burn (tensor library and Deep Learning Framework, in Rust, by Tracel AI). Runs on CPU, CUDA, Metal, wgpu. Includes an interactive TUI that trains a model live and visualizes depth attention evolving from uniform to selective in real time. Repo link and more on what is implemented in the comments.

Congrats to the Kimi.ai team! This is awesome. Great to see this level of research coming from open-source frontier model labs. I liked the paper so much I built a Rust implementation of it ;) Full AttnRes + Block AttnRes with two-phase inference, built using Burn (tensor library and Deep Learning Framework, in Rust, by Tracel AI). Runs on CPU, CUDA, Metal, wgpu. Includes an interactive TUI that trains a model live and visualizes depth attention evolving from uniform to selective in real time. Repo link and more on what is implemented in the comments.

abdel

95,651 views • 5 months ago

Certain people love cherry picking and showing off an badly bugged model in our game before the art refresh. They wont show you what she looks like now. This is a rigged model running in engine.

Certain people love cherry picking and showing off an badly bugged model in our game before the art refresh. They wont show you what she looks like now. This is a rigged model running in engine.

Grummz

158,608 views • 1 year ago

I loved learning about the ways Ryan Paganetti used data to give his NFL teams a real edge, what it was like being on the headset in Super Bowl XXXIX, and this story on what it was like working for Pete Carroll 👇

I loved learning about the ways Ryan Paganetti used data to give his NFL teams a real edge, what it was like being on the headset in Super Bowl XXXIX, and this story on what it was like working for Pete Carroll 👇

Scott Barrett

29,043 views • 1 month ago