Want to know what all this microsoldering was for and which silicon bugs we had to exploit? Come to our (@stacksmashing, @cryptotx and me) wallet.fail talk at #35c3. fahrplan.events.ccc.de/congr…

Today @StackSmashing and I are announcing our most thorough course on glitching ever. We call it “Applied Glitching of Real-World Hardware Targets”. The course has been entirely designed to be taught live with many live demos against actual targets. advancedsecurity.training/tr…

Found myself opening Photoshop just to remove a background from an image. In fact I think statistically background removal is what I use Photoshop for the most. But guess what? There's an open source AI for that (~17 models). Adding to @comput3ai soon: github.com/comput3ai/c3-remb…

A bunch of conferences are moving to an online format. We wrote down our thoughts about how online trainings should be and what you should watch out for. advancedsecurity.training/po…

Updated my list of recommended lab equipment to include the Rigol MSO5074. My current recommendation for an entry to medium level oscilloscope. The refresh is a bit slow at times but everything else is awesome. toothless.co/equipment

We just improved our TRELLIS Gradio app and docker container. It now boots in about 40 seconds instead of 2 minutes on our GPU instances. github.com/comput3ai/c3-dock…

Happy to see people trying the coding models on @comput3ai. People don't realize how far open source models have come. Open source models get updated as well. The gap to frontier is definitely shrinking. We built some coding agents internally that were already pretty impressive.

Got claude code running on one of @comput3ai's B200's today. The B200s are running a Kimi K2 Q8 with 128K (the max) of context. With 4 "threads".

That moment when the list of bugs you found is way too long for one #35c3 talk.

Just updated our FPGA and Glitching classes. Get them both for the price of just one training. ast.fyi/bundles

Long time coming! We‘ll be offering not only my four day class, but classes people have been asking us for forever. I’ll be teaching an intro to FPGAs class using Lattice FPGAs and the open source tool chain in addition to the other excellent classes we have.

We're gonna play around with hosting our own Kimi K2 instances this Thursday on @comput3ai. Noon Eastern. Mark your calendars.

Boom! Here's where we got those models up and running (and how). We'll be giving away 10 free GPU hours to those who join today, so make sure you join! 👉piped.video/jrYe2DXozqQ

People always ask how big a model really is and what it takes to host it. Forget marketing slides—here’s the napkin math you can use in the field. At FP8, one parameter equals one byte. A 1 B model needs 1 GB of vRAM. That’s it. The GPUs that used to push pixels are now just expensive memory sticks: H100 80 GB H200 140 GB B200 180 GB Those numbers are per card and you almost always run eight in a box. So a 40 B model already pushes you past any desktop card. Add the KV cache and even an 80–96 GB RTX 6000 Pro or single H100 is gasping. Take Qwen3-235B-A35B. The “A35B” means only 35 B are active at once, a trick to survive narrow pipes. Still, the full 235 B must live somewhere. Round up to 235 GB plus KV. Two H200s (2×140 GB) or four H100s (4×80 GB) will hold it—barely. And the new 256 K context upgrade? That KV cache just doubled. (Nobody hosts this at scale except @comput3ai—more later.) Now drop the coding variant, Qwen3-Coder-480B. It's Qwen3 with additional training to be good at coding tasks. Same footprint math, but extra layers mean extra KV. Eight H200s will serve it, but you’ll be red-lining. DeepSeek-R1 clocks in at 671 B. That’s 671 GB naked. Eight H100s give 640 GB—game over. Eight H200s give 1.14 TB, so it fits. Then DeepSeek drops a new checkpoint with more layers and the KV cache inflates again. You start pricing two 8-GPU H200 nodes. Kimi K2 is 1 T—one trillion parameters. One terabyte of vRAM minimum. Moonshot’s own doc says “don’t try this on fewer than sixteen H200s,” i.e., two 8-way boxes. The B200 finally looks relaxed: 8×180 GB = 1.44 TB. That leaves enough headroom for the 128 K KV cache and still lets you breathe. Rule of thumb: if the model won’t fit, you cannot ship. So grab the B200s—we’ve got them warm and waiting at @comput3ai.

We have our FPGA Hacking and Intro Glitching classes coming up. We have European session in May and June and US in June and July. Is this the right course for you? Here's a sneak peak at Intro to Glitching. By day 3 you'll be able to do this to NXP LPCs, STM32s and NRF52s.

I’m incredibly humbled and excited to hear that our participants are already in the process of recovering the cryptocurrency they’ve had locked on their wallets for many years. You can do it too and if nothing else this was an epic hardware hacking class! advancedsecurity.training/tr…

Last week, we brute-forced a PIN, did timing analysis of strcmp, built a couple RiscV's in Verilog and glitched a password check. We'll be doing it all again June 26-29 US Eastern time. Check out our subscriptions to get 50% off.

Deepseek v3.1, B200s.

Today we’re announcing our new event format. Based on years of doing trainings at various events. Our goal is to make this by far the best format for trainers and participants. If you’re interested in training get in touch with me or @StackSmashing.

Coolest course we’ve done to date and the one that took the most advantage of our streaming platform. Recordings are now also available. Check it out.

By popular demand, we rescheduled our PUF and Anti-Tamper class for next month, which means it's not too late to register! Come and learn about PUFs and Tamper circuits from me and @cryptotx. advancedsecurity.training/tr…

Here's kimi! (streaming in ~50 mins @comput3ai)

Had a great time in Paris giving the @toothlessco Hardware Hacking Training to @Synacktiv got lots of positive feedback! And they even gave me a beret!

Tomorrow, we decided we'll be streaming the webinar to Twitter as well as other platforms. If you want the recording or to ask questions you can still register on our website webinar.advancedsecurity.tra… In the meantime, here's brute forcing a pin purely in logic form our FPGA class

Been using kimi-k2 (via @comput3ai) on my self-hosted open-webui. Absolutely glorious. Haven't opened up claude, chatgpt or grok.

People always ask me why I have so many bad things to say about PUFs. Well you can now find out in our upcoming class with @cryptotx advancedsecurity.training/tr…

Try us out, get 60 mins for free launch.comput3.ai/

I canceled chatgpt, cursor, ... do I cancel claude and just use @comput3ai now?

Friends don't let friends dual FT2232 - the FT4232 exists for a reason. Four channels, four devices, four things, four times the fun! Check out the full recording from OUR FREE WEBINAR on our favorite hacking devices from AliExpress ast.fyi/aliexpress #hardwarehacking

In case you wondered what my class is like as one of our Training Live Streams, here you go

It always surprises me that people don't realize that hugging face spaces have a git repo accessible linked directly to them. It contains the exact gradio code you need to run the space yourself. All you need is an Nvidia GPU.

So for this of you who asked about my next training in Europe, here it is. We’ll be next door to Offensivecon and we have group discounts on that hotel! See you in Berlin.

If LLMs have so many issues with whitespace, how did python become the language of choice?

To speed up running and downloading gigantic models on gigantic gpus, here's a Dockerfile the llama.cpp as a base, uses huggingface_hub's multi threaded downloader to download the models and then run them. Also included a traefik docker-compose (pro tip) github.com/comput3ai/c3-llam…

We had one cancelation for our sold out training in Berlin. Let me know if you want the slot. toothless.co/events/berlin-s…

BREAKING NEWS: you can do useful things with a Raspberry Pi! 🎉 New class announcement 🎉 Secure Edge Development with MQTT and the Raspberry Pi. 100% LIVE on our industry leading streaming platform.

Went on the @stephanlivera Podcast to talk about hardware security in crypto and specifically hardware wallets. You can check that out here: stephanlivera.com/episode/56… #hardware #crypto #hardwaresecurity #hardwarewallets #bitcoin

Today we're launching our Hardware Hacking Bootcamp on the Advanced Security Private Discord! We'll be covering recommended equipment, tips, tricks, and much more. Starts Monday, there's a 20% discount until then. Don't miss out! 🛠️🔒 #HardwareHackingBootcamp

With the fallout from the #ledger messaging fiasco, many wallets have asked for help in enhancing their security and transparency. That's what we do - we can help! In case you're unfamaliar with our work, here's a case study of the Trezor One. #cryptocurrency #trezor #blockchain

TIL: most USB C to USB C are not USB 3 and the most reliable way to get guaranteed USB 3 is to grab a thunderbolt 3 cable. I don’t know why I didn’t think about that before.

Ported unirig to a self-contained dockerized gradio app. Enjoy metaverse fam! github.com/comput3ai/c3-unir…

Crowd pleaser: glitching the LPC cortex-M series microcontrollers. Watch until the end and see us dump flash.

Life forced me to use python3 today 😖

🔧 Learn about hardware security testing with FPGAs! 🛠️ On November 12th, join our hands-on training on FPGAs and get our glitching and fault injection course as part of the bundle 🔬 Couple seats left, don't miss out! 📚 ast.fyi/bundles

If you haven't written HDL code since university - don't worry! We'll start out slow. If you're there live, we'll have a CTF going with notifications to the stream for when you solve assignments. Don't miss this opportunity. May 8-11 (EU), June 26-29 (US). #hardwarehacking

Looking forward to playing around and benchmarking these on the next live stream.

This year we’re offering our European spring training the week of Offensivecon. It will be just next door. Come and learn what it takes to glitch microcontrollers like the stm32 and what impact this may have on your products.

Devin here. Could you not tell my boss?

Not surprised.

Classic hack: hardware timing analysis of a password strcmp using FPGAs. BTW, we added another set of dates. Use code TWITTER at checkout and check out our subscriptions!

Getting ready for tomorrows FREE WEBINAR - Ali Express Stocking Stuffers. It'll be pretty, pretty, pretty good. Check it out here: webinar.advancedsecurity.tra…

If you're curious what that looks like.

My most epic road trip ever starts tomorrow.

From today's stream, we did it all using @comput3ai. Thanks to @Roolznft for the assets. (1/5)

For all those that have been asking me for an EU session: we’ll have a session of Hardware Hacking with FPGAs for clients in CET and neighboring timezones on March 28. Completely virtual with remote hardware for you to use - no hardware for you to purchase advancedsecurity.training/tr…

If someone has a four day class they’d like to teach in Berlin and you’re based in Europe: DM me.

Like i said, summer is when all the coding models launch. huggingface.co/Qwen/Qwen3-Co…

We (@comput3ai) let you run datacenter GPUs for $1/hr. You get one hour for free when you sign up: launch.comput3.ai

Our FREE WEBINAR starts in 3.5 hours. You can still register - it'll be fun! And you'll get some nice hardware out of it. #hardwarehacking #hardwaresecurity webinar.advancedsecurity.tra…

Best Stream ever. As a security expert.

Excited to be announcing our FIRST FREE WEBINAR of 2023. It'll be on JTAG vs. SWD. The differences as well as some of our favorite JTAG programmers. webinar.advancedsecurity.tra… #webinar #jtag #swd #hardwarehacking #embeddedsecurity

Our 4-day class at our favorite security conference. Hope to see you there!

🔑 Brute forcing a 16 character password would take 10^19 years ⏲ You can try a side-channel attack on the strcmp ⚡ Or just glitch right when the password is being checked. Get our FPGA and Glitching courses together for the price of one training.

People ask how we keep things entertaining during our streams ✅ Funny usernames ✅ CTFd for every training ✅ STREAM NOTIFICATIONS Get our FPGA and Glitching courses live or recorded together for the price of one ast.fyi/bundles

Fun fact: OpenOCD error messages call it the JayLink. Not joking.

I'll just leave this here: timing analysis of a password check. Still just over a week, to register for EU/CET session of Hardware Hacking with FPGAs May 8-11. We'll also have a US Eastern session June 26-29.

Turns out when you install AMD Radeon drivers on windows it steals the ctrl + shift + c hotkey for turning your webcam on/off. WTF AMD. Here's to losing an hour trying to copy some text from the terminal.

I think adding 15 min deployments of new models directly from huggingface paid off in spades.

Still some room for tomorrows FREE WEBINAR on the Raspberry Pi Pico and where you can use it in your day to day flows. We'll keep things to 30 mins, you'll be able to attend during lunch! See you tomorrow at 12:00 Eastern #hardwaresecurity #hardwarehacking webinar.advancedsecurity.tra…

Decided to package all our most relevant trainings for hardware wallet and cold storage engineering teams. We've brought back our one-of-a-kind wallet class for a limited time only. And don't miss the last episode of the @stephanlivera podcast where we discuss all of this.

I found myself #glitching 3 or 4 times this week and for that I went ahead and used clips from our classes. I decided I'd share it here as well. If you're interested we've bundled all of this with a nice discount over here (happy 4th of July everyone!): ast.fyi/bundles

Every time claude runs out of context I feel like I'm dealing with a body double.

Had some fun glitching ⚡ some new targets recently including the NRF52. Grab our bundle and get our FPGA and Glitching classes together and join our upcoming FPGA class live on November 12th. ⌛Limited time offer: ast.fyi/bundles

🔥This week at @comput3ai we’re dropping our epic new media instances 🤯that include voice cloning, animating speakers, and karaoke-style subtitles all running on private GPUs. Wanna try it out? DM @comput3ai to level up your content game 🚀✨ Want to see an example? 👇

I tried Ubuntu 25.04 for like a month and a half because it builds docker containers (especially cuda) faster than Docker Desktop on wsl2. But it still isn't the year of Linux on the desktop is all I'll say.

Using litellm-proxy locally as a proxy, but we'll roll this out for everyone to use on @comput3ai. You'll just have to override your anthropic base url.

🚀 Our Hardware Hacking Bootcamp launches today! 💥 Join us to explore cool equipment, hack with lab gear, and dive into FPGAs! 😎 Don't miss out – REGISTER NOW while we still have a couple of slots left! 🔒💻 JOIN NOW!👉 ast.fyi/bootcamp

Today's the last day to join our Bootcamp with access to May and June's content. Every month we do something completely new! Here's a quick summary of what it's all about. ast.fyi/bootcamp

People think a Raspberry Pi Pico can do everything an FPGA can do. Nope! We'll be discussing when you absolutely need an FPGA for hardware hacking this month as part of our Hardware Hacking Bootcamp ast.fyi/bootcamp

So @cryptotx and I recorded a nice Podcast on why you should train with @advancedsecio. You should check it out. piped.video/iWKLO0MV7zM

We (@comput3ai) are splitting up servers and letting users get full access to an entire GPU to run whatever AI workloads they want. I think the biggest problem with local AI is the hardware isnt powerful enough and an entire 8xGPU server is overkill. Separate isolated containers

Today at 12:00PM Eastern 🕛 don't miss our 📅 monthly bootcamp webinar 📡! Join us for your last opportunity to access all the content released in May as well as all the upcoming content in June. See you there! 👉 ast.fyi/bootcamp

Open CUDA on github, go to pull requests, go to issues. Look for Blackwell support. Still glad it's here, but so much stuff broke.

You'll get 1hr. Can confirm.

From building custom container orchestration, to learning how to run GPUs in every single AWS region, the end result is launching AI GPU workloads in under 3 seconds. What a ride!

Digikey should have discounted shipping on that second order where you add everything you forgot on the first one.

Man flying easyJet is like pulling teeth

What $15 device can replace $1000 worth of debugging hardware? Find out during our Hardware Hacking Bootcamp. ast.fyi/bootcamp

Fun times playing with all the stuff we built at @comput3ai

I'm going to make a model that just removes whitespace from other models. It will be the most used AI.

The things you have to buy to test your state of the art video streaming platform ...

Why is the best way to access AI claude on the command line. Why do all the mobile apps suck so much? Why cant these gigantic companies track the state of my query while running in the background.

I wish blackwell didn't break so many pytorch containers.

I'm getting flashbacks of setting up lasers at the optics department. Stay safe!

Today is the last day to get early bird rates on our training which will run next door to Offensivecon. We just have a couple of seats left! toothless.co/events/berlin-s…

Upcoming trainings in Canada, Germany and the US. Sign up now to get early-bird rates.

Last day to register for tomorrow's FPGAs and Verilog in One Day. People who take our trainings, get one month of the bootcamp for free It'll also be available as a recording for those who can't join us live. See you tomorrow! advancedsecurity.training/tr…

7. Same idea, but with the STM32 and JTAG. We're using OpenOCD to check whether or not JTAG is enabled. When we see that JTAG is enabled, we're in. This one is actually a real-world example the Trezor One hardware wallet.

I have to say vibe coding kubernetes may be the most underappreciated thing. Curl, ssl error, figure out what's wrong with ingress, check the tls cert, check the secrets.