Tim Reha

Full-stack Creative Technologist product marketing pro #realtime #events #generativeai #digitalmarketing #socialmedia #video #virtualproduction

[steve @gpusteve]

kimi k3 at 3.8x higher throughput and 71% lower cost @AMD mi355x vs b200 btw

[Tim Reha @timreha]

@ana_piligrim Cool. Keep sharpening the creativity and tools! Brake boundaries

[Tim Reha @timreha]

@gpusteve @AMD Super cool

[Chirag Asarpota @ChiragAsarpota]

Bro, this is actually a huge AMD moment.

Kimi K3 is so massive that it needs 16x B200s across two NVIDIA servers, but it fits inside one 8x MI355X AMD server because AMD gives you much more HBM memory per GPU.

That single AMD node hit:

- 952 tok/s total throughput
- 118 tok/s for a single user
- Nearly 4x the throughput per node of the 16× B200 setup
- Better performance per dollar than both B200 and B300

But the craziest part is that ROCm mostly worked out of the box.

@wafer_ai only had to make a few relatively small fixes. No months of kernel engineering. No custom kernels at all.

Even the slow time-to-first-token problem came down mostly to one attention kernel not loading because Kimi had 12 heads instead of a supported shape. They simply padded 12 to 16, used AMD’s fast existing kernel, and cut cold-prefill time by roughly 2–3x

AMD’s bet on packing more HBM into each server is going to become extremely important as frontier open models keep getting larger.

If AMD keeps improving ROCm and day-one model support, data-center operators will have to seriously consider these GPUs.

The CUDA moat isn’t dead yet, but this definitely puts a big dent in it.

[Pablo Vela @pablovelagomez1]

So much of the complexity for multimodal/robotics data is just getting the right format that's both flexible and performant. I dig into it here in the context of 3d reconstruction. I'm hoping to build an opensource arkit like pipeline and explain the first bit for the data pipeline here!

Heres a video of what the final data looks like in a registered catalog

[Léo @LeoKharon]

New framework: Kick down your robot, it will get back up every time 🥋

Chinese startup RoboParty is a Beijing startup founded April 2025 by Huang Yi, originally shipping ROBOTO ORIGIN, the world's first full-stack open-source bipedal humanoid.
They released UFO: Unsupervised Reinforcement Learning Framework for Humanoid Control.

DEFINITIONS -> what differs is where the learning signal comes from:
- SUPERVISED: humans supply the right answers (labels), the model imitates them.
- UNSUPERVISED: no answer key, the model finds structure in raw data on its own.
- REINFORCEMENT LEARNING: no answer key either, the model tries things and a reward scores each attempt.
→ UNSUPERVISED RL: trial and error where the agent invents its own rewards, instead of engineers hand-writing one per task.

REPRESENTATION LEARNING: compress raw states into a useful internal map.
TEMPORAL DISTANCE: distance on that map is "how many steps from A to B."
CONTRASTIVE: trained by pulling together what's close in time, pushing apart what isn't.
-> CONTRASTIVE TEMPORAL-DISTANCE REPRESENTATION LEARNING: the model builds an internal map of body states where distance means how many steps it takes to get from one to another. It is trained by contrast: states that occur close together in a movement get pulled together in the map, randomly paired states get pushed apart.

UFO is an open-source training framework that teaches humanoid robots skills, like getting up, walking, goal-reaching, teleoperation, without reference motions -> no motion-capture or human-video demonstrations to imitate.

Its core is TeCH, a contrastive temporal-distance representation-learning algorithm: the robot explores, builds pseudo-goals by temporal rolling, and learns goal-conditioned policies from a single unified progress reward.

One framework trains five different robots (Unitree G1/H1, RoboParty RP0/RP1, AgiBot X2) with automatic config conversion in ~2–3 hours per robot!

The real novelty here "no demonstrations at all".
No data-collection arms race,the dominant humanoid-locomotion recipe is tracking: imitate mocap/retargeted-human reference trajectories.
The robot self-generates goals from its own exploration and learns from a progress reward, needing zero reference motion data.

Everybody else is fighting over data acquisition, while this team just teleports out of the race entirely (inb4 "competition is for losers 💀 ).

This strategy reminds me of the DeepSeek playbook applied to robots: open-source the whole stack to become the global default and commoditize everyone else.
RoboParty is giving away hardware and now control software (UFO) to be the Android of humanoids.
Yet another reason for the US to ban Chinese open models perhaps 🥶 ?

What I also really like about this approach is the cross-embodiment infrastructure, one framework trains Unitree G1/H1, RoboParty RP0/RP1, and AgiBot X2 with automatic configuration conversion.
Just like Physical Intelligence, RoboParty seems to place itself as a neutral hardware agnostic middle man.

Also woth mentioning: their ability ot perform stable skill injection, e.g. adding a cartwheel without forgetting how to walk.
A common failure of RL humanoid policies is that teaching a new agile skill destabilizes the existing ones (catastrophic forgetting).
UFO claims you can inject rare motions (cartwheel) without collapsing learned behavior.
If it holds, that's a significant incremental/continual skill-learning!

But again, I have to underline it: no arXiv, no external validation, no success-rate numbers.
-> robotics badely needs an independent unbiased evaluator imho.

Still, look at that cool demo: robot is getting kicked and pushed around (serious disturbance) during teleoperation (controlled the person at the back wearing the VR headset), and still managed to always get back up.
This is some serious demonstration of stability and robustness!

[Tim Reha @timreha]

@kyzoroXX @AcdNrg Bingo

[KyzoroX @kyzoroXX]

https://t.co/JGVBQ8PZzO

[ViralRush ⚡ @ViralRushX]

This mobile drone show system can automatically charge, deploy, and launch drones wherever it parks.

[wafer @wafer_ai]

🚨 BREAKING:

these engineers figured out how to serve Kimi K3 on @AMD MI355X at 952 tok/s/node and 118 tok/s single stream!

this crushes B200 by 3.8x in aggregate throughput/node and 1.3x in single stream decode + beats B300 on performance per dollar (48 vs 33 tok/s/$)

See how in the thread.

[Massimo @Rainmaker1973]

Julian Voss-Andreae is a quantum physicist-turned-sculptor.

His work is heavily influenced by his background in science and his blending figurative sculptures can vanish in front of our eyes.

[Object Zero @Object_Zero_]

Weapons of Mass Production

Today there are a handful of single assets that gate huge sectors of the economy. The criteria should be something like >50% of the global supply passes through this one cool thing.

• Falcon 9
Commercial access to space is 90% via the Falcon 9 launch system.

• Ulsan shipyard
Large high quality FPSOs and LNG tankers are almost exclusively manufactured here.

• Muroran forge press
Giant forged ingots >200 tons are mostly made in Japan and by this giant forge press.

• Zeiss vacuum grinder
The mirrors that allow ASML to do frontier EUV lithography are made by 1 machine.

There are a few others but you get the idea, what else should be on the list?

[explos1ve @explosss1ve]

HOUDINI + MCP COULD TURN ONE T-SHIRT INTO A $15,000 FASHION CAMPAIGN.

The shirt isn't modeled.

It grows from thousands of loose fibers, broken threads, and floating fragments that slowly weave themselves into fabric.

A fashion brand could use this as the opening shot for a collection, then turn the same procedural setup into product films, social ads, billboards, and an entire digital campaign.

That’s where the money is.

Not in selling one 3D animation.

In building a system that can generate a hundred variations without rebuilding the whole scene every time.

Houdini handles the physics.

MCP makes the workflow faster.

The final video looks like experimental art.

The pipeline could be a business.

[Tim Reha @timreha]

@tengizlab Are there open source models for this?

[Tim Reha @timreha]

@Alacritic_Super Flex payload

[Praveen Kumar Verma @Alacritic_Super]

What if aircraft were built like open-source software?

Project Caribou is an ambitious open-source effort to build a 200 kg-class heavy-lift hexacopter that anyone can study, improve, and contribute to. The long-term goal is to accelerate the development of larger open-source aircraft and eVTOL platforms.

The project targets a ~200 kg maximum takeoff weight with an ~100 kg payload, uses an 18S power architecture, and is built around an ArduPilot-based flight stack. The repository includes CAD models, PCB designs, documentation, and engineering workflows that are evolving in the open.

If you are interested in drones, aerospace, robotics, or embedded systems, this is one of the most exciting open-source engineering projects to follow.
https://t.co/ECIYXbQ1jc

[Tim Reha @timreha]

@slipgatecentral What are you using for the eye tracking latency reduction?

[Vadim B @slipgatecentral]

still polishing my #threejs parallax engine
stacking so many layers on the screen heats up my hardware faster than Nomad on max settings

Speed feels better tho, that's what I'm after.

[Marwa ElDiwiny @MarwaEldiwiny]

FreeCAD is underrated.

[Tengiz @tengizlab]

Inspecting a 3D-Printed Turbine Blade

This isn't an X-ray scan. It's a neutron computed tomography (CT) scan of an additively manufactured Inconel 718 turbine blade.

Unlike conventional inspection methods, neutron CT allows engineers to visualize both the external geometry and complex internal features without cutting the part open. This is especially valuable for components with internal cooling channels that cannot be inspected directly.

The scan is then compared with the original CAD model, allowing engineers to verify printing tolerances, detect internal defects, evaluate surface roughness, and confirm that the manufactured part matches the intended design before it enters service.

Source: Oak Ridge National Laboratory