Calling all Mathletes, this one is for you. We’ve been asked to show the math behind our MoE claims. So we did. Our analysis confirms: On GPUs, expert parallelism creates severe communication overheads that dwarf computation and make MoE training painfully slow. At
AI HARDWARE
-
Humanoid Robot Sprinting: Can It Reliably Recover From Falls?
By
–
The lab mentioned developed a humanoid for sprinting, with very explosive leg movements. https://
newshub.sustech.edu.cn/en/html/202508
/46874.html
…
What's unclear from the video is if this is a fluke or can reliably help the robot get up after falling. -
Voyager SDK 1.4.0 Released: Enhanced Performance Edge AI
By
–
Voyager SDK 1.4.0 is out there and ready. Boosted performance, smoother deployment & more models in the #VoyagerSDK zoo, all built to power AI at the edge with Metis. Be great to hear about your experience with it
-
CUDA graphs and action input models improve performance consistency
By
–
Making everything one big CUDA graph helped wall clock consistency a lot on a laptop, but we still fought with power management. Looking forward to using an Nvidia Spark in the future. It isn’t in the repo code, but the single biggest win I have seen is an “action input” model
-
Battery Issues Were Known Before: What Happened?
By
–
Disons qu'on était au courant pour les batteries… https://t.co/O4ejenGsJP
— Olivier Rimmel 🕊️ (@OlivierRimmel) 15 septembre 2025Disons qu'on était au courant pour les batteries…
-
AGI Architecture: Centralized Systems Outperform Distributed Colonies
By
–
The correct metaphor is whether you want your AGI to run on one GPU rather than a redundant cluster. Large integrated ant colony singletons tend to outperform and replace ant species that form small, competing colonies
-
LFM2-VL-450M Tiny AI Model Now Available in Smart Glasses
By
–
LFM2-VL-450M in your glasses! Tiny models unlock completely new use cases. Really proud to see LFMs being adopted in a completely new form factor.
-
Europe Launches JUPITER Exascale Supercomputer
By
–
Europe enters the exascale supercomputing league with ‘JUPITER https://
ec.europa.eu/commission/pre
sscorner/detail/en/ip_25_2029
…
#digitaleu #jupiter #computing #innovatiok #technology @DigitalEU @elaniazito @ArturHabant @BetaMoroney @Fabriziobustama @CurieuxExplorer @Shi4Tech @mvollmer1 @mikeflache @smaksked -
Tinygrad Build Quality Assessment in GPU Guide
By
–
gotta admit, i've been getting increasingly curious about tinygrad's build quality as i wrap up the Buy a GPU guide and website
-
Nvidia vs Cisco: Bubble or Justified Valuation?
By
–
À gauche (Cisco, années 2000) :
On voit que le prix de l’action (courbe bleue) s’est envolé alors que les bénéfices réels (courbe orange) stagnaient. Résultat : énorme décalage → c’était une bulle. Et quand elle a éclaté, le prix est retombé brutalement. À droite (Nvidia,