Proud to be spotlighted by @GoogleCloud “
@SakanaAILabs
, a Tokyo-based startup that leverages NVIDIA GPUs on Google Cloud, along with services like Filestore and Persistent Disk, to power its research activities and host its foundational models.”
HARDWARE
-
Sakana AI Labs Spotlight on Google Cloud Platform
By
–
-
SambaNova Cloud Achieves Fastest Token Generation Speeds
By
–
Breaking news: we're still the fastest. 😎
— SambaNova (@SambaNovaAI) 3 octobre 2024
Running at full-precision, SambaNova Cloud delivers top speeds of 2470 tokens per sec on Llama 3.2 1B and 1566 tokens per sec on 3B, independently verified by @ArtificialAnlys 🏎️💨
Start developing ⤵️Breaking news: we're still the fastest. Running at full-precision, SambaNova Cloud delivers top speeds of 2470 tokens per sec on Llama 3.2 1B and 1566 tokens per sec on 3B, independently verified by @ArtificialAnlys Start developing
-
Shenzhen breaks world record with 8100 synchronized drones
By
–
El pasado 4 de septiembre la ciudad china de Shenzhen estableció un nuevo record Guinness mundial con el vuelo simultaneo de 8100 drones.
— Juan Merodio (@juanmerodio) 3 octobre 2024
El espectáculo es muy sorprendente, logrando crear increíbles patrones en el cielo nocturno 😍 pic.twitter.com/mtC5a8fc3dEl pasado 4 de septiembre la ciudad china de Shenzhen estableció un nuevo record Guinness mundial con el vuelo simultaneo de 8100 drones. El espectáculo es muy sorprendente, logrando crear increíbles patrones en el cielo nocturno
-
Oura Ring 4 competes with Samsung Galaxy Ring in wearable health tech
By
–
Oura just announced the Oura Ring 4, which comes with "smart sensing" features that pick up on blood oxygen, heart rate, and breathing. The Finnish startup now faces competition from tech giant Samsung, which launches its Galaxy Ring earlier this year.
-
Untether AI Partners Ampere for Energy-Efficient AI Inference
By
–
We've joined the AI Platform Alliance with @AmpereComputing
! See our partnership in action at #Yotta2024 – we're showcasing our record-breaking speedAI240 Slim PCIe card. Visit us in #601F, Oct 7-9, to see the future of sustainable, high-performance #AI. https://
untether.ai/untether-ai-jo
ins-forces-with-ampere-to-provide-arm-based-energy-efficient-ai-inference-solutions/
… -
RAS Blackwell GPU Isolation and Host Boot Issues
By
–
RAS is blackwell+, so TBD. no single GPU isolation, whole host gets booted.
-
Training Machine Learning Models on 10k H100 GPUs
By
–
"How to train a model on 10k H100 GPUs?"
has now been immortalized on my blog: https://
soumith.ch/blog/2024-10-0
2-training-10k-scale.md.html
… -
HBM Memory in Switches: Critical Infrastructure for AI Scale
By
–
one more thing to add.
at this scale, we also have to adjust the actual packet routing algorithms in our switches and NICs, to be able to load-balance well. Did you know switches have to have significant HBM memory as well (not just GPUs) because as the packets queue up, they -
Training Models at Scale: 10K+ H100 Infrastructure Guide
By
–
quick mini-post I wrote for @francoisfleuret broadly summarizing the things one needs to do to train models on 10k+ H100s.
