I would love to see Sohu-like ASICs eventually come to desktop/consumer market! For now it seems focused on cloud inference, but the tech is exciting!
HARDWARE
-
NVIDIA Launches Podcast on Nemotron 70B Open Source Model
By
–
Uh oh, @NVIDIAAI is starting a podcast!
— Latent.Space (@latentspacepod) 2 novembre 2024
and it’s a good dive into one of the sleeper hit open source models of the year – Nemotron 70B.
its incredible that nvidia funds entire research and alignment teams to do actual llm research, rather than just selling chips. extreme… https://t.co/iK8VjyFGw8Uh oh, @NVIDIAAI is starting a podcast! and it’s a good dive into one of the sleeper hit open source models of the year – Nemotron 70B. its incredible that nvidia funds entire research and alignment teams to do actual llm research, rather than just selling chips. extreme
-
Cerebras and Groq achieve impressive 2000 tokens per second speed
By
–
2000 token / second running Llama 3.1 70b. Thats insane! I have high hopes for Cerebras and Groq. Especially when reasoning models like o1 take much longer to "think".pic.twitter.com/OvLJpK5rc8
— Chubby♨️ (@kimmonismus) 1 novembre 20242000 token / second running Llama 3.1 70b. Thats insane! I have high hopes for Cerebras and Groq. Especially when reasoning models like o1 take much longer to "think".
-
Ray-Ban Meta Glasses Rose Lens Availability Feature
By
–
My glasses are Ray-Ban Meta.
Rose lenses are not an option I'm aware of. -
IBM z16 Launch: Key Differences from z15 Revealed
By
–
How is @IBMZ 16 different from z15? Here is PJ Catalano's elevator pitch at #IBMTechXchange.
— Helen Yu (@YuHelenYu) 1 novembre 2024
It brought back the fond memory of #IBMZ15 launch in #NYC in 2019. That's where I met @avrohomg @StevenDickens3 @Kevin_Jackson @craigmullins
The IBM z16™ represents a significant… pic.twitter.com/xGbdkCBPBLHow is @IBMZ 16 different from z15? Here is PJ Catalano's elevator pitch at #IBMTechXchange. It brought back the fond memory of #IBMZ15 launch in #NYC in 2019. That's where I met @avrohomg @StevenDickens3 @Kevin_Jackson @craigmullins The IBM z16™ represents a significant
-
Edge Computing Models: Necessity and Importance
By
–
Its even necessary, to have models that run on the edge
-
Google TPU Evolution: SeaStar to JellyFish Hardware Journey
By
–
Amin Vahdat and I gave a brief presentation about the origin and development of Google's TPUs over ~a decade at this week's Google TGIF meeting. Presenter costumes were encouraged, so I went as a SeaStar (codename for TPUv1 ) & Amin went as a JellyFish (codename for TPUv2).
-
NVIDIA HOVER: Efficient Motor Skills Learning with 1.5M Parameters
By
–
NVIDIA GEAR lab is breaking new ground. With just 1.5M parameters, HOVER proves that mastering complex motor skills doesn’t require huge models. Using NVIDIA simulation suite, which accelerates physics by 10,000x, humanoids can learn a year’s worth of motion in under an hour. pic.twitter.com/LjWJhw0Fpl
— Chubby♨️ (@kimmonismus) 1 novembre 2024NVIDIA GEAR lab is breaking new ground. With just 1.5M parameters, HOVER proves that mastering complex motor skills doesn’t require huge models. Using NVIDIA simulation suite, which accelerates physics by 10,000x, humanoids can learn a year’s worth of motion in under an hour.
-
Etched Sohu ASIC Enables 4K 30fps AI Video Generation
By
–
This is amazing!
— Marek Rosa | European🇪🇺 | South African🇿🇦 (@marek_rosa) 1 novembre 2024
Sohu, new Transformer ASIC from Etched, will soon enable 4k 30fps video generation & playable AI-generated games 👍🙂 https://t.co/Y5dM1r6p4UThis is amazing! Sohu, new Transformer ASIC from Etched, will soon enable 4k 30fps video generation & playable AI-generated games