They have devices planned for next seven years. I am not worried.
AI HARDWARE
-
Semiconductor Fabs: The Most Complex Machines Ever Built
By
–
I visited many factories and headquarters and a semiconductor fab is the most complicated machine I have seen. By far.
-
NeRF Technology Enables Walking Through Personal Memory Experiences
By
–
Wait until they turn into a NeRF and then you walk around your own memories in some future glasses. Or your kids will. Or your grandkids.
-
New Insta360 Camera Delivers Impressive Video Capabilities
By
–
These videos are of the same thing. I love my new @insta360
. -
Andes Tech and Tenstorrent Partnership at RISC-V Conference
By
–
@tenstorrent x @Andes_Tech @jimkxa w/ Dr Charlie Su and Franklin Lin at Andes @risc_v Con
-
Nvidia Partners with Snowflake for LLM GPU Infrastructure
By
–
Nvidia teams up with Snowflake for large language model AI Snowflake is extending its Snowpark offering with GPU instances https://
zdnet.com/article/nvidia
-teams-up-with-snowflake-for-large-language-model-ai/
… $NVDA $SNOW #investing #AI #generativeAI #database -
Inflection AI Sets New Generative AI Standard with H100 GPU Cluster
By
–
We’re proud to have helped set a new standard for Generative AI by building a cluster with 3,584 @NVIDIA H100 GPUs in the debut MLPerf benchmark. Our system completed the massive GPT-3-based training benchmark in under eleven minutes. Read more: https://
blogs.nvidia.com/blog/2023/06/2
7/generative-ai-debut-mlperf/
… -
NVIDIA H100 Cluster Breaks MLPerf GPT-3 Training Record
By
–
We set a new industry standard for MLPerf with our cluster of NVIDIA H100s. It completed the massive GPT-3-based training benchmark in less than 11 minutes.
-
Groq Spotlight Event on Large Language Models at Scale
By
–
Don't miss our next #GroqSpotlight! Watch the live event focused on #LargeLanguageModels at scale on LinkedIn at http://
groq.link/lgs3or YouTube at http://
groq.link/ytgs3. -
High-speed tensor core inference with massive memory bandwidth
By
–
So, you have a 200 TB/s fire hose of recycling information, which you would have to “pump up” over many cycles from much slower conventional memory, but it could then feed straight in to a field of tensor core registers for real time model inference. Fun to think about!