Our founder and CEO Jensen Huang returns to the stage of the world's largest professional graphics conference #SIGGRAPH2023 on August 8 to deliver a live keynote. https://
nvda.ws/3DJ8R2k
HARDWARE
-
Jensen Huang Keynotes SIGGRAPH 2023 Graphics Conference
By
–
-
CoreWeave: Largest Private GPU Operator in North America
By
–
"With over 45,000 GPUs in our fleet, we are the largest private operator of GPUs in North America." Numbers feel right to me https://
coreweave.com/blog/we-are-co
reweave
… -
Groq Language Processor Llama2 70B Performance Spotlight Event
By
–
If you are interested in throughput and low latency for large language models, don't miss our next GroqSpotlight: Groq Language Processor™ Llama2 70B Sneak Peek. http://
groq.link/gsaugust
#LLM #llama2 #largelanguagemodel -
Cerebras Condor Galaxy 1: First 4 ExaFLOPS AI Supercomputer
By
–
Cerebras Tech Talk! Join us on August 9th at 11:00 AM PT as we will share details about Condor Galaxy 1 (CG-1), the first of nine interconnected 4 exaFLOPS AI supercomputers to be built in partnership with G42. Register here: https://
hubs.li/Q01Z-ghM0 -
GPU Announcements Are the New Fundraising Milestone
By
–
If you want media attention nowadays, just start telling people you want/are ordering/have a lot of GPUs. It’s the new “I raised $100m round”
-
GPU Power Consumption: High Current Flow at Low Voltage
By
–
I didn’t really appreciate that, because they operate at under one volt, high end GPUs are continuously flowing over 500 amps through the die at full load.
-
CoreWeave Raises $2.3B Debt Collateralized by Nvidia Chips
By
–
https://
reuters.com/technology/cor
eweave-raises-23-billion-debt-collateralized-by-nvidia-chips-2023-08-03/
… cc @aashaysanghvi_ -
NVIDIA’s 55x Growth Since 2015: GPUs Dominating AI Market
By
–
Overheard in the hallowed hallways of AI compute: "I have been hearing GPUs are not designed for AI since 2015 – during that time NVIDIA 10x (at least) its market cap" I actually checked. It's not 10x since 2015, it's…55x!
-
Loading Model Weights into GPU Memory with Ray Object Store
By
–
How did we do it? By loading the model weights into memory once before training begins and inserting them as numpy arrays into the #Ray object store, we can then zero-copy read the weights directly from shared memory into each GPU worker process.