@AMD acquires @mipsology to strengthen its AI inference software capabilities https://actuia.com/actualite/amd-acquiert-mipsology-pour-renforcer-ses-capacites-logicielles-dinference-ia/?time=1242
… #AI #ArtificialIntelligence #Business
AI HARDWARE
-
AMD Acquires Mipsology to Strengthen AI Inference Capabilities
By
–
-
Google TPUv4 Pods: Water Cooling and Dynamic Optical Switching
By
–
A delightful 2m40s tour through one of our TPU datacenters, looking at some things that make the TPUv4 pods special, like water cooling, & custom optical circuit switching that enables dynamic reassembly of up to 64 racks (64 chips each) into different shape supercomputers.
-
Hand Towel Dispensers Still Struggle While LLMs Advance
By
–
y'all out here doin these Large Language Models i'm just trying to get this hand towel dispenser to recognize my hands
-
CPU Programming Complexity Versus GPU Performance Models
By
–
high-performance CPU programming IMO is harder than GPU programming.
GPU performance model is a lot more uniform and easy to understand than CPU performance which has so many edges and discontinuities and so many layers of caches and a much more complex instruction pipeline. -
Google’s Hardware-Software Stack Strategy for AI Dominance
By
–
We don't really have a lot of details on the latter, but it represents a tantalizing opportunity to offer a very Google-only stack—hardware optimized for its software—to recapture its dominance in AI. And, in this case, avoid repeating history for the third time.
-
Groq Demonstrates LLM Performance at Smoky Mountains Conference
By
–
Stop by our table today or tomorrow at @ORNL
's annual Smoky Mountains Computational Sciences & Engineering Conference to see our demo of best-in-class performance for #LLM applications running on the @GroqInc Language Processing Unit™! -
Google Announces General Availability of Cloud TPU v5e
By
–
Excited to see this announcement about the general availability of our latest Cloud TPU offering, based on the TPU v5e system!
-
Nvidia Launches JAX-Based LLM Framework at Google Next
By
–
Nvidia is announcing a large language model framework built on top of JAX and OpenXLA at Google Next. No specific details but this has the potential to have a lot of very drastic implications. More on it later in the newsletter today.
-
GPU Concentration Divides AI Innovation Capabilities
By
–
For the confused: Semianalysis is a semi-conductor analysis newsletter that just published a long piece about GPU-rich vs GPU-poor companies, arguing that only GPU rich companies can do important work in AI. It also included a chart about the growth of Google’s large and
-
Fine-tuning Llama 2 7B on Single GPU A100 and Google Colab
By
–
You can finetune a 7B Llama 2 model on a single GPU A100 for example, so you could even use Google Colab. But in general, most universities and companies are supportive of open source these days.