Great purchase! I’m still on M1 Max.
AI HARDWARE
-
Disney Deploys Bipedal Robot for Star Wars Park Experience
By
–
Disneys new Bipedal robot spotted at a Disney World Park. Seemingly with Star Wars roleplaying cast. pic.twitter.com/Bg5GHIATy2
— Linus ✦ Ekenstam (@LinusEkenstam) 31 octobre 2023Disneys new Bipedal robot spotted at a Disney World Park. Seemingly with Star Wars roleplaying cast.
-
LLaMa2-70B Training Computation Corrected Calculations
By
–
Corrected Calculations (the original claim doesn't change, as it goes from 5e+24 -> 1.1e+24 and is still two orders of magnitude behind):
LLaMa2-70B was trained on 2T tokens, and 1.7m hours of A100 GPU time. At a HFU of ~60%, LLama2-70B took ~1.1e+24 flops (1.7m * 312TFlops -
Apple Patents Matte Black Finishing Technique for Devices
By
–
Apple was granted a patent last month on a matte black finishing technique. I guess we're seeing it first hand here. Looks great.
-
Solar Panel Efficiency and H100 GPU Power Consumption Analysis
By
–
No, the amount of incident sunlight is around 1000 watts per square meter peak, but a solar panel will only get about 220 watts out of that. A fully loaded H100 system is 10 kW for 8 GPUs, not counting facility cooling. Probably budget 2kW / GPU, so 10,000 GPUs would be 20 MW, or
-
Cerebras Accelerates 175B LLM Training with PyTorch Single Device
By
–
Cerebras presented at the PyTorch Developer Conference and covered how we accelerate Large Language Model (175B+) training using PyTorch, Torch-MLIR, and the simplicity of single device training without a single torch.distributed instruction Watch here:
-
Multiple Companies Building High-Performance AI Compute Clusters
By
–
more than two companies have clusters with those FLOPs. quite a handful.
-
LLaMa2-70B Training Costs: Computing Efficiency Analysis
By
–
Calculations:
LLaMa2-70B was trained on 2T tokens, and 3.3m hours of A100 GPU time. At a HFU of ~60%, LLama2-70B took ~4.4e+24 flops (3.3m * 624TFlops bfloat16 * 60% HFU). -
NVIDIA AI Model Wins Sign Language Recognition Kaggle Challenge
By
–
NVIDIAN Christof Henkel clinched the top spot in an American Sign Language recognition challenge on @Kaggle with an #AI model that detects fingerspelled characters and translates them into text. #NVIDIAlife https://
blogs.nvidia.com/blog/2023/10/2
3/nvidia-life-kaggle-ai-contest-sign-language-recognition/?ncid=so-twit-477997
… -
Predibase AI Cloud Now Supports High-End GPUs for LLM Fine-tuning
By
–
The @Predibase AI Cloud is the most efficient, cost-effective way to #finetune and serve open-source #LLMs and we now offer high-end GPUs (#A100s / #H100s) so you can train even the largest LLMs like #Llama-2-70B. Get started here! https://
pbase.ai/40lQcEf