#Web3’s Post

The evolution of Nvidia GPUs from 1995-2020

Google announces Gemma, its open LLMs that can run locally Google has released 2B and 7B parameter open models called Gemma The lightweight LLMs were built using the research and technology used to create the Gemini models Gemma can run locally on PCs and is optimized to run on NVIDIA GPUs #gemma #googleai #geminiai

See how you can take advantage of the powerful GPU-accelerated optimization capabilities offered by NVIDIA cuOpt on an #OCI VM instance accelerated by NVIDIA GPUs. https://lnkd.in/e6-Wc3x8

Few things to consider if NVIDIA is truly going after the custom DC AI ASIC market: 1/ it justifies the market for DC AI ASICs. Been all GPU architecture so far. 2/ 2 custom chip and 15 merchant silicon companies out there. Lot more competitive market. 3/ Nvidia already ships tiny ASIC blocks on its DC GPUs. (Tensor cores) 4/ Nvidia should/ would play the SW compatibility card.

Here’s a video of an NVIDIA DGX SuperPOD rack with DGX GB200 systems. Each liquid-cooled rack features 36 NVIDIA GB200 Grace Blackwell Superchips consisting of 36 NVIDIA Grace CPUs and 72 Blackwell GPUs, all connected as one with NVLink switching. #gtc24 #ai #supercomputers #datacetner #nvidia StorageReview.com NVIDIA Jordan Ranous

Introducing Lambda 1-Click Clusters: On-Demand GPU Clusters featuring NVIDIA H100 Tensor Core GPUs with Quantum-2 InfiniBand. No long-term contract required. https://lnkd.in/eiY-ebq7

In this video I am talking about the remarkable benefits of using GPUs for CFD. For more than 15 years we are working with NVIDIA GPUs. That is why I am looking forward to be at the NVIDIA GTC 2024 conference from 18th to 22nd of March. I am happy to meet you there. Please DM me or write an e-mail to eugen.riegel@numeric.systems.

In Case You Didn't Know One important factor that elevates GPUs for deep learning applications is their inherent ability for extensive parallel processing. GPUs feature a highly parallel architecture, characterized by numerous cores capable of executing simultaneous computations. Take, for instance, the NVIDIA RTX A5000, equipped with an impressive 8,192 CUDA cores, a big contrast to the 24 cores present in the Intel i9 CPU processor. This parallelism proves particularly advantageous for the repeated calculations intrinsic to deep learning, such as matrix multiplications and convolutions. The substantial number of CUDA cores in GPUs allows for the efficient handling of parallel tasks, establishing them as a powerhouse in accelerating deep learning workflows. #nvidia #gpu #deeplearning #cuda #intel

I've been playing with matrix multiplication on GPUs, using a Hilbert curve to improve L2 cache utilization, and seeing if I can get non-embarrassing performance out of it. Seems this approach can outperform Nvidia's cuBLAS for matrix sizes of 16384x16384 and above. Blog post here with open source code on GitHub for anyone who'd like to mess with it too. #gpu #cuda #cplusplus https://lnkd.in/e4QvmSVR

Taiwanese startup Neuchips displayed an add-in board that enables desktop computers to run large language models like the 7 billion-parameter version of Llama 2. The Evo PCIe AI accelerator is optimized for transformer networks to provide comparable performance to GPUs while consuming less electricity (55 watts versus an Nvidia RTX 4080’s 320 watts). The card will be available later this year at an undisclosed price. Versions outfitted with four or more chips are on the company’s roadmap. #nvidia #ai #artificialintelligence #machinelearning @Neuchips #llm #gpt #generatieveai