Forbes’ Post

Cerebras Systems Inc. was founded just nine years ago but seems to benefit massively from the recent AI computing boom. It's innovated in ways that appear to put the current-gen H100 and the upcoming GB200 die to shame with a "single, enormous chip" cable of up to 900,000 GPU cores - such as with its CS-3 chip. The Cerebras CS-3 chip absolutely dwarfs the double die size of the huge GB200, and is the size of a steering wheel, requiring two hands to hold. It's been described by the manufacturer as the "world's fastest and most scalable AI accelerator" which is purpose-built to "train the world's most advanced AI models". https://lnkd.in/ebhSWZ94

Nvidia introduced a new supercomputer, dubbed the DGX H100, at the 2024 Supercomputing Conference. This model is based on Ampere Next architecture and is designed to support artificial intelligence tasks. It boasts a peak performance of 6 exaFLOPS and is equipped with up to 4096 Tensor Cores. The supercomputer also features 256GB of HBM2E memory, capable of up to 8 TB/s of memory bandwidth. With its combination of high-performance computing capabilities and power-efficient design, the DGX H100 is positioned to be a powerful tool for AI researchers and developers. https://lnkd.in/gdWX2SZp

NVIDIA just released the first #H200 Tensor Core GPU to OpenAI! If you haven't checked out its performance stats already click the link below. This machine is a beast. If the only real restraint on how fast AI can learn is its processing capacity, then with the advancements in GPU hardware, human roles in the technology field will soon become obsolete. All the accumulated knowledge you took centuries to gather, AI will take less than a year. Coupled with its speciality in pattern recognition and access to unlimited amounts of data, it's literally impossible for anyone to outperform it in this field. Going by the trend, a general improvement in AI is usually accompanied with a significant improvement in other complementary technology fields. We should expect to see a significant improvement in #robotics and #3Dprinting. If one of the biggest hurdles in #quantumcomputing, quantum decoherence, ends up getting solved, then mankind will no longer be the most intelligent species on earth. We will end up creating technological beings superior to ourselves. This sounds like science fiction right now but it's true. 🤣🤣 The ability of quantum computers to analyze data simultaneously rather than in a sequence would make human thinking seem dim! By the time we came up with any course of action, AI would have calculated every possible scenario and came up with possible countermeasures for all of them. https://lnkd.in/d9q89H6G

The new model called the H200 from NVIDIA, the world’s most valuable #chipmaker has added capabilities to use high-bandwidth memory in better coping with large data, fuelled its dominance in the AI #computing #market. READ on: https://lnkd.in/dTAMUuVR #electronicsmanufacturing #electronicsengineering #electronicsindustry  #electroniccomponents #manufacturing #manufacturingindustry #manufacturingengineering #manufacturingtechnology #technology #technologynews #technologyinnovation #technologysolutions #aeis #aeissg #sg

TensorWave Advances AI Technology with New AMD Accelerators Revolutionizing the AI Hardware Market, TensorWave is setting the stage for substantial changes in AI processing power by integrating AMD’s latest Instinct MI300X AI accelerators into its systems. These accelerators are touted as more efficient alternatives to NVIDIA’s established Hopper H100 AI GPU. Expanding its hardware infrastructure, TensorWave is working towards acquiring a sizable fleet of 20,000 AMD Instinct MI300X AI accelerators by year’s end, spread across two of its data centers. Additionally, the company is on a trajectory to roll out cutting-edge liquid-cooled systems by 2025. #amd TensorWave

‘As far as performance is concerned, Isambard-AI is expected to achieve over 200 FP64 PetaFLOPS for high-performance computing that requires accurate calculations and simulations, and will also deliver over 21 ExaFLOPS for AI inference and training workloads that use lower precision. Performance of the supercomputer represents a tenfold improvement over the U.K.'s previous fastest supercomputer, according to Nvidia. "Isambard-AI represents a huge leap forward for AI computational power in the U.K.," said Simon McIntosh-Smith, a Bristol professor and director of the Isambard National Research Facility. "Today, Isambard-AI would rank within the top 10 fastest supercomputers in the world and, when in operation later in 2024, it will be one of the most powerful AI systems for open science anywhere."’ https://lnkd.in/gCJppBBQ

Have you heard about the groundbreaking Wafer Scale Engine 3 (WSE-3) from Cerebras? 🤯 This amazing technology will change how we train large language models. It overcomes the performance and memory issues that traditional GPUs and CPUs have struggled with for a long time..   The WSE-3 contains 4 trillion transistors and 900,000 AI cores into a single, monolithic chip. 🔥This massive scale allows it to train LLMs with up to 24 trillion parameters, pushing the limits of traditional architectures.   To learn more about Cerebras and how it compares to industry giant Nvidia, checkout this article: https://lnkd.in/gcjx3RCW   Have you had the opportunity to work with Cerebras' chips or train LLMs on their systems? I'd love to hear your experiences and insights!💬

AI is this year's buzzword, and business use cases continue to increase. Ultimately, LLMs and compute-heavy workloads are just 1s and 0s machine language and assembly language on servers. If your critical business needs require these workloads to be on-premise, consider looking into Supermicro. Supermicro chassis have liquid cooling capabilities, easy access to GPUs, and PSUs with 3+kW sized for future upgrades. Easy access to your GPUs is a differentiator for your DC operation techs and admins. When running 24x7 mission-critical AI clusters, failure is inevitable. World Wide Technology is an Elite Partner (Highest level) with NVIDIA. We have deep expertise working with A100 and H100 GPUs and the DGX platform. https://lnkd.in/ggMzXpwi #WWT #Nvidia #DGX #H100 #A100 #HPC #AI #ML

Reads of the week Engineering — AMD reveals the MI325X, a 288GB AI accelerator built to battle Nvidia's H200 https://lnkd.in/de-Vxmyw Repos — Enrich your Slack alerts with contextual observability data, helping on-call engineer investigate faster: https://lnkd.in/dQ2xDTVn AI — Hitachi and Microsoft Announce Billion-Dollar AI Partnership: https://lnkd.in/dmt-cSCb Self Development — An Algorithmic Solution to Insomnia: https://lnkd.in/d9s5kAxC Nice Reads — It wasn’t me - Snowflake denies responsibility: https://lnkd.in/d9sgbByT How a Self-Published Book Broke ‘All the Rules’ and Became a Best Seller: https://lnkd.in/dqQf5rrf All Spotify Car Things Are About to Become E-Waste: https://lnkd.in/dBft-aYw Ticketmaster confirms hack which could affect 560m: https://lnkd.in/dvd9a8sy Night-vision lenses so thin and light that we can all see in the dark: https://lnkd.in/dXdKR2XT

Exciting news! We're incredibly proud to be NVIDIA's preferred partner for UQD and UQDB Quick Disconnect Couplings, helping build #AI factories and #datacenters that drive the next industrial revolution. These cutting-edge components are an integral part of the NVIDIA GB200 NVL2 platform, designed to deliver unparalleled performance in large language model inference, retrieval-augmented generation, and #dataprocessing. “Danfoss’ focus on innovative, high-performance quick disconnect and fluid power designs makes our couplings valuable for enabling efficient, reliable and safe operation in data centers,” said Kim Fausing, president and CEO of Danfoss. “As a vital part of NVIDIA’s AI ecosystem, our work together enables data centers to meet surging AI demands while minimizing environmental impact”. Curious about how we're #EngineeringTomorrow in the AI and data center industries? Check out the full story here: https://lnkd.in/gunA4J4h #datacentersolutions #fluidconveyance #fluidpower #quickdisconnectcouplings #datacentercooling #AIrevolution