➀ Google showcases photos of its NVIDIA Blackwell GB200 NVL AI server racks for its AI cloud platform; ➁ The racks use liquid-cooled GB200 AI GPUs; ➂ Each GB200 chip features 1 x Grace CPU and 1 x B200 AI GPU for up to 90 TFLOPs of FP64 compute performance.

➀ Elon Musk unveils Tesla's new supercomputer cluster named 'Cortex' at Giga Texas, powered by 100,000 NVIDIA H100/H200 AI GPUs. ➁ The cluster is designed for video training of FSD and Optimus projects with massive storage requirements. ➂ The cooling system for the cluster is extensive, requiring 130MW of power and set to expand to over 500MW within 18 months.

➀ Nvidia has been using a dedicated supercomputer for the past six years to train and refine DLSS models; ➁ This infrastructure has contributed to the continuous improvement of DLSS across its various iterations; ➂ The training process involves analyzing failures and retraining the model with ideal visuals and challenging cases.

➀ ExaAILabs has deployed its Exacluster, a cluster based on Nvidia's H200 GPUs for AI and HPC; ➁ The cluster aims to build a search engine superior to Google's; ➂ The Exacluster includes 144 H200 GPUs, 192 96-core CPUs, and 36TB of DDR5 RAM.

➀ Cerebras推出新一代晶圆级AI加速器CS-3，其速度比前代快两倍；➁ Cerebras的WSE-3将高速SRAM均匀分布在整个芯片表面，每个核心都能单周期访问快速内存；➂ Cerebras与Tri-Labs合作，利用晶圆级技术加速模拟，如材料科学和药物研究。

➀ The Param Rudra supercomputer, named after Lord Shiva, is a significant development in India's scientific capabilities; ➁ It is designed to support scientific research and development in India; ➂ The supercomputer is expected to be operational by 2025; ➃ It features high-speed computing and advanced data analysis capabilities; ➄ The project is part of India's strategy to become a global leader in technology and innovation.

➀ India's prime minister Narendra Modi dedicated three new supercomputers, known as 'PARAM Rudra'; ➁ The machines use local designs for servers, interconnects, and cooling, but do not feature the planned 96-core Arm CPU; ➂ The National Centre for Radio Astrophysics and the S.N. Bose Centre for Basic Sciences have revealed some details about their systems; ➃ The Inter-University Accelerator Centre has not disclosed the specifics of its machine, but has previously teased a three-petaflop machine using Intel Xeon 2nd Gen processors; ➄ A third generation of Rudra machines is planned to use a 96-core processor designed by India's Centre for Development of Advanced Computing.

➀ The JUPITER Exascale Supercomputer has commenced installation with the Modular data center, set to be a unique European Exascale supercomputer using all NVIDIA CPUs and GPUs. ➁ The system is expected to house around 125 BullSequana XH3000 racks in 50 containers, with the finished system consisting of approximately 24,000 NVIDIA GH200 Superchips. ➂ JUPITER will utilize NVIDIA Quantum-2 InfiniBand networking and is expected to be a significant win for Arm in the HPC space.

➀ The HPC market, including servers and cloud, was valued at $372 billion in 2023, with a slight decrease from 2022. ➁ The on-prem HPC server market was $150 billion in 2023, expected to grow to $163 billion in 2024. ➂ HPC cloud market is projected to grow strongly, with a 21% increase in spending on public cloud resources for HPC workloads. ➃ HPC/AI storage continues to be the fastest-growing element, accounting for about 21% of on-prem HPC spending. ➄ Quantum computing market is expected to reach $15 billion by 2026, with a 22.1% annual growth rate.

➀ NVIDIA's DGX SuperPOD architecture is designed for advanced AI model training, inference, and HPC tasks. ➁ The H100 SuperPod consists of 256 GPUs interconnected via NVLink and NVSwitch, with a reduce bandwidth of 450 GB/s. ➂ The GH200 SuperPod integrates GH200 GPUs with Grace CPUs, utilizing NVLink 4.0 for enhanced connectivity and scalability. ➃ The GB200 SuperPod, featuring GB200 GPUs and Grace CPUs, aims to support larger-scale AI workloads with a 576 GPU configuration.