HPC News Bytes 20250616: Wave of AI Hardware-Software from AMD, Fujitsu Chip, SIGHPC SC25 Travel Grants, HPC Guru Signs Off

A happy day after Father’s Day! It was a week full of news, what with ISC and major vendor activities, here’s a quick (8:52) run-through: AMD’s new GPUs, ROCm 7.0 software and Helios rackscale system; Fujitsu Monaka chip, SIGHPC Travel Grants for SC25, HPC Guru signs off

NVIDIA Reports Blackwell Surpasses 1000 TPS/User Barrier with Meta’s Llama 4 Maverick

NVIDIA said it has achieved a record large language model (LLM) inference speed, announcing that an NVIDIA DGX B200 node with eight NVIDIA Blackwell GPUs achieved more than 1,000 tokens per second (TPS) per user on the 400-billion-parameter Llama 4 Maverick model. NVIDIA said the model is the largest and most powerful in the Llama 4 […]

AI Inference: Meta Collaborates with Cerebras on Llama API

Sunnyvale, CA — Meta has teamed with Cerebras on AI inference in Meta’s new Llama API, combining  Meta’s open-source Llama models with inference technology from Cerebras. Developers building on the Llama 4 Cerebras model in the API can expect speeds up to 18 times faster than traditional GPU-based solutions, according to Cerebras. “This acceleration unlocks […]

Oracle Cloud Deploys NVIDIA Blackwell GPUs

Nvidia said Oracle has stood up its first wave of liquid-cooled NVIDIA GB200 NVL72 racks in its data centers. NVIDIA Blackwell GPUs are now being deployed on NVIDIA DGX Cloud and Oracle Cloud Infrastructure (OCI) to develop ….

NVIDIA to Manufacture AI Supercomputers in U.S.

NVIDIA today said it is working with manufacturing partners to design and build factories that will produce NVIDIA AI supercomputers — i.e., “AI factories” — entirely in the United States… NVIDIA said that within four years, it plans to produce up to half a trillion ….

Google Launches ‘Ironwood’ 7th Gen TPU

Google Cloud today introduced its seventh-generation Tensor Processing Unit, “Ironwood,” which the company said is it most performant and scalable custom AI accelerator and the first designed specifically for inference. Ironwood scales up to 9,216 liquid cooled chips linked via Inter-Chip Interconnect ….

Lightmatter Announces Passage L200, Co-Packaged Optics for AI

MOUNTAIN VIEW, Calif.– Photonic supercomputing company Lightmatter announced Passage L200, which the company said is the first 3D co-packaged optics (CPO) product. Designed to integrate with the latest XPU and switch silicon designs, Passage L200 supports unprecedented AI performance scaling by eliminating interconnect bandwidth bottlenecks, the company said. The L200 3D CPO family includes both […]

CoolIT Announces 4000W Liquid Cooling Coldplate

Calgary, AB, March 13, 2025— Liquid cooling company CoolIT Systems announced a 4000W-ready single-phase direct liquid cooling (DLC) coldplate – which the company said more than doubles the previously accepted limits of single-phase direct liquid cooling. “CoolIT continues to lead the industry in performance.  We are thrilled to show silicon leaders that single-phase DLC will continue […]

@HPCpodcast: Dr. Ian Cutress on the State of Advanced Chips, the GPU Landscape and AI Compute, Global Chip Manufacturing and GTC Expectations

Just before GTC (and for the 100th episode of the @HPCpodcast, and this one sponsored by liquid cooling company CoolIT) ….

Report: 64,000 Nvidia GB200s for Stargate AI Data Center in Texas

A report from Bloomberg states that some 64,000 of Nvidia’s flagship Blackwell GB200 chips will be purchased by OpenAI and Oracle as part of the Stargate AI data center project announced in January at a ceremony ….