HBM2E
PulseAugur coverage of HBM2E — every cluster mentioning HBM2E across labs, papers, and developer communities, ranked by signal.
-
SambaNova details SN50 AI accelerator with focus on bandwidth utilization
SambaNova has unveiled new technical details about its SN50 RDU, a dedicated AI accelerator designed for high-efficiency and low-latency inference. The SN50 features a dataflow architecture with a large on-chip SRAM and…
-
Nvidia CMP 170HX mining GPUs hacked to unlock 64GB VRAM
A software modification called CMP Unlocker has been developed to restore previously disabled VRAM on Nvidia's CMP 170HX cryptocurrency mining GPUs. This hack allows the cards to access up to 64GB of VRAM, an eightfold …
-
SiPearl's Rhea CPU enters lab testing, targeting late 2026 availability
SiPearl has successfully received the first silicon samples of its Rhea CPU, a significant milestone for Europe's sovereign high-performance computing (HPC) efforts. The Rhea1 processor, featuring over 80 Arm Neoverse V…
-
Intel's canceled 'Arctic Sound' AI GPU prototype surfaces
An engineering sample of Intel's canceled 'Arctic Sound' data center GPU, part of the Xe-HP family, has surfaced. This dual-tile processor, originally intended for the AI market, features 32GB of HBM2E memory and was de…
-
Idle GPU power cost driven by CUDA context, not VRAM
Researchers have quantified the energy cost of keeping AI models loaded on GPUs, a practice known as "model parking." Their study found that the primary energy drain comes from the CUDA context, which adds 26-66W of idl…