D-Matrix's Raptor accelerator, detailed in an ISCA 2026 paper, pairs a TSMC 4-nanometer logic die face-to-face with a custom DRAM die to reach up to 100 TB/s of bandwidth per card, with the paper repo…
At the Hot Chips '26 conference in California, new AI chips from OpenAI, Intel, Meta, and others promise cheaper and faster token generation at lower power consumption, signaling a shift away from Nvi…
Nvidia and Cerebras are touting benchmark results at Hot Chips that show inference speeds of 3,400 tokens per second on Gemma 4 31B, but The Register argues these figures are unattainable in real prod…
Intel's upcoming Datacenter GPU, codenamed Crescent Island, features a 480 GB LPDDR5X memory subsystem, the highest capacity of any GPU or AI accelerator, enabling it to run Deepseek-V4 Flash entirely…
At the 37th Hot Chips conference, OpenAI unveiled its custom inference chip Jalapeño, claiming it outperforms NVIDIA's GB200/GB300 systems with 1.5–1.9× more work per watt at peak throughput, 1.7–3.6×…
Meta Platforms Inc. unveiled its MTIA 400 accelerator at the Hot Chips conference, designed for both LLM training and ad-serving inference, delivering 12 petaFLOPS of MXFP4 compute at 1.7 GHz, which i…
Cerebras Systems introduced the CS-4, the fastest AI accelerator in the industry, at Supernova 2026, and detailed its architecture at the Hot Chips conference in Palo Alto. The CS-4 is the first syste…
OpenAI and Broadcom's Jalapeño inference chip, unveiled Tuesday at the Hot Chips conference, delivers 1.5 to 1.9 times more work per watt and 1.7 to 3.6 times lower latency on tested models, according…
Cerebras Systems detailed the rack architecture for its CS-4 AI system at the Hot Chips conference on August 25, introducing the Nexus platform that integrates three WSE-3 Turbo processors with modula…
OpenAI's first in-house inference chip, "Jalapeño," outperforms Nvidia's Blackwell and Rubin in throughput and energy efficiency, according to SemiAnalysis tests presented at the Hot Chips conference.…
OpenAI's Jalapeño inference chip outperformed Nvidia's Blackwell system on Semianalysis's InferenceX benchmark, delivering more tokens per user and more throughput per kilowatt, according to results p…
OpenAI unveiled its Jalapeño AI accelerator at the Hot Chips conference on Tuesday, claiming the custom chip, developed with Broadcom, delivers 1.5x to 1.9x higher peak throughput and 1.7x to 3.6x low…
Nvidia Corp. introduced a new networking concept called 'scale-in' at the Hot Chips event in Palo Alto, California, claiming it as a 'fifth' pillar of AI networking that unifies north-south traffic by…
IBM (NYSE: IBM) announced at the Hot Chips conference the first dual-architecture mainframe processor, designed to run both IBM and Arm compute platforms in future IBM Z and LinuxONE systems. This pro…
Cerebras Systems unveiled its fourth-generation CS-4 rack, doubling the performance of the CS-3 by increasing clock speeds and power delivery on the same 5nm wafer-scale engine (WSE-3), enabling doubl…
SRAM serves as the staging ground for KV Cache values written from HBM in LLM inference, and its size directly determines the maximum local data available to local compute, according to a technical de…