#techwatch
November 17, 2026 at 11:29 PM
Expedera and Cadence have shown how two processor designs can share an AI workload before a chip reaches fabrication. Expedera’s Origin Evolution NPU runs network segments, and Cadence’s NeuroEdge 130 handles jobs such as preprocessing and operator fallback. www.jonpeddie.com/techwatch/ex...
Expedera and Cadence pair inference engines
Simulation shows how two processors divide work.
www.jonpeddie.com
November 19, 2026 at 11:47 PM
Intel and SambaNova have demonstrated an AI inference system that divides prompt processing and token generation between two accelerator architectures. Intel’s D-Series GPUs handle prefill, while SambaNova’s SN50 processors manage decode. www.jonpeddie.com/techwatch/in...
Intel and SambaNova split inference workloads
Air-cooled architecture separates prefill and decode processing.
www.jonpeddie.com
November 17, 2026 at 6:27 PM
Q.ANT has introduced its NPU Gen 2+ photonic processor at SC26 in Chicago, extending optical computing into discrete Fourier transforms. The company demonstrated a hybrid system combining the photonic processor with an AMD GPU to run a large language model. www.jonpeddie.com/techwatch/q-...
Q.ANT brings Fourier transforms into light
Photonic processor joins AMD GPUs for AI inference.
www.jonpeddie.com
November 18, 2026 at 6:03 PM
AMD’s MI455X and Nvidia’s Rubin give infrastructure buyers two approaches to cloud AI. AMD supplies 432 GB of HBM4 per GPU; Nvidia supplies 288 GB and distinguishes sparse inference from dense training throughput. We compare the processors. www.jonpeddie.com/techwatch/am...
AMD and Rubin demand careful comparisons
AMD and Rubin demand careful comparisons
www.jonpeddie.com
November 16, 2026 at 4:28 PM
Nvidia announced Rubin HPC Nov. 16 as part of the Vera Rubin platform, with availability expected in 2028. It is aimed at climate, engineering, genomics, & finance simulation. Memory, FP64, power, and pricing details remain undisclosed, but here's what we do know. www.jonpeddie.com/techwatch/nv...
Nvidia Rubin HPC targets scientific simulation
Over 50,000 CUDA cores target scattered data.
www.jonpeddie.com
November 16, 2026 at 11:34 PM
AMD offers cloud buyers three MI400 processors for training and inference. The MI450X and MI455X add memory options to Helios, and the MI430X combines scientific computing with AI. We examine what those choices mean for software developers & infrastructure budgets. www.jonpeddie.com/techwatch/am...
AMD targets cloud training and inference
Memory choices sharpen the Helios proposition.
www.jonpeddie.com
November 16, 2026 at 3:42 PM
【週刊海外テックWatch】データセンターが軍事標的に ロシア・中東の事例が示すクラウドの備え - クラウド Watch https://cloud.watch.impress.co.jp/docs/column/techwatch/2154704.html
November 9, 2026 at 7:14 AM
Your next camera or sensor may need enough intelligence to recognize an event and act locally. STMicroelectronics gives developers that option via the STM32N6x7, a microcontroller family that combines application control, neural acceleration, and image processing. www.jonpeddie.com/techwatch/st...
STMicroelectronics brings neural inference into microcontrollers
Local vision puts memory planning first.
www.jonpeddie.com
November 9, 2026 at 5:09 PM
Meta’s Muse Spark Open combines 816 billion parameters with a sparse architecture that activates 42 billion per token. Intel has announced day-zero support on its upcoming GPU D-series and current Xeon processors. www.jonpeddie.com/techwatch/fo...
Four Intel GPUs host Meta’s Muse Spark
High-capacity memory brings large models into air-cooled servers.
www.jonpeddie.com
November 9, 2026 at 4:10 PM
Innatera has opened its Developer Program, giving engineers access to Pulsar neuromorphic processors, software tools and evaluation hardware through DigiKey. The program combines Python-based development with event-driven computing for always-on sensor applications. www.jonpeddie.com/techwatch/in...
Innatera opens Pulsar to sensor developers
Neuromorphic silicon reaches developers through DigiKey.
www.jonpeddie.com
November 6, 2026 at 3:33 PM
November 6, 2026 at 1:36 AM
Positron has published specs for Asimov, its custom inference ASIC due in 2027, and for Titan, the system built from four or eight of them. www.jonpeddie.com/techwatch/po...
November 5, 2026 at 5:36 PM
Sparxity, formerly SpiNNcloud, has installed its JUBIS sparse-computing system at Germany’s Jülich Supercomputing Centre, bringing 656,640 processing cores to one of Europe’s major HPC facilities. SpiNNaker2 technology executes irregular, event-driven workloads. www.jonpeddie.com/techwatch/sp...
Sparxity brings sparse computing to Jülich
SpiNNaker2 system targets energy-efficient AI and HPC, and renames itself.
www.jonpeddie.com
November 2, 2026 at 7:04 PM
🤖 #MSFT Stance:
Reason: MSFT steady above 200‑MA, but OBV bearish and MACD still lagging. Patience pays—wait for a clear bullish pivot before committing. #TechWatch #QuantTrading #Stocks #TechStocks
October 28, 2026 at 3:14 PM
TetraMem just taped-out its MLX300, a 12nm analog AI processor that stores model weights in multi-level resistive RAM and computes inside those same cells, with an Andes RISC-V vector processor alongside. www.jonpeddie.com/techwatch/te...
TetraMem 12nm analog chip
MLX300 computes inside multi-level RRAM cells.
www.jonpeddie.com
October 28, 2026 at 8:43 PM
【週刊海外テックWatch】Anthropicが「AIワーム」を予告 「恐怖をあおる」との批判も - クラウド Watch https://cloud.watch.impress.co.jp/docs/column/techwatch/2149191.html
October 20, 2026 at 7:05 AM
A 10-person team in Romania wants to retire the triangle. Gacriva Technologies builds Earth Renderer on hopfions, knotted toroidal math structures, and pairs it with low-bandwidth streaming and its own neural texture compression. Here's what they built. www.jonpeddie.com/techwatch/ga...
Gacriva’s hopfions rethink rendering and streaming
Romanian start-up builds math-based graphics pipeline.
www.jonpeddie.com
October 20, 2026 at 9:37 PM