@edsonbellido.bsky.social3/4 CORE expands beyond design into EDA, IP and more, covering the full value chain from research to manufacturing for customers in Japan and overseas. Would your team tape out early to a new 2nm foundry on ecosystem strength alone? #FoundryWars #AIHardware
The IEEE Design Program addresses the memory wall bottleneck, teaching engineers how to optimize modern AI chips for resource-constrained edge deployments.
@futuregearai.bsky.socialNew arXiv paper explores normative competence in LLMs using multi-agent community debates, finding baseline agents fail to learn community norms even when doing so would improve outcomes. Relevant to consumer AI…
#AIAlignment #LLMResearch #AIHardware #ConsumerAI
https://arxiv.org/abs/2610.10906
@autonainews.comSK Hynix Hybrid Bonding Struggles While Micron Taiwan Strike Vote Looms
Samsung quietly cleared Nvidia's HBM4E bar—just as rivals Hynix and Micron hit major walls.
#AI #AIHardware #Semiconductors
https://autonainews.com/sk-hynix-hybrid-bonding-struggles-while-micron-taiwan-strike-vote-looms/
@futuregearai.bsky.socialDynaTE is a hardware-software co-design that adapts accelerator execution to per-token requirements during diffusion LLM decoding, aiming to cut overhead that prior dLLM accelerators left on the table. The paper frames it…
#DynaTE #dLLM #AIHardware #DiffusionLLMs
https://arxiv.org/abs/2610.11284
@securityonline.bsky.socialNvidia launches the DGX Station for Windows, a desktop AI supercomputer featuring the Grace Blackwell Ultra superchip to run trillion-parameter models locally.
#Nvidia #DGXStation #AIHardware #Supercomputer #MachineLearning
@edsonbellido.bsky.social1/4 Intel just moved inside Applied Materials' EPIC Center as a founding partner. Equipment vendor and chipmaker now share one R&D roof, and TSMC, Samsung, and Micron sit in the same building. 🏗️ #Semiconductors #AIHardware
![SemiRecap Infographic [STRATEGIC_PARTNERSHIP_MOU]: Applied Materials, Intel Partner on Next-Gen Chip Manufacturing. A 2.5D vector cartoon illustration on a solid pure white background (#FFFFFF) showing two modular buildings docking across a cyan collaboration bus. The Applied Materials EPIC Center module contains cleanroom bays labeled FEOL, BEOL, and PACKAGING, filled with Deep Maroon (#450d16) process tools and warm gold wafer handling arms, with a glowing Foveros 3D stacked die highlighted at the packaging bay. The Intel Hillsboro R&D campus module links into a secure dedicated bay interface marked with partner labels. Minimal clean structural annotations identify each partner and the shared interface.](https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:nxdiuv7626cjwxguzjlh4o3b/bafkreie4hge6zjwg4cebj6izstqbcbxd4hllkpon2wzmudkwv3cv57fdj4@png)
@tmtabor.ioIntel markets Gaudi as the open alternative to CUDA. This week, three new open-weight models dropped that I can't run on it. #AIHardware
@businessam.bsky.socialChipfabrikant TSMC profiteert van groeiende vraag naar AI-hardware en ziet omzet sterk stijgen #TSMC #AIhardware #semiconductors #kunstmatigeintelligentie #technologie
@hardwarebriefcom.bsky.socialAnthropic Develops Custom AI Chip with Samsung
@futuregearai.bsky.socialA new edge AI architecture (AO-DA) splits persona and logic into separate inference paths on a single INT4 base model with hot-swappable LoRA adapters, aiming to keep on-device LLM reasoning robust against persona-heavy…
#EdgeAI #OnDeviceLLM #AIHardware #LLMAgents
https://arxiv.org/abs/2610.09772
@futuregearai.bsky.socialarXiv paper shows LLMs can be trained to tolerate unreliable hardware, with error resilience growing alongside model size. Results from 40,000 GPU-hours suggest large models learn internal error-correcting…
#AIhardware #FaultTolerance #LLMs #EfficientInference
https://arxiv.org/abs/2610.10311
@hardwarebriefcom.bsky.socialSamsung Completes HBM4E Testing for NVIDIA
@edsonbellido.bsky.social2/4 NVHBM moves the memory controller off the GPU compute die into the custom base die, plus a custom I/O PHY. NVIDIA claims up to 30% more bandwidth, 15% lower HBM power, and up to 67% less PHY package area vs standard HBM4E. #AIHardware #Chips
@edsonbellido.bsky.social2/4 The constraint: aggressive performance and cost targets, zero appetite for unnecessary risk. Ho called the Turin device strong, and partners already had experience with it. A proven host shortened the design cycle. #AIHardware #ChipDesign
@edsonbellido.bsky.social3/4 AMD holds 46% of server CPU revenue, so its nearest competitor is its own last generation. PCIe 6.0 reaches PCs around 2030; MRDIMM may never leave the datacenter. When stickers drift from real prices, what does a flat $14,904 signal? #AIHardware #Datacenter
@edsonbellido.bsky.social3/4 Micron is the only US HBM maker, working with Nvidia on the industry's first custom HBM. A $250B buildout spans campuses in Clay, New York and Boise, Idaho, while SK Hynix and Samsung expand in Korea. Memory costs are lifting iPad and MacBook prices. #AIHardware #HBM
@michel-k-tech.bsky.socialNvidia's 64GB DGX Spark brings its compact Grace Blackwell AI desktop down to $4,999, but local AI hardware is still clearly aimed at pros.
@futuregearai.bsky.socialA new system called SparseCraft uses a language model in a closed loop to co-optimize sparse accelerator hardware and software, editing Chisel RTL and schedules on Gemmini across 15 iterations with…
#SparseComputing #HardwareDesign #AgenticAI #AIHardware
https://arxiv.org/abs/2610.05037
@edsonbellido.bsky.social2/4 The stack: Supermicro AS-1116CS-TN + AMD EPYC 9355, an Alveo U250 FPGA running Wolley CXL 3.2 controller IP, and 4x 1GB PERSYST STT MRAM UDIMMs. Carrier: PCIe Gen3 x8. Writes up to 100x faster than NAND SSDs, nanosecond class cache line access. #CXL #AIHardware