Astera Labs has announced Orion, a rack-scale reference design that lets a neocloud or enterprise pick its GPUs and CPUs after the rack is built. Accelerators arrive as PCIe plug-in cards, and a 320-lane PCIe switch fabric scales a pod from 8 to 160 of them, so an operator can deploy the rack once and

Supermicro announced an expanded line of Open Compute Project Accepted systems, racks, and rear door heat exchangers, all delivered through its Data Center Building Block Solutions (DCBBS) program and on display at booth B4 at the OCP Global Summit in San Jose through October 15. The common element is the rack: the 21-inch ORv3 and

MSI is showing a 48RU NVIDIA MGX rack at the OCP Global Summit in San Jose this week, built from four liquid-cooled CG681-S6093 servers stacked under an NVIDIA Spectrum-4 SN5600 switch and an SN2201 out-of-band switch. Each 6U server takes up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition liquid-cooled GPUs, so the rack

Running large language models locally has become much easier over the past couple of years, but a gap remains between downloading a model and having a useful desktop environment for it. Jan describes itself as a ChatGPT alternative, offering an open-source desktop app with a familiar chat-style interface that keeps most of the setup inside

IBM and Red Hat said on October 6 that Lightwell, the open-source security program the two companies funded with $5 billion in May, has found, fixed, and backported patches for more than 400 previously unknown vulnerabilities in widely used Java libraries. The same announcement made Lightwell Clearinghouse generally available, so enterprise customers can now submit

Marvell announced its second-generation Structera X CXL memory expanders at the OCP Global Summit, moving the line to six memory channels, four-host sharing, and a PCIe 6.0 x16 host link. The Structera X 3406 drives 18 DDR4 RDIMM slots for up to 9TB behind one controller, and the Structera X 3506 drives 12 DDR5 RDIMM

IBM announced a self-hosted deployment option for IBM Bob on October 1, bringing its agentic software development platform to on-premises, private cloud, sovereign cloud, and air-gapped environments. Bob runs on infrastructure the customer manages, with model inference either on supported models the customer installs locally or, in hybrid configurations, through approved external model services. IBM

Microsoft unveiled Azure Maia 300 today at the OCP Global Summit, the third generation of its in-house inference accelerator and the first built as a multi-chiplet part. Each accelerator pairs a TSMC N2P compute die with eight HBM4 stacks for 288GB of memory at 18TB/s inside a 950W TDP, and a Maia 300 rack holds

Intel today gave Crescent Island its product name at the OCP Global Summit. The Intel GPU D-series is built on the Xe3P architecture and pairs up to 480GB of LPDDR5X with 1.5TB/s of memory bandwidth and 1.3 petaflops of FP8 compute, all inside a 350W air-cooled PCIe card. The 480GB figure applies to partner-built cards;