AMD's upcoming EPYC 9006-series 'Verano' processors, designed for AI server workloads, will utilize a new SB1 socket that is smaller than existing SP7 and SP8 sockets. The CPUs will feature up to 72 Zen 6 cores and a 24-channel LPDDR5X memory system, tailored for hosting AI accelerators rather than general-purpose computing. AMD has not yet disclosed detailed specifications for the SB1 socket or the exact configurations planned for Verano processors.
tomshardware.com
· 2026-10-08
OpenAI is deploying its new Jalapeño AI chip in racks alongside AMD's EPYC Turin CPUs, each configured with 1.5TB of memory, rather than Nvidia's newer Vera processor. Richard Ho, OpenAI's VP and Head of Hardware, told Tom's Hardware that Turin was chosen for pragmatic reasons, calling Vera 'a little bit behind' in maturity and noting partners already had experience with AMD's chip.
tomshardware.com
· 2026-10-02
A pre-launch Geekbench 7 benchmark shared on X shows OpenAI's dots agent, powered by GPT-6 Astra, running on a nine-core slice of an AMD EPYC 9V74 virtual machine with roughly 9.73GB of memory. The leaked run scored 9,435 multi-core, higher than six post-launch runs found in Geekbench's public database ranging from 8,135 to 8,991. Dots launched Tuesday at OpenAI's DevDay for Pro and Business Premium users, giving each agent its own cloud computer, browser, and preloaded software including Blender, GIMP, and Python.
tomshardware.com
· 2026-09-30
AMD has published preliminary pricing for its upcoming Epyc 9006 Venice server processor lineup, spanning 31 models from an eight-core Epyc 9106 to a 256-core Epyc 9996 flagship priced at $14,904. The chips span two platforms, SP7 for high memory bandwidth and capacity and SP8 for greater PCIe connectivity, with dual-socket SP7 systems capable of housing 512 Zen 6 cores. AMD says the prices are not yet finalized.
techspot.com
· 2026-09-30
AMD has released full pricing for its EPYC 9006 (Venice) server processor lineup, spanning 8 to 256 cores and ranging from $700 to $14,904. The SP7 platform tops the range with the 256-core EPYC 9996 at $14,904, while the 64-core EPYC 9556 starts the lineup at $8,008, according to a list obtained by StorageReview.
tomshardware.com
· 2026-09-29
Phoronix's Michael Larabel reports that a clean x86-64 defconfig Linux kernel build now takes about 15 seconds on his dual AMD EPYC 9575F workstation, down from over 22 seconds before recent Kbuild patches. The speedup stems from work by Linux memory-management developer Lorenzo Stoakes, who used AI/LLM assistance to remove compilation bottlenecks, alongside continued hardware gains.
tomshardware.com
· 2026-09-27
Researchers Evan Hoffman and Tae Kim probed Meta's Muse AI agent and got it to reveal its infrastructure: each user sandbox runs on AMD EPYC 9D25 Turin CPUs with two dedicated cores and 8GB of memory, using Ubuntu 24.04. Muse's sandboxes have no GPUs, since Meta reportedly routes inference through separate GPU servers, and Hoffman says the agent could be prompted to attempt risky actions like setting up SSH access to its own VM.
tomshardware.com
· 2026-09-25
AMD released a white paper expanding on its earlier performance claims for the upcoming EPYC 'Venice' server CPUs, reiterating that a 96-core Venice chip is about 20% faster per-core than Nvidia's 88-core Vera on SPEC CPU 2026's Integer Rate test. The paper also details throughput results showing the 256-core EPYC 9996 outperforming Intel's Xeon 6980P by 2.37x and Vera by 2.24x, plus a 78% generational gain over the prior 192-core EPYC 9965.
tomshardware.com
· 2026-09-18
A developer benchmarked how Intel, AMD, Arm and Apple processors handle IEEE 754 subnormal floating-point numbers—the tiny values near zero that require special handling. Testing double-precision arithmetic on Intel Granite Rapids and Emerald Rapids Xeons, AMD Zen 5, AWS Graviton, and Apple M4 Max chips, the results showed Intel processors suffer dramatic slowdowns, sometimes 40-plus times slower, when computations involve subnormal values compared to normal ones.
lemire.me
· 2026-09-15
Jerry James benchmarked an 8-GPU NVIDIA RTX PRO 6000 Blackwell system paired with an AMD EPYC 9555 CPU, offering 768GB of combined GDDR7 memory. Rather than splitting huge models like Llama-3.1 405B or DeepSeek-R1 671B across all eight PCIe cards, which introduces heavy latency, the team found the setup performs best running independent single-GPU model instances in parallel.
gpupartner.com
· 2026-08-31
GoKawiil