Tech news in 3 minutes

Intel Foundry Nabs A Custom ASIC Win with Fortinet

3 h ago

NVIDIA Vera CPU takes center stage this week as the company releases its deepest technical disclosure yet on the upcoming server processor, including a full architectural whitepaper and fresh SPEC CPU 2026 benchmarks, ahead of Intel’s earnings and AMD’s Advancing AI event. The new details reveal that Vera features 88 Olympus CPU cores with spatial multithreading (176 threads), a 10-wide decoder fed by a 16-instruction fetch unit, 64KB L1 instruction cache and 96KB L1 data cache per core, 2MB L2 per core, and a 164MB unified L3 cache. Memory bandwidth reaches up to 1.2TB/s via LPDDR5X, with NVLink-C2C interconnects at 1.8TB/s and PCIe Gen6/CXL 3.1 support. The Olympus core employs memory renaming, value prediction, and move elimination to boost instruction-level parallelism, backed by 18 execution pipes (8 integer, 6 vector/SIMD). TDP ranges from 250W to 450W. While NVIDIA positions Vera as a broader server CPU—not just a GPU companion like Grace—the benchmarks show competitive but not universally dominant performance against Intel Xeon and AMD EPYC. The disclosures underscore NVIDIA’s push to grow in the lucrative server CPU market, aiming to pair Vera tightly with Rubin GPUs and dissuade customers from using rival processors.