Nvidia’s first desktop PC chip, the GB10 Superchip system-in-package (SiP), is set to launch this month with Asus leading the charge on its Ascend GX10 Grace Blackwell platform. The compact platform promises data center-level performance in a workstation and edge deployment suitable for AI workloads.
The GB10 SiP integrates a high-performance CPU with 10 Arm Cortex-X925 cores running up to 3.90 GHz, alongside a Blackwell GPU capable of delivering 1 PetaFLOPS of FP4 compute throughput for AI workloads. A unified memory architecture and massive FP4 throughput are key advantages over traditional CPU-GPU setups.
However, leaked performance figures from Geekbench suggest that the GB10 may not be as powerful as expected in general-purpose compute tasks. It trails behind Qualcomm’s Snapdragon X Elite and Apple’s M3 processor in similar tests.
Asus’ Ascend GX10 mini-PC is expected to launch on July 22, with a starting price of $2,999. The system is designed for AI development and comes with Nvidia’s GB10 Superchip platform.
Nvidia emphasizes the GB10’s advantages over traditional CPU-GPU setups but acknowledges that the GPU portion may be weak compared to their dGPU products. This could disappoint users not specifically looking to use the chip for AI workloads.
In contrast, Asus’ Ascend GX10 mini-PC is priced lower than Nvidia’s DGX Spark, with identical configurations except for smaller storage. The Ascend GX10 has been reported at a price of $2,999, making it an attractive option for those in the market for a compact AI workstation.
The GB10 Superchip is expected to be a key component in Nvidia’s strategy to dominate the AI PC market. With its focus on leading-edge performance and premium pricing, the company aims to capitalize on the growing demand for AI workstations.
Source: https://www.tomshardware.com/tech-industry/artificial-intelligence/nvidias-first-desktop-pc-chip-lands-this-month-asus-leads-with-ascend-gx10-grace-blackwell-desktop-platform