Nvidia just jumped into the desktop computer chip business with their brand new GB10 Grace Blackwell platform. The tech giant wants to grab a piece of the artificial intelligence market that everyone seems crazy about these days. Asus will start selling computers powered by these chips on July 22nd. The company plans to bring back their Ascend GX10 mini computer at the same time. Pricing details remain fuzzy but Nvidia has shared some hints about costs.
The GB10 packs serious punch with a 20-core processor and graphics card built specifically for AI work. Ten high-speed Arm Cortex-X925 cores can run up to 3.9 GHz for demanding tasks. Ten energy-saving Cortex-A725 cores handle lighter work to save power. The graphics side uses Nvidia's Blackwell design that delivers 1 PetaFLOPS of computing power. This setup targets people working with large language models and generative AI programs.
Memory performance stands out as a major selling point for this new platform. The system features 128GB of fast LPDDR5X memory running on a 256-bit connection. This setup provides up to 273 GB/s of bandwidth for moving data around quickly. Apple's M4 Pro chip offers similar memory performance for comparison. The unified memory helps AI models run more efficiently than traditional setups.
Nvidia aims this chip at AI researchers and developers who need powerful workstations in smaller packages. The expected price sits around $3000 based on similar Nvidia systems. Dell, HP, and Lenovo are also preparing their versions of these AI-focused computers.
The GB10 packs serious punch with a 20-core processor and graphics card built specifically for AI work. Ten high-speed Arm Cortex-X925 cores can run up to 3.9 GHz for demanding tasks. Ten energy-saving Cortex-A725 cores handle lighter work to save power. The graphics side uses Nvidia's Blackwell design that delivers 1 PetaFLOPS of computing power. This setup targets people working with large language models and generative AI programs.
Memory performance stands out as a major selling point for this new platform. The system features 128GB of fast LPDDR5X memory running on a 256-bit connection. This setup provides up to 273 GB/s of bandwidth for moving data around quickly. Apple's M4 Pro chip offers similar memory performance for comparison. The unified memory helps AI models run more efficiently than traditional setups.
Nvidia aims this chip at AI researchers and developers who need powerful workstations in smaller packages. The expected price sits around $3000 based on similar Nvidia systems. Dell, HP, and Lenovo are also preparing their versions of these AI-focused computers.
