News from the AI & ML world
DeeperML
@www.nextplatform.com
//
Nvidia is expanding its AI dominance by opening up its NVLink ecosystem to rivals. CEO Jensen Huang unveiled NVLink Fusion at Computex 2025, allowing non-Nvidia CPUs and accelerators to integrate with Nvidia's AI infrastructure. This move allows customers to use CPUs, GPUs, and accelerators from other vendors alongside Nvidia GPUs, all tied together using Nvidia’s high-bandwidth NVLink interconnect. Key partners include MediaTek, Marvell, Fujitsu, and Qualcomm, and the program allows customers to use CPUs, GPUs, and accelerators from other vendors alongside NVIDIA GPUs, all tied together using NVIDIA’s high-bandwidth NVLink interconnect.
This strategic positioning ensures Nvidia remains central to AI data centers, even when customers use rival chips. Huang stated that NVLink Fusion enables the building of "semi-custom AI infrastructure, not just semi-custom chips," emphasizing that every such system runs through Nvidia's ecosystem, including its GPUs, interconnects, and software stack. The resulting intellectual property package is called NVLink Fusion, meaning that those making custom CPUs or accelerators can license the NVLink port designs and the memory atomic protocol that rides on top of it to allow a kind of NUMA memory sharing akin to that which has been available for CPUs for several decades now.
Nvidia is also partnering with Foxconn to build an AI supercomputer in Taiwan, which will help fuel Taiwan's innovation. The supercomputer will utilize 10,000 Nvidia Blackwell GPUs. Taiwan’s National Science and Technology Council will use this supercomputer to deliver AI cloud computing muscle to Taiwanese organizations, making it easier for the country to adopt AI tech across private and public institutions. TSMC, Nvidia’s main supplier, intends to use it for R&D with the AI supercomputer said to have “orders-of-magnitude faster performance, compared with previous-generation systems”.
References :
This strategic positioning ensures Nvidia remains central to AI data centers, even when customers use rival chips. Huang stated that NVLink Fusion enables the building of "semi-custom AI infrastructure, not just semi-custom chips," emphasizing that every such system runs through Nvidia's ecosystem, including its GPUs, interconnects, and software stack. The resulting intellectual property package is called NVLink Fusion, meaning that those making custom CPUs or accelerators can license the NVLink port designs and the memory atomic protocol that rides on top of it to allow a kind of NUMA memory sharing akin to that which has been available for CPUs for several decades now.
Nvidia is also partnering with Foxconn to build an AI supercomputer in Taiwan, which will help fuel Taiwan's innovation. The supercomputer will utilize 10,000 Nvidia Blackwell GPUs. Taiwan’s National Science and Technology Council will use this supercomputer to deliver AI cloud computing muscle to Taiwanese organizations, making it easier for the country to adopt AI tech across private and public institutions. TSMC, Nvidia’s main supplier, intends to use it for R&D with the AI supercomputer said to have “orders-of-magnitude faster performance, compared with previous-generation systems”.
Share:
References :
- Maginative: NVIDIA Opens Its NVLink Ecosystem to Rivals in Bid to Further Cement AI Dominance
- The Next Platform: Nvidia Licenses NVLink Memory Ports To CPU And Accelerator Makers
- www.tomshardware.com: Nvidia teams up with Foxconn to build an AI supercomputer in Taiwan
Classification:
- HashTags: #AI #NVIDIA #Supercomputing
- Company: Nvidia
- Target: Taiwan
- Product: NVLink
- Feature: NVLink interconnect tech
- Type: AI
- Severity: Informative