Computex 2023 started today in Taipei, Taiwan, with several presentations from technology companies and one of them is NVIDIA, which announced the new Spectrum-X Ethernet switch, a network platform dedicated to innovations involving artificial intelligence in the cloud.
In today’s presentation NVIDIA also announced that Cumulus Linux, pure SONiC and NetQ SDKs will be compatible with Spectrum-X, meaning developers will be able to transition to the new platform with ease.
Spectrum-X is built on top of the NVIDIA NVIDIA BlueField-3 DPU, which is expected to improve the performance of AI cloud networks by up to 1.7 times with high energy efficiency, consistent and predictable performance in a multi-tenant environment.
Gilad Shainer, Senior Vice President of Networking at NVIDIA, commented on the release saying:
Transformative technologies such as generative AI are forcing all enterprises to push the boundaries of data center performance in pursuit of competitive advantage. NVIDIA Spectrum-X is a new class of Ethernet networking that removes barriers to next-generation AI workloads that have the potential to transform entire industries.
Additionally, NVIDIA says this is the world’s first 51TB/s internet switch built specifically for AI networking. Therefore, it can direct the traffic as best as possible to reduce the memory buffer.
It also allows developers to build software-defined, cloud-native AI applications thanks to the acceleration provided by NVIDIA’s own platform. With this, it is possible to reduce the response time of applications with generative artificial intelligence.
According to NVIDIA, the plans are:
Using Spectrum-X1 on Israel-1, a hyperscale generative AI supercomputer to be deployed in its Israeli data center on Dell PowerEdge XE9680 servers based on the eight-GPU NVIDIA HGX™ H100 platform, BlueField-3 DPUs and Spectrum-4 switches .
Another key benefit is that Spectrum-X can optimize bandwidth usage by offering lossless data transfer with 200 200Gb/s ports on just one switch, or up to 16,000 dual-tier ports to support exponential AI growth in cloud with reduced latency.
In addition to these news, NVIDIA also announced the DGX GH200 supercomputer dedicated to generative AI, a new technology that makes game NPCs interact with the player in real time and even an alliance with MediaTek to create a new mobile processor.
