Serving technology enthusiasts for over 25 Years. TechSpot is the place to go for tech advice and analysis you can trust.
Nvidia created NVLink in 2014. It was developed as an alternative technology to connect GPUs and other computing components in servers and data centres. NVLink-connected device use mesh-style communication, instead of relying solely on a central hub. Nvidia has introduced a new enhanced solution for the next-generation of AI data centres.
Nvidia revealed a new generation NVLink interconnect at this year’s Computex. Called NVLink Fusionis a technology that supports semi-custom AI infrastructures and data center deployments. The company has already partnered up with a list of industry leaders who are working to adopt the new technology in their products.
Alongside Nvidia’s recently introduced Blackwell Ultra graphics card, NVLink Fusion plays a key role in Nvidia’s vision for “AI factories.” NVLink Fusion has been integrated into two compute dense rack models: GB200 NVL72 & GB300 NVL72. Together, NVLink Fusion with Nvidia’s specialized Racks can deliver a data bandwidth of 1.8 TB/s for each GPU – 14x faster than traditional PCIe Gen5 connection.
| Third Generation | Fourth Generation | Five Generation | ||||
|---|---|---|---|---|---|---|
| NVLink bandwidth per GPU |
600 GB/s | 900 GB/s | 1,800 GB/s | |||
| Max. number of Links per GPU |
12 | 18[1919[1919659016] | [1919659016][1919659016][1919659016][19[19 | [1919][196590][19][19][19][19][19][19][19][19]18][19]19]18][19HuangstatedatComputexthatdatacenterswerebeingreimaginedre-architectedandAIisbeingembeddedintoalmosteverycomputingplatformforthefirsttimeindecadesNvidiapartnersforNVLinkFusionincludeMediaTek(AsteraLabs)Marvell(AlchipTechnologies)AsteraTechnologiesSynopsysandCadence(19659023)FujitsuQualcommTechnologiesandMediaTekarealsousingthesolutiontointegrateNvidiaGPUswiththeirCPUsenablingthemtodevelophigh-performancecomputingsystemNvidiastatedthat[19HuangstatedatComputexthatdatacenterswerebeingreimaginedre-architectedandAIisbeingembeddedintoalmosteverycomputingplatformforthefirsttimeindecadesNvidiapartnersforNVLinkFusionincludeMediaTek(AsteraLabs)Marvell(AlchipTechnologies)AsteraTechnologiesSynopsysandCadence(19659023)FujitsuQualcommTechnologiesandMediaTekarealsousingthesolutiontointegrateNvidiaGPUswiththeirCPUsenablingthemtodevelophigh-performancecomputingsystemNvidiastatedthat
NVLink Fusion could be scaled in massive cloud provider configurations, putting “millions” GPUs in charge of any ASIC solutions. Sassine Gahzi, CEO of Synopsys, said that data centers are becoming AI platforms which need to move large datasets faster and in the most efficient manner. NVLink Fusion will offer ASIC and silicon makers a wide range of customization options. Nvidia offers a highly integrated product stack through its Mission Control orchestration component. This unified control centre allows enterprise customers to automate the complex tasks needed to run a modernized AI Infrastructure. Is Nvidia’s AI-focused strategy bad for gamers?
|
