NVLink
NVLink is a communications protocol developed by Nvidia. NVLink specifies a point-to-point connection between a CPU and a GPU and also between a GPU and another GPU.[1] NVLink products introduced to date focus on the high-performance application space.
Use in the datacenter
On 5 April 2016, Nvidia announced that NVLink will be implemented in the forthcoming Pascal microarchitecture-based GP100 GPU, to be used e.g. in the Tesla P100 products, which are also aimed at deep learning in the datacenter.[2]
Use in supercomputing
The US Department of Energy contracted Nvidia and IBM to building two more supercomputers named "Summit" and "Sierra".[3] These will use NVLink for the Node interconnect, while InfiniBand Dual Rail EDR (23 GB/s) shall be used for the System interconnect.[4] They shall combine Volta with POWER9.
Use in off-the-shelf hardware
It remains to be seen whether low-budget custom tailored solutions such as e.g. the Fastra II will take advantage of NVLink. Fastra is said to be mainly limited by Nvidia's Scalable Link Interface (SLI).
See also
- Intel QuickPath Interconnect
- PCI Express
- HyperTransport
- Point-to-point (telecommunications)
- InfiniBand
- Message Passing Interface
- INK (operating system)
- Compute Node Linux
References
- ↑ "What Is NVLink?". Nvidia. 2014-11-14.
- ↑ "Inside Pascal: NVIDIA’s Newest Computing Platform". 2016-04-05.
- ↑ "Whitepaper: Summit and Sierra Supercomputers" (PDF). 2014-11-01.
- ↑ "Nvidia Volta, IBM POWER9 Land Contracts For New US Government Supercomputers". AnandTech. 2014-11-17.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||