IBM has gained a significant reputation through its quantum computing research and brain-like smart chip named True North, and now it is introducing an additional technology that will complement its current power saving lineup.
The rather innovative Nvidia’s NVLink interface in IBM’s new server enables you to boost almost five time speedy communication between CPUs and GPUs (graphic processor units). Inside IBM’s new S822LC server provides a prime interconnect which is based on Nvidia’s homegrown NVLink technology that has been working since years.
The latest NVLink is of course an upgrade to PCI Express 3.0, is developed aiming easy communication between a GPU and the rest of the components of the system. While Nvidia’s GPU builds up is Pascal architecture oriented to support NVLink, on the other hand, IBM has added the NVBus support in its chip.
While it is the talk about the main as well as primary hardware vendors who managed shipments of NVLink-compatible server, IBM is one of the very first ones. As per experts, Nvidia has x86 servers from tech giants like Hewlett Packard Enterprises, Dell and Cray. The homegrown server “DGX-1” by Nvidia, which has eight Tesla P100 GPUs, is already is being shipped with the NVLink interface.
The GPU is used by the servers as the co-processors to fasten the applications and speed pipe to the CPU processing at the faster pace. “Faster processing helps run databases, as well as fraud detection and engineering-related applications”, said Dylan Boday, senior offering manager for Linux on Power infrastructures at IBM.
Kinatica, an in-memory database application which harnesses the GPUs’ computing power is a good and considerable example here. Boday further describes the applications, “designed to work with the NVLink interconnect”.
At present, it is only Tesla P100 GPU that is going to be benefited by roaring NVlink data transfer speeds, and with the 2U server up to 4 GPUs can be easily plugged in. But the Linux server also has a PCI-Express 3.0 interconnect for GPUs. Having double sockets, this power server supports up to 20 CPU cores and 1 TB of memory, and has storage slots for hard drives and SSDs. The other additional features include:
• The three PCI-Express slots
• An extra interface named CAPI (coherent accelerator processor interface) for facilitating attachment of custom chips such as FPGAs (field programmable gate arrays).
Though the server will be shipped and be available worldwide very soon, IBM has not provided the price for the server yet.
The Power8 server is the first version of NVLink of the company and IBM has very futuristic plans to introduce Power9 chips and servers in order to compliment NVLink 2.0 which is expected to be impressively faster.
The another release by IBM is the two socket S822LC Big Data server. This does not have the NVLink interface, but is based on the older Kepler architecture and significantly supports up to Nvidia K80 GPUs. Loaded with five PCI-Express 3.0 and four CAPI slots, it supports up to 20 CPU cores, up to 512GB of memory, and 96TB storage in multiple slots