Nvidia has announced the world’s first high-speed GPU interconnect, opening the door for a new wave of Exascale supercomputers.
The Nvidia NVLink will be integrated into the company’s future Pascal-based graphics cards, boosting GPU and CPU data sharing speeds by five to 12 times what they can achieve today.
This will address an existing bottleneck that is holding back speeds, and will result in new supercomputers that are between 50 and 100 times faster than anything available now.
The interconnect was co-developed by IBM, which will also be introducing it to its new POWER CPU lineup.
The interconnect will replace PCI Express (PCIe), the existing interface on most modern computers. PCIe does not take full advantage of CPU memory, making the need for a new high-bandwidth interconnect essential. NVLink is also more energy-efficient than PCIe.
An additional Unified Memory feature allows the programmer to treat CPU and GPU memories as one block of memory, freeing up developer time.
“NVLink technology unlocks the GPU’s full potential by dramatically improving data movement between the CPU and GPU, minimizing the time that the GPU has to wait for data to be processed,” said Brian Kelleher, SVP of GPU Engineering at Nvidia.
The technology is expected to be available in 2016. IBM and Nvidia are not the only ones working on unified technology. AMD has also been working on something similar called HUMA.
- How long will Moore’s Law last?