Nvidia has unveiled the Pascal series of GPUs which will debut in 2016, so don’t hold your breath.
Nvidia collaborated with IBM to create a new faster GPU-CPU interconnect, enabling GPUs and CPUs to communicate 5-12 times faster than they can now. GPU-CPU communication is an essential part of supercomputing and such a development can increase development by leaps and bounds in the field. How fast? 80 GB per second. The current 16 GB per second seems paltry compared to this.
NVLink is supposed to replace the current CPU-GPU connector, which is PCIe. The new interconnect will also prove helpful in fields like machine learning, cloud computing, etc.
The interconnect will debut in Nvidia’s next gen GPU line, starting with the Pascal series of GPUs. The new high speed interconnect is called ‘NVLink’
“Today’s GPUs are connected to x86-based CPUs through the PCI Express (PCIe) interface, which limits the GPU’s ability to access the CPU memory system and is four- to five-times slower than typical CPU memory systems,” Nvidia explained. “PCIe is an even greater bottleneck between the GPU and IBM Power CPUs, which have more bandwidth than x86 CPUs. As the NVLink interface will match the bandwidth of typical CPU memory systems, it will enable GPUs to access CPU memory at its full bandwidth… Although future Nvidia GPUs will continue to support PCIe, NVLink technology will be used for connecting GPUs to NVLink-enabled CPUs as well as providing high-bandwidth connections directly between multiple GPUs.”
Wait, more interesting bits coming!
NVIDIA agreed that GPUs are one of the biggest and most power hungry part of any set up. So, NVIDIA is going to be using something called 3D memory packaging: Building chips over chips. This makes the whole setup smaller WHILE increasing power efficiency by up to four times.
“We’re going to, for the first time, build chips on top of other chips.”
Pascal also has a common/unified memory space for the CPU and the GPU, so now developers don’t have to worry about splitting resources between the two. This is a bit similar to AMD’s hUMA architecture.
Pascal will replace the current Maxwell architecture. But we have to wait for some time as it isn’t coming anywhere near our mitts anytime soon.
Faster GPUs. While being 4 times as much power efficient. All while also being compact. All three at one go, great going NVIDIA!