NVIDIA Announces "Pascal" Next-Gen GPU Family
While NVIDIA's recently-announced Maxwell graphics architecture is doing terrific under Linux, Pascal was announced today as their next-generation GPU come 2016.
Pascal will be the successor to Maxwell and it's looking to be super-great. NVIDIA's Pascal GPU will feature stacked 3D memory, which are DRAM chips with wide interfaces and on the same package as the GPU. Stacked 3D memory will allow for faster graphics memory performance while Pascal will also feature Unified Memory support between the GPU and CPU, a new NVLink interface, and a Pascal Module.
NVLink is what NVIDIA's touting as the world's first high-speed GPU interconnect that is much faster than today's PCI Express 3.0. NVLink is a high-bandwidth interconnect and will be dramatically faster in communicating with the CPU and its memory. Besides being faster than PCI Express 3.0, it's also supposed to be more power efficient and fit into smaller form factors than a PCI-E x16 slot. Details on NVLink can be found via this press release. IBM will reportedly be shipping hardware with NVLink for Exascale computing but it's not known what other vendors may decide to get on board.
NVLink comes in at 5 to 12x faster than PCI Express 3.0, 3D Memory should be 2 to 4 X the memory bandwidth and size potential, and the Pascal module should be a third the size of a PCI Express graphics card.
Other initial NVIDIA Pascal details can be found via this NVIDIA blog post.
Pascal will be the successor to Maxwell and it's looking to be super-great. NVIDIA's Pascal GPU will feature stacked 3D memory, which are DRAM chips with wide interfaces and on the same package as the GPU. Stacked 3D memory will allow for faster graphics memory performance while Pascal will also feature Unified Memory support between the GPU and CPU, a new NVLink interface, and a Pascal Module.
NVLink is what NVIDIA's touting as the world's first high-speed GPU interconnect that is much faster than today's PCI Express 3.0. NVLink is a high-bandwidth interconnect and will be dramatically faster in communicating with the CPU and its memory. Besides being faster than PCI Express 3.0, it's also supposed to be more power efficient and fit into smaller form factors than a PCI-E x16 slot. Details on NVLink can be found via this press release. IBM will reportedly be shipping hardware with NVLink for Exascale computing but it's not known what other vendors may decide to get on board.
NVLink comes in at 5 to 12x faster than PCI Express 3.0, 3D Memory should be 2 to 4 X the memory bandwidth and size potential, and the Pascal module should be a third the size of a PCI Express graphics card.
Other initial NVIDIA Pascal details can be found via this NVIDIA blog post.
15 Comments