The AMD platform is fully supportive of PCIe 4.0, from consumer to data center, with graphics cards, processors, and motherboard chipsets all available. Intel initially refused to say so, but the body is also very honest, in addition to the FPGA and other areas have supported PCIe 4.0, the second generation of Aoten storage, the future consumer-grade platform will also be imported. On the NVIDIA side, Huang Renxun has just unveiled a new generation of Ampere architecture, focusing on aI, data centers, high-performance computing and other market areas, but also the first introduction of PCIe 4.0.
For support of PCIe 4.0, Huang Renxun did not mention in his speech, the official information is also very low-key, only on ampere’s architecture map marked with the words “PCIe 4.0 main interface”, while the Tesla A100 calculation card specificationtable list listed PCIe 4.0 64GB/s.
In addition to pcIe 4.0, the Tesla A100 computing card also supports the third generation NVLink interconnect bus, with up to 12 routes per GPU and up to 600GB/s bandwidth, compared to the second generation 6,300GB/s, and the first generation only 4-way, 160GB/s.
In addition, NVIDIA also released the next generation workstation/personal supercomputing DGX A100, which integrates 15TB PCIe 4.0 NVMe SSD SSDs, but does not disclose specific manufacturers and brands.
As to whether the future RTX 30 series game cards also support PCIe 4.0, there is no definitive information, but there is no reason not to.