What to do if the GPU memory is not enough? New technology scales through PCIe with sub-100 nanosecond latency

This site reported on July 5 that with the support of the Korea Institute of Science and Technology, Panmnesia recently demonstrated a new technology that allows the GPU to directly access system storage through the Compute Express Link (CXL) interface, which also means that the GPU can use System storage serves as its video memory.

The company named this PCIe attached graphics memory solution CXL GPU Image, with a latency of less than 100 nanoseconds, which is significantly faster than traditional solid-state drives.

Note from this site: CXL is a protocol that works on PCIe links, but the technology must be recognized by ASIC and its subsystems. Since there are no CXL logic structures and subsystems in the GPU to support DRAM and high-speed SSD endpoints, there is no way to simply add a CXL controller to the technology stack.

Panmnesia achieved this by developing a series of hardware layers supporting all key CXL protocols integrated into a unified controller.

The above is the detailed content of What to do if the GPU memory is not enough? New technology scales through PCIe with sub-100 nanosecond latency. For more information, please follow other related articles on the PHP Chinese website!