IBM and Nvidia want to connect graphics cards and SSDs faster with Big Accelerator Memory.
Big Accelerator Memory is intended to connect GPU and SSD directly to each other.
The two manufacturers IBM and Nvidia, together with several universities, have developed a new architecture that is intended to combine the advantages of graphics chips and SSD storage. The Big Accelerator Memory (BaM) baptized technology should be able to process large amounts of memory faster. With an Nvidia graphics card or a hardware accelerator, the load on the processor should be reduced. At the same time, the access times to the memory can be reduced in this way. An NVMe SSD should serve as a counterpart, which can store and make the data available quickly.
No CPU dependency
Since BaM abstracts the GPU threads, massive data structures can be accessed in a relatively short time. Nvidia already sees several areas of application for BaM in its own company. The technology relieves the processor, since many work steps that are normally taken over by the CPU are no longer necessary and instead lead directly to the SSD. A dependency on the CPU is simply no longer necessary.
With BaM, the graphics card can write its data directly to the slower SSD and access new data at the same time. There is no intermediate storage. Currently only a prototype of the BaM design has been developed. But even this should serve as a viable alternative to conventional solutions. Nvidia and IBM want to release BaM as open source. Other companies could also benefit from the new technology. AMD has established a similar system with Radeon Solid State Graphics. However, both technologies are not yet relevant for everyday PC use. In data centers, however, the technology could save time and electricity costs with large amounts of data.
SSD technology: Everything you need to know when buying