is leading the creation of a memory standard for the PCs with AI

Jensen Huang, the co -founder and general director of NVIDIA, is convinced that in the future most users will have a “supercommer of artificial intelligence (AI) Personal “. At the beginning of last January he led the presentation in the CES of Las Vegas (USA) of Project Digitsa very compact personal computer capable of executing models of up to 200,000 million parametersand therefore bigger than GPT-3.

This computer is mainly intended for researchers, developers and students, although a good part of the latter can hardly invest the $ 3,000 (about 2,870 euros) that costs the most economical review of this machine. His heart is a soc GB10 that integrates a GPU with Blackwell architecture and a 20 -core CPU grace with ARM architecture. He works side by side with 128 GB of unified DDR5X type and low consumption type, although NVIDIA available memory standards do not seem to convince him for this scenario of use.

Nvidia leads the development of the Socamm memory standard

Projects Digits is just the spearhead. Presumably in the future NVIDIA will launch other personal computers even with greater capacities in The field of AIbut it seems that they will not use any of the memory technologies currently available. According to Sedailythe company headed by Jensen Huang has allied with South Korean companies Samsung and Sk Hynix, and also with the American Micron Technology, to develop a new memory standard known as Socamm (System on Chip Advanced Memory Module).

These three companies are The biggest memory chips manufacturers of the planet, so there is no doubt that they are the best allies to which Nvidia can resort. Especially if the Socamm standard is being devised without Jedec’s participation (Joint Electron Device Engineering Council), which is the global organization that It is responsible for the development of standards used by the semiconductor industry and microelectronics. Apparently it is just what is happening: Nvidia, Samsung, SK Hynix and Micron for the moment are not having Jedec.

In any case, what we know right now about Socamm memories, beyond the fact that they will be used in the next batch of personal computers for AI, paints very well. Although this information has not been officially confirmed by any of the companies involved in their development, it seems that the Socamm modules are being designed on the basis of LPCAMM memories (Low-Power Compression Attached Memory Modules).

According to Sedaily, the Socamm standard will be very efficient from an energy point of view; It will have more I/O ports than conventional LPCAMM and DRAM memories (up to 694 ports); will allow to easily expand memory initially installed in the equipment for AI; And finally, these modules will have a physical size, which not storage capacity, much more restrained than conventional dram modules. If this standard contributes to the hardware for the most accessible to all, welcome. However, it is still early to trust that it will be so.

Image | Nvidia

More information | Sedaily

In Xataka | The 20 most important personal computers in the history of technology