NVIDIA plans to move to multi-GPU design for their future graphics cards

GPU manufacturers are taking a look at a new approach. Rather than creating even larger GPU dies for their future cards the proposed design will feature several smaller dies that work together. NVIDIA has recently talked about MCM package (Multi-Chip-Module package) where multiple GPU, memory, and controller chips are placed on one single chip.

NVIDIA has tried this approach but not on a consumer-grade product. The MCM design proposes using multiple smaller GPU dies and lots of DRAM dies which will be connected through ridiculously fast I/O.

You can check all available NVIDIA products over here: http://amzn.to/2inSKeA

The company tested this design by simulating the performance of 256 streaming multiprocessors based on MCM with 64 SMs per GPU (four GPUs total). This resulted in 16 383 CUDA cores which is three times as much as the flagship Volta GPU. The performance gain of the MCM-GPU was 45.5% compared to te monolith GPU.

AMD is already using a similar technology for their upcoming EPYC processors. They will also follow the same approach for their next generation GPU architecture set to debut next year.

This may be the next big leap in the GPU world allowing much faster and better performing graphics cards than the current large die design.

Meanwhile, you can check all available NVIDIA products over here: http://amzn.to/2inSKeA

Abonnieren
Benachrichtigung von
guest
0 Comments
Inline-Rückmeldungen
Alle Kommentare anzeigen