New Technology Offers Massive VRAM Boost for GPUs
A little-known company called Panmnesia, supported by the Korea Advanced Institute of Science and Technology (KAIST), has developed a groundbreaking approach to expanding GPU memory using the PCIe interface. This innovation surpasses existing solutions, especially in terms of latency.
How It Works:
Panmnesia's CXL IP technology allows expansion of VRAM on GPUs by utilizing SSDs and memory connected via PCIe. This can significantly increase the amount of available video memory, boosting performance in demanding tasks like machine learning and high-performance computing.
Key Advantages:
- CXL 3.1 Compatibility: Seamless integration with existing systems without major modifications.
- Deceptive Efficiency: The GPU's memory subsystem is "tricked" into believing it's using system memory when it's actually leveraging DRAM or NAND connected through PCIe.
- Unmatched Speed: Offers two to three times lower latency compared to existing solutions and Unified Memory Architecture (UWM). For context, Samsung's CXL-Proto has a latency of 250ns, while Panmnesia achieves double-digit nanosecond latency.
The Catch:
While promising, it's unclear whether AMD and NVIDIA will embrace this technology. Its adoption could potentially impact the sales of their high-end GPUs.
Learn More: More details about this development can be found on the Panmnesia website.