New tech promises to add terabytes of memory (or SSD storage) to any GPU — but there's a big catch

Panmnesia CXL controller
(Image credit: Panmnesia)

KAIST startup Panmnesia (the name means “the power to remember absolutely everything one thinks, feels, encounters, and experiences”) claims to have developed a new approach to boosting GPU memory.

The company's breakthrough allows for the addition of terabyte-scale memory using cost-effective storage media such as NAND-based SSDs while maintaining reasonable performance levels.

However, there's a catch: the technology relies on the relatively new Compute Express Link (CXL) standard, which remains unproven in widespread applications and requires specialized hardware integration.

Technical challenges remain

CXL is an open standard interconnect designed to efficiently connect CPUs, GPUs, memory, and other accelerators. It allows these components to share memory coherently, meaning they can access shared memory without requiring data to be copied or moved, reducing latency and increasing performance. 

Because CXL is not a synchronous protocol like JEDEC’s DDR standard, it can accommodate various storage media types without requiring exact timing or latency synchronization. Panmnesia says initial tests have shown that its CXL-GPU solution can outperform traditional GPU memory expansion methods by more than three times.

For its prototype, Panmnesia connected the CXL Endpoint (which includes terabytes of memory) to its CXL-GPU via two MCIO (Multi-Channel I/O) cables. These high-speed cables support PCIe and CXL standards, driving efficient communication between GPU and memory.

Adoption, however, may not be straightforward. The GPU cards may need additional PCIe/CXL-compatible slots, and significant technical challenges remain, particularly with the integration of CXL logic fabric and subsystems in current GPUs. Integrating new standards like CXL into existing hardware involves ensuring compatibility with current architectures and developing new hardware components, such as CXL-compatible slots and controllers, which can be complex and resource-intensive.

While Panmnesia's new CXL-GPU prototype potentially promises unparalleled memory expansion for GPUs, its reliance on the emerging CXL standard and the need for specialized hardware could create obstacles for immediate widespread adoption. Despite these hurdles, the benefits are clear, especially for large-scale deep learning models which often exceed the memory capacity of current GPUs.

Panmnesia CXL-GPU prototype

(Image credit: Panmnesia)

More from TechRadar Pro

Wayne Williams
Editor

Wayne Williams is a freelancer writing news for TechRadar Pro. He has been writing about computers, technology, and the web for 30 years. In that time he wrote for most of the UK’s PC magazines, and launched, edited and published a number of them too.

Read more
d-Matrix Corsair card
Tech startup proposes a novel way to tackle massive LLMs using the fastest memory available to mankind
Artificial Intelligence
Here's why 100TB+ SSDs will play a huge role in ultra large language models in the near future
Sandisk 3D Matrix Memory
Sandisk's revolutionary new memory promises DRAM-like performance, 4X capacity at half the price
HBM vs HBF memory
A GPU or a CPU with 4TB HBM-class memory? Nope, you're not dreaming, Sandisk is working on such a monstrous product
CAMM2 modules
CAMM2 memory modules promise significant advancements in memory technology with impressive read and write speeds
Half man, half AI.
Yet another tech startup wants to topple Nvidia with 'orders of magnitude' better energy efficiency; Sagence AI bets on analog in-memory compute to deliver 666K tokens/s on Llama2-70B
Latest in Pro
Finger Presses Orange Button Domain Name Registration on Black Keyboard Background. Closeup View
I visited the world’s first registered .com domain – and you won’t believe what it’s offering today
Racks of servers inside a data center.
Modernizing data centers: an efficient path forward
Dr. Peter Zhou, President of Huawei Data Storage Product Line
Why AI commonization is so important for business intelligent transformation and what Huawei’s data storage has to offer
Wix automation
The world's leading website builder aims to save businesses time with new tool
Data Breach
Thousands of healthcare records exposed online, including private patient information
China
Juniper patches security flaws which could have let hackers take over your router
Latest in News
Google Pixel 8a in aloe green showing
Google Pixel 9a benchmark link teases the performance of the upcoming mid-ranger
Quordle on a smartphone held in a hand
Quordle hints and answers for Monday, March 17 (game #1148)
NYT Strands homescreen on a mobile phone screen, on a light blue background
NYT Strands hints and answers for Monday, March 17 (game #379)
NYT Connections homescreen on a phone, on a purple background
NYT Connections hints and answers for Monday, March 17 (game #645)
Apple iPhone 16 Pro HANDS ON
Leaked iPhone 17 dummy units may have given us our best look yet at all four models
A super close up image of the Google Gemini app in the Play Store
It's official: Google Assistant will be retired for phones this year, with Gemini taking over