Nvidia’s Big Accelerator Memory: an SSD for your GPU
We’re on the eve of the next generation of GPUs, but it’s hard to stay focused on what’s right in front of us. At the end of this month, Nvidia is expected to launch the highly anticipated RTX 3090 Ti, which has a good chance of being the most powerful gaming GPU we’ve ever seen. But the company has also teamed up with university researchers and IBM to unveil new technology that already has us looking to the future of graphics cards.
According to The Register, BaM, or less exciting, Big Accelerator Memory is Nvidia’s new technology that allows a GPU to transfer to a computer’s SSD without having to go through the CPU. This would allow graphics cards to operate much more independently of processors, while reducing the load on those processing chips. Conceptually, this seems somewhat similar to AMD’s approach with the Radeon Pro SSG which had 2TB of SSD attached via M.2, which had a greater focus on gamers and content creation.
The case document for BaM explains the benefits of using this new system:
“We show that (1) BaM infrastructure software running on GPUs can identify and communicate fine-grained accesses at a high enough throughput to fully utilize the underlying storage devices, (2) even with large SSDs. public, a BaM system can support application performance that is competitive with a much more expensive DRAM-only solution, and (3) the reduction in I/O amplification can provide significant performance benefits .
All of this means that GPUs wouldn’t necessarily have to rely as much on their own GDDR6 memory when they have direct access to an SSD, and it removes the CPU middleman, which speeds up transfers and performance. It also doesn’t require a proprietary API like Microsoft’s DirectStorage solution. Instead, these GPUs will transfer data using PCIe, RDMA, and a specially designed Linux kernel driver. This allows SSDs, even consumer-grade ones, to read and write directly to the GPU. All in all, it looks like a win for computing performance across the board.
Being a GPU-based development, it’s easy to get a gaming-centric view of this technology, but that’s not necessarily the case. While it’s very likely that the efficiency improvements will lead to benefits for PC gaming, where it’s really likely to shine is for complex and large workloads. These tasks are where the GPU having direct access to tons of SSD storage is most likely to benefit. But that doesn’t mean I don’t want to see him render really nice video games either.