Graphics card producers are always looking for new ways to improve their GPUs, and a new technology that could be coming might be one of the most impressive improvements yet. Experiments are showing that the GPU SCM with DRAM is able to produce gains of up to 12.5x faster than the typical high-bandwidth memory.
SCM, or storage class memory, is seeking to improve memory bandwidth within GPUs. While mainly used within RAM sticks and SSDs, this technology could make the leap over to GPUs. The research was conducted by the POSTECH research team at Soongsil University. They found that a variation on Intel’s 3DXPoint memory, SCM was more powerful and cheaper so long as a DRAM cache and hardware-based memory tagging system was set up. It was tested on an Nvidia A100 GPU, and performance gains of up to 12.5x faster were found.
Cheaper and better GPUs could be on the way with new tech experiments
Not only was the GPU 12.5x faster, but it also had an 89.3% power reduction. With these combined, the technology could be used to create better graphics cards at lower prices. This kind of power is impressive, and with fierce competition within the graphics card market, companies will be eager to get their hands on it. It could even replace HBM altogether, which was recently used on the AMD Vega lineup of GPUs.
The technology is very promising. However, it’s still in its infancy. The early stages are just taking place for SCM research in GPUs. It will be a long time, if ever, that we see it coming to actual products. However, with such impressive results, it’s likely we’ll see more developments. GPUs are one of the most essential components of the best gaming PCs, so it makes sense that more research will be conducted. Better technology is always on its way, and SCM is just one of them.
Until then, gamers will be waiting patiently for more news. HBM is still costly to make and comes with drawbacks such as overheating. Hopefully, more experiments will be able to solve these problems.
