Faster Than A Speeding Velocipede
Microsoft have shared more details on just how they managed to increase the I/O of the new Series X over 70 fold from a mere 68MB/s to 4.8GB/s. As you likely suspect, the use of an NVMe SSD for storage is a major part of the boost, but it simply can’t explain it all, indeed according to Microsoft it accounts for a boost of 2.4 GB/s of raw I/O throughput. Microsoft also looked at how to use hardware to more effectively stream textures, which has a huge effect on the very specialized purpose of the console.
A new algorithm called BCPack will run in tandem with the more commonly known LZ decompressor, hardware accelerated specifically to deal with moving textures from memory to your eyeballs. The most unique solution they implemented is called Sampler Feedback Streaming, which Microsoft developed for the XBox Series X to make transferring textures from storage to the GPU significantly more efficient. They determined a way to break up textures in a way that allows them to send only as much data as is necessary for the GPU to properly render on screen.
Previous generations loaded the entire texture from memory, as the architecture precluded the ability to load only partial texture data and this also helps provide the impressive boost we will see on the new XBox Series X. Ars Technica delves into a few of the other tricks that Microsoft implemented which you can check out here, to keep you busy until July 23rd.
Back in March, Microsoft announced what it was calling the "Xbox Velocity Architecture" for the Xbox Series X, and we got excited about the prospect of data streamed from long-term storage at advertised rates up to 4.8GB/s (compared to just 68MB/s on the Xbox One). Today, Microsoft offered some new details about how, exactly, the company got such high speeds from the hardware's I/O systems.