Microsoft DirectStorage 1.1: GPU quickly decompresses game files

Games should load faster with DirectStorage. Version 1.0 of Microsoft’s Low-Level API addressed CPU overhead and I/O performance. Microsoft has now announced the next step with DirectStorage 1.1, which introduces decompression of game data on the GPU. This ensures even better performance.

This is behind DirectStorage

The idea behind DirectStorage is to speed up the flow of data from mass storage to the graphics card. Game files should be transferred faster by better utilizing the performance potential of NVMe SSDs and decompressed directly on the GPU without going through the CPU, which, among other things, significantly reduces load times. Current video consoles have already proven it and the DirectStorage API is practically the answer to the Velocity architecture of Xbox Series X only now for PC.

In the first step with DirectStorage 1.0, only the transfer of data from the SSD to the graphics card was initially accelerated by making better use of the high transfer rates of today’s NVMe SSDs through more I/O requests. A lot of potential was wasted here before.

Step 2: Decompress via GPU

The second step follows with DirectStorage 1.1: the decompression of the files by the GPU. As is well known, graphics processors are predestined for parallel computing and can perform these tasks much more efficiently than CPUs. The latter is alleviated by the redistribution of work. Combined with the accelerated storage performance, game data should end up in the graphics card’s VRAM and thus on the player’s screen much faster.

With a “heavily optimizedAs an example, Microsoft is now demonstrating the potential of DirectStorage 1.1. Using GPU decompression with the new GDeflate compression standard to transfer the 5.65 GB game files with 667 models and 3335 textures, the process takes only 0.8 seconds. Throughput reaches around 7 GB/s and maximum CPU utilization is only around 15 percent.

In the right part of the image, on the other hand, the classic approach with the decompression of the same data per CPU (via zlib) illustrated. Here, the process takes 2.36 seconds, almost three times as long. According to this, the performance of 2.4 GB/s is only a third. The CPU, on the other hand, is used at 100 percent at its peak. How representative the demo is for upcoming games remains to be seen. Microsoft notes that performance may vary depending on workload and hardware.

Requirements for DirectStorage 1.1

However, Microsoft does not reveal which SSD, CPU and GPU were used in each case. To noticeably benefit from DirectStorage, an NVMe SSD with high performance should be used. In principle, it also works (with small advantages) with SATA SSDs; Hard drives, on the other hand, should hardly benefit as they are too slow.

On the GPU side, all models that support DirectX 12 and Shader Model 6.0 should be able to use GPU decompression. However, Microsoft recommends graphics cards that support DirectX 12 Ultimate. Partners AMD, Intel, and Nvidia are already working on drivers that will be optimized for the new technology.

DirectStorage with GPU decompression can be used on both Windows 11 and Windows 10. However, additional “I/O stack optimizations” will only be available on Windows 11, thus Microsoft makes more of an argument for its latest system. operative.

In the following video, Microsoft explains DirectStorage in detail:

First games next year

By the end of the year, Microsoft wants to give game developers DirectStorage 1.1 with GPU decompression. So the first games to benefit from this aren’t expected until next year. As is well known, the action RPG Forspoken is said to be one of the first games with DirectStorage support. This has since been pushed back to January 24, 2023. A tech trailer already hinted at the accelerated loading times.