DirectStorage is Microsoft's solution to modernizing how SSDs are used in games. As it stands, the SSDs of today are very limited by software when it comes to data transfer speeds, and DirectStorage is simply a way to remove that bottleneck and get games to use the full force of NVMe/PCIe drives. It even allows your GPU to access your SSD directly, bypassing your CPU entirely. You've probably heard that this is going to be great for cutting down loading times, and that's where most of the rhetoric has been focused.

But DirectStorage isn't just about loading times, and in fact, I'd argue loading times aren't even the primary focus of this new technology. What DirectStorage is really about is attaching your SSD to your GPU, and increasing graphics performance for higher framerates. This could be really great for PC gaming and allow gamers to take full advantage of PCIe 5.0 SSDs (which are currently not that useful for gaming), or it could also be yet another thing that makes modern GPUs even more abominable.

How DirectStorage can make your SSD as important as your CPU and GPU

MSI-Spatium-M570 installed on a motherboard.

Obviously, SSDs and GPUs are pretty different pieces of hardware. You store games on one and run games off the other, and it's not really obvious that a faster SSD means better gaming performance outside of loading times. But all GPUs have one thing: memory, or more specifically VRAM. At its core, VRAM is just a data storage medium, which is what an SSD is, and while GDDR is pretty different from NAND flash, they both have one thing in common: bandwidth.

Memory bandwidth is a key specification for GPUs. It's determined by a combination of memory bus width, the frequency of memory chips, and how many memory chips are in use since utilizing an entire memory bus requires a certain number of chips. Low-end GPUs usually have at least 100GB/s of bandwidth at their disposal, and higher-end GPUs can just touch 1TB/s. The bandwidth of SSDs is just what transfer speeds they can hit, which is something that SSD manufacturers tend to advertise.

Now, you might be aware that PCIe 5.0 SSDs will theoretically cap out at 16GB/s, a far cry from even 100GB/s, but that's just raw data. The secret sauce of DirectStorage is the fact that it can use the GPU rather than the CPU for data decompression, so compressing data straight from the SSD and giving it to the GPU is now possible. This also has an additional side effect of reducing the load on the CPU.

Microsoft hasn't yet announced how much compression DirectStorage will be capable of, but from what I can gather it will probably be five times the compression or a little more. That's essentially five times the bandwidth, which means the fastest PCIe 5.0 SSDs of the future can offer up to 80GB/s of extra bandwidth.

For lower-end and midrange GPUs, this is a pretty big deal. 80GB/s of extra bandwidth would be a 30% improvement for the RTX 4060, and roughly 25% improvement for the RX 7600. The benefit is less clear for high-end GPUs, but it would at least be a little extra. PCIe 6.0 SSDs will increase this to 160GB/s of extra bandwidth, and that'll be an even bigger deal. Heck, a PCIe 4.0 SSD in a handheld gaming PC would be a massive boost for integrated graphics, which usually require high-end RAM to perform well, even in the case of the Steam Deck.

DirectStorage could be a great thing, or another reason to impede future GPUs

The Nvidia RTX 4060 Ti graphics card.
Source: Nvidia

One thing I'm worried about with DirectStorage is that it could become less of a nice-to-have sort of feature and more of a your-performance-will-be-awful-if-you-don't-have-a-fast-SSD sort of gimmick. One reason why I'm particularly worried about this is that GPU manufacturers (mainly Nvidia and AMD here) have been trying to save money by slimming down the memory on GPUs for a while now, and while much of that has to do with capacity, it's also about bandwidth.

In recent years, we've seen some pretty fast GPUs come out with shockingly small memory bus widths. For instance, 256-bit or even 192-bit buses tended to be pretty standard for the midrange, but now midrange cards like the RX 7600 and the RTX 4060 are at just 128-bit buses, which used to be reserved for some of the slowest GPUs in many series. Even the RTX 4090 has only a 384-bit-wide bus, the same as the RTX 3080 Ti.

Of course, there are good reasons for Nvidia and AMD to want to cut down bus width: it makes GPUs smaller and thus cheaper to produce, and wider buses require more memory chips to populate the full width and get the full performance. Still, this is ultimately a cost-saving measure, and when saving as much money as possible is the goal, it ends up creating incentives to compromise a product's integrity just shy of crossing the line. Nvidia has done that with its VRAM on recent GPUs, and in some games, you can actually run out of VRAM on a $400+ GPU like the RTX 3070.

My biggest concern is that DirectStorage will become a requirement to get a normal amount of performance out of GPUs if this technology becomes the standard or even just widespread enough. It would be like an even worse version of the RX 6400, a low-end GPU that requires PCIe 4.0 for good performance because AMD put the bare minimum amount of lanes on it. If everyone needs a top-end PCIe 5.0 SSD to get a GPU running as expected, it'll be yet another barrier to entry for gamers with less means to buy hardware.

The technology is undeniably cool though

While I am worried about what incentives DirectStorage might create for GPU design, I do think it's something that could have the power to really improve gaming. Obviously, game devs will be able to extract lots of use out of DirectStorage in the Xbox, where the hardware is easier to work with, but from what I've heard it sounds like this technology will be plug-and-play, so it should work well on PC with a variety of hardware configurations.

I'm particularly excited about what DirectStorage could do for integrated graphics like in handheld gaming PCs. The ROG Ally has only about 100GB/s in memory bandwidth, and DirectStorage could add an extra 25GB/s or so with the fastest 2230-sized PCIe 4.0 SSDs. That could be up to 40GB/s if faster 2230-sized drives come out. Think about how fast the iGPU in your laptop or mini PC could get with a fast PCIe 4.0 or 5.0 SSD (once they're cool enough to run without heatsinks). It could really boost the viability of low-end, integrated GPUs for gaming, and I'd love to see that.