What is Nvidia RTX IO? Guide and Reveal

The Reveal of RTX IO

After months of teases and waiting, Nvidia finally unveiled its new generation of graphics cards, the RTX 3070, 3080, and 3090. Alongside the announcement of these impressively powerful and comparatively inexpensive cards, Nvidia talked about some tech coming alongside them. Perhaps the most interesting of these was RTX IO, Nvidia’s answer to the growing sizes of games and capabilities of NVMe SSDs.

Right now, your PC rarely, if ever, uses its NVMe SSD at full capacity. Unless you’re working in heavy content creation, you’re not getting the most out of it, and even then you might not be. The APIs that manage storage can’t keep up with the speeds NVMe drives are capable of, especially on PCIe Gen 4.

And so here comes RTX IO. Nvidia says that RTX IO is a “suite of technologies that enable rapid GPU-based loading and game asset decompression, accelerating I/O performance by up to 100x over traditional hard drives and storage APIs.”

What’s DirectStorage?

RTX IO is, in short, the meeting of Nvidia’s RTX hardware and Microsoft’s DirectStorage API. DirectStorage for Windows is the vaunted Xbox Velocity Architecture, but for PC. Microsoft explains that current storage APIs were never optimized for the high number of IO requests made by modern games. “Even with super-fast PC hardware and an NVMe drive, games using the existing APIs will be unable to fully saturate the IO pipeline, leaving precious bandwidth on the table.”

DirectStorage is pretty early on, and many PC configurations won’t support it right now, Microsoft says. A similar comparison for that might be the PlayStation 5’s storage expansion solution. Sony will let you drop in your own NVMe SSD, but it has to be a Sony-approved one for PlayStation 5 games to load off it to ensure there’s enough bandwidth available to function as expected. Similarly, lower-quality NVMe SSDs, certain CPUs, and other hardware may act as bottlenecks that would prevent DirectStorage from functioning.

Even without Nvidia’s RTX IO, DirectStorage promises to be huge for PC gaming as it expands the bandwidth from hundreds of IO requests across a 50MB/s streaming budget to a GB/s budget and tens of thousands of IO requests per second.

Summed up, DirectStorage gives developers direct and less-restricted access to storage, letting them make many more IO requests at any given moment than they can with existing storage APIs.

RTX IO takes out the middle man

Nvidia then takes this and does what every podcast advertisement you’ve heard for mattresses and underwear seems to promise: it cuts out the middle man. RTX IO uses DirectStorage’s increased IO bandwidth and the much faster storage of NVMe SSDs and puts the RTX card in charge of them, offloading “dozens of CPU cores’ worth of work to your RTX GPU” according to Nvidia. It takes the data from your storage and puts it directly in the much faster RTX memory and gives the data to the RTX GPU to sort out.

The RTX GPU will be able to pull compressed data directly from your NVMe storage, where its in its smallest and easiest-to-move form, and let the RTX card decompress it. It skips the CPU and system memory entirely. So with this, your rig will be moving smaller amounts of data and then decompressing them on faster, bespoke hardware; each element speeds up the process of loading both in loading screens and when streaming data from the drive in things like open-world games.

Nvidia says RTX IO will enable “near instantaneous game loading,” improved frame rates, and reduced texture pop-in and game stuttering. Microsoft and Nvidia say that this technology–both DirectStorage on its own and RTX IO as well–will allow games to use heavier compression, reducing overall game size as well.

Future Technology (slow-burn technology)

This is going to be somewhat of a slow-burn technology, though. Nvidia hasn’t said for sure that it’ll be available on RTX 20-series cards, though that seems like it’s very possible. But it’ll also require greater adoption of NVMe SSDs as PC storage solutions and of PCIe Gen 4. It doesn’t seem like PCIe Gen 4 is required, though.

It’ll also require game developers to be on board, to some degree. This stuff works in the upcoming PlayStation 5 and Xbox Series X because developers can target the hardware and APIs in those systems directly. They know what hardware will be available and what that hardware is capable of. A game developer can’t develop a game for a person with an RTX card and a high-end NVMe SSD when there are people out there with AMD cards and games stored on rotational media. AMD will have to present its own IO solution, too, then. That also seems likely considering that AMD is behind the SoCs in the upcoming consoles.

It’ll be huge when it gets here

For those of us with brand new systems, we’ll likely see improvements from RTX IO and DirectStorage immediately. The improvements will only continue, though, as developers start expecting that decompression hardware to be present on GPUs and can build games with it in mind. It seems likely, too, that a suite of other technologies and APIs will begin to accompany this. Developers could have an SSD-optimized installation for those with the hardware to do it, and a bulkier standard installation for the rest of us. It could be something we enable manually or it it could be automated.

There’s no doubt that Nvidia’s 30-series GeForce RTX video cards are a huge jump over the 20-series cards, but RTX IO could be the low-key game changer here in terms of how we play and enjoy games.

4.9 18 votes
Article Rating
Subscribe
Notify of
guest

18 Comments
Newest
Oldest Most Voted
Inline Feedbacks
View all comments
mike
mike
2 years ago

Why stop there, let’s put an nvme socket on the GPU

Gastec
Gastec
2 years ago
Reply to  chuckula

The CPU “shunt” is a.k.a latency/lag.

Adnan
Adnan
2 years ago
Reply to  tfp

Which would be weird because outside of the blistering fast SSD, the PS5 is inferior to the Xb1X specs wise.

tfp
tfp
2 years ago
Reply to  Sweatshopking

Maybe they are unimpressed with the Xbox Next but like the PS5?

tfp
tfp
2 years ago
Reply to  Krogoth

It seems to me like you are working under the assumption that games are already written to decompress graphics related data to System RAM and games use as much RAM as is available or reasonably available at an OS level to do this. I do not believe this is the case with most games in general and I can’t think of one game off hand that does this effectively so load times drop as system memory massively increases.   Then there is the consideration of memory management once the graphics data is moved to your video card, does the game continue… Read more »

JohnnyNismo
JohnnyNismo
2 years ago

This is the most exciting new technology for me. Microsoft’s DirectStorage API will revolutionize PC gaming. We can all be pretty certain that AMD will have this tech in RDNA 2 since the XBX said they’re using it. The next few years of gaming will be interesting.

Jared Hohman
Jared Hohman
2 years ago

What’s to stop current game engine from doing this? Also, aren’t textures already sent to GPU compressed?

Krogoth
Krogoth
2 years ago
Reply to  tfp

How so then? It isn’t like UDIMM capacity is a problem on desktop systems. 32GiB is easily obtainable and doesn’t break the bank. That gap will grew with next generation platforms. UDIMM also has far superior throughput than flash. Throughput on PCIe 4.0 SSD media is barely on par with JEDEC-spec DDR2.
 
Like I said, it is a solution looking for a problem outside of certain niches like Optane memory.
 
 

Sweatshopking
Editor
2 years ago
Reply to  The lost cat

Personally, I’m surprised how every comment I’ve seen mentions the ps5’s memory but not the xbox’s which has been detailed and this is literally carried over from.

The lost cat
The lost cat
2 years ago
Reply to  Krogoth

Or, just putting their own name on DirectStorage, which was announced with the Xbox Series X and hasn’t gotten to PC yet.

Krogoth
Krogoth
2 years ago
Reply to  chuckula

It seems like it is really meant for professional crowd. Nvidia marketing is trying to spin it off to gaming crowd and trying to “Metoo” on PS5’s memory hierarchy.

RandomGenericDude
RandomGenericDude
2 years ago
Reply to  tfp

I actually agree with you but a comment that simply says wrong is just “wrong” 😛
 
Educate Krogoth by explaining that any solution that frees up the CPU from decompression and memory accesses means that it has more time to spend on other stuff so it can only be a boon.

chuckula
chuckula
2 years ago
Reply to  Keith Patrick

It sort of is, but from the vague description being provided it’s more GPU-specific for the purposes of loading compressed texture data directly to the GPU. The bytes still have to shuffle through the CPU’s PCIe controller due to the point-to-point nature of PCIe, but other than acting as a shunt for bits the CPU is basically hands-off, which is supposed to boost the performance.

Keith Patrick
Keith Patrick
2 years ago

It sounds like DMA

tfp
tfp
2 years ago
Reply to  Krogoth

Wrong

Krogoth
Krogoth
2 years ago

This looks like another solution for looking for a problem outside of certain niches. Upcoming PS5/XSX have small memory pools which is why they are even bothering with such solutions.
 
 

KeillRandor
KeillRandor
2 years ago

Impressively powerful? 30% over a 2080ti for the 3080 is impressive? We’ll see what their ray-tracing is like, I suppose…
 
As for the price – that’s because their 2k series was overpriced to hell – which shows how important it is for everyone that AMD sort their, errr, stuff out.

chuckula
chuckula
2 years ago

io_uring: Less marketing, moar powah.
 
https://lore.kernel.org/io-uring/[email protected]/T/#u

18
0
Would love your thoughts, please comment.x
()
x

Pin It on Pinterest

Share This

Share this post with your friends!