Next-Generation NVMe SSD and I/O Technology [PC, PS5, XBSX|S]

Which might come to Windows PCs well past the RTX 3000's active life on the shelves, but that's probably a different discussion.

You're right. The discussion is about the technical capability of RTXIO, not when it will be released.

Is nvidia's own marketing material enough evidence for you?

a3Bz83v.jpg

As has already been discussed at length, the slide does not explicitly state that data transfers are done via P2P DMA direct from SSD to GPU without any involvement whatsoever from the CPU. While that may be the case, it may also be the case that the slide is put together like that to represent the reduced workload on the CPU rather than it's elimination entirely. As it stands, no special CPU/motherboard requirements have been mentioned despite the question being asked and details of other requirements being given. Time will tell of course if there are additional requirements, but at present there's no basis for assuming that there will be. And even if there are, it's likely most modern systems will already be supported. We already know all Zen and newer AMD systems support P2P DMA over the root complex.


Because not every compression format gains performance with parallel execution. Kraken uses 2 threads (!!!) per file, meaning you're either decompressing hundreds/thousands of different textures at the same time or you don't gain anything from running it through GPGPU.
And now you're just assuming RTX graphics cards can decompress BCPack through GPGPU just because they're both related to DirectStorage, without knowing if BCPack decompression is even effective through GPGPU. Microsoft is using dedicated decompression hardware for BCPack, so where's the evidence that BCPack is effective on GPGPU?

Please don't try to put words in my mouth. I haven't assumed that Nvidia are using anything specific. I've merely presented an alternative compression option that achieves a 2:1 compression ratio in response to your implication that it's not possible outside of Oodle based solutions.

The fact is that Nvidia are claiming a 2:1 compression ratio will be attainable at 14GB/s output on their GPU's. That means that whatever compression scheme they're using, it's parallelizable for execution on the GPU. Whether that's BCPACK, Kraken, or something else entirely is irrelevant to the core point. So again, unless you're claiming that Nvidia are lying, I don't see what there is to argue about here.

Of course it's relevant. Are the 14GB/s achievable using a widely adopted industry standard in the PC space? With how many files decompressing at the same time (i.e. how dependent on parallel execution is that throughput)? Or is it only achievable using a nvidia-proprietary texture format that decompresses at 14GB/s when using a nvidia GPU but then can only decompress at 2GB/s if you're using a GPU from another vendor?
nVidia purposely omitting the texture compression format can be a very telling move on their part. They don't even need to be lying for the real-life performance of RTX IO being completely different to their statements and/or having repercussions to the PC ecossystem.

So your argument essentially boils down to: "14GB/s is possible with RTXIO but only if developers use the right compression scheme that allows them to achieve a 2:1 compression ratio" okay.... I think we agree then.

You can argue all day long that no games will support that compression scheme, or that it'll be proprietary, or that it won't work well with other vendor GPU's, but as that's entirely baseless speculation I don't see much point in discussing it. Better to wait and see what happens.
 
And if you use Oodle Texture with RTX IO? Or if you use some other type of texture encoder that uses RDO optimization? What RAD is using to increase compression ratios isn't exclusive to Oodle Texture or the PS5. There are a number of encoders that support different super compression techniques.
Beside's RAD's other compression formats, Oodle Texture AFAIK works with Zlib (which can only use one thread per file BTW). Not knowing what compression formats RTX IO can use means


So your argument essentially boils down to: "14GB/s is possible with RTXIO but only if developers use the right compression scheme that allows them to achieve a 2:1 compression ratio" okay....
No, my argument is:
1 - For all you know 14GB/s may only be possible in an unrealistic scenario that will never come to light (well it does hit 14GB/s* ), and
2 - only through draconian anti-consumer methods (let's get developers to use our proprietary compressor that runs on a single thread made of x87 instructions if a non-nVidia GPU is detected)

*if you're loading 2000 textures at the same time

Yet people are comparing RTX IO to the new consoles as if "nVidia already has something better Pc MaStErRaCe RuLeS". The later which is of course the messaging that nvidia's marketing teams want to pass.
 
No, my argument is:
1 - For all you know 14GB/s may only be possible in an unrealistic scenario that will never come to light (well it does hit 14GB/s* ), and
2 - only through draconian anti-consumer methods (let's get developers to use our proprietary compressor that runs on a single thread made of x87 instructions if a non-nVidia GPU is detected)

*if you're loading 2000 textures at the same time


Okay, so you are saying that Nvidia is lying. Or at least being quite deceitful in it's claims about RTX IO performance. And you're basing this conclusion on.... pure speculation. So not a technical discussion anymore then.

At least we seem to have moved on from your original position of "Nvidia's 14GB/s is comparable to Sony's 22GB/s though. So progress is being made.

Yet people are comparing RTX IO to the new consoles as if "nVidia already has something better Pc MaStErRaCe RuLeS". The later which is of course the messaging that nvidia's marketing teams want to pass.

The comparison that is being made rightly compares the claimed compressed throughput from each vendor. Neither has actually demonstrated real world throughput measurements / benchmarks (showing short or zero load times in games where we have no idea how much data is being transferred doesn't tell us anything about transfer rates), yet you seem to be taking Sony's claims at face value while assuming Nvidia is being deceitful. I can't imagine why someone would want to do that.
 
Beside's RAD's other compression formats, Oodle Texture AFAIK works with Zlib (which can only use one thread per file BTW). Not knowing what compression formats RTX IO can use means

Decompressors can use general-purpose CPUs or specially designed for proprietary hardware like the PS5. If GPU decompression becomes a thing with DirectStorage, then it would advantageous for RAD to support GPU based decompressors. It would allow them to support multiplatform games across their entire development environment. Even then there is nothing stopping Nvidia from supporting RDO optimized texture decompression on its GPUs.

Yet people are comparing RTX IO to the new consoles as if "nVidia already has something better Pc MaStErRaCe RuLeS". The later which is of course the messaging that nvidia's marketing teams want to pass.

Even if Nvidia's RTX IO offered 1400 GB/s, PC developers would be forced to support what the average PC gamer uses, so dev won't initially exploit RTX IO to the fullest (outside of loading or fast travel). That would require that type of performance to be standard across the market which it will never be anytime soon. Not in a way that a first party Playstation dev can when making a PS5 game.

Games that fully exploit the SSD will show up on the PS5 first followed by the Xbox. It will happen on the PC when a vast majority of the PC gamers moved to an SSD.
 
Last edited:
I do not know where this graphic comes from. The OP on reddit did not say. This was for a 49GB game.


I understand the HDD numbers as the write and read speeds may be different and govern the transfer speed dependent on the scenario. But why the varying numbers when transferring between SSDs? Isn't the write speeds of the drives governing the transfer speed and the write speeds should be the same if we are talking about the internal and expansion XSX drives.

Otherwise, wouldn't it indicate a difference in performance on the XSX. Games will write out faster to the internal vs. the external affecting features like quick resume.
 
Last edited:
I understand the HDD numbers as the write and read speeds may be different and govern the transfer speed dependent on the scenario. But why the varying numbers when transferring between SSDs? Isn't the write speeds of the drives governing the transfer speed and the write speeds should be the same if we are talking about the internal and expansion XSX drives.

Depends on the exact model of external SSD -- most external USB SSDs write speeds are substantially less than the read speeds. At least I'm assuming they're using normal USB SSDs and not the NVME Card. But then again they slightly mislabel the internal storage too. So I can't tell exactly what they're doing.
 
Guilty until proven innocent.

I wouldn't want to live in your world. Sounds scary.

I was merely stating what some are doing the whole time. We have nothing to test or go by other then PR and slides from NV, Sony, MS etc. If NV is to be lying, so is Sony.

You can argue all day long that no games will support that compression scheme, or that it'll be proprietary, or that it won't work well with other vendor GPU's, but as that's entirely baseless speculation I don't see much point in discussing it. Better to wait and see what happens.

Those are his wishes, what he ultimatetly wants to happen. It's his vision, and discusses after that.

Yet people are comparing RTX IO to the new consoles as if "nVidia already has something better Pc MaStErRaCe RuLeS". The later which is of course the messaging that nvidia's marketing teams want to pass.

Ah, theres where the problem lies, i see the anger. Your plastic box just got surpassed by the PC which always dynamically evolves. Well, your doing the wrong race, your always gonna loose out.
 
Depends on the exact model of external SSD -- most external USB SSDs write speeds are substantially less than the read speeds. At least I'm assuming they're using normal USB SSDs and not the NVME Card. But then again they slightly mislabel the internal storage too. So I can't tell exactly what they're doing.

Must be a USB drive. As 49 GBs isn't a whole lot considering we expect the XSX to write out 13.5 GBs of data to the SDD followed by reading in 13.5 GBs out of data in under 10 seconds. While these SDD to SDD transfer rate are ~180-350 MB/s which would take at least 40 secs for 13.5 GBs to be written out to the SDD.
 
Must be a USB drive. As 49 GBs isn't a whole lot considering we expect the XSX to write out 13.5 GBs of data to the SDD followed by reading in 13.5 GBs out of data in under 10 seconds. While these SDD to SDD transfer rate are ~180-350 MB/s which would take at least 40 secs for 13.5 GBs to be written out to the SDD.
If accurate, this is slower than I hoped for but does demonstrate that there is a signifiant advantage to using a SSD. What I'm also after is load times from an external HDD and SSD. If the load times are also a third I think I'll be investing an external SSD, rather than a HDD.
 
If accurate, this is slower than I hoped for but does demonstrate that there is a signifiant advantage to using a SSD. What I'm also after is load times from an external HDD and SSD. If the load times are also a third I think I'll be investing an external SSD, rather than a HDD.

BRiT has you covered

The good news is if you have a USB SSD and plenty of unpatched games, you won't need to load them on the NVME to see benefits. There obviously are bottlenecks elsewhere, as the performance numbers using USB SSD should be ~5x slower than internal NVME, but performance is limited from other aspects.

Extrapolating what Borderlands3 loading times would be if I used my external SSD on Series X ends up at some impressive numbers... so much where I thought I was doing the numbers wrong. Then I looked at ArsTechnica and noticed an update to their article, which aligns with my rough estimates.



Update from Ars Technica

[Update, 7pm EDT: Since this article's publication, I've re-run this gamut of tests upon confirming that the external USB 3.1 drive I'd used on both Xbox One X and Xbox Series X was not the highest-speed drive I had available. I blame my mix-up on how my slim Seagate drive runs silently and outperforms Xbox One X's built-in mechanical drive. I've since added a fourth loading time to each test, as run on a WD Blue 1TB 3D NAND drive connected via a USB 3.1 adapter.

The newly added scores (marked in gray) are for the faster drive as connected to Xbox One X. Notice I didn't list scores for this drive on Series X. That is because its results are nearly identical to Series X's built-in NVME 4.0 drive, within a margin of error of 1-2 seconds. This is very good news, should you wish to enjoy cheaper higher-speed storage for older software on Series X.]​


series_x_load_revise.001-1440x1080.jpeg
 
Wow is this the first time some of you folks seen a PC marketing slide? The joy of the flexibility of the PC platform for the marketeer is that any claim can be theoretically true with just the right parts. NV have long had a habit of showing only the most flattering of benches in slides (as do AMD and Intel PC GPU marketing is a dirty biz).

For me the proof in the PC as ever will be when anyone can test their tech away from NDAs and with release drivers. We're not that far from the days when renaming 3Dmark.exe would cost you 20% in scores. I suspect like most GPU innovations that long term this will be great but frankly you won't see any benefit for at least 2-3 years until it's a part of the stable Win10 releases outside TWIMTBP doing it for devs
 
I find it quite puzzling Nvidia didn't debut or show a demo of their RTX SSD/IO solution in action. They have access to Samsung memory/storage tech and more than likely has early access to Microsoft's DirectStorage API like most developers and/or driver development teams. I have this strange feeling something is missing at the motherboard level (bios) on facilitating or mapping these request.
 
@PSman1700, I hope Nvidia's RTX I/O solution is good, but for PC gaming it needs to be widely adopted to be more effective. Hopefully, MB manufacturers, Nvidia, AMD and Intel are creating some type of next-generation SSD/IO standards that aren't necessarily tied to one company's particular product's IP (that never works out well).
 
Last edited:
Just like every other SSD hyperbole PR then.

You're not furthering the conversation in a constructive manner. I'm all about PC gaming, but I'm not blinded by it either. Let's have actual articles, videos and talks surrounding such tech... that's bound to be great on all sides.
 
Yes, we will have to wait and see what happens. It will take time for this ssd tech to reach to the masses, but so do most new (and especially) expensive products.
It's most likely that, if you want the most sophisticated SSD solution that devs optimize for, you'd probably want a PS5 to enjoy those things this year. Anyway, lets see what happens ;) (before that, it's all based on what corporations promise us, and every one just assumes their favorite brand does the best and their opponent fails).
 
Back
Top