BlueXImpulse
Banned
slowest IO award
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
NVIDIA announces RTX I/O - PCIe 4.0, on-GPU storage decompression, collaberating with Microsoft on DirectStorage
- Thread starter Bernd Lauert
- Start date
slowest IO award
thicc_girls_are_teh_best
Member
Okay so ask yourself this: why would they go down on IPC gains from an older arch to a newer arch, when "even" AMD have made big gains from old to new?
It's an idea that fundamentally makes no sense, which is why it was ridiculous you posted that same idea to begin with. No company designs a new product to deliver worst performance than their previous product especially when the new one is meant to replace the older one altogether.
Craig avatar. Not surprised.
Last edited:
They don't.
Instead of saying: our arch is 1.5-2x more efficient because of IPC gains.
They just inflated all their numbers 2x for no reason.
But it's a better performance.
2080 is 10TF, 3080 is 15TF there is a +50% increase from TF and there are IPC gains so it's another +25-50% from these gains which gets us to the +75-100% over 2080
"would the industry follow with all that custom hardware or follow the Microsoft hybrid approach."
What's the hybrid approach? all you are practically arguing is that Playstation should use Direct X's directStorage just because nvidia has to support it in some form considering it's a PC card.
The hardware is all custom even on this nvidia GPU. The API abstraction layer is different as it has always been. I don't understand what you mean by the hybrid approach instead of custom hardware. You mean they should have just used slow drives with higher latency? What custom feature are you referring to that you think they shouldn't have included?
888
Member
Hell I’ll give you a windows 7 pro key that can be updated to 10 pro for free.
thicc_girls_are_teh_best
Member
Okay but hear me out: my issue is with you saying 30 Ampere TF = 15 Turing TF. Everything you're saying in this comment is completely contradictory to what you said in that earlier statement. What you said would imply that Ampere is worst than Turing despite being newer than Turing because even with higher numerical performance it is somehow delivering worst efficient and effective performance than the older architecture.
That literally makes no sense, unless you actually meant to say something more along the lines of 30 TF Turing = 15 TF Ampere, which would be more accurate (possibly a bit overboard, but much more on-point).
Also I just noticed this:
You're taking Sony's lossy compression ratio figure (3.99:1 or something like that) and comparing it to Nvidia's lossless compression figure (2:1). That's how they get 7 GB/s raw, 14 GB/s compressed. They didn't actually give a lossy compression figure for us to compare to Sony's.
...but seeing as how they are doing the equivalent of 24 Intel CPU cores (what type, we don't know. Could be i3 for all we know), I figure the ratio would be quite high. But that depends on what type of compression algorithm they are running.
The headroom is absolutely there, though.
Last edited:
Robot Carnival
Member
haha, thanks, but already got the key from M1chl's link. appreciate the thought thou.
I've noticed that "troll" posts get much more attention here. Think about it.
Not really. I'm comparing the algo implementation for NV and Sony. Where Sony uses 9 cores and NV - 24 cores.
The algorithm itself doesn't care about what compression rate was there, it just decompresses blocks of data.
But to decompress at higher speed you need a better algo, otherwise you will die a small death on each additional ps of latency.
And I couldn't resist to hint that NV sw engineers are worse than Sony's. Which is kinda true, because all the NV sw so far was pretty heavy (except DLSS which is kinda good, but DL is a new software tech, where programmer's skill doesn't matter, you can read Andrej Karpathy on that)
martino
Member
Last edited:
pasterpl
Member
So I was rewatching Nvidia's presentation and there are lots of features there very similar to what MS did with Xbsex;
Nvidia Reflex - Xbox Dynamic Latency Input
Nvidia DLSS/AI Cores - Xbox DirectML
RTX IO - it even works with DirectStorage
NVCache - using RAM and SSD to support VRAM (MS mentioned that XVA will offer memory multiplier)
Nvidia Reflex - Xbox Dynamic Latency Input
Nvidia DLSS/AI Cores - Xbox DirectML
RTX IO - it even works with DirectStorage
NVCache - using RAM and SSD to support VRAM (MS mentioned that XVA will offer memory multiplier)
LostDonkey
Member
200gb games huh....
I though games were getting smaller
I though games were getting smaller
jimbojim
Banned
But it's the slowest common denominator.
Ascend
Member
Not the same thing for consoles as for PC. Even with an SSD, a PC would need a lot of steps...;But its just faster load times bro
from SSD to CPU for decompression
From CPU to RAM
Then from RAM to VRAM
Then from VRAM to the GPU
Now, they can go from SSD to GPU/VRAM, bypassing the system RAM and CPU.
Consoles have had a unified RAM pool exactly to avoid having so many steps. So for the consoles, the SSD obviously helps compared to a hard drive, but, it is still not as ground breaking as it is for PC, because they already could go from storage to RAM(which is also VRAM).
I hope you're aware that DirectStorage was announced for the XSX like a few centuries back.
Last edited:
Lethal01
Member
They are saying even with direct storage it will be slower that Pc's with RTX and PS5.
EverydayBeast
thinks Halo Infinite is a new graphical benchmark
It's easy to root for PC gaming because they're trying to get better and figure out how to get better quickly.
llien
Member
I bet it was Microsoft pushing it in NV dirction (allowing the same API to run on both console and windows is in their interests) and if so, it would also happen with RDNA2 chips.
Ascend
Member
I guess hating on the XSX is still a thing on here... Anyway...
It's inaccurate to say that it will be slower than PCs with RTX. RTX has zero to do with DirectStorage. nVidia's is good at marketing their stuff. They added a decompressor on their GPU to enable DirectStorage and called it RTX I/O. You can have an RTX and you will still be slower than the XSX until DirectStorage is actually leveraged on PC. And even then, there's SFS which is still custom to the XSX.
Let me just say this. Even if the XSX turns out to have the slowest I/O solution in practice, it is still a more powerful console than the PS5 overall, and both are going to be vastly inferior to high end PCs, but more than capable enough compared to mid range PCs.
thicc_girls_are_teh_best
Member
It actually isn't, but I can see how yesterday's news was hard for some of you guys. It's okay, this is how you cope. I'm here for you.
Never mind the fact Nvidia is leveraging MS technology and certain people are trying to isolate Series X in the discussion when Nvidia's solution destroys Sony's pretty notably too. It has much more hardware decompression capability than Sony's (24 Intel CPU cores vs. 9 AMD Zen 2 cores), so it has a much higher ceiling. 7 GB/s raw was just given as a baseline, it will be able to easily support much faster drives in the future when they become available.
Of course none of this would need to be specified if some folks didn't try taking every single opportunity to shit on a given platform brand, but they can't seem to help themselves. There are clinics to help with addiction
Except you've been posting "troll" posts in relation to tech discussion that always lean a certain way, for months now. It's out of habit by this point.
Is this another one of your "troll posts", too? I'm genuinely asking because it's hard to detect sarcasm through text-only especially in the current environment of next-gen console discussion.
You literally misread the numbers and their context. You mentioned Sony's 22 GB/s, which is a lossy compression ratio, and compared it with NV's 14 GB/s, which is a lossless compression ratio. And you've also ignored that in terms of sheer hardware capability NV curbstomps Sony here, so they have much more headroom to expand and support even faster drives in the future.
I know this probably sucks for certain people, but the truth is both MS and Sony have already been bested in terms of SSD I/O before the new systems have even launched. NV were sitting well, biding their time, and came out with some banger cards. Just get over it. Yes I know, "paper specs" and the so, but literally anyone can look at NV's figures and see they simply have the much more robust I/O setup when it comes to sheer power. If DirectStorage is as big of a rewrite for file I/O access as it's been hinted (MS's own research papers into stuff like FlashMap suggest this is the case, provided if FlashMap is the basis for DirectStorage), then that also greatly benefits NV in terms of efficiency, too.
Don't see why any of this is surprising or startling to certain people who are seemingly desperate to cling onto any "performance advantage" to gloat about. PCs have always been at the forefront of performance ever since arcades tapered off on that front near the end of the '90s, and it'll remain that way well into the future. Consoles have always offered the best performance-to-price value, and it'll remain that way well into the future. I mean for goodness sake a 3080 costs the price alone you can very likely pick up a PS5 or Series X for, that should tell people enough right there!
But apparently having that advantage isn't enough for some people. It bear repeating though: NV simply have the most capable I/O on the market....or will be on the market, by the time these cards launch later this year. And in the very likelihood AMD are designing something similar with their RDNA2 cards, we can also see at least some of their cards either rival or best Sony's setup, too, around the end of the year, though probably not to the degree NV seems set to do. None of this suddenly makes Sony's or even Microsoft's console SSD I/O implementations worst or bad; it just means they won't be the king of the hill WRT to all upcoming I/O implementations. That's all.
I will say, though, that these developments are more or less proof of what I was stating well before in what MS's vision for file I/O advancements was, and the fact their approach is focused on scalability and stackability. This benefits them and companies like AMD and Nvidia in the console and PC space. Sony have a very strong I/O solution but it's seemingly a lot more insular and more built to their specific system design. That makes it less scalable/stackable though (I wouldn't be surprised if AMD too inspiration from their and MS's I/O and have implemented some weird hybrid I/O acceleration hardware in Big Navi cards, however).
Nvidia's GPUDirectStorage is basically their implementation of MS's DirectStorage. However Nvidia products look like they'll be the first to market with the feature (similar to VRS).
The parallels aren't surprising when you take that into consideration. However we can clearly see they have beefier hardware to utilize/support it. This is the kind of scalability and stackability MS had in mind, their FlashMap papers also support this (and get into a lot very technical details on specific I/O problems and how they resolve them. I wouldn't be surprised if DirectStorage and aspects of XvA are using the FlashMap research and going with a more "marketable" (?) name).
Last edited:
thicc_girls_are_teh_best
Member
The CPU doesn't factor into this at all because NV's solution is offloading work from the CPU. They kind of specify exactly this in the slides.
So it makes no sense to be referring to CPU when the CPU isn't doing any of the decompression work for the cards. All of it is handled through the GPU (btw NV have had ARM & FPGA cores in their GPUs for a while now, at least on some models, to handle access of the data and such, along with using HBCCs).
It makes a lot of sense.
Since NV could not optimize their CPU algo well enough to use less cores.
Unless their "24 cores" is just pure PR bullshit pulled out of their ass. Which one is it?
jimbojim
Banned
Yesterdays news were hard for some of us? LOL. Actually, quite the opposite. Even if Nvidia 30xx series are vastly more powerful than consoles, their compression/decompression method is still not par with PS5 solution. Maybe in 40xx series.
Last edited:
thicc_girls_are_teh_best
Member
Wrong. They listed a 7 GB/s drive because that's the limit for the fastest drive currently on the market. They actually have much more hardware for decompression/compression than Sony's solution, and they had to do this in order to future-proof it.
Basically, 14 GB/s compressed data on a 7 GB/s raw drive is the lossless standard. With various lossy compression ratios that amount can scale much higher. And PC SSDs will only keep getting faster as NVMe expands the standard to support faster drives in the future.
That's why I said yesterday's news was hard for some certain folks; the reality of it makes a lot of things very clear.
You should learn how to read the data better, because you're doing a piss-poor job and it's obvious now you aren't "trolling"; this is just the rhetoric you want to push. That's sad.
CPU, or GPU? Pick one. You're getting your terminology messed up right off the bat. Their I/O decompression/compression solution is equivalent to 24 Intel cores. If that's marketing talk, then so is Sony's, and so is Microsoft's. NV's solution is future-proofed for where they expect PC drives to go into the future.
I.e, PC SSDs will get much faster than 7 GB/s limit currently imposed by current NVMe specifications, and they wanted a solution to have the headroom for those faster drives. This is obvious to anyone looking at the graphs without a console slant in mind. So you've chosen to continuously look at this incorrectly.
It's funny to assume NV couldn't optimize their stuff on the software side when they have the best image reconstruction techniques in the industry at moment with DLSS 2.0 and the such. That's a very software-heavy thing to support as you need to train strong data models to implement it.
Just time for you guys to accept this. Things are as they are. Neither MS nor Sony have any massive advantages over PC tech now (aside perhaps audio, potentially), except in terms of value proposition. Which has always been the strongest strength of consoles anyway, so the performance you're getting from Series X and PS5 at their prices is absolutely amazing and a steal.
They aren't outperforming Nvidia (or higher-end AMD) cards at all, that much is clear. But they are giving more "bang for the buck" relative to the market they serve and that's where they are strongest.
Last edited:
mckmas8808
Mckmaster uses MasterCard to buy Slave drives
I don't understand the bolded. Who "REALLY" thinks a $500 device will outclass a $1500 device? Like how many people on this forum "REALLY" think that?
jimbojim
Banned
So, since you use numbers. XSX is then the slowest common denominator. PS5 doesn't need to use GPU cycles for compression crap. Nvidia use software solution, just like XSX, PS5 use completely hardware solution, so PS5 is ahead of both actually. Also PS5 SSD goes max to 22 GB/s. With Oodle Kraken is on average 17 GB/s. Nvidia's problem is latency too. Also sustainability of SSD's speed at higher temperatures on PC. PC SSDs has 2 priority levels, PS5 has 6 of them. Just saw Gameseeker's post on REEE :
NVIDIA announces RTX I/O - PCIe 4.0, on-GPU storage decompression, collaberating with Microsoft on DirectStorage
There is no doubt the PC will ultimately match and exceed the PS5 on raw bandwidth, due to sheer brute force and spending a LOT more money on the problem. Latency will be a different matter. Sony really, really thought out how to minimize the latency end-to-end and how to minimize cache...
www.resetera.com
Last edited:
NullZ3r0
Banned
Take your fanboy tantrum elsewhere. XVA uses DirectStorage. Dev adoption benefits the Series X.
Dr Bass
Member
First off, there was no fanboy tantrum lol. You need to chill a bit.
Secondly, you're right, it looks like my initial assessment might have been wrong. But again, the X/S hardware is done, nothing that is newly developed hardware wise (like what they are doing with Nvidia it seems) will be able to be added to the X because it's done. But yeah I think I was a bit wrong. Everyone gets things wrongs sometimes, it's not a problem to admit it when you do.
El Pistolero
Banned
It is tempting for some people to lump the Xbox Series X and the PS5 together to try and negate the advantage the latter has over both the Xbox and the PC in terms of I/O throughput. And, no, the Nvidia solution is still not as good as what Sony/AMD have worked on.
Nope. That's not what the slide says.
It says: to decompress a 14GB/sec stream (7GB/sec raw) you will need 24 CPU cores.
Already addressed that: DL is not a regular sw eng. You need data scientists and a shitload of data but programming skills required are rather low. That's why it's such a hot topic.
I don't care.
Again I point out poor messaging and huge marketing bullshit under a "tech" disguise.
It doesn't devalue the tech itself. It devalues the bullshit dealers.
thicc_girls_are_teh_best
Member
You'd be surprised.
That's why I kept saying this doesn't change the fact consoles offer the best performance-for-price value proposition. That isn't enough for warriors, though. They always need more.
TheContact
Member
what benefit does it get passing through the NIC?
NV has some tech for feeding the GPU through NIC interface. Added benefit: you can certify drives this way.
But it's just an image. It may be all wrong.
thicc_girls_are_teh_best
Member
Still didn't learn how to read the graphs correctly I see.
Doesn't change the fact NV have the best solution for this on the market and for the foreseeable future.
Feelings before facts?
You seem to be oblivious to messaging that downplays well-understood factors of various system designs, when the messaging comes from a preferred source however. It's been an M.O of yours for a while now.
Since you never actually balance out your discussion points either, it actually comes off as you devaluing the tech itself. I.e you are always quick to downplay particular tech (especially if it seems to somehow challenge anything PS5 has going for it), and then just...leave it at that. Such is perceivable by many as devaluing the tech, whether you see it that way or not.
More goalpost shifting. Beautiful. Love it!
They are all mix of hardware and software, you think the APIs run on sheets of paper?
Only if you don't understand how the tech works.
Nope. Oodle Kraken RDO 40 tests showed results of 3.99:1 compression ratio.
Assumptions, assumptions, assumptions. They make an ass out of you, not me. That's how the saying goes. Also it seems like you've done zero looking into what DirectStorage (and related patented technologies) are meant to fully address on PC side. Priority levels are expanded, for starters. GPUDirectStorage will implement this so that makes your point moot (like all of your other points tbh).
Just saw Gameseeker's post on REEE :
NVIDIA announces RTX I/O - PCIe 4.0, on-GPU storage decompression, collaberating with Microsoft on DirectStorage
There is no doubt the PC will ultimately match and exceed the PS5 on raw bandwidth, due to sheer brute force and spending a LOT more money on the problem. Latency will be a different matter. Sony really, really thought out how to minimize the latency end-to-end and how to minimize cache...
So now we're trusting randoms on ResetEra again when they say something that is absolutely definitely not placating to a given side simply to score brownie points, because they said some words that confirm preexisting biases?
Welp wrap it up folks, we only need to listen to "Gameseeker" and guys like Matt on Era for now on...even when they are wrong and/or try downplaying anything they perceive as negative to their brand of choice
Last edited:
mckmas8808
Mckmaster uses MasterCard to buy Slave drives
So what would that 3.99:1 ratio mean for the PS5 as far as compression goes? It takes it from 5.5 GB/s to what? 12 GB/s?
geordiemp
Member
People seem to lap up the marketing PR
Its available for preview in 2021 sometime.....
And the speed is the decompression possible if the file and IO system is not a bottleneck....., lets see how it works first shall we.
Rewriting win10 file system and IO for games....I will see you guys next year maybe.........
Just wait till next E3.....
Its available for preview in 2021 sometime.....
And the speed is the decompression possible if the file and IO system is not a bottleneck....., lets see how it works first shall we.
Rewriting win10 file system and IO for games....I will see you guys next year maybe.........
Just wait till next E3.....
Last edited:
PaintTinJr
Member
It is more likely that the GPU will get the data using the same mechanisms it would without RTX !O- through a RAM resident driver client/server configuration - as normal. But with the new change being that the RTX IO NIC will offloading bulk transfer driver work that normal goes through the CPU - because it was either resident in RAM and prepared for transfer to VRAM, or storage mapped in RAM like a clipmap/megatexture structure and constantly streaming into VRAM.
Once the game engine kicks off a transfer between storage and VRAM, the co-processor NIC (with SSD attached) will most likely see(snoop) the transfer and takeover the request, while keeping the GPU driver(the client) informed of the RTX IO NIC(server) to GPU (Server) transfer progress - with all the complexity of the SSD filesystem and data - packing and packet-ing - handle by the NIC. The cpu overhead of 50% of a cpu core is likely for reading every packet header of the NIC to GPU transfer to ensure everything is work as intended, and to stay informed incase either server reports a warning or error, or fails to respond before the packet expiry time.
The reason a NIC is a useful solution in the PC space is because graphics are already client/server and a PC NIC will have the lowest latency and data resilience of any DMA bus device in the system, as see by pinging your own IP address or the loopback address, which should report less than 1ms
DonJuanSchlong
Banned
thicc_girls_are_teh_best
I wouldn't even waste my time. I see the same handful of the biggest Sony fanboys, trying to shit up every thread, as per usual. Not worth trying to explain facts and reasoning to those who only have one agenda.
thicc_girls_are_teh_best
Member
No, it'd be to 21.945 GB/s or something like that. So more or less the figure Sony provided.
Yep I've noticed that too. I just hate dishonesty is all. And some of these people are otherwise knowledgeable but they have an agenda always present and can't keep it from twisting their own logic around into stating flat-out inaccurate and wrong information.
I'd love to discuss the strengths of both systems more openly in a way that doesn't have to resort to shitting on one or the other, but when a certain tone is set in a threat that creates a runaway effect of a narrative that's counter-productive to such, usually it ends up with trying to correct the points of discussion instead. That's automatically going to shift things another way.
There's no need for it but some people can't help themselves. Oh well; I'll just keep refuting that type of stuff with correct information when I have the time. And it's not like I've only done this in "favor" of MS, either; ITT I've had to do it for Nvidia and I've done it for Sony multiple times as well. But it's not out of some brand loyalty; it's because I actually like multiple options and want to discuss them all in the best and accurate light.
...Because right now I'm probably actually leaning to picking up a 3080 when I get around to upgrading my rig later in the year. But that doesn't mean I suddenly think PS5 or Series X are trash. It's just been extremely weird seeing some PS fanboys use 3070/3080/3090 as an attack on Series X, goes to show where their true intents are at.
Last edited:
jimbojim
Banned
What's the problem? Since you said fanboys, i will said it too. looks like PC fanboys trying to spread crap how Nvidia compression/decompression method is better than PS5, which is 100% wrong. Looks like you, PC fanboys ( or Nvidina fanboys ) has an agenda too. It's easy to blame others, isn't it?
Last edited:
DonJuanSchlong
Banned
I'm guessing you felt compelled to respond, as you're a confirmed fanboy. I wouldn't have to see your custom tag to realize, but it's helpful that it's so obvious. It must be a hard pill to swallow, since you and your bandwagon buddies kept making threads, shitting on anything positive about pc, etc. Now that you guys have been shut down, and put in your place, you guys still try to downplay the superior hardware. What gives? Should I just admit the ps5 is the strongest piece of hardware? Not even the ps6 will be as strong? When will you guys learn to give up?
jimbojim
Banned
Really, who gives a crap about my tag. Us, guys, shut down? For what? I'm shitting on PC? LOL, i've said couple of times that what Nvidia is doing is a step in right direction for I/O. And, no, Nvidias I/O solution isn't better than PS5, and that's a 100% fact. Maybe it will be 100% hardware level in 40xx series. Now it is a software based. So, it's OK to say that Nvidia has a better I/O solution than PS5, which is wrongly, of course. But when someone saying the opposite, immediately someone must be labeled as fanboy. LOL. What a hypocrisy.
Last edited:
No, it's thank you Microsoft.
jimbojim
Banned
Still didn't learn how to read the graphs correctly I see.
Doesn't change the fact NV have the best solution for this on the market and for the foreseeable future.
Feelings before facts?
You seem to be oblivious to messaging that downplays well-understood factors of various system designs, when the messaging comes from a preferred source however. It's been an M.O of yours for a while now.
Since you never actually balance out your discussion points either, it actually comes off as you devaluing the tech itself. I.e you are always quick to downplay particular tech (especially if it seems to somehow challenge anything PS5 has going for it), and then just...leave it at that. Such is perceivable by many as devaluing the tech, whether you see it that way or not.
More goalpost shifting. Beautiful. Love it!
They are all mix of hardware and software, you think the APIs run on sheets of paper?
Only if you don't understand how the tech works.
Nope. Oodle Kraken RDO 40 tests showed results of 3.99:1 compression ratio.
Assumptions, assumptions, assumptions. They make an ass out of you, not me. That's how the saying goes. Also it seems like you've done zero looking into what DirectStorage (and related patented technologies) are meant to fully address on PC side. Priority levels are expanded, for starters. GPUDirectStorage will implement this so that makes your point moot (like all of your other points tbh).
So now we're trusting randoms on ResetEra again when they say something that is absolutely definitely not placating to a given side simply to score brownie points, because they said some words that confirm preexisting biases?
Welp wrap it up folks, we only need to listen to "Gameseeker" and guys like Matt on Era for now on...even when they are wrong and/or try downplaying anything they perceive as negative to their brand of choice
LOL more goalpost shifting. Nvidias I/O solution isn't best on the market. Assumptions that SSD read speed on PC is reducing because higher temperature? It's not an assumption. That's always happening on PC. So, we are trusting some randoms on ERA or here. LOL. What are you then?
GameSeeker
is here too. Oodle Kraken showed us results bigger than 15 GB/s. And that's a fact tooMore than 15 actually.
Couple of examples on these pages :
Next-Gen PS5 & XSX |OT| Console tEch threaD
And there you go, the reason why 2.5 x is left vague. LOL. Timdog will be posting about effective 2.5 x 16 GB memory next. MS really need to clean up their act. It's very simple concept, and they have explained it very well. Nothing is left vague...
Next-Gen PS5 & XSX |OT| Console tEch threaD
https://www.ea.com/games/fifa/fifa-21/news/fifa-21-faq-release-date "Sony do not have equivalent of Smart Delivery" Turns out that was non-sense too, it supports cross-buy as well as cross-save across same platforms. Let's see what next FUD we have to deal with next. So is dual entitlement...
PlayStation 5 System Architecture Deep Dive |OT| Secret Agent Cerny OT
Don't trust any pricing rumors until it's directly from Sony.
www.resetera.com
But hey, we are randoms, yo
Last edited:
GameSeeker
Member
Note that games that use RTXIO aren't coming anytime soon. See article: https://wccftech.com/rtx-io-and-directstorage-are-coming-but-itll-be-a-while-yet/
The conclusion of the article says:
"It sounds like both Microsoft and NVIDIA are working to address what Tim Sweeney called out when he said the PlayStation 5's storage architecture was way ahead of the standard one in place for PC games. That's great news, but there is a catch: Microsoft is 'targeting' a release for the DirectStorage API at some point next year in the hands of developers, and that's only as a preview. In all likelihood, we'll have to wait until 2022 before we see some games actually taking advantage of both RTX IO and DirectStorage."
The conclusion of the article says:
"It sounds like both Microsoft and NVIDIA are working to address what Tim Sweeney called out when he said the PlayStation 5's storage architecture was way ahead of the standard one in place for PC games. That's great news, but there is a catch: Microsoft is 'targeting' a release for the DirectStorage API at some point next year in the hands of developers, and that's only as a preview. In all likelihood, we'll have to wait until 2022 before we see some games actually taking advantage of both RTX IO and DirectStorage."
So the year cross generation ends perfect timing and unreal 5 games are out again perfect timing. It be nice sooner but it won't be needed until then when next generation is really going.
PaintTinJr
Member
When looking at this solution in context of what XsX and PS5 listed, it feels that the numbers aren't entirely representative, even if we just take the first info about it being lossless decompression.
The console solutions discussed openly that lossless can't be entirely useful to the majority of game data because a lot of that data is textures that already exist in a lossy block encoded compression format, and that to get decent benefits for that data requires other pre-processing like RDO(rate distortion optimization that will reduce PSNR of the texture), or the use of some form of block translation process that helps the lossless compressor compress better - but a process that then needs undone after decompression inside the GPU, either by an ASIC or CU. And so even the notion that 7GB of raw SSD data would represent 14GBs of data read - once decompressed - isn't entirely a given...unless that is, that the data is ultra setting raw game assets. But then the question would be: how does that 14GB then fit inside most of the RTX 30 range's VRAM - when only the 3090 has more GDDR6X memory than 10GB?
And surely if a XsX or PS5 game is going to be 10.5GB of mostly block compressed assets (at a minimum), then wouldn't the PC version of that game (with ultra level assets) be like 2 or 3 times that size when losslessly decompressed - like 30GB? Which again, would have a VRAM size issue for even the RTX 3090.
The other part that also might not compare favourably to the console solutions is the GPU compute cost of the decompression - maybe specialized Turing/Ampere compute units being commandeered and a lot of the cache bandwidth to decompress and copy out such large data to VRAM - which is maybe an indication of why an RTX 2060 would be the minimum card to support the RTX IO - but then the question is whether the RTX IO card will be cheaper than an a RTX 2060 to make that a viable option for the wider PC gaming audience.
With AMD having had eyes on the PS5's APU IO complex throughout, it should be very interesting if they follow suit with a DriectStorage NIC card solution like Nvidia, or find a way to offer a PCIe card with a full SoC at Zen2/Rdna2 with NVMe SSD to get closer to the PS5 solution.
The console solutions discussed openly that lossless can't be entirely useful to the majority of game data because a lot of that data is textures that already exist in a lossy block encoded compression format, and that to get decent benefits for that data requires other pre-processing like RDO(rate distortion optimization that will reduce PSNR of the texture), or the use of some form of block translation process that helps the lossless compressor compress better - but a process that then needs undone after decompression inside the GPU, either by an ASIC or CU. And so even the notion that 7GB of raw SSD data would represent 14GBs of data read - once decompressed - isn't entirely a given...unless that is, that the data is ultra setting raw game assets. But then the question would be: how does that 14GB then fit inside most of the RTX 30 range's VRAM - when only the 3090 has more GDDR6X memory than 10GB?
And surely if a XsX or PS5 game is going to be 10.5GB of mostly block compressed assets (at a minimum), then wouldn't the PC version of that game (with ultra level assets) be like 2 or 3 times that size when losslessly decompressed - like 30GB? Which again, would have a VRAM size issue for even the RTX 3090.
The other part that also might not compare favourably to the console solutions is the GPU compute cost of the decompression - maybe specialized Turing/Ampere compute units being commandeered and a lot of the cache bandwidth to decompress and copy out such large data to VRAM - which is maybe an indication of why an RTX 2060 would be the minimum card to support the RTX IO - but then the question is whether the RTX IO card will be cheaper than an a RTX 2060 to make that a viable option for the wider PC gaming audience.
With AMD having had eyes on the PS5's APU IO complex throughout, it should be very interesting if they follow suit with a DriectStorage NIC card solution like Nvidia, or find a way to offer a PCIe card with a full SoC at Zen2/Rdna2 with NVMe SSD to get closer to the PS5 solution.
Last edited:
thicc_girls_are_teh_best
Member
So now I'm a PC fanboy? xD FOH, stop getting caught so up in your feelings you lose sight of the bigger picture at discussion here.
Gameseeker (whoever that is; they are just another poster to me, same as anyone else on any other forum) might have an account here, but they aren't posting here. If they want to discuss their points, they can post here whenever they want. If they want to debate anyone here (such as myself), they can freely drop by and do so. Point is, they're most likely a grown adult and can speak for themselves if they wish, they don't need you doing it for them (poorly).
PaintTinJr
I think you're reading too much into some things and looking at other points of this wrongly. PS5's SSD can decompress up to 22 GB of data per second which also clearly outstrips its GDDR6 physical memory amount, yet that was never raised as a concern. It doesn't ultimately matter if these systems can stream in more data than they can fit in RAM because if that data is only being briefly used and then expunged, then there's no conflict.To answer some of your other concerns, here's some speculation from function at B3D that touches back on earlier speculation several users here have had regarding XvA:
Most bolded emphasis mine. The one thing up for contention is the part I asterisked , but seeing as how RTX I/O is in fact reading the data from the storage, processing it, then placing it in VRAM, this is a possibility of something else they could be doing on the GPU end, meaning they're using some sort of HBCC as well. It'd also tie back into a few other speculation that's been going around for the past several months, plus certain things that were mentioned in the FlashMap R&D papers.
Anyways...compute cost on the GPU for decompression, at least in Nvidia's case, won't be all that much and if it was something too resource-depriving they simply would have forgone including it as a feature in the first place. It's also not like they don't have the muscle to handle such; they've routinely been outperforming AMD in this area for over a decade now and it seems like they're going to continue holding that crown for top-of-class GPUs on the market going into the future.
You guys also need to keep in mind that since AMD is involved in the PC space and are collaborating not just with Sony but also MS, they are probably going to roll with prioritizing SSD I/O integration that targets compatibility with DirectStorage. They will take features in that regard from both MS and Sony and probably utilize them as some hybrid that still complies with DS feature sets.
All this does show one thing though: that the I/O solutions between MS and Sony are indeed not apples-to-apples, and have their strengths and weaknesses. CPU overhead and latency aren't weaknesses to MS's approach tho like some users have been adament on trying to imply (their "weakness" is more in the fact that the scalability and stackability means lower-end SSDs can also benefit but they will always offer lesser performance than higher-end SSDs supporting the same thing). That's not me implying Sony's solution is weak in those areas, either, but a "weakness" to it could be lack of scalability with other hardware-based solutions, especially if AMD's approach moreso takes after MS's.
Granted AMD already had the Radeon SSG cards which were doing this stuff earlier on, but that's actually a reason why I think their solution will take liberally from both MS and Sony rather than just going cold on one and hot on the other.
Similar threads
- Locked