• 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.

How much difference does the improved SSD + io make in series S/X and PS5?

ZywyPL

Banned
But again, you're incredibly wrong about SSD being a "workaround". It's essential. If anything, RAM is a workaround for main storage being too slow.

You could also say that a GPU is a workaround for the CPU being too slow to render graphics... A platform is a sum of its parts, where each has it's specifically designed purpose, not just one, it's like a chain that's only as strong as the weakest link (as seen with Jaguar CPU last gen), but at the same time making a single link 10x stronger than others doesn't change much, if anything.

And like I said, it's not me, you have MS themselves telling you exactly why they went with the SSD, what it's purpose and features. Next year the consoles will get totally smoked by NV Lovelace and AMD Zen4 and no magic SSD trick will be able to close that gap, not even close, the PCs are doing circles around consoles already. And we already see the PS5-only games heavily relying on the mentioned SSD tech, like DS Remake that doesn't look anything the PS4 wouldn't handle at 1080p30, and R&C which up until pointed out by the community manager on Twitter nobody would've been able to tell the assets are being constantly streamed in and out as the camera moves. That's what I'm trying to say the whole time, under the hood the games will run different, sure, but the end users won't even know/notice it, just like if there was simply a lot more memory and everything was already loaded and waiting to be used instead.
 

Allandor

Member
The Series S has 8gb for games and the XSX has 13.5gb, the PS5 likely has a similar amount to the XSX.

Now theres been more concern over the seriesS's smaller amount, but all the current gen systems do have small gen on gen RAM increases.

So how much will the ssd and better io alleviate this issue?

Will the SSD be able to be used like RAM as cerny and Microsoft have said?

Will these consoles have the equivalent of like 64gb ram + HDD? or better, or worse?

Because if the better io and ssd is as good as they say, games made exclusively for these consoles may be better then we anticipate?
The SSD bandwidth is good enough for most stuff expect maybe for some really fancy edge-cases.
Sony used brute force speed (which is way easier to use) and MS used high speed + a few hardware things to more granular stream stuff into memory (reduce the things that should be streamed). I really don't see a big problem with series s, as all streaming-related stuff works especially well on that console (lower quality assets -> more effective available SSD bandwidth and less memory needed).
I really don't think we will see any SSD bandwidth related bottle-necks this gen, this bottle-neck is just gone with this generation. A bigger problem will be the size of higher res and more diverse assets. Better compression and no duplicates is nice to reduce the size but the quality-jump over time will make games much, much bigger over time.
 

RoadHazard

Gold Member
You could also say that a GPU is a workaround for the CPU being too slow to render graphics... A platform is a sum of its parts, where each has it's specifically designed purpose, not just one, it's like a chain that's only as strong as the weakest link (as seen with Jaguar CPU last gen), but at the same time making a single link 10x stronger than others doesn't change much, if anything.

And like I said, it's not me, you have MS themselves telling you exactly why they went with the SSD, what it's purpose and features. Next year the consoles will get totally smoked by NV Lovelace and AMD Zen4 and no magic SSD trick will be able to close that gap, not even close, the PCs are doing circles around consoles already. And we already see the PS5-only games heavily relying on the mentioned SSD tech, like DS Remake that doesn't look anything the PS4 wouldn't handle at 1080p30, and R&C which up until pointed out by the community manager on Twitter nobody would've been able to tell the assets are being constantly streamed in and out as the camera moves. That's what I'm trying to say the whole time, under the hood the games will run different, sure, but the end users won't even know/notice it, just like if there was simply a lot more memory and everything was already loaded and waiting to be used instead.

The Jaguar was one of the weak links last gen, the HDD was another. Staying with that, or even a slow SSD (~500MB/s) would have severely limited what games this generation can be. We haven't seen much of that yet, but we will.
 

Sosokrates

Report me if I continue to console war
Nobody has ever said the SSD can be used "as RAM", but with faster data streaming you don't have to preload data as far in advance. Instead of loading what you MIGHT need 1 minute from now, you can load only what you need in the next 5 seconds, which means you waste a lot less RAM and can use it much more efficiently for the stuff you actually want to display. So you effectively get more memory per moment of gameplay, making the on-paper small jump from 8 to 16GB larger than it first seems.
If nobody said the ssd can be used "as RAM" why mention this?
 

BootsLoader

Banned
The Series S has 8gb for games and the XSX has 13.5gb, the PS5 likely has a similar amount to the XSX.

Now theres been more concern over the seriesS's smaller amount, but all the current gen systems do have small gen on gen RAM increases.

So how much will the ssd and better io alleviate this issue?

Will the SSD be able to be used like RAM as cerny and Microsoft have said?

Will these consoles have the equivalent of like 64gb ram + HDD? or better, or worse?

Because if the better io and ssd is as good as they say, games made exclusively for these consoles may be better then we anticipate?
This is a fantasy based on….SSD.
 

Sosokrates

Report me if I continue to console war
You said it in your OP, as if anyone who knows anything has ever claimed that. Which they haven't.

You need to re-read my OP, I said "like RAM"

Mark Cerny said: "on PS5 though the SSD is very close to being like more RAM"

@12.40
 

3liteDragon

Member
You need to re-read my OP, I said "like RAM"

Mark Cerny said: "on PS5 though the SSD is very close to being like more RAM"

@12.40

It is, it's like the console has access to 100GB of graphics memory because of how robust their I/O setup is and how efficiently RAM's being used in gameplay due to how fast their custom SSD is. That's why they went ham on the whole I/O thing & the "SSDs don't improve graphics" crowd have no idea what they're talking about. This has huge ramifications for future titles, especially open-world titles.
 
Last edited:

PaintTinJr

Member
Specifically looking at PlayStation - where through Cerny, they've stated how it will change things - it seems like I'm forever reading that the PS5 is just a console made with PC parts, yet where the IO decompression and SSD are concerned, I think it is actually, far from that.

For me the PS5 is the accumulation of all PlayStation have learnt from the simpler PS1 and PS4 successes, and finally delivered their third stream processor console - one and two being the PS2 and PS3 which respectively had a lack of memory, crappy IO/storage bandwidth and too low a CPU clock, and for the PS3, a lack of memory, crappy storage bandwidth. With both needing complex software innovation and so had too high a learning curve.

With the IO complex and raw SSD read bandwidth, the PS5 seems perfectly balanced as a stream processor console - without any weak areas like the Ps2 and PS3 - particularly, as the contention for data is mostly offloaded from the primary controlling CPU thread, and co-processed by the IO complex and DMAs direct into unified ram, making the normal efficiency cap on homogeneous multicore in a PC not present in the PS5 - when used as a stream processor console, as opposed to being used as a entry level, high-end PC with PC multi-plat game engines.

If doing a comparison with similar systems to the Ps5, but with slower SSDs and less effective IO decompression(and more latency), we could probably compare them like two different shaped frusta or pyramids - from how clipmapping works.

In the PS5 pyramid it has a large enough base, to make the height feel unnoticeable as a modest gradient Whereas, in lesser IO/SSD systems, the pyramid is still the same height, but has a noticeable gradient/smaller base. And in systems with lesser IO/SSD, but more RAM, if we use the cut nose pyramid (frustum) for the example, the top layer of the frustum gets bigger - and so does the base, proportionally - but the gradient of the frustum edges remain unchanged as they represent the effective IO decompression, storage speed and latency.
 

Loxus

Member
Nobody has ever said the SSD can be used "as RAM", but with faster data streaming you don't have to preload data as far in advance. Instead of loading what you MIGHT need 1 minute from now, you can load only what you need in the next 5 seconds, which means you waste a lot less RAM and can use it much more efficiently for the stuff you actually want to display. So you effectively get more memory per moment of gameplay, making the on-paper small jump from 8 to 16GB larger than it first seems.
I think it came from what Cerny said in Road to PS5.

"On PlayStation 5 though the SSD is very close to being like more RAM.

f:id:keepitreal:20200329140011j:plain


Typically it's fast enough that when you realize you need a piece of data you can just load it from the SSD and use it, there's no need to have lots of data parked in system memory waiting to potentially be used.

A different way of saying that is that most of Ram is working on the game's behalf."


PS5 SSD can reach 8-9 GB/s, similarly to DDR2 & 3. And when looking at DDR4 RAM, it has a transfer rate of 17 - 21 GB/s. PS5 decompression can reach 22 GB/s.
What are the data transfer rates for DDR, DDR2, DDR3 and DDR4?
hPZMKyB.png


With all the stuff going on with the PS5 hardware.
fPPxpJe.jpg

f:id:keepitreal:20200329140503j:plain


It isn't impossible to think the SSD can be used as RAM for slower CPU tasks.
 

ReBurn

Gold Member
I think it came from what Cerny said in Road to PS5.

"On PlayStation 5 though the SSD is very close to being like more RAM.

f:id:keepitreal:20200329140011j:plain


Typically it's fast enough that when you realize you need a piece of data you can just load it from the SSD and use it, there's no need to have lots of data parked in system memory waiting to potentially be used.

A different way of saying that is that most of Ram is working on the game's behalf."


PS5 SSD can reach 8-9 GB/s, similarly to DDR2 & 3. And when looking at DDR4 RAM, it has a transfer rate of 17 - 21 GB/s. PS5 decompression can reach 22 GB/s.
What are the data transfer rates for DDR, DDR2, DDR3 and DDR4?
hPZMKyB.png


With all the stuff going on with the PS5 hardware.
fPPxpJe.jpg

f:id:keepitreal:20200329140503j:plain


It isn't impossible to think the SSD can be used as RAM for slower CPU tasks.
At best feels like it would allow for some really fast caching to support asynchronous background task processing but if Cerny was comparing it to RAM feels like more magic SSD hyperbole.
 
Last edited:

Allandor

Member
It isn't impossible to think the SSD can be used as RAM for slower CPU tasks.
No, this isn't possible. You have a nice bandwidth, yes but the latencies are much, much higher.
Also for CPU task write speed (next to latencies) is needed and that is just not there on PS5. The write speed is really, really low for an SSD. But that is not important for a console. Just forget those "use it as RAM" thing. It would also really "burn out" the nand chips over time.
The best thing you can make with the SSD (and that is really the god part of this gen) is just load shortly before it is needed things into memory (and only those things) and save memory this way.
 

RoadHazard

Gold Member
I think it came from what Cerny said in Road to PS5.

"On PlayStation 5 though the SSD is very close to being like more RAM.

f:id:keepitreal:20200329140011j:plain


Typically it's fast enough that when you realize you need a piece of data you can just load it from the SSD and use it, there's no need to have lots of data parked in system memory waiting to potentially be used.

A different way of saying that is that most of Ram is working on the game's behalf."


PS5 SSD can reach 8-9 GB/s, similarly to DDR2 & 3. And when looking at DDR4 RAM, it has a transfer rate of 17 - 21 GB/s. PS5 decompression can reach 22 GB/s.
What are the data transfer rates for DDR, DDR2, DDR3 and DDR4?
hPZMKyB.png


With all the stuff going on with the PS5 hardware.
fPPxpJe.jpg

f:id:keepitreal:20200329140503j:plain


It isn't impossible to think the SSD can be used as RAM for slower CPU tasks.

Sure, and that's the point, that it lets you use RAM much more efficiently instead of wasting most of it on "parked" data that you might not even need. It still needs to go through RAM though, but just before it's needed instead of a minute before.

But RAM is also used as working memory for data that needs to be temporarily stored, altered, and read back out. You're not gonna be doing that on the SSD. So the comparison doesn't hold all the way.
 
Last edited:

Loxus

Member
No, this isn't possible. You have a nice bandwidth, yes but the latencies are much, much higher.
Also for CPU task write speed (next to latencies) is needed and that is just not there on PS5. The write speed is really, really low for an SSD. But that is not important for a console. Just forget those "use it as RAM" thing. It would also really "burn out" the nand chips over time.
The best thing you can make with the SSD (and that is really the god part of this gen) is just load shortly before it is needed things into memory (and only those things) and save memory this way.
Take that up with Cerny, he said it.
BgdvyGl.jpg

Even in the image Cerny used, it says, "low latency/high bandwidth access in some ways like RAM.

It already stored in the SSD, instead of putting it in GDDR memory. Just let the GPU/CPU access and take the data directly from the SSD when possible. That's basically what your doing with RAM.

With the SSD, Cerny removed the bottlenecks and created dedicated decompression hardware.

Cerny didn't just make the SSD fast, but 100 times faster. I know it's hard to believe, but Mark Cerny ain't no normal man. Especially when looking at his patents.

Many people thought Einstein was crazy when he first proposed some of his theories. Those theories still hold true to this day.

Just look pass the console/war for once and look at the bright side of the tech.
 

Loxus

Member
Sure, and that's the point, that it lets you use RAM much more efficiently instead of wasting most of it on "parked" data that you might not even need. It still needs to go through RAM though, but just before it's needed instead of a minute before.

But RAM is also used as working memory for data that needs to be temporarily stored, altered, and read back out. You're not gonna be doing that on the SSD. So the comparison doesn't hold all the way.
Which is why I said slower CPU tasks. Data that is already stored uncompressed and ready to use as is, shouldn't be an issue.
 

Allandor

Member
Take that up with Cerny, he said it.
BgdvyGl.jpg

Even in the image Cerny used, it says, "low latency/high bandwidth access in some ways like RAM.

It already stored in the SSD, instead of putting it in GDDR memory. Just let the GPU/CPU access and take the data directly from the SSD when possible. That's basically what your doing with RAM.

With the SSD, Cerny removed the bottlenecks and created dedicated decompression hardware.

Cerny didn't just make the SSD fast, but 100 times faster. I know it's hard to believe, but Mark Cerny ain't no normal man. Especially when looking at his patents.

Many people thought Einstein was crazy when he first proposed some of his theories. Those theories still hold true to this day.

Just look pass the console/war for once and look at the bright side of the tech.
We know this here and discussed it for month. But the fact is, the SSD is still way, way worse than RAM. Yes you can save some reads but that is all you can get from it. Latencies are way better than the latencies of HDDs but comparison to RAM is still not appropriate. It saves much RAM to just load stuff before it is needed. That is where the "RAM"-thinking already ends. If you request a resource before actually rendering it, you loose time. So you need the resource before you render the frame. That is way better than what you can do with HDD where you need the resources buffered in RAM for (at least) seconds before you render a frame. Else you would get something like pop in of details.
Just don't get to much out of the "in some ways more like RAM". Having SSDs mean much, much less buffering of content. And in this way, the SSD is used like the RAM before (with HDDs). But there the similarities already end.


The nvidia picture just shows what nvidia want to make in the future (and would also work with a normal SSD). Instead of letting the CPU decompress stuff for the GPU, the GPU gets the stuff directly into it's memory and uses it own cores (shaders or whatever nvidia has on board) and decompresses it there. This saves CPU cycles, and at least 2 routes through the PCIe "port". Last gen consoles already saved those copies because those had one pool of memory. But PCs still copy the stuff from the SSD to main memory, the CPU decompresses it and copies it over to the GPU memory. GPUs are now more than capable enough to take over the CPU part for decompression and so even much bandwidth and CPU cycles can be saved. But at the same time the GPU has more to do (bandwidth usage on GPU side is more or less the same).
This would also work with HDDs, but as HDDs are so slow it wouldn't make a difference. With SSDs data can just be delivered much faster and eliminate the bottlenecks on that side so this concept makes sense on PC now.
 
Last edited:

Loxus

Member
We know this here and discussed it for month. But the fact is, the SSD is still way, way worse than RAM. Yes you can save some reads but that is all you can get from it. Latencies are way better than the latencies of HDDs but comparison to RAM shows that someone does not know what he is talking about. It saves much RAM to just load stuff before it is needed. That is where the "RAM"-thinking already ends. If you request a resource before actually rendering it, you loose time. So you need the resource before you render the frame. That is way better than what you can do with HDD where you need the resources buffered in RAM for (at least) seconds before you render a frame. Else you would get something like pop in of details.
Just don't get to much out of the "in some ways more like RAM". Having SSDs mean much, much less buffering of content. And in this way, the SSD is used like the RAM before (with HDDs). But there the similarities already end.


The nvidia picture just shows what nvidia want to make in the future (and would also work with a normal SSD). Instead of letting the CPU decompress stuff for the GPU, the GPU gets the stuff directly into it's memory and uses it own cores (shaders or whatever nvidia has on board) and decompresses it there. This saves CPU cycles, and at least 2 routes through the PCIe "port". Last gen consoles already saved those copies because those had one pool of memory. But PCs still copy the stuff from the SSD to main memory, the CPU decompresses it and copies it over to the GPU memory. GPUs are now more than capable enough to take over the CPU part for decompression and so even much bandwidth and CPU cycles can be saved. But at the same time the GPU has more to do (bandwidth usage on GPU side is more or less the same).
This would also work with HDDs, but as HDDs are so slow it wouldn't make a difference. With SSDs data can just be delivered much faster and eliminate the bottlenecks on that side so this concept makes sense on PC now.

I would of said the same thing if Cerny hadn't done all these things to remove all the SSD bottlenecks.

dOcMLb0.jpg

NKIzYhO.jpg


It isn't just the SSD, it's the SSD + the custom I/O unit that makes it possible.
 

Boglin

Member
Looks at pc …which has had ssds for years
Yeah I’m guessing not much outside of quicker loading times

Don’t believe all the marketing spin
I think you're right that you won't see much outside of quicker loading times on PC. Currently a 500MB/s sata drive and a 5GB/s nvme drive perform closely for game loading times but development of both direct storage and RTX IO will rectify this. PCs have had SSDs for over a decade but these improvements are only just now coming out and I think it's because PCs don't see much of a benefit outside of initial loading times.

For consoles however, I believe this will help memory utilization just as much as it helps loading times. PCs have never needed to have a crutch developed to improve memory utilization in games because they would simply raise the minimum requirements for memory instead.

I do think there are some weird semantic issues around what is considered a big upgrade for these consoles. I'm pulling numbers out of my ass here but hypothetically, do you think people would consider it a big deal to free up 4-6GB of memory by not needing to keep assets in reserve due to the utilization of the new I/O and SSD? What difference would that make in DF comparisons?
 

Boglin

Member
I would of said the same thing if Cerny hadn't done all these things to remove all the SSD bottlenecks.

dOcMLb0.jpg

NKIzYhO.jpg


It isn't just the SSD, it's the SSD + the custom I/O unit that makes it possible.
This helps to utilize the full bandwidth of the SSD which is obviously massively important but the latency will still be hundreds of times slower than that of the memory.
 
They help significantly by being able to regularly stream in a higher dataset during gameplay, which should help with maintaining a higher level visual presentation on a more consistent basis.

It's my belief very few games have yet to significantly increase their data streaming demands over last gen in a way that will properly utilize these SSD. There has perhaps only been a marginal increase at best, or no increase at all. They're likely doing the same old, but just faster thanks to the SSD speeds, so the benefit right now is maybe mostly being seen in load times.

Games would need to be designed with the larger streaming datasets in mind. PS5 went with a super fast SSD that'll make even the loading in of models notably faster than it should be on the Series X without any special techniques. The same is true also for texture data.

And on the Series X side of things, Microsoft went with a slower SSD, but they designed their GPU around techniques such as Sampler Feedback Streaming, which is specifically designed to significantly cut down on the amount of work thee hardware needs to do in loading textures. It leads to a scenario where the Series X could get much better efficiency on its available system memory while simultaneously significantly decreasing the amount of texture data the SSD needs to load into memory. So in such a scenario, at least for texture data, Microsoft doesn't need an SSD any faster than the one they already possess. But the very big win in this scenario outside of effective SSD performance is the RAM savings because saving RAM can directly impact game performance or allow them to use the saved memory to use even higher quality textures than they would have otherwise been able to get away with.

The demo starts at 7:35. I'm not certain, but Halo Infinite could be using this technology.

 

Boglin

Member
Isn't RTX IO just nvidias branding of DirectStorage? AFAIK there is no hardware change.
It's not adding new hardware but it is moving the processing to the more appropriate existing hardware. Adding new hardware wouldn't be feasible on PC because all of the hardware companies would need to agree on which compression/decompression algorithms to use.
 
That's exactly what directstorage is.

I guess it's using different hardware for decompression than before. Which in this case is the GPU so it really isn't BS that you have to shift the decompression to something else to avoid the CPU from suffering because of it. On PCs it will be the GPU and on consoles it will be the custom decompressors that they have.
 

Dream-Knife

Banned
I guess it's using different hardware for decompression than before. Which in this case is the GPU so it really isn't BS that you have to shift the decompression to something else to avoid the CPU from suffering because of it. On PCs it will be the GPU and on consoles it will be the custom decompressors that they have.
No I'm not saying it's BS at all, just that RTX IO and directstorage are the same thing.
RTX I/O keeps the data coming from the SSD (which is moved through by the DirectStorage API) compressed while being sent to the GPU for decompression.
That's what directstorage is. Your card does the decompression instead of the cpu.

Make me wonder why they have their own name for it.
 

3liteDragon

Member
That's what directstorage is. Your card does the decompression instead of the cpu.

Make me wonder why they have their own name for it.
Yea I just remember what I read about it, might be the same thing or there could be differences idk, would have to do some research.
 
I think you're right that you won't see much outside of quicker loading times on PC. Currently a 500MB/s sata drive and a 5GB/s nvme drive perform closely for game loading times but development of both direct storage and RTX IO will rectify this. PCs have had SSDs for over a decade but these improvements are only just now coming out and I think it's because PCs don't see much of a benefit outside of initial loading times.

For consoles however, I believe this will help memory utilization just as much as it helps loading times. PCs have never needed to have a crutch developed to improve memory utilization in games because they would simply raise the minimum requirements for memory instead.

I do think there are some weird semantic issues around what is considered a big upgrade for these consoles. I'm pulling numbers out of my ass here but hypothetically, do you think people would consider it a big deal to free up 4-6GB of memory by not needing to keep assets in reserve due to the utilization of the new I/O and SSD? What difference would that make in DF comparisons?
people would like to think we will get some mindblowing games with wizzbang graphics....reality will be ....faster loading time
i have seen enough diagrams from graphics card makers and cpu companies to realise not a great deal has changed in 10 years outside of multithread cpus and ray tracing graphics cards
 

Boglin

Member
people would like to think we will get some mindblowing games with wizzbang graphics....reality will be ....faster loading time
i have seen enough diagrams from graphics card makers and cpu companies to realise not a great deal has changed in 10 years outside of multithread cpus and ray tracing graphics cards
Your answer is exactly what I'm getting at with the semantic issue. If I claim 6GB would make a big difference then you'll interpret that the way you choose to, in this case assuming I mean mindblowing wizzbang graphics. A person can be a tech nerd and see 6GB of additional usable memory as a huge gain in efficiency but another person only sees it as an imperceptible texture and resolution bump. These two people will end up at odds and speaking passed each other because they are approaching the benefit with different perspectives and expectations.
 

PaintTinJr

Member
No, this isn't possible. You have a nice bandwidth, yes but the latencies are much, much higher.
Also for CPU task write speed (next to latencies) is needed and that is just not there on PS5. The write speed is really, really low for an SSD. But that is not important for a console. Just forget those "use it as RAM" thing. It would also really "burn out" the nand chips over time.
The best thing you can make with the SSD (and that is really the god part of this gen) is just load shortly before it is needed things into memory (and only those things) and save memory this way.
With regard to the PS5's IO complex with an effective LLC (esram), in addition to the SSD's custom controller using 6 priority levels - and uses an 8 core ARM board with its own memories IIRC - that is a simplified take that doesn't really stack up IMHO.

Looking back at the use of RAM + edram and esram in the 360 or X1, and the results of the games/resolutions/fidelity we have nearly 2 decades worth of physical evidence to say that using those small memory apertures did make the RAM work "sort of like" VRAM, which seems semantically no different to what Cerny said about hardware like the io complex/ssd controller and ssd facilitating latency hiding to work like memory.

The 6 priority levels, in a satellite board, means that cancelling requests for optimistic data requests via the slower priorities incurs no burden for the Zen2 CPU or IO complex AFAIK, and the cache scrubbers will also dove tail with the IO's latency hiding while streaming data just in time.
 

Md Ray

Member
Isn't RTX IO just nvidias branding of DirectStorage? AFAIK there is no hardware change.
RTX IO brings the hardware component (GPU, the alternative to custom IO complex that consoles have) for decompression.
DirectStorage brings the software component (new storage API stack) to utilize PCIe-based SSDs to their full potential and eliminate load times.
 
Last edited:

Allandor

Member
With regard to the PS5's IO complex with an effective LLC (esram), in addition to the SSD's custom controller using 6 priority levels - and uses an 8 core ARM board with its own memories IIRC - that is a simplified take that doesn't really stack up IMHO.

Looking back at the use of RAM + edram and esram in the 360 or X1, and the results of the games/resolutions/fidelity we have nearly 2 decades worth of physical evidence to say that using those small memory apertures did make the RAM work "sort of like" VRAM, which seems semantically no different to what Cerny said about hardware like the io complex/ssd controller and ssd facilitating latency hiding to work like memory.

The 6 priority levels, in a satellite board, means that cancelling requests for optimistic data requests via the slower priorities incurs no burden for the Zen2 CPU or IO complex AFAIK, and the cache scrubbers will also dove tail with the IO's latency hiding while streaming data just in time.
There is so much wrong with this post I don't know where to start.

A simplified try:
1) Cache Scrubbers are there to minimize Cache <-> RAM synchronization without impacting to much (small updates instead of big ones) have nothing to do with the SSD and will just save some cycles and bandwidth.
2) Priority level don't decrease latencies. They increase latencies as they are always an addition in a additional, minimalistic processor and not a replacement. They give just some requests a priority so everything else waits a bit longer if there is no room to process them at that time. But that is minimal. The big latency difference comes from access times SSD vs RAM. While RAM is measured in ns SSD latency still comes close to ms (HDDs are much worse though).
3) The edram or esram in the old xboxes were only for the GPU to use to reduce access to the main RAM pool and have big bandwidth for operaptions accessing that additional memory. Also this is content-free memory as only the GPU access them (yes the xbox one could theoretically access the esram but that was quite slow and not recommenced as it slows down the GPU access times). Also keep in mind the edram/esram was quite big (a 10MB and 32MB) but still to small for modern times. Those tiny caches for some co-processors are needed e.g. for decompression etc so decompression doesn't hurt main memory bandwidth to much (while the cpu & GPU also access the memory). The xbox series X APU has <70MB of caches does that say anything about SSD access times or using the SSD as RAM? No it doesn't.
4) even without looking at the latencies, write speed is still way to low
5) Just look at the benches of the PS5 processor kits. Those get the performance of a ryzen 1800x with boost clocks higher than the CPUs of the PS5 or Xbox. But the GDDR6 memory really hurts the CPU performance. We know from the Zen2 mobile CPUs that the cache reduction didn't hurt the performance so much. But memory latencies are essential for Ryzen processors. Even more as the cache is so "small".

Please, stop throwing buzzwords around. Buzzwords are there for marketing purpose and nothing else.
 

Dream-Knife

Banned
RTX IO brings the hardware component (GPU, the alternative to custom IO complex that consoles have) for decompression.
DirectStorage brings the software component (new storage API stack) to utilize PCIe-based SSDs to their full potential and eliminate load times.
DirectStorage implies that the GPU will do the decompression regardless.
 

CuNi

Member
Of course it will translate, once games start being made with the new consoles in mind.

Then, no, it doesn't mean infinite detail, but it will mean far more variety etc, and no more tricks to hide loading. Also smaller game sizes, because assets won't need to be repeated over and over again.

But again, you're incredibly wrong about SSD being a "workaround". It's essential. If anything, having lots of RAM is a workaround for main storage being too slow. We could do with a much smaller amount otherwise.

With it not being translated he meant graphics fidelity. Ofc Game-Mechanics will change, like you said less hiding loading screens etc. but removing loading screens is not improving graphics fidelity.
When it comes to "looks" everything is still bound by CPU/GPU power. You could have infinite bandwidth but you would not be able to push Real-Life Graphics simply because the GPU is not able to handle it.
Smaller game size is not because SSDs are 5GB/s but because they work differently than a HDD. This could have been achieved on a normal SATA3 SSD too (the removal of asset duplication obviously).

So yes, SSD enables new gameplay mechanics, like the Portal hopping in Ratched and Clank, but the game is not looking as good as it is because of SSD but because of CPU/GPU advancement. If you'd have developed R&C on a normal SSD or even HDD, you could achieve the same graphics fidelity but with loading screens in-between portal jumps.


tl;dr
Yes, SSDs and fast I/O are "game-changers" but not "graphics-changers" but we are still all happy that we got them as we all look forward to the crazy new gameplay mechanics that they enable to be made.
 

RoadHazard

Gold Member
With it not being translated he meant graphics fidelity. Ofc Game-Mechanics will change, like you said less hiding loading screens etc. but removing loading screens is not improving graphics fidelity.
When it comes to "looks" everything is still bound by CPU/GPU power. You could have infinite bandwidth but you would not be able to push Real-Life Graphics simply because the GPU is not able to handle it.
Smaller game size is not because SSDs are 5GB/s but because they work differently than a HDD. This could have been achieved on a normal SATA3 SSD too (the removal of asset duplication obviously).

So yes, SSD enables new gameplay mechanics, like the Portal hopping in Ratched and Clank, but the game is not looking as good as it is because of SSD but because of CPU/GPU advancement. If you'd have developed R&C on a normal SSD or even HDD, you could achieve the same graphics fidelity but with loading screens in-between portal jumps.


tl;dr
Yes, SSDs and fast I/O are "game-changers" but not "graphics-changers" but we are still all happy that we got them as we all look forward to the crazy new gameplay mechanics that they enable to be made.

It will translate into visuals as well, because you will be able to have much more detailed scenes etc since you can load more data in any given moment. The main thing limiting that in open world games last gen was not the GPU. Take a look at the GDC talk on Spider-Man if you haven't. The reason that game has tons of repeating assets and rather low variety in the environments is not because the GPU couldn't render more complex scenes, it's because the data simply couldn't be loaded fast enough.

Of course the GPU also becomes a limiting factor, but mostly in other ways (number of pixels shaded, etc).
 
Last edited:

CuNi

Member
It will translate into visuals as well, because you will be able to have much more detailed scenes etc since you can load more data in any given moment. The main thing limiting that in open world games last gen was not the GPU. Take a look at the GDC talk on Spider-Man if you haven't. The reason that game has tons of repeating assets and rather low variety in the environments is not because the GPU couldn't render more complex scenes, it's because the data simply couldn't be loaded fast enough.

Of course the GPU also becomes a limiting factor, but mostly in other ways (number of pixels shaded, etc).

The last bolded part is exactly what we are talking about. The fast paced movement of Spiderman was a game mechanic. So there bandwidth was the bottleneck.
But that's a symptom of game design and game mechanics. Like I said, take Ratched and Clank as a example. You could achieve most likely the very exact same game if you'd put loading screens on the rifts.
Ofc that would destroy the flow of the game etc. but that's not the point. Bandwidth only matter for graphics if you decide on a fast paced game, which is a game mechanic decision. If you aim for best graphics, you'll always get your GPU close to 100% utilization. The only difference would be that on a HDD the game would be slow paced, something like Death Stranding where with a SSD you could go for a more fast paced game like Spiderman or even a flight sim because you'd see constantly changing geometry.

But all of the above is a game mechanic decision. It doesn't allow you to create better looking games. It allows you to create equally good looking games with different gameplay.
 

RoadHazard

Gold Member
The last bolded part is exactly what we are talking about. The fast paced movement of Spiderman was a game mechanic. So there bandwidth was the bottleneck.
But that's a symptom of game design and game mechanics. Like I said, take Ratched and Clank as a example. You could achieve most likely the very exact same game if you'd put loading screens on the rifts.
Ofc that would destroy the flow of the game etc. but that's not the point. Bandwidth only matter for graphics if you decide on a fast paced game, which is a game mechanic decision. If you aim for best graphics, you'll always get your GPU close to 100% utilization. The only difference would be that on a HDD the game would be slow paced, something like Death Stranding where with a SSD you could go for a more fast paced game like Spiderman or even a flight sim because you'd see constantly changing geometry.

But all of the above is a game mechanic decision. It doesn't allow you to create better looking games. It allows you to create equally good looking games with different gameplay.

Yes, but if you do want a game where you can move faster than very slowly, the visual level you can achieve will be severely limited by an HDD. We even see that in slow-paced linear last-gen games, where you need to wait in elevators or slowly squeeze through crevasses to allow the data for the next area to load in.
 

Lysandros

Member
With regard to the PS5's IO complex with an effective LLC (esram), in addition to the SSD's custom controller using 6 priority levels - and uses an 8 core ARM board with its own memories IIRC - that is a simplified take that doesn't really stack up IMHO.

Looking back at the use of RAM + edram and esram in the 360 or X1, and the results of the games/resolutions/fidelity we have nearly 2 decades worth of physical evidence to say that using those small memory apertures did make the RAM work "sort of like" VRAM, which seems semantically no different to what Cerny said about hardware like the io complex/ssd controller and ssd facilitating latency hiding to work like memory.

The 6 priority levels, in a satellite board, means that cancelling requests for optimistic data requests via the slower priorities incurs no burden for the Zen2 CPU or IO complex AFAIK, and the cache scrubbers will also dove tail with the IO's latency hiding while streaming data just in time.
Good post. What is the source for the flash controller having 8 arm cores? (By the way these aren't the cores of the I/O complex of course.)
 

Dream-Knife

Banned
DirectStorage and RTX IO are two different things that work in tandem. They aren't the same thing with different brandings like you assumed.
NVIDIA RTX IO plugs into Microsoft’s upcoming DirectStorage API, which is a next-generation storage architecture designed specifically for gaming PCs equipped with state-of-the-art NVMe SSDs, and the complex workloads that modern games require. Together, the streamlined and parallelized APIs, specifically tailored for games, allow dramatically reduced IO overhead and maximize performance/bandwidth from NVMe SSD to your RTX IO-enabled GPU.

Specifically, NVIDIA RTX IO brings GPU-based lossless decompression, allowing reads through DirectStorage to remain compressed while being delivered to the GPU for decompression. This removes the load from the CPU, moving the data from storage to the GPU in its more efficient, compressed form, and improving I/O performance by a factor of 2.
 
Top Bottom