Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

[megre]

What about Intel Optane, could MS get a deal with Intel, and that way be faster then Sony?

I really hope they could. I believe Optane can also go up to 25gb/s. I am of the opinion that neither has an advantage if both are using nand chips.

 

[manux]

One can buy nvme4 ssd for pc that will have 6-7GB(byte)/s read speeds. It just doesn't do squat for loading times. Optane falls to same category. Loading time/asset streaming is complicated problem that's not solvable by just having large number in specifications.

 

[megre]

One can buy nvme4 ssd for pc that will have 6-7GB(byte)/s read speeds. It just doesn't do squat for loading times. Optane falls to same category. Loading time/asset streaming is complicated problem that's not solvable by just having large number in specifications.

Both have the potential to solve this problem. It's just unbelievable to me that it's only Sony that seemingly have solved it.

 

[Deleted member 11852]

One can buy nvme4 ssd for pc that will have 6-7GB(byte)/s read speeds. It just doesn't do squat for loading times. Optane falls to same category. Loading time/asset streaming is complicated problem that's not solvable by just having large number in specifications.

It's boggling that more folks don't understand the I/O chain, nor the steps involved using that I/O and where CPU and GPU computation fits into that chain, that transforms data read from disc into an actual game.

 

[Nisaaru]

Why would you need a controller when the SSD "cold storage" only job is to load to the "warm storage"?

You still need a PCI->Flash controller which translates IO jobs/memory accesses and remaps dead pages.

 

[MrFox]

It's boggling that more folks don't understand the I/O chain, nor the steps involved using that I/O and where CPU and GPU computation fits into that chain, that transforms data read from disc into an actual game.

I like that we now have benchmarks with multiple nvme in raid0 on PC that still show 20secs load times. That should be the biggest eye opener.

 

[iroboto]

I like that we now have benchmarks with multiple nvme in raid0 on PC that still show 20secs load times. That should be the biggest eye opener.

That has to do with the fact that it's not loading in parallel. not because it's the speed of the drive.
If games are designed to load like SATA3 (sequentially) then nvme will never be taken advantage of.

 

[Remij]

It's boggling that more folks don't understand the I/O chain, nor the steps involved using that I/O and where CPU and GPU computation fits into that chain, that transforms data read from disc into an actual game.

I'd love to learn more about it as I only have a fairly basic understanding. Any recommended reading/viewing you could link me? I'd appreciate that.

 

[goonergaz]

That escalated quickly...


I never mentioned production cost nor subsidy rates.
I just think 600€ one year from now wouldn't be that big of a deal. Certainly not as game breaking as 600€ in 2006.

The 2013 consoles were developed during a massive financial crisis within the western countries. Next generation consoles were not, so we should all expect higher launch prices.

Save for a low end SKU with less RAM and slower GPU (that I still think is a bad idea BTW), we shouldn't expect any console to cost less than €500.



I don't take pleasure in people getting banned. Do a pledge bet instead.

Maybe I read too much into the reply, but I’m obviously aware PS3 was 599 and Sony learnt the lesson last gen so they won’t repeat it this gen...they won’t increase the entry level by 50% in just a few years - that’s ridiculous.

Add to that Sony have been quite vocal regarding costs, I totally expect PS5 to be more but not 50% more!

 

[megre]

That has to do with the fact that it's not loading in parallel. not because it's the speed of the drive.
If games are designed to load like SATA3 (sequentially) then nvme will never be taken advantage of.

If games are to be developed with SSD in mind then both PS5 and XSX are going to benefit. It's the key to next-gen for both.

 

[PSman1700]

I really hope they could. I believe Optane can also go up to 25gb/s. I am of the opinion that neither has an advantage if both are using nand chips.

Damn, 25gb/s, why doesn't Sony go with this Optane Hope it gets cheaper so i can have it sometime, could be for good use in non-gaming situations too i guess.

PS5 will have more TFs, HBM memory and 2-3x faster SSD at lower price.

Sony is killing it, MS better pack in and forget about competing, and sent their resources elsewhere.

One can buy nvme4 ssd for pc that will have 6-7GB(byte)/s read speeds. It just doesn't do squat for loading times. Optane falls to same category. Loading time/asset streaming is complicated problem that's not solvable by just having large number in specifications.

There will probably come a solution sometime though, i can imagine. Thing is that the gaming market is kinda small compared to what those SSD's are being put to use for.

Both have the potential to solve this problem. It's just unbelievable to me that it's only Sony that seemingly have solved it.

Where did they announce this, specs, and information that they have a superior solution, or is it just guesswork?

That has to do with the fact that it's not loading in parallel. not because it's the speed of the drive.
If games are designed to load like SATA3 (sequentially) then nvme will never be taken advantage of.

The much critized star citizen already utilizes an SSD to an extend that a normal HDD is unplayable. It is a small look in to what is going to come.

If games are to be developed with SSD in mind then both PS5 and XSX are going to benefit. It's the key to next-gen for both.

Dito for PC, mechanical spinning HDD's are a thing of the past, same for non-RT GPU's, and less then 8 core CPU's, storage solutions go forward too, finally.

Another thing, that XSX thing, where you can exit a game, and jump in another time, anywhere in the game, is that the same idea as 'savestates' in emulators? For example, i have PCSX2, can do a savestate with a hotkey, and the next day, i just press a hotkey and i'm exactly at that point in the game where i left/did the savestate, there's no loading times or anything.

 

[turkey]

Another thing, that XSX thing, where you can exit a game, and jump in another time, anywhere in the game, is that the same idea as 'savestates' in emulators? For example, i have PCSX2, can do a savestate with a hotkey, and the next day, i just press a hotkey and i'm exactly at that point in the game where i left/did the savestate, there's no loading times or anything.

Next gen: bringing the Bosskey

 

[Osamar]

You have an idea though how much that is? I mean MS surely is not buying a top of the line off the shelf SSD for cost reasons.

I have used that SSD as an example because it has the same controller as the supposed PS5 devkits specs.

Why need SONY or MS to use exotic untested (RERAM) and/or expensive (OPTANE) solutions when the problem is solved? You will use much less resources bypassing all the breaks that we see on PC with SSDs, than developing system with exotic ram.

I have no idea how much it is, but I am sure that buying thousands of controllers and NAND chips is much more cheaper than the price of the SSD.

 

[PSman1700]

Why need SONY or MS to use exotic untested (RERAM) and/or expensive (OPTANE) solutions

Because that's what we like, for some reason, exotic things.

 

[manux]

I'd love to learn more about it as I only have a fairly basic understanding. Any recommended reading/viewing you could link me? I'd appreciate that.

Anandtech article is good read before considering how to make game utilize ssd efficiently: https://www.anandtech.com/show/2738 . Optane would be simpler than ssd to understand as it has less restrictions on data access.

 

[BRiT]

Please remember the ReRam thread is over here: https://forum.beyond3d.com/threads/...nd-cant-reram-bring-to-a-console-spawn.61480/

Please remember the ReRam thread is over here: https://forum.beyond3d.com/threads/...nd-cant-reram-bring-to-a-console-spawn.61480/

DejaVu all over again.

 

[Deleted member 11852]

I'd love to learn more about it as I only have a fairly basic understanding. Any recommended reading/viewing you could link me? I'd appreciate that.

You can google primers for how PCs work and how all the individual components are connected via the various buses but a good start the Wikipedia article on the northbridge, which leads on the CPU, southbridge and the traditional I/O. From this you'll get an understanding of how data pulled from disc gets to the RAM, the CPU and the GPU.

In terms of how this arrangement presents challenges, I'd recommend consuming as many GDC technical talks as possible because these are usually delivered by game devs and they will explain the challenges their games are faced with. Presentations on engines or games which stream would be favourite. Lots of these are linked in these forums so you'll have to hit search!

 

[megre]

With the CPU potentially being the bottleneck and not the bandwidth, why not allot a portion of the SSD as a scratchpad where games will be uncompressed and unloaded? It will free up CPU bandwidth and resources at run-time.

I'm talking about a possible 1TB 4gb/s SSD. Is it not a good idea to reserve 64gb - 128gb as a scratchpad to where pertinent and immediate assets will be constantly loaded and unloaded to? The system can have a simple secondary chip for the decompressing job too.

Imagine decompressing 2gb/s of data at run-time. How much CPU or GPU bandwidth and resources do you need for that? At load-time I assume all CPU and GPU cores and full bandwidth can be used for decompression, but at run-time you don't have that luxury. Now if your data is uncompressed then you don't have that problem.

 

[Deleted member 11852]

With the CPU potentially being the bottleneck and not the bandwidth, why not allot a portion of the SSD as a scratchpad where games will be uncompressed and unloaded? It will free up CPU bandwidth and resources at run-time.

Break down how you arrived a this conclusion. To start off, I'd recommend working out the CPU/RAM bandwidth, GPU/VRAM bandwidth, RAM/VRAM (PC only - moving either way), and then look at the bandwidth of the southbridge and northbridge. How is the solid-state attached to the system?

I'm talking about a possible 1TB 4gb/s SSD. Is it not a good idea to reserve 64gb - 128gb as a scratchpad to where pertinent and immediate assets will be constantly loaded and unloaded to? The system can have a simple secondary chip for the decompressing job too.

You don't want any immediately-needed game assets residing only on solid state storage under current implementations. It's way too slow.

 

[megre]

Break down how you arrived a this conclusion. To start off, I'd recommend working out the CPU/RAM bandwidth, GPU/VRAM bandwidth, RAM/VRAM (PC only - moving either way), and then look at the bandwidth of the southbridge and northbridge. How is the solid-state attached to the system?

That's quite a task for a non-dev like me. I'll tell how I came up with that question though so you know where I'm coming from.

I ran a search on google to know what it is that bottlenecks games to load fast these days. A lot of the answers I got is that the bottleneck has shifted to CPU now because it takes awhile to decompress assets.

Then I remembered what JoeJ told me in PM. Now don't mind that we were talking about ReRAM, I won't go there.

Assuming magic compression exists, CPU could uncompress insane details to ReRam so it is available quickly, e.g. when player turns view.
Then my initial Megatexture argument would make more sense again, if we make some assumptions:
* decompression very expensive, so need to cache full environment around player, not just what's currently on screen. 10 x more data.
* Sub millimeter texture resolutions everywhere. 100x100 more data?

Now I recognize that ReRAM may be too expensive and maybe not ready yet. So I thought about a regular SSD doing the same thing that JoeJ told me. A 1TB SSD at 4GB/s alloting a small portion of it at about 64-128GB functioning as a scratchpad where assets can be uncompressed so that immediately-needed game assets can be used without ever needing to steal CPU bandwidth and cores or GPU resources. A 4GB/s SSD can fill a 16GB RAM in 4 seconds. I don't know how much it'll take if the data is compressed or how much CPU bandwidth and resources it needs to compare to the uncompress scenario.

You don't want any immediately-needed game assets residing only on solid state storage under current implementations. It's way too slow.

I see. You can forgive me then for wanting a cost-effective SSD cache to happen.