• 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.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.

Corndog

Banned
Matt on the other forum who has a track record that goes back years, and has access to both dev kits has said the difference in I/O to be significant.

Another verified insider here shared the same opinion.

Tim Sweeney’s comments and U5 demo.

Etc

I mean at this point if you believe otherwise it’s because you have too much affection for Xbox and can’t let that image gain form in your mind.

Some people believe PS5 to have a more powerful gpu so, hey, there are those too.
What does any of that have to do with me implying what he said was opinion since it was not based in actual facts? He said BCPack was in Xbox one. Did you actually read what he wrote?

Also please show where I have stated ps5 ssd is not faster then the series x ssd.
 

bitbydeath

Member
Similar to the shape from road to ps5, no?
Or that’s the shape of the PS5. It looks similar to the one on Cerny’s presentation.

I thought it resembled a seven on my phone but looking at it closer on my computer it isn't.
Close to the BC image but more rounded.

SQx9xKJ6aXa3w4BZzgUy3h.jpg


EXwxfOKUwAAVgWg
 

Bo_Hazem

Banned
An NVMe controller isn’t a generic programmable chip that can wholesale be changed. There is potential to upgrade its firmware, but if you want it to run some entirely different protocol or handle more work than originally designed you’re likely to run into problems.

It seems you’ve convinced yourself that this is how it all works by finding what you wanted to find and having it fit what you’ve already decided is true, but this directly contradicts what Cerny himself has said, so don’t go posting this as some kind of confirmed fact.

Cerny emphasized the word “true” when saying NVMe only has two true levels of priority. That could mean fundamentally at the hardware level, even if there is software on top that arbitrates requests.

Until you know what was meant by that, you’re just speculating while contradicting when he said the main chip will arbitrate the extra levels of priority when using a PC NVMe drive.

Secondly, the idea that PS5 would update the firmware on an NVMe drive you bought sounds a bit daft and assumes the user doesn’t want to remove it and use it elsewhere at a later date.

If adding queues in software is all it takes, why would Cerny say the main custom chip has to arbitrate the extra queues and the NVMe expansion drive needs a little more speed to do that?

You’re posting speculation, not fact.

Great explanations. This lines up with the concerned attitude of Mark Cerny when he was referring to top-tier 7GB/s NVMe m.2 PCIe 4.0 that would need extra work from the I/O to compensate for its lack of 4 extra priority levels, and wasn't sure if that speed and method will match the internal SSD results. He was pretty clear when he said wait after launch which means 2021 when probably something Sony will be cooking:

(Timestamped, thanks to Rusco Da Vino Rusco Da Vino again)


EDIT:
I agree the 12 channel interface will be a balance between getting the 5+ GB/s speed required and the cost of the flash memory. There could also be an element of heat management, too. If you can pull 5+GB/s over 12 different chips, that might spread heat better than less channels over more exotic chips. Flash memory performance degrades with heat.
I wouldn’t be surprised if spreading the load over 12 channels and 12 slower chips actually means they can keep the flash memory cool enough to sustain peak performance.
Something necessary if your game is built around those speeds.

That's also what I thought about, heat management and sustained peak speeds. Also doesn't that mean as well stable speeds of data as every module has its own channel? Meaning for one game data it can be scattered in more than 1 module which can result in latency if not each module has its own exit channel. Mark Cerny was clear that he didn't go over 12-channel to avoid extra costs. Meaning he might go for 14-15 channels just to reach ~1TB?
 
Last edited:

Md Ray

Member

PaintTinJr

Member
Let's say BCPack with that name doesn't exists but part of BC exists since many time ago and was available for the current gen of xbox and PC:

https://docs.microsoft.com/en-us/windows/win32/direct3d11/texture-block-compression-in-direct3d-11
https://docs.microsoft.com/en-us/wi...programming-guide-resources-block-compression

Even before was not called BC was DXT

https://docs.microsoft.com/en-us/windows/win32/direct3d9/using-compressed-textures

Probably BCPack include a new kind of compression maybe not and is only hardware (which itself is a good improvement).

Let me just quote one line of BC of Direct 10


That sounds familiar right.

Is so common than now you can some of this BC you can use in any big commercial engine:
https://docs.unity3d.com/es/2018.4/Manual/class-TextureImporterOverride.html
https://docs.unrealengine.com/en-US...extures/TextureCompressionSettings/index.html

That means BCPack is exactly the same than BC, not I am sure not because put a nice name to already existing technology and then saying is a very important
part of you new console will be not correct even dishonest.
Even if the functionality can or has been achieved already, I get the impression that decompression for BCPack (Block compress pack) is actually hw accelerated in the texture filtering of the XsX GPU, so that the when Xbox say their sfs(?) is unique to xbox and has 2-3x gain, they are refering to BCpack hw filtering and double counting the BCpack compression gain. But in reality, if the texture in a double block compressed format can be decompressed as late as the filtering, then that is a very good memory saving that other platforms won't get - even if the point is moot if the lossy compression reduces image quality and is competing against 8K raw textures on PS5.
 
Guys whats happening I've been playing space haven all day now I read all those things that seem to point at something thats about to happen.
Tension is building up.
I need to know whats going on!
 
Guys whats happening I've been playing space haven all day now I read all those things that seem to point at something thats about to happen.
Tension is building up.
I need to know whats going on!

Bro, i'm sitting here working from home, frantically refreshing this thread because in the last hour there has been a surge in posts here.

Have no idea whats going on but also losing my shit right now LOL,

WHATS GOING ON GUYS ?????!!!!!!

giphy.gif
 
Last edited:
Doesn’t the exact same thing apply to ps5? Since I’m sure PS4 used some form of texture compression. I know PC Cards use compression as well.
Yes also apply, PS4 use Zlib and PC also have many formats. But for now use compression like Kraken to that bandwidth quantity without IO specialized will need
many CPU cores for a PC maybe is no a problem if has 16 cores but for a console that is not an option that is why the have reduce even eliminate the use of CPU
for decompression.

CD2HJkS.jpg



The difference is Cerny don't put a extravagant name to use Kraken, he just said we support now also Kraken which is 10% better than zlib, already the zlib is around 50% we have
to remember here we care not only how much data is compressed also how much of data is compressed and how much can be decompressed for second and not all
the data can be compress equal ratio.

With Kraken:
5.5 (raw bandwidth)*1.5*1.1=9.075 (the max normal bandwidth)

With only ZLib:
5.5*1.5=8.25 (close to the 8 GB/s)

From those number is where Cerny say between 8-9 Gb/s, with XSX the operation should be a little different as BCPack is only used for textures from this number they got
4.8 GB/s.

Of course both chips for decompression in both consoles care manage more data but in real situation in not probably never reach its full potential.
 
Even if the functionality can or has been achieved already, I get the impression that decompression for BCPack (Block compress pack) is actually hw accelerated in the texture filtering of the XsX GPU, so that the when Xbox say their sfs(?) is unique to xbox and has 2-3x gain, they are refering to BCpack hw filtering and double counting the BCpack compression gain. But in reality, if the texture in a double block compressed format can be decompressed as late as the filtering, then that is a very good memory saving that other platforms won't get - even if the point is moot if the lossy compression reduces image quality and is competing against 8K raw textures on PS5.
My point is BCPack is just a library they improved and renamed again which the last version already exists also in PC, the main change is now they hardware to support it regarding . But people
put crazy because they heard things like 'BCPack will reach and overcome and exceed the bandwidth of PS5' that is a nonsense.

With SF is a similar situation the difference is that is already compatible with Nvidia turing so is not exclusive (SFS is part of SF).
 

Corndog

Banned
My point is BCPack is just a library they improved and renamed again which the last version already exists also in PC, the main change is now they hardware to support it regarding . But people
put crazy because they heard things like 'BCPack will reach and overcome and exceed the bandwidth of PS5' that is a nonsense.

With SF is a similar situation the difference is that is already compatible with Nvidia turing so is not exclusive (SFS is part of SF).
Maybe a few people say it will make it faster then ps5 but most understand it is just an optimization. Promoting it as a benefit is not saying something like BCPack will eliminate the bandwidth difference between the consoles.
 

Entroyp

Member
My point is BCPack is just a library they improved and renamed again which the last version already exists also in PC, the main change is now they hardware to support it regarding . But people
put crazy because they heard things like 'BCPack will reach and overcome and exceed the bandwidth of PS5' that is a nonsense.

With SF is a similar situation the difference is that is already compatible with Nvidia turing so is not exclusive (SFS is part of SF).

I haven’t been following this thread for months, what’s Behind BCP textures making up for 2x the disk speed? How is that even possible?
 
Status
Not open for further replies.
Top Bottom