Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

The 3.16:1 definitely isn't optimal because in an earlier post he showed an example of another texture set that provided a 3.99:1 compression ratio when using kraken + oodle texture, which on the PS5 would actually saturate the maximum 22GB/s of the decompressor's output.

View attachment 5572


"Close to 2:1" is what Kraken does without Oodle Texture, so saying Kraken + Oodle Texture only averages at 2:1 doesn't make much sense.
View attachment 5573

These are contextless comparisons. Are these entire games? Or specific hand picked data sets? Are they using the same level of Kraken compression that the PS5 decoder is designed to handle?

The bottom line is that despite the individual comparison charts showing compression ratio's for specific texture sets, they are absolutely explicit in the statement:

"Sony has previously published that the SSD is capable of 5.5 GB/s and expected decompressed bandwidth around 8-9 GB/s, based on measurements of average compression ratios of games around 1.5 to 1. While Kraken is an excellent generic compressor, it struggled to find usable patterns on a crucial type of content : GPU textures, which make up a large fraction of game content. Since then we've made huge progress on improving the compression ratio of GPU textures, with Oodle Texture which encodes them such that subsequent Kraken compression can find patterns it can exploit. The result is that we expect the average compression ratio of games to be much better in the future, closer to 2 to 1."

There is no room for interpretation there. And those numbers also align 100% with Sony's own marketing materials. Why would Sony undersell their consoles capabilities while at the same time the developers of Kraken and Oodle Texture make incorrect statements about their own softwares effectiveness? It makes no sense to assume anything other than exactly what RADGameTools and Sony are telling you right here and via all other official marketing material.
 
Last edited by a moderator:
These are contextless comparisons.
There's plenty of context given in the blogposts and the questions answered by the author and other RAD devs in the bottom.

Are these entire games? Or specific hand picked data sets?
They are full datasets of textures from actual games, as explained in the blogposts.


The bottom line is that despite the individual comparison charts showing compression ratio's for specific texture sets, they are absolutely explicit in the statement:
And so are the results provided by the author, all of which showing Kraken providing a near 2:1 compression ratio without Oodle Texture.

What he's talking about in that specific paragraph is total game size, as he mentions installation volumes and Kraken as a generic compressor. What he compares in the graphics are texture datasets.
In a game there's a bunch of data for which Kraken and Oodle Texture do virtually nothing, such as audio and video files. Those have a compression ratio of 1 because they don't compress at all, yet they're probably packed into Kraken files anyway, given the nature of the IO complex.

Kraken + Oodle Texture provide a >3x compression ratio for textures, so the IO complex provides the system with textures at >16GB/s.
It's when they average textures with a bunch of barely compressible data that they reach a 2:1 compression.



And those numbers also align 100% with Sony's own marketing materials. Why would Sony undersell their consoles capabilities while at the same time the developers of Kraken and Oodle Texture make incorrect statements about their own softwares effectiveness?

Sony's numbers precede Oodle Texture, whose results add up to Sony's initial estimates. It's right there in the text you quoted.
 
There's plenty of context given in the blogposts and the questions answered by the author and other RAD devs in the bottom.


They are full datasets of textures from actual games, as explained in the blogposts.



And so are the results provided by the author, all of which showing Kraken providing a near 2:1 compression ratio without Oodle Texture.

What he's talking about in that specific paragraph is total game size, as he mentions installation volumes and Kraken as a generic compressor. What he compares in the graphics are texture datasets.
In a game there's a bunch of data for which Kraken and Oodle Texture do virtually nothing, such as audio and video files. Those have a compression ratio of 1 because they don't compress at all, yet they're probably packed into Kraken files anyway, given the nature of the IO complex.

Kraken + Oodle Texture provide a >3x compression ratio for textures, so the IO complex provides the system with textures at >16GB/s.
It's when they average textures with a bunch of barely compressible data that they reach a 2:1 compression.





Sony's numbers precede Oodle Texture, whose results add up to Sony's initial estimates. It's right there in the text you quoted.

https://cbloomrants.blogspot.com/2020/09/how-oodle-kraken-and-oodle-texture.html

And the text comes from this blog article of the same Charles Bloom, oodle kraken with oodle texture will gives an average around 11GB/s. And this is after he gave some examples. This is the last blog post about the PS5.
 
Looks great. Waiting to be impressed by Nanite in something without rocks though.
It will be very interesting to see how it works on hard surface models.
If it can keep edges well, it would be amazing for lot of things. (Thinking spaceship designed similarly to Star Citizen.. lot's of well tweaked vertex normals etc.)
 
It's when they average textures with a bunch of barely compressible data that they reach a 2:1 compression.

Excellent, so we're now in full agreement. So if I may I would just like to update your earlier statement as follows (updates in red):

Single Core CPU decoding of zlib tops out at ~800MB/s and Kraken at ~1.8GB/s, although multicore will go much higher. The Series X decompressor has a ~7GB/s maximum output with BCPack and the PS5's can go all the way up to 22GB/s, but with Kraken + Oodle Texture it supposedly averages around 11GB/s.
 
Surely 11GB/s is plenty fast enough anyway.

It's the difference between loading a level in a second Vs half a second. When there's a bigger VRAM pool then faster is better, but based on the hardware it's more than sufficient.
 
Last edited by a moderator:
It will be very interesting to see how it works on hard surface models.
If it can keep edges well, it would be amazing for lot of things. (Thinking spaceship designed similarly to Star Citizen.. lot's of well tweaked vertex normals etc.)

It should be awesome for spaceships. Nanite doesn't lose any triangles from the original model. If it can cope with a bike* down to a modelled chain and cassette, it'll be fine with whatever granular detail's on a spaceship.

*Epic Brian randomly shoved one into the first demo during the Nanite talk
 
Single Core CPU decoding of zlib tops out at ~800MB/s and Kraken at ~1.8GB/s, although multicore will go much higher. The Series X decompressor has a ~7GB/s maximum output with BCPack and the PS5's can go all the way up to 22GB/s, but with Kraken + Oodle Texture it supposedly averages around 11GB/s.

I did mention it was for single core in that post. I even acknowledged that if they use Selkie the CPU decompression speed can get pretty high, at the cost of a lower compression ratio (which is less of a problem in the PC because storage is simpler to upgrade).

I think the confusion stemmed from the fact that when you mention Kraken + Oodle Texture, I'll assume you're talking about texture compression, not texture + everything else.
 
It should be awesome for spaceships. Nanite doesn't lose any triangles from the original model. If it can cope with a bike* down to a modelled chain and cassette, it'll be fine with whatever granular detail's on a spaceship.

*Epic Brian randomly shoved one into the first demo during the Nanite talk

Now you've got me dreaming of a nanite based Start Trek game. Star Trek Legacy is long overdue a sequel!
 
Everybody blamed me that i'm a fool that i'm addicted to micro polygons and that polygons comes first before raytracing if you want have nice looking worlds.

Now after nanite this view has changed lol :mrgreen:

Has it though? Most of the people that felt RT is king still feel that RT is king. Most of the people that felt like RT was just part of the picture still feel like RT is just part of the picture. I'm not sure much has changed WRT to how people view RT.

Regards,
SB
 
I have not seen anything really of substance in bcpack, do you have a link to the source of this?
There isn't.
I just picked the 2:1 compression ratio that was announced in some tweets which leads to 4.8GB/s, and with a lossy pre-process method like Oodle Texture I put another ~50% on top of it.
I do recall seeing the 7GB/s number somewhere, but I'm assuming it was calculated like mine.

Though it needs to be said this is for textures only. Average data will probably go below this number, like it does on the PS5 using Kraken + Oodle Texture.
 
Back
Top