Playstation 5 [PS5] [Release November 12 2020]

Based on the numbers we have for near lossless, it seems quite more savings than "making the final package a bit smaller"(whatever that means)

It means someone wants BCPack to be better, somehow.


Funny thing is RDO probably boosts the SeriesX's effective I/O throughput by a lot too, without any discernible impact in texture quality, but saying Oodle is bad takes precedence.
 
A lot of this really cool technology of Oodle applies to all platforms. It's only the Kraken hardware decompression which is specific to PS5. I think they even have GPU assisted decompression on last-gen hardwarel ike PS4/XboxOne and PCs.
 
A lot of this really cool technology of Oodle applies to all platforms. It's only the Kraken hardware decompression which is specific to PS5. I think they even have GPU assisted decompression on last-gen hardwarel ike PS4/XboxOne and PCs.
Zlib decompressor, seems no much used for being slow?.
 
Massive overengineering for some things, while leaving other things lacking (cough memory bandwidth on PS4 Pro and PS5 cough), yep that sounds like Cerny.

Also this sub is weirdly obsessive with Oodle. It just makes the final package a bit smaller guys, that's it, and it's lossy compression on top of the already lossy block compression formats as it is. Honestly only packing Z-lib kinda sucks, for compression of the game without the CPU getting involved in decompression Microsoft's BCPack is probably a good deal better. Deflate compression ratios for textures suck, you need image specific compression schemes to get good ratios. BCpack could easily fit another 4:1 lossless compression or better with realtime asic decompression, which the z-lib just doesn't match.

The pending question we had about the PS5 was whether significantly over 2:1 compression was going to be seen on some real world data and if so how often, which led to other questions: Is it a good balance to design the entire pipeline for 22GB/s, and is it a large area on the die. We now have benchmarks with real world data using the exact same technology in the PS5. We know geometry can compress 3:1, and we have 4:1 compression on some real world datasets. We now have the proof it's a usable figure. This is true regardless of the total average of a game install size.

It's disingenuous to claim that the silicon is massively overengineered unless the newly defined goal post becomes... a peak of 22GB/s is not useful anyway, and 6+GB/s should be enough for everybody.

What we do not have, is BCPack benchmarks on the same dataset, because nobody knows what BCPack even is. We have a peak figure of 6+GB/s according to MS interviews. If that is correct, it cannot improve bandwidth even if it compressed 1000:1. But this is a technical PS5 thread, not a comparison or rumor thread.
 
Last edited:
A lot of this really cool technology of Oodle applies to all platforms. It's only the Kraken hardware decompression which is specific to PS5. I think they even have GPU assisted decompression on last-gen hardwarel ike PS4/XboxOne and PCs.

Yeah, sony has licensed it for ps4 as well as ps5.

But iirc the whole big news came out because of the savings with kraken and oodle combined(which is a combination on console unique to ps5 right?)
 
But iirc the whole big news came out because of the savings with kraken and oodle combined(which is a combination on console unique to ps5 right?)

From all the material provided on Kraken they describe it as around 15% more compression than others. In this situation the ratios seemed to be 3.99:1 vs 3.36:1, so about 18% more space savings.

Where Kraken really shines is the speed of decompression. I wish all systems had Kraken decompression hardware, including PCs.
 
I wish all systems had Kraken decompression hardware, including PCs.

I think it's intresting that only PS5 has it, so comparisons between platforms become less dull. Going to be intresting to see what developers can come with for ideas with the different hardware in each console.

Isnt it third party middleware? Surely pc can use it fine?

Yes but Kraken is a hardware feature, atleast as in PS5?
 
Isnt it third party middleware? Surely pc can use it fine?

Yes. But it's not built into PC hardware. This is one of the unique aspects of PS5 hardware where it really shines.

On the PC you'd need to shift some CPU/GPU cores to do the decompression (just like using it on the PS4/XO/XBSX). Short of AMD building in acceleration instructions that could be used for Kraken decompression, the PCs will be behind the PS5. I'm thinking along the lines of the AES instructions being created for the acceleration of SSL/VPN/Cryptography uses.
 
Anyone can license Kraken and use it. What the PS5 has is a hardware decompression block that supports Kraken and zlib, and it's much faster than any number of Zen2 cores (because decompression can only use a couple of threads). Throwing more cores at decompression won't make it any faster, or at least not even close to the throughput of the SeriesX let alone the PS5.
Also, IIRC Sony purchased a Kraken license for every PS5 devkit out there, so every PS5 developer can use it.
 
Frees up GPU usage, as i cant imagine doing this compression on the cpu is ideal, you can't force everyone on a threadripper build. How many TF's would you need to decompress on the GPU to match the PS5, if even possible? Explains why forbidden west can implement ray tracing at such advanced/dense graphics.
 
Anyone can license Kraken and use it. What the PS5 has is a hardware decompression block that supports Kraken and zlib, and it's much faster than any number of Zen2 cores (because decompression can only use a couple of threads). Throwing more cores at decompression won't make it any faster, or at least not even close to the throughput of the SeriesX let alone the PS5.
Also, IIRC Sony purchased a Kraken license for every PS5 devkit out there, so every PS5 developer can use it.
Pretty sure Fabian said it was like 3-5GB/s on a single core with a Skylake CPU @ 3Ghz and you can throw more cores at it.. so presumably it would make it faster.

 
Frees up GPU usage, as i cant imagine doing this compression on the cpu is ideal, you can't force everyone on a threadripper build. How many TF's would you need to decompress on the GPU to match the PS5, if even possible? Explains why forbidden west can implement ray tracing at such advanced/dense graphics.

I haven’t seen anywhere that HFW is using ray tracing. Not that it matters, if it does end up as good as that trailer then that’s amazing.
 
A lot of this really cool technology of Oodle applies to all platforms. It's only the Kraken hardware decompression which is specific to PS5. I think they even have GPU assisted decompression on last-gen hardwarel ike PS4/XboxOne and PCs.

Yes, Oodle can improve the performance (either compression ratio or optimise for faster decompression) of zlib, LZ4 and 7z/LMA-based decompression algorithms.

Anyone can license Kraken and use it. What the PS5 has is a hardware decompression block that supports Kraken and zlib, and it's much faster than any number of Zen2 cores (because decompression can only use a couple of threads).

The industry standard remains zlib and kraken is approx 11x faster decompressing. It's worth noting that many decompression algorithms can use more than two cores/threads though, it depends on how the data was compressed.If you're compressing data using an oldskool file-by-file basis, which obviously has drawbacks, you decompress as many files in parallel as you can spawn decompression threads.

oodle260_typical_combined_small.png
 
Last edited by a moderator:
Pretty sure Fabian said it was like 3-5GB/s on a single core with a Skylake CPU @ 3Ghz and you can throw more cores at it.. so presumably it would make it faster.



Kraken uses two threads, and the second thread provides a 70% speedup.


Kraken can decode on two threads!



Kraken has an unusual capability to decode normal Kraken compressed data with two threads. This allows for about a 1.7× speedup with 2 threads, which puts Kraken decode speed even farther out of the ballpark.
No special preparation of the data is needed (such as the chunking that is done for more primitive threaded decoders), and no compression ratio is lost. That is, existing Kraken-compressed data can be directly loaded by the threaded Kraken decoder.

We provide pre-threaded decoders ready to go in the Oodle SDK, or you can easily run it on your own threads.

I don't know who Fabian is, but throwing "more cores" at Kraken seems useless if we go by Kraken's own documentation. You can throw one more, and you get a sizeable 70% improvement (meaning 5-8.5GB/s). We don't even know if throwing another core (instead of just using the second thread on that Skylake). Also, 5GB/s on a CPU core would put all the console makers' claims a blatant lie regarding the performance of their dedicated decompressors. Is he really talking about decompression and not Oodle Texture RDO recoding?

In that same thread, he also implies they got decompression working on GPU compute shaders for the PS5. Which is odd because the PS5 shouldn't be using compute shaders to decompress Kraken.
 
Also, 5GB/s on a CPU core would put all the console makers' claims a blatant lie regarding the performance of their dedicated decompressors.
Not been following this, but is Sony/ms claims specifically regarding kraken.
If not then why would it mean their lieing even if these particular figures were true?
 
Not been following this, but is Sony/ms claims specifically regarding kraken.
If not then why would it mean their lieing even if these particular figures were true?

If one Skylake core at 3GHz does 3-5GB/s, then one Zen2 core at 3.5GHz should do more than that. Sony is claiming the decompressor corresponds to 9x Zen2 cores. Unless these hypothetical cores are working at 300MHz, claiming the decompressor is doing the same as 9 Zen2 cores is a lie.
 
Can't Nvidia and AMD just license Kraken (or BCPack) and put its functionality onto the Ampere/RDNA2 SoC as a small decompression block? Surely this would be more sensible than asking gamers to upgrade to a 12-core CPU to match console speeds.

Is such a thing even possible? Sorry for my noob question.
 
Back
Top