RS-232 should suffice, with good enough compression.....
Clearly it's whatever is on the back of the PS5 - duh
Playstation 5 [PS5] [Release November 12 2020]
- Thread starter BRiT
- Start date
Clearly it's whatever is on the back of the PS5 - duh
You can even let it stand both ways...
D
Deleted member 13524
Guest
That poster is assuming the PS5 can use RDO on the fly. I don't know if that's true.
Though if it does, those numbers would explain why the decompressor has a ~24GB/s bus to the RAM, which sounded like a massive overengineering when Cerny said it would be doing 8-9GB/s on average.
Yes, it uses RDO on the fly. It's a preprocessing step and works naturally with kraken without any additional processing required when decompressing it.
I hope this could also mean smaller game size than expected. The SSD iops allowing deduplication of streaming data, the RDO on textures, and the 10% better compression of kraken, could mean the games remain closer to PS4 sizes. Hopefully mostly below 100GB install sizes even for the big AAA.
Last edited:
D
Deleted member 13524
Guest
Did someone on Sony's side confirm this? Is there something on the I/O Complex doing that pre-processing?
I do see those two coprocessors that could be doing that, but I don't know of any confirmation on Sony's part.
No it's how RDO works. It makes the data compress better with kraken or zlib. The preprocessing is actually done when packaging the game, it's basically part of the compression step during development. It requires nothing on the console, and it also works on PS4. The blog explaining Oodle Texture confirms this.Did someone on Sony's side confirm this? Is there something on the I/O Complex doing that pre-processing?
I do see those two coprocessors that could be doing that, but I don't know of any confirmation on Sony's part.
The IO processors are for moving and mapping data around and managing the file system without CPU intevention. Cerny said those are worth the equivalent of one or two zen2 cores in terms of processing requirement. In addition to the kraken block which is worth 9x zen2 cores.
I'm expecting a large die area for the IO complex, you don't get 11x zen2 cores of effective power without using a lot of silicon. Even if it has the ASIC kind of efficiency. Sony really banked on the storage for next gen, and it should show up on the BOM as more silicon for storage, and less silicon for the CUs. Maybe balancing out as a similar BOM as XBSX.
Last edited:
D
Deleted member 13524
Guest
Ok I think I get it now. Oodle Texture RDO simply repackages the already-BCn-compressed to increase compression rate, but the resulting textures are still in BCn format for GPU use. And when working with Kraken it provides an additional 20-50% compression rate. So assuming a ~50% compression rate from Kraken, and up to 50% of RDO-over-Kraken, we get 0.5*0.5=0.25. Assuming a 5.5GB/s throughput from the SSD to the decompressor block, we're then getting 5.5/0.25 = 22GB/s for textures.
I.e. Sony had to know RDO was coming. They weren't going to make a ~24GB/s bus from the IO Complex to the RAM to pass an average 8.5GB/s of effective data using Kraken, just to have Oodle now releasing by surprise a texture pre-processing software that takes full advantage of that bus.
Last edited by a moderator:
Interestingly the 8.5 GB/s average speed stated by Cerny (in real game, not short ideal benchmark) is without this Oodle texture RDO improvement.
D
Deleted member 13524
Guest
Exactly. His statements were a curve ball to the console's effective throughput with textures. He was actually sandbagging with the already impressive 8.5GB/s number.
Again, it’s because while you may not need 20GB of data to be transferred in 1 second, you may very well need, say, 3GB to be transferred in 0.15 seconds.
Another interesting thing is that the power required for decompression is nearly 5 times less than zlib.
So reaching that speed if they stayed with zlib would have required an impossibly gigantic amount of silicon and power. Maybe they got that speed "for free" when they made the decision to switch to kraken in 2017. Altough an ASIC and CPU code are very different, maybe it's not that dramatic in an ASIC.
ooKraken8:
3.99:1
1.3 enc MB/s
1919.3 dec MB/s
zlib9:
3.36:1
19.1 enc MB/s
438.9 dec MB/s
So reaching that speed if they stayed with zlib would have required an impossibly gigantic amount of silicon and power. Maybe they got that speed "for free" when they made the decision to switch to kraken in 2017. Altough an ASIC and CPU code are very different, maybe it's not that dramatic in an ASIC.
ooKraken8:
3.99:1
1.3 enc MB/s
1919.3 dec MB/s
zlib9:
3.36:1
19.1 enc MB/s
438.9 dec MB/s
Very fast Loading or have very fast portal... It means you can load something in less than a second.
One very practical example right in front of my eyes.
I’m playing NFS Heat at the moment as I got it discounted and it’s a real blast (recommended!) and...
It does have relatively “Fast loading” for PS4 standards, but the way Frostbite deals with that is by loading the different LOD stages literally in front of you. From a complete mess of a low poly model with no textures, morphing right in front of you into the higher LOD model you need, with textures going from 3 squares to higher res. It takes about a second or more and it’s quite hilarious to see. This will disappear.
I’m playing NFS Heat at the moment as I got it discounted and it’s a real blast (recommended!) and...
It does have relatively “Fast loading” for PS4 standards, but the way Frostbite deals with that is by loading the different LOD stages literally in front of you. From a complete mess of a low poly model with no textures, morphing right in front of you into the higher LOD model you need, with textures going from 3 squares to higher res. It takes about a second or more and it’s quite hilarious to see. This will disappear.
Inuhanyou
Veteran
Im asking seriously here guys...why isnt anyone talking about this? Like smaller game sizes even faster ssd speed than officially advertised...like this seems to be a real next gen feature and its just a compression algorithm.
Well yeah. Thats what those of us who dont care for the iteration model dislike. Might as well call this a ps6 and fully take advantage of it.
Well yeah. Thats what those of us who dont care for the iteration model dislike. Might as well call this a ps6 and fully take advantage of it.
Last edited:
Frenetic Pony
Veteran
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 point is it reduces the need for SSD bandwidth, enabling more effective ssd bandwidth.with a multiplier.
Similar threads
- Locked
- Deleted member 7537
- Games Industry
- Sticky