Digital Foundry Article Technical Discussion [2020]

Yes and still 7GB/s throughput on the fastest PCIe3 drives so nothing to sneeze at.

 

I know is too early, but the upgraded Witcher 3 would be a good candidate to beta test DirectStorage on PC.

 

Nvidia probably has Vulkan extensions. Just a guess on my part. So it's DirectStorage compatible, but maybe has some of its own features and works in Vulkan as well.

 

Isn't that just the maximum speed of the decompression block? Like, it can theoretically write 22GB/s but it can't actually decompress that much per second, so it's ability to write out decompressed data will never be the bottleneck. Or maybe I misunderstood the specs. All of the Sony released specs I've seen simply state the drive speed as 5.5 GB/s raw and 9 GB/s compressed.

 

It depends on data. Sony said the typical decompressed data rate would be somewhere around 8GB/s. PS5 is capable of that 22GB/s but it requires data that compresses exceptionally well.

 

I get that it's theoretically possible that the decompression block can output 22GB/s, but all of the official verbiage from Sony has been something along the line of "if the data compressed particularly well", but does it? Isn't the ratio of Kraken+RDO something like 3:1 vs uncompressed under ideal conditions? But that combo somehow achieves 33% better results on PS5 somehow? I just don't know how that claim holds water.

I rewatched the Road to PS5 just to get Cerny's quote, and he states that the "custom I/O unit is cable of handling over 5GB of Kracken format input data a second. After decompression that typically becomes 8 or 9 GB but the unit itself is capable of outputting as much as 22 GB a second if the data happened to compress particularly well." That leaves some room for interpretation. It's sort of like me saying "I can punch a hole through concrete, if my fists happen to be made of steel". They aren't. So I can't.

I still think 22 GB/s is the maximum throughput of the I/O block and that will never bottleneck the system, because the compression formats it supports can't compress data that small to achieve that number. To be clear, I don't think this is a bad thing. This is how you design hardware so that it isn't tripping over itself.

 

To get compression that high, the data must be extremely repetitive or patterned in a very predictable way. Like compressing a table full of largely the same values for instance. That's just unlikely to be the case with most data.

 

If there is something like mp3's inside game data that would likely get 0 compression and output would be 5.5GB/s. Because the compression is so heavily data dependent sony gave the 8GB/s number which is based on their existing game assets. 22GB/s is real and can happen and ps5 can do that but that kind of data is not your typical game asset.

Any compression number is pretty much uncomparable between each other as long as different parties don't agree on data sets to be used for measurement.

 

Right. I get that. Back in the day we had fun passing around a zip file with a .txt in it labeled as CD keys but it was just the letter "A" repeated enough where this tiny zip file decompressed to 400MB. So maybe i guess in an ideal world we would get something like that. But maybe not, because to achieve 22 GB/s you would either have to saturate the input with enough compressed data It exceeds the ideal compression ratio of Kraken+RDO, which is 3:1. Ideal, not average. To achieve 22 GB/s you would have to decompress 5.5GB/s at 4:1, 33% more data than the idea situation. Or you could decompress at a higher ratio, if the hardware supports that. But we don't know if it does.

Not only that, you would need that type of data to occupy the totality of the 22 GB of output, or maybe the 5.5 GB of input. That would depend on where the bottleneck resides. We only have Cerny's words on this, the 22 GB is a theoretical output. Based on what? a 4:1 maximum compression ratio that the hardware supports? Or maybe that supports 10:1 but it can only output 22 GB/s because that's the speed it can write to RAM. Or maybe it only supports 3:1 but it can output 22 GB/s, and Cerny was just telling us how fast it could move data to RAM. But at that rate 22 isn't achievable based on the input speed.

I don't take Cerny's statement about the I/O block being able to output 22 GB/s as a statement about ability, more a statement about how they designed the I/O block to not be the bottleneck. It can handle the data that the SSD is going to provide, and not get in the way.

 

Tommy McClain

 

lol Alex, made me spit my coffee out this morning.

 

Great job Alex. Here's the article...

https://www.eurogamer.net/articles/...stered-your-first-look-at-console-ray-tracing

Tommy McClain

 

The profile top right corner reads DX11 64bit build?

Very good info on this video on ray tracing in general. Great deep dive here.
Great to see the different performance deltas between configurations.

Seeing hybrid ray tracing happening is incredible. Great discussion overall.

Shadows and self shadow is so apparently on the remaster vs the original

 

DR FP16 of the PS4 Pro could help here since they would not need FP32 precision at that resolution for RT.

 

So, we can now see a preview on xbox one X what to expect from RT on xbox series S ^^

 

My expectations are higher for it given the RT hardware and increased cpu power. But yes this is a good preview

 

Next gen graphics on PC!

 

Beasts, they've known since March. Interesting compare here.

watching this, felt like I was experiencing an episode of Dark.

 

it's been confirmed that the new CoD is going to run at 120fps on next gen consoles, including the Series S, btw

 

perhaps the best Xbox console ever, at least the most console-ish one from MS, it took them almost 20 years but they finally got it