• 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.
T

Three Jackdaws

Unconfirmed Member
So some speculation from an old tweet



Note what Matt is teasing here, can intersetion engines calculate bounce across CUs ? You can only do that if CUs are sharing more information.

Is this an L1 infnity cache tease or more to do with ps5 custom cache handling ?

There is more to be learned, and some people already know the answer and we dont, so frustrating.

allot of Post processing techniques performance is about speed, caches and sharing data across Cus. @Elog this is for you.

I think the simplest explanation is the best. Ray bounces scale better with higher clocks rather than CU's and the number of rays scale better with more CU's.
 

avenovah

Member
These guys are especially praised by critics.

49996616377_7ac824cf67_o_d.jpg
 

geordiemp

Member
That is an interesting tweet, because rays have to be calculated within the frame cycle. There's just not enough time to go off-chip for that data. However, could he be referring to L0 or L1 cache (referring to the XSX GPU), where that data could be shared across a DCU or shader engine, or would he be referencing a massive L2 or even L3 cache? L3 seems like it would introduce latency issues on a highly iterative process, so I have to assume L2 is as the speculation can go. That would be infinity cache level.

In reading multiple sauces the past month, it sounds like RT actually generates a lot of data, so I'm thinking fatter L1 or L2 cache. And if you have the die space to buff the L1 cache size, why not do it to the L2 instead, so that you can share that data across multiple shader engines? My only question then, is what are the relative latencies as you move to higher levels of cache. There's a point of diminishing returns, where sharing data between CUs runs into a latency bottleneck. Ultimately, if PS5 uses DCUs, then a shared L0 between CUs lets you iterate on the same data with 2 ray accelerators, without having to make multiple trips up the tier ladder. Then let the "coherency engine" manage that data further up the chain. Theoretically, fewer stalled CUs. Or maybe I'm thinking about this incorrectly.

The RDNA1 bandwidths are below, so ps5 will be in the 4TF/s league for L1. Different ball game, even if it takes a few cycles to ctross share, its >>>>>> memory.

Sharing all CU and L1 caches takes 4 mm2 on ps5 according to AMD silicon costs.


jOONlBg.png
 

FunkMiller

Member


The more we see revealed as we get closer to the launch of these two consoles, the more it feels like XBox Series X is a more powerful continuation of the current generation, while Ps5 is a clear change to a new one.

Not suggesting one strategy will be more successful than the other, but that’s becoming the difference between the two.

And I certainly know which approach fill me with more excitement.
 
Last edited:
The more we see revealed as we get closer to the launch of these two consoles, the more it feels like XBox Series X is a more powerful continuation of the current generation, while Ps5 is a clear change to a new one.

Not suggesting one strategy will be more successful than the other, but that’s becoming the difference between the two.

And I certainly know which approach fill me with more excitement.

Both are next gen consoles, both have next gen thought process going into their respective hardware and controller. Controller on the PS5 feels like a whole new generation compared to the Series X though, with the haptic feedback, adaptive triggers and mic. DualSense will be a game changer in immersion!
 

geordiemp

Member
I think the simplest explanation is the best. Ray bounces scale better with higher clocks rather than CU's and the number of rays scale better with more CU's.

LOL no, I saw that banded about a while ago trying to simplify ray tracing very poorly. More cU gives more capacity, but also more issues in occupancy, and even more for BVH.

More BVH puts even more strain on CU needing none private shader array data. PC parts will go the extra mile and add L2 as well into the mix, but I bet the extra MB is a combined one, but that is my guess, lets wait.......

One day left.
 
Last edited:

MastaKiiLA

Member
PS5 = F1 Car
XSX = nascar
I wouldn't go that far. As an F1 fan, and former NASCAR fan, F1 cars obliterate all comers. They don't always hit the same top speed, but over a single lap, an F1 car is seconds down the road. And I mean against an Indycar or LMP1. Against a NASCAR car, it would be a bloodbath. I'd compare it to an F1 car with a V8/V10, versus one of the current-era hybrid cars with a turbocharged V6 hybrid engine. Smaller engine, but it more than makes up the difference with boost and electrics.

EDIT: As an analogy. Not that I can say which direction the performance advantage goes with the 2 systems. But it's a more fair comparison for gearheads like me.
 
Last edited:

3liteDragon

Member
The more we see revealed as we get closer to the launch of these two consoles, the more it feels like XBox Series X is a more powerful continuation of the current generation, while Ps5 is a clear change to a new one.

Not suggesting one strategy will be more successful than the other, but that’s becoming the difference between the two.

And I certainly know which approach fill me with more excitement.
I have said it a few times and the first time I got some nasty PMs here for it.

The XSX is an amped up Xbox One and thats still great.

PS5 is a next gen console.
 

FunkMiller

Member
I have said it a few times and the first time I got some nasty PMs here for it.

The XSX is an amped up Xbox One and thats still great.

PS5 is a next gen console.

The more we see revealed as we get closer to the launch of these two consoles, the more it feels like XBox Series X is a more powerful continuation of the current generation, while Ps5 is a clear change to a new one.

Not suggesting one strategy will be more successful than the other, but that’s becoming the difference between the two.

And I certainly know which approach fill me with more excitement.

Great minds think alike.

...or rather, one great mind, and one fucking idiot who really should get back to work, and stop repeatedly watching people opening a bloody box.
 
Last edited:

jose4gg

Member
The question for the long term will be how well supported is the additional controller and the ui features like the 'game guide videos' that dev's will have to do just for the PS platform.

Fro the controller I think they will very well be supported because it's an improvement over an existing tech like a rumble, the triggers setup should be a very basic API, when the gun is like this, have this setup. The UI features are the thing I'm a little bit concerned, that really requires another employee doing all that configuration, text, etc.
 

Shmunter

Member
Always thought voice coil actuators could be a game changer, given the levels of granularity that can be programmed into their function with relative ease, and a much, much higher degree of accuracy than rumble. I’m glad to see that they are pushing the technology.
One of the reviewers mentioned it has the crisp precision of high end phones. And I’m the guy that uses the iPhone timer setting scroll haptic as a party trick.

And that’s just the haptics. Then the force feedback triggers push the package to a whole new stratosphere.
 
Last edited:
Status
Not open for further replies.
Top Bottom