SUBSTANCE ENGINE

[3dilettante]

I do not interpret that post as meaning the same as the statement that the cache is 50% faster.

That post is discussing the higher coherent bandwidth of the link to memory, which would not impact a workload that doesn't require that much bandwidth or is significantly limited by other things like compute.

 

[Shifty Geezer]

Agreed. The only thing we have from Sony was the KZ slide from way back in February AFAIK.

1.6 GHz from VGLeaks that got everything else right, and 1.6 GHz for XB1, the same fundamental architecture, until it got the public upclock.

 

[DrJay24]

Why limit then the PS4's CPU to 20GB/s of bandwidth when you have 7 cores at developers disposal instead of 6 CPU cores like on the Xbox One, but those cores can use up to 30GB/s on the X1?

Because CPUs don't traditionally eat up system memory bandwidth? High end CPUs, far faster then the 8 Jaguar cores have been using ~20GB/s for years. They put a lot of money into the memory for these consoles for the GPUs, where bandwidth matters the most.

 

[Shortbread]

*hypothetically* *this is not fact* *just some funky thoughts*

PS4: 2MB/s per core (14MB/s 7 cores) x 1.6GHz per core = 3,200MB/s per core (22,400MB/s total)

XB1: 2MB/s per core (12MB/s 6 cores) x 1.75GHz per core = 3,500MB/s per core (21,000MB/s total)

So obviously from my funky math, XB1 CPU is able to handle more data during cycles because of the higher clock rates; however the PS4 CPU is capable of processing more throughput data, if, the seven cores are available for developers.

Anyhow, scaling across the cores seems to be a better fit for these systems, IMHO… rather than risking unwanted effects of increasing CPU clock speeds in small enclosed box environments.

 

[onQ]

Maybe the PS4 is able to use all 8 cores for games & use the secondary chip for the OS.



so 8 x 1.6 vs 6 x 1.75.

 

[FATBOT]

Maybe the PS4 is able to use all 8 cores for games & use the secondary chip for the OS.

so 8 x 1.6 vs 6 x 1.75.

Nope. The secondary CPU wouldn´t be fast enough for that.

 

[DrJay24]

Maybe the PS4 is able to use all 8 cores for games & use the secondary chip for the OS.



so 8 x 1.6 vs 6 x 1.75.

It would likely be 14.5 vs 12, they do use 0.5 increments in the graph.

 

[Shortbread]

Nope. The secondary CPU wouldn´t be fast enough for that.

I wouldn’t be surprised if the PS4 OS is exclusively running off the ARM processor, only reserving certain time/resources from the Jaguar CPU during heavy OS/APP/Game interactions. PS4 OS seems fairly light (like vita / cell phone OS light) compared to the XB1’s cluster of OSs.

 

[KKRT]

I wouldn’t be surprised if the PS4 OS is exclusively running off the ARM processor, only reserving certain time/resources from the Jaguar CPU during heavy OS/APP/Game interactions. PS4 OS seems fairly light (like vita / cell phone OS light) compared to the XB1’s cluster of OSs.

So, what would be reason for KZ:SF to use only 6 cores, instead of 8?

 

[onQ]

So, what would be reason for KZ:SF to use only 6 cores, instead of 8?

Are you basing that off of the old slides?

 

[KKRT]

Are you basing that off of the old slides?

No, i'm basing that on Digital Foundry's article from few weeks ago:
http://www.youtube.com/watch?feature=player_detailpage&v=DMO4X9leTG8#t=7
http://www.eurogamer.net/articles/digitalfoundry-the-making-of-killzone-shadow-fall

 

[Shortbread]

So, what would be reason for KZ:SF to use only 6 cores, instead of 8?

As far as we know, the six available cores "then" could have changed from that pdf back in May. Lots of improvements have been made to KZ:SF, so possibilities of using more cores could have happened. Looking at what the Substance Engine graph is showing (implying)… is that PS4 developers might have access to the other cores. Anyhow, we don’t have enough of “anything” to make a solid conclusion on how the PS4 Jaguar cores are allocated.

 

[KKRT]

As far as we know, the six available cores "then" could have changed from that pdf back in May.

What May? We have data from November. KZ:SF is using still 6 cores, like in February demo.

 

[onQ]

Maybe this has more to do with HSA & not needing to go through the OS/driver.

 

[Aeoniss]

Seems like the PS4 CPU holds up reasonably well against an i7 here.

There are alot of unknowns- frequency for both being one of them. I'm assuming (and it makes since given the info) that this truly is measured per 'core'.. If that is the case, then in this specific operation the PS4 CPU with 6-7 available cores (taking into account OS reserves) would be close to 75% as powerful as this i7 correct?

I can only guess that this measurement takes into account most of the overheads involved with running the CPU's.

Doesn't this make the PS4 fairly competitive for a dedicated gaming unit?

 

[Shortbread]

What May? We have data from November. KZ:SF is using still 6 cores, like in February demo.

I’m not debating it doesn’t – I’m just debating about the “possibility” of them doing that. That being said, KZ:SF engine was built off the previous PS3 builds - not a new engine from the ground up. So the game code used in KZ:SF may have just required 6 Jaguar cores, as the Cell provided six SPE’s for the previous builds. It would make sense too - that GG felt comfortable enough to get something out for the PS4 launch without having to rewrite all the gaming code for the additional Jag-cores. The only thing they needed to do was update art/texture assets, better lighting, additional geometry, etc… that the GPU could handily do by itself.

 

[Cyan]

I do not interpret that post as meaning the same as the statement that the cache is 50% faster.

That post is discussing the higher coherent bandwidth of the link to memory, which would not impact a workload that doesn't require that much bandwidth or is significantly limited by other things like compute.

That explains it in a better manner. I wonder why the CPU of the Xbox One doesn't perform as PS4's CPU when they are basically the same and it has received an upclock.

It would be good if the CPU at least gets on par, which means that after the initial years of cross-gen releases no multiplatform game will be gimped.

I don't doubt that in the future the console will perform quite well, technically wise, after playing some exclusive games myself -Forza 5, Ryse, KI, Max the Course of Brotherhood....-. However I wonder if virtualisation might eat some CPU cycles.

 

[DuckThor Evil]

Like, totally.

Totally clearly, like obviouslay

(My post was like clearly sarcastic. Like ya. )

I know you didn't say it by saying it . It was not an impossible suggestion imo. Jaguar could be clocked to 2.0 Ghz. I personally don't believe that to be the case, but I guess it was worth to be mentioned even if sarcastic.

 

[joker454]

They mean texture effects. The substance demos are very impressive. The shaders support parameters like 'age' and you can easily adjust age of a material and have it computed in realtime, adding decay.

Very cool, although seems like this tech will be more useful on mobile and tablets compared to consoles. From the benchmarks it looks like it wont take too long for mobile cpu's to catch up to the new consoles, and with Unity support this tech may really take off in the mobile world.

 

[pjbliverpool]

Seems like the PS4 CPU holds up reasonably well against an i7 here.

There are alot of unknowns- frequency for both being one of them. I'm assuming (and it makes since given the info) that this truly is measured per 'core'.. If that is the case, then in this specific operation the PS4 CPU with 6-7 available cores (taking into account OS reserves) would be close to 75% as powerful as this i7 correct?

I can only guess that this measurement takes into account most of the overheads involved with running the CPU's.

Doesn't this make the PS4 fairly competitive for a dedicated gaming unit?

A single Haswell core at 3.2ghz is 3-4 times faster than a single Jaguar core at 1.6ghz full stop. These numbers just dont make make any sense for a single core unless the implementation is absurdly inefficient on the i7.

I also don't see why they would measure performance from only a single core when testing a multithreaded application on an array of multicore cpu's of differing core counts. It makes no sense. What purpose would the table have if its not showing a real world usage scenario?