PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)

[jonabbey]

Ow, my brain. Hope a good translation of that comes along.

 

[yewyew]

I came here to relax but that translated article gave me even more headache !



Where does it say that ?

Did I read the article correctly ?

* CPU can send data directly to the shader cores (bypassing the caches), and the GPU result "sent back". They can use up to 10 GB/s coherent bandwidth without touching the 176GB/s incoherent access ? So compute jobs can run without messing up the graphics cache ?

* Cache entries can be tagged to write out to memory. What does it mean ? Don't cached entries get written out to memory upon flushing anyway ?

EDIT: The CUs would be like the SPUs in this case.

There's a leak on compute rings on the GPU that interconnect with the CPU.

http://www.vgleaks.com/orbis-gpu-compute-queues-and-pipelines/

 

[patsu]

Hmm... you mean the CU writing back into the ring buffer ?

 

[Cjail]

Further in the interview, it appears that they're going for a mixed approach in regards to installs. Partial downloads/installs for more immediate playback, but they also mention being able to play off the disc directly. I had assumed both the Next Box and PS4 would require an install (masked by the partial install process) to keep the experience consistent between retail and digital.

PS3 already has partial install, just not for every title.
Hopefully partial install on PS4 will really bring substantial benefits.

 

[Arwin]

The last active state of the memory will have to be maintained given what they've announced at the unveil. Caching it to hard drive wouldn't provide the solution they were aiming for. So, at least the memory will still need power for that alone I suspect. Also, are you sure the custom chip has its own memory? I didn't really get that impression from the interview.

I am actually guessing that standby will still be writing memory state to harddrive while in standby mode. So the CPU and GPU are switched off, and the ARM or whatever chip takes over and writes all RAM to HDD ...

Being able to read CDs, by the way, suggests they're leaving the option open for you to put Playstation 1 game discs into the PS4?

Further in the interview, it appears that they're going for a mixed approach in regards to installs. Partial downloads/installs for more immediate playback, but they also mention being able to play off the disc directly. I had assumed both the Next Box and PS4 would require an install (masked by the partial install process) to keep the experience consistent between retail and digital.

I'm still guessing this will be mostly the same across titles, as BD read speed just isn't fast enough, but you never know. Some games may just not need it, and then it's a waste of HDD space.

Interesting stuff about the CPU to GPU connection. We had a graph earlier where we saw a pipeline overview from CPU Core down.

 

[arijoytunir]

http://www.gamasutra.com/view/news/...campaign=Feed:+GamasutraNews+(Gamasutra+News)

gamasutra will run a technical article on ps4 with mark cerny soon .

 

[Love_In_Rio]

ugh...so, where are the standard GPU enhancements?. I don´t see anyone anywhere ( 8 ACEs is in the Sea Islands ISA doc as a posibbility, so AMD standard stuff ).

 

[Scott_Arm]

I was talking about the CPU & GPU working together as one creating a Cell like Processor

You're right. It's pretty much just like Cell. Let's all start calling it Cell 2.0.

 

[Gradthrawn]

PS3 already has partial install, just not for every title.
Hopefully partial install on PS4 will really bring substantial benefits.

Sorry, I should have been more clear. By partial installs, I was more referring to starting the game after only part of the game has been installed to HDD, while the rest gets installed in the background (full install with partial install start). Perhaps similar to what the UC series does seamlessly, but taking it a step further and continuing to install the entire game to HDD and not a mix of HDD/ODD streaming like UC is now. I was assuming all games would have to be installed, and the partial install to start the game would be used to ease the pain/user wait time (similar to how partial downloads of games would be used as mentioned at the conference).

I am actually guessing that standby will still be writing memory state to harddrive while in standby mode. So the CPU and GPU are switched off, and the ARM or whatever chip takes over and writes all RAM to HDD ...

Being able to read CDs, by the way, suggests they're leaving the option open for you to put Playstation 1 game discs into the PS4?

That would be the safer route in terms of keeping the state through power failures and would allow for immediate standby mode (going into standby, at least). The problem with writing the memory state to HDD is that the entire memory state has to then be read from HDD back to memory when resuming. The difference between putting Windows in Standby/Sleep vs Hibernate, for example. If I recall correctly, in a post reveal interview, a Sony rep mentioned that a users's game state would be lost if power is lost while in standby (I'll have to verify the source of that, though). For those 2 reasons (speed of resume, and that state would be lost with power), I'm guessing they won't write the memory state to HDD for standby mode.

 

[Shifty Geezer]

I was talking about the CPU & GPU working together as one creating a Cell like Processor

Cell processor wasn't CPU+GPU as one. PS3 was CPU<>GPU using Cell to do graphics tasks. Cell was a discrete CPU with fast vector units and local storage with necessary manual memory management. There was a direct Cell <> RSX link, but I don't recal the BW, and AFAIK (but I could be very wrong) most interaction between the two components was by sharing data in RAM. Cell reads data, writes to RAM, RSX reads from RAM and processes and writes output. Potentially Cell reads that output from RAM and writes changes. I don't think anyone has ever gone into detail on what data is sent direct from Cell to RSX though.

 

[patsu]

http://www.gamasutra.com/view/news/...campaign=Feed:+GamasutraNews+(Gamasutra+News)

gamasutra will run a technical article on ps4 with mark cerny soon .

The fact that he spent so much of his personal time working on the question of just what hardware should go into the box made Cerny realize something important: "I probably have more passion about the next generation than anybody inside the Sony Computer Entertainment world."

Hopefully they answer the questions in that AV Watch Impress article.

 

[patsu]

Sorry, I should been more clear. By partial installs, I was more referring to starting the game after only part of the game has been installed to HDD, while the rest gets installed in the background (full install with partial install start). Perhaps similar to what the UC series does seamlessly, but taking it a step further and continuing to install the entire game to HDD and not a mix of HDD/ODD streaming like UC is now. I was assuming all games would have to be installed, and the partial install to start the game would be used to ease the pain/user wait time (similar to how partial downloads of games would be used as mentioned at the conference).

That would be Progressive Download/Install.

That would be the safer route in terms of keeping the state through power failures and would allow for immediate standby mode (going into standby, at least). The problem with writing the memory state to HDD is that the entire memory state has to then be read from HDD back to memory when resuming. The difference between putting Windows in Standby/Sleep vs Hibernate, for example. If I recall correctly, in a post reveal interview, a Sony rep mentioned that a users's game state would be lost if power is lost while in standby (I'll have to verify the source of that, though). For those 2 reasons (speed of resume, and that state would be lost with power), I'm guessing they won't write the memory state to HDD for standby mode.

In this area, Jeff Rigby has an interesting post on GAF regarding TrustZone, which the AV WachImpress article acknowledged (according to GAF):
http://m.neogaf.com/showpost.php?p=51675942

 

[patsu]

ugh...so, where are the standard GPU enhancements?. I don´t see anyone anywhere ( 8 ACEs is in the Sea Islands ISA doc as a posibbility, so AMD standard stuff ).

Do you mean standard enhancements or PS4 specific enhancements ?

Cell processor wasn't CPU+GPU as one. PS3 was CPU<>GPU using Cell to do graphics tasks. Cell was a discrete CPU with fast vector units and local storage with necessary manual memory management. There was a direct Cell <> RSX link, but I don't recal the BW, and AFAIK (but I could be very wrong) most interaction between the two components was by sharing data in RAM. Cell reads data, writes to RAM, RSX reads from RAM and processes and writes output. Potentially Cell reads that output from RAM and writes changes. I don't think anyone has ever gone into detail on what data is sent direct from Cell to RSX though.

The Uncharted and KZ2 presentations had some insights about their buffer layout so that the RSX and Cell could work together.

The Cell just send commands via the RSX command queue, perhaps like the queues in Orbis.
Someone sent me a link to the leaked PS3 dev documents on the net.


In Cell, the PPU also loads program and data pointers into the SPUs, and the latter DMA data in independent of RSX.

In Orbis, the CPU can send program and data (pointers) to the GPU via the queues and buffers. It seems that the CUs can process compute tasks without messing up the graphics cache as if they are independent compute units ? But of course, the compute units have to split their time, perhaps optimized to prevent stalls.

I also wonder if the Onion, Onion+ and Garlic bandwidth are completely independent. In PS3, the useful numbers are here: http://www.ps3devwiki.com/wiki/RSX
It looks like Cell could send data to RSX (IOIFO) @ 20GB/s and vice versa (IOIFI) @ 15GB/s.
Memory bandwidth ranges from 16MB/s to 25GB/s depending on the pools and components involved.

 

[Gradthrawn]

That would be Progressive Download/Install.



In this area, Jeff Rigby has an interesting post on GAF regarding TrustZone, which the AV WachImpress article acknowledged (according to GAF):
http://m.neogaf.com/showpost.php?p=51675942

Thank you! I was searching for a better name/terminology but coming up lost. Progressive download/install makes much more sense.

The jeff_rigby post you linked to about the snapshot boot is quite interesting. Further clarification from someone about the implementation and potential feasibility of it (in the upcoming consoles) would be quite useful. That could potentially invalidate my statements about the memory state being written to HDD, I think.

 

[Cjail]

@Gradthrawn
Oh sorry I missed your point.
Anyway I see progressive download working well IF internet connection is not the limiting factor.

 

[Love_In_Rio]

Do you mean standard enhancements or PS4 specific enhancements ?

I mean PS4 specific/custom enhancements that are not standard AMD. As far as we know this is a glorified Kabini and maybe more ACEs, but increment in the number of ACEs is not something to take the world apart as is also a Sea Islands feasible thing. With RSX at least they increased the texture caches size.

And having today a Battlefield 4 demo running in a 7990 doesn´t make me think a standard 7850 will run it with the same bells and whistles. I want some custom stuff to "believe" it is better than a vanilla 7850/60.

 

[onQ]

You're right. It's pretty much just like Cell. Let's all start calling it Cell 2.0.

I know you're being sarcastic but it's more like the Cell then you think.

Cell processor wasn't CPU+GPU as one. PS3 was CPU<>GPU using Cell to do graphics tasks. Cell was a discrete CPU with fast vector units and local storage with necessary manual memory management. There was a direct Cell <> RSX link, but I don't recal the BW, and AFAIK (but I could be very wrong) most interaction between the two components was by sharing data in RAM. Cell reads data, writes to RAM, RSX reads from RAM and processes and writes output. Potentially Cell reads that output from RAM and writes changes. I don't think anyone has ever gone into detail on what data is sent direct from Cell to RSX though.

I'm not saying that the Cell was a CPU + GPU I'm saying that the PS4 SoC with the CPU & GPGPU together is creating a Cell like Processor.

Jaguar Cores as the PPU & the GPGPU CU's as the SPE's.


People are looking right at it but still can't see what's happening.

 

[Love_In_Rio]

I know you're being sarcastic but it's more like the Cell then you think.





I'm not saying that the Cell was a CPU + GPU I'm saying that the PS4 SoC with the CPU & GPGPU together is creating a Cell like Processor.

Jaguar Cores as the PPU & the GPGPU CU's as the SPE's.


People are looking right at it but still can't see what's happening.

You are right. In a sense i think PS4´s, and even more Durango´s, APU is like a super vector capable CPU. But the problem is that there is no also a discrete GPU...and so these socs must do the rendering instead of only the compute. And when you fight against Sandy Bridges ( and soon super vector capable Haswells ) paired with 7990s...your APU will struggle to cope with the quality of the Battlefield 4 demo we have seen today.

I am really a little desappointed they demoed it in a 7990 and not in PS4. Its 6 months at least for new consoles to launch ( when suppossely there will be also a 8990 in the market ) and i already have thoughts of buying a new PC instead of a new console...

 

[patsu]

I mean PS4 specific/custom enhancements that are not standard AMD. As far as we know this is a glorified Kabini and maybe more ACEs, but increment in the number of ACEs is not something to take the world apart as is also a Sea Islands feasible thing. With RSX at least they increased the texture caches size.

And having today a Battlefield 4 demo running in a 7990 doesn´t make me think a standard 7850 will run it with the same bells and whistles. I want some custom stuff to "believe" it is better than a vanilla 7850/60.

The only thing we "know" now are:

* high priority VSHELL queue

* Onion+ compute link and perhaps additional cache/buffer management to prevent unwanted cache effects.

 

[Love_In_Rio]

The only thing we "know" now are:

* high priority VSHELL queue

* Onion+ compute link and perhaps additional cache/buffer management to prevent unwanted cache effects.

If onion was in Fusion, onion + will be in Temash/Kabini, or little steps will have been given in the HSA progress...As 3dilettante said onion + must be something derivative from going from VLIW to GCN.

The Vshell queue for OS messages is not something very attractive from a game rendering point of view.