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

[Bagel seed]

Some noteworthy parts

Cerny: For us, GPGPU is a very important feature. We believe, and want to be able to use a more practical purpose GPU. Therefore, in many items, I made ​​a technology that is already customized for very current.
Is one example ....... If the CPU and GPU to exchange information, CPU information is input, in a typical PC architecture, the GPU should not clear the cache once, you will have to read. even if the GPU returns the result of the operation to clear the cache for the GPU, it does not pass the result to the CPU then you should not be. We create a bypass the cache. Here pass through, the result of the operation of the GPU. By using this mechanism, the shader core GPU, I can pass the data received directly from the main memory. In short, I'm able to bypass the L1 · L2 cache GPU. Of course, you can also write as well as read data. Very quickly, this bandwidth for seconds is 10GB /.

We have a bit of L2 cache also tagged. Is called "VOLATILE" tag we. In the cache, we have to be able to control by marking the others do not have this attribute. If you use this tag is written directly to memory. As a result, it can be used to efficiently handle all the graphics cache.

Dat Custom Chip!

Cerny: custom chip, the second corresponds to the so-called south bridge. However, a CPU built-in. Where power is always supplied, even when the power is turned off PS4, is monitoring all the IO system. Built-in CPU and Southbridge governs all processes and downloads, and access to the hard disk. Of course, even when the power is off.

Ito: The role of the second custom chip, there is also a consideration of environmental issues. When you download in the background, and has launched a main CPU each time, power consumption can become so large, it is done only in the second custom chip. It is also because there is one purpose of strict regulations with respect to power consumption, in Europe, and, in this way we may be able to support in particular.

Some aspects of the measures of network bandwidth: Cerny. Has the effect of realizing large downloads smoothly even when you download the background, there is not enough bandwidth.
However, the large of all, it is that can significantly reduce the time until you can play the game. At the same time, we also attempt to reduce the size of the initial download. While you download when you play only the first few GB, continues to play, I will download the rest. Of course, even when the main power is turned off, followed by the rest of the download.
First anyway, is to significantly reduce the download time before playing.

 

[MrFox]

So the custom chip is basically a complete low-power SoC which acts as the whole south bridge, I guess that solves the issue of standby power, both the GDDR5 and liverpool can be completely powered off. I'm curious about what kind of local memory they used for it.

 

[Gradthrawn]

So the custom chip is basically a complete low-power SoC which acts as the whole south bridge, I guess that solves the issue of standby power, both the GDDR5 and liverpool can be completely powered off. I'm curious about what kind of local memory they used for it.

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.

Lots of stuff in there! I wish I could read japanese.

The drive doesn't have a CD pickup at all

The article appears to have been updated/corrected. The correction notes that it does not support the playback of audio CDs, specifically, and not just simply CDs. A strange omission, regardless. How much can the necessary audio CD playback licenses possibly cost?

[Correction]
From SCE, correction function for reading CD of PS4. Had been described in the first occurrence article, "does not have the ability to read the system CD as an optical drive, it only supports DVD · Blu-ray" and, as the "drive read performance of the BD / DVD / CD It has changed to "do not support playing music CD as a function of the PS4. With it, I have to modify the representation.

The OS is BSD (no real surprise, but finally confirmed)

Why weren't you surprised by the OS being based on BSD? It appears to be the first time they've done that (if I'm reading it correctly).


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.

Cerny said mechanism is intended for simple split download, time would not take to develop. It works well as a positive type game downloads, the title provided by the Blu-ray disc and.

Cerny: divided into several chunks of data logically, I'll just write a script special. Moreover, customization is possible in various ways how to write a script. For example, you can download before the single player, and I can download the multiplayer before ....... Although it is a story related to it .... Memory system was 16 times greater than in the age PS3, Blu-ray drive transfer speed is not faster only a few times. Therefore, you can use the method to a combination of scripts and mass similar to the approach of the earlier, logical data, and transfers to the hard drive only the main part of the game from the Blu-ray drive, start the game. That way, you can play from your hard drive fast and comfortable short load time. Of course, you can also play while reading the data directly from the Blu-ray giant.

Even with the translation difficulties, its quite an interesting interview.

 

[itsmydamnation]

i wonder if the APU still has this limitation

http://amddevcentral.com/afds/assets/presentations/1004_final.pdf slide 35

where the CPU read from the GPU memory is 1Gbps.

 

[3dilettante]

I'm not certain if that constraint can be optimized, or that it's an APU restriction.
Accesses to uncached memory by the CPU force buffers to be written out, and it breaks the pipelining done by the rest of the memory subsystem.

Other papers comparing Llano to a discrete CPU and GPU show the same very bad bandwidth, claiming that uncached reads must be done one at a time.

 

[MrFox]

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 don't know I was assuming wildly, here's my thinking: It seems to imply the southbridge is a separate chip specifically to allow the SoC to be powered down. I agree, RAM needs to be kept at least refreshed, but I think GDDR5 is self-refreshing in the powered-down state taking an extremely low amount of power. To allow that state, the SoC at the opposite side must be powered down, including the memory controller. So they have this small CPU inside the south bridge, which have access to all the IO, including the wifi, ethernet, USB, HDD, ODD, all locally... and the only thing missing is a small memory pool for code and data. Just a few MB embedded would be enough, or a cheap PiP like the Vita main ram. Otherwise they'd have to power on the whole path through the SoC, mem controller and memory, just for that (this CPU's program space would also need to be from the GDDR5). I'm also thinking this custom chip could be on a low-power process like a cellphone chip (another reason to need it as a separate chip).

Why weren't you surprised by the OS being based on BSD? It appears to be the first time they've done that (if I'm reading it correctly).

Because they said quite some time ago that the OS would be a Linux-type instead of a simple low-level OS like the PS3.
BSD is the only one that is similar to linux, and the BSD license is the best one for commercial use, allowing them to keep the source secret. Linux is a very bad choice because it's GPL (having to release all the source code at every firmware update would be a security nightmare), and all the other alternatives are too primitive (uber light kernels, focused on RT) so nothing else like linux.

 

[yewyew]

I'm also wondering the value of a GPU seeing a CPU's L1 cache. That's very low level, like the data the CPU thread is currently working on before being written out to RAM, or just written out. Synchronising jobs so the GPU is wanting data just when the CPU is working on it can't be easy (I don't understand APU differences at all, BTW ).

It's possibly for the CU's to address the CPU directly, since each thread has its own L1 and instruction cache, to give it some Cell-to-RSX support in reverse.

 

[onQ]

Do what? Cap the RAM BW? At first it sounds a bit crazy, but if we compare to PS3 which had 8 CPUs and more float power on a more limited bus, perhaps Sony's experience was enough to say 20 GB/s is all that's needed for the CPU, especially if it's not handling some of the workloads Cell was? Sony actually have more experience of 8 cores in a console than AMD at this point and this may be a sensible cost saving. Not sure what other benefits there are (less heat, smaller buses, equals cost saving to me).

I'm also wondering the value of a GPU seeing a CPU's L1 cache. That's very low level, like the data the CPU thread is currently working on before being written out to RAM, or just written out. Synchronising jobs so the GPU is wanting data just when the CPU is working on it can't be easy (I don't understand APU differences at all, BTW ).

It's possibly for the CU's to address the CPU directly, since each thread has its own L1 and instruction cache, to give it some Cell-to-RSX support in reverse.

@Shifty Geezer remember what I said about the PS4 SoC being like The Voltron of The Cell Processor or Larrabee when the CPU & GPGPU form together?

I'm looking at the Design of the PS4 SoC & it's looking like the Voltron of The Cell Processor or Larrabee when the CPU & GPGPU form together.

Notice all the 8's?

• 
  [*]8 Jaguar Cores
  
  [*]The GPGPU will have 8 Compute-only pipelines
  
  [*]Each pipeline has 8 queues of a total of 64
  
  [*]Rumor of 8 very wide vector engines attached to the CPU
  
  [*]all using 8GB GDDR5 Unified Memory ( not really part of my argument but it does have 8)
  

Seems like it's all made to work together as a powerful CPU if a dev choose to use it all for Computing.

I don't seem so crazy now do I?

 

[arijoytunir]

@onQ also there is no extra vector processor like the recent rumours suggested!

 

[onQ]

@onQ also there is no extra vector processor like the recent rumours suggested!

I'm thinking that maybe someone confused the way the CPU cores can work with the GPGPU in 8 compute pipelines for the CPU having 8 vector engines or maybe that's how the CPU/GPGPU connection will be used.

 

[Scott_Arm]

@Shifty Geezer remember what I said about the PS4 SoC being like The Voltron of The Cell Processor or Larrabee when the CPU & GPGPU form together?



I don't seem so crazy now do I?

What does any of this have to do with Cell or Larrabee or Voltron?

 

[patsu]

what is costly about an IP camera? for 1gbps switching its dirt cheap. the benefit is distance and that can be a huge one.

If it's something high speed and general, then we can hook up other stuff to the base PS4 unit. ^_^

Hope it's not tied to USB host. They implemented LightPeak over USB, so Sony definitely has the knowhow to do something interesting here.

 

[patsu]

Some noteworthy parts

[Translated PS4 stuff]

Dat Custom Chip!

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

It's possibly for the CU's to address the CPU directly, since each thread has its own L1 and instruction cache, to give it some Cell-to-RSX support in reverse.

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.

 

[onQ]

What does any of this have to do with Cell or Larrabee or Voltron?

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

Cerny: For us, GPGPU is a very important feature. We believe, and want to be able to use a more practical purpose GPU. Therefore, in many items, I made ​​a technology that is already customized for very current.
　Is one example ....... If the CPU and GPU to exchange information, CPU information is input, in a typical PC architecture, the GPU should not clear the cache once, you will have to read. even if the GPU returns the result of the operation to clear the cache for the GPU, it does not pass the result to the CPU then you should not be. We create a bypass the cache. Here pass through, the result of the operation of the GPU. By using this mechanism, the shader core GPU, I can pass the data received directly from the main memory. In short, I'm able to bypass the L1 · L2 cache GPU. Of course, you can also write as well as read data. Very quickly, this bandwidth for seconds is 10GB /.
　We have a bit of L2 cache also tagged. Is called "VOLATILE" tag we. In the cache, we have to be able to control by marking the others do not have this attribute. If you use this tag is written directly to memory. As a result, it can be used to efficiently handle all the graphics cache.
　These mechanisms are a mechanism for making harmonize the graphics processing and arithmetic processing, to work together efficiently. It is "harmony" in Japanese say. By customizing the cache and bus these large, successful on PS3 "(Note I: SPU Runtime System) SPURS" are trying to reproduce. SPURS is a mechanism to virtualize resources such as SPU, to management autonomously.
　In similar feeling any x86-64 GPU, the PS4 is hardware available resources at various levels. There are eight pipes, this idea has eight queues each operation. In each queue, you can perform arithmetic and physics middleware, and workflow made ​​themselves. It is, in a state that was carried out at the same time graphics processing it.
　These features, in a big way in the launch titles are not used. However, more broadly, in the whole life of the console, to be used in many games, according to the years passed, this kind of functionality should be going to be more and more important things.

 

[BoardBonobo]

Why weren't you surprised by the OS being based on BSD? It appears to be the first time they've done that (if I'm reading it correctly).

The PS3 uses portions of BSD in it's OS (according to the about bit) and an nmap sniff maps it to being a BSD system. So I would guess that it safe to assume they would stick with a known and pretty secure OS for the PS4.

 

[BoardBonobo]

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

Not forming a giant robot that destroys all enemies? Shame...

 

[patsu]

I vaguely remember PS3 GameOS only uses the BSD network stack, not the entire thing.

 

[patsu]

So the custom chip is basically a complete low-power SoC which acts as the whole south bridge, I guess that solves the issue of standby power, both the GDDR5 and liverpool can be completely powered off. I'm curious about what kind of local memory they used for it.

Rumor says 16GB Flash memory for system use.

Then again, rumors also said 14 + 4 and independent compute unit.

Not forming a giant robot that destroys all enemies? Shame...

It's CPU + background CPU + GPGPU. I wonder how BSD handles them seamlessly.

 

[jonabbey]

I vaguely remember PS3 GameOS only uses the BSD network stack, not the entire thing.

Yeah, the credits on PS3 talk about NetBSD, but I also thought they just used user space stuff? They certainly had a discrete hypervisor..

Good on them if they are using a BSD kernel on the PS4. It's a great choice.

 

[patsu]

Back to the same old question. Which browser ? ^_^

I have no idea if they only used user space stuff from NetBSD.