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

[patsu]

I believe it's proprietary and not stock, which to me alludes from Mark Cerny's interview, stating they couldn't get HSA and such on time so they had to make their own...

Granted, running is not the translate cannot automated text brings the understanding disaster.

GAF commented that Orbis does not just include GCN but it extends the latter. *If* they are correct, then Cerny probably meant Sony added something to (try to) achieve some features in a future GCN revision.

I would be surprised if they whipped out their own GCN in such a short time.

 

[patsu]

Do you guys thing that the Jaguar CPU will be a bottleneck in extracting all the theoretical performance from the graphics side of the APU?

I am curious about this too.

The developers will optimize for Orbis, so they will likely program around the specs.

The Orbis CPU has dedicated h/w units to offload some mundane work (audio, video, decompression, loading levels in the background via the extra ARM chip).

The GPU can perform more compute tasks compared to a traditional PC setup. I have seen path finding, raycasting driven compute jobs, fluid and hair stuff, etc. on AMD slides.

I think the proof is in the pudding. Developers have to slap their code on it to really know the hardware shortfalls.

I am curious what difference can 8GB GDDR5 bring to the table though.

 

[Silent_Buddha]

I'm guessing he misspoke. He could mean the CUs are less likely to go unused with the added compute only pipelines adding jobs to the queue. This is the "efficiency" that was touted a while back for Durango.

You have to look at what onQ was replying to in order to see the proper context.

Of course it can do compute and graphics at the same time, like every other desktop GPU that's been on the market for the past few years.. Love_In_Rio is saying that you don't get the full rendering power of a 7850-level GPU if you're using some of the CUs for something else.

Yes, efficiency gives you some gains as has been speculated WRT Durango. That isn't in question. The same way adding in things to facilitate HSA as seen in Bonaire which are likely in the Durango GPU also lead to increased efficiency.

The fact that he's replying to that in an attempt to prove it wrong implies that he thinks that doing Compute on a CU resource isn't going to make it unable to work on Graphics workloads on that same resource.

It's not so much me misinterpreting onQ, but that he's seriously misinterpreting Cerny and we're all trying to point that out to him. Never mind the whole nonsense about Jaguar + GCN GPU = Glorified Cell nonsense which started the whole thing.

Regards,
SB

 

[DrJay24]

You have to look at what onQ was replying to in order to see the proper context.

I read it, but I read "the CUs can do Graphics & Computing at the same time efficiently" as the aggregate 18CUs, not a single CU doing both at the same time. Without the additional compute only pipelines you would have less efficient use of the CUs. I don't want to speak for him, what he says is ambiguous enough not to jump to conclusions and start a dog pile.

 

[AlphaWolf]

If what you understand him saying isn't possible, he likely meant something else.

 

[Shifty Geezer]

II'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.

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

Because it's not. SPEs were autonomous processors with full programmability and a vector compute engine. CUs are massive SIMD arrays with robust yet limited programmability and various non-compute functional extras (TMUs, etc). The architecture is not Cell. It's homogenous multicore CPU and discrete 'GPU' with data sharing allowing for 'GPGPU' algorithms, versus heterogeneous CPU with low-power full CPU and vector strong, autonomous (once kickstarted) lite CPUs. They aren't even remotely the same in terms of software design either. On Cell, you design your code as much as possible as SPE jobs, run on SPEs. 3.2 GHz of linear processing speed meant linear jobs could be given to the SPEs and they'd do a good job on them. PPE often handles job scheduling, telling which SPE to load which program when, and then the SPE goes off as a discrete CPU core and does it. If your analogy were correct, Liverpool would have the 8 CPU cores creating jobs for the CUs to process, and once loaded, those CUs doing their own thing.

Liverpool (GCN APU) is something new. It's more comparable to a separate vector compute unit hanging off the CPU, in the same way FPUs used to be additional silicon to the CPU. The CPU cores execute programs and can farm out vector-heavy workloads to the CUs. Algorithms can be designed that use both processing pools simultaneously on the same data, but in different ways playing to their different strengths.

 

[patsu]

In the Japanese interview, Mark Cerny mentioned that they learned from their Cell experiences to build PS4. In this sense, onQ, the PS4 borrows or inherits from PS3.

As I understand, both machines have some sort of libGCM API to issue commands and synchronize them from the CPU to GPU.

Compute tasks are handled by the CPU's vector engines, as well as the GPU's compute units. In the latter case, it's hetereogeneous like the SPUs. The CPU has to "talk a different language" to make the CUs work for them (and may be vice versa). However the nature of the SPUs is very different from the GPU's CUs.

Orbis is also programmable using Playstation GL, which extends from HLSL (used in RSX). [EDIT: Oh wait, I think RSX also supports PSGL]

The big difference is the memory is unified, and so there should be less copying. What's more vague now is how closely the CPU and GPU can work together, like the Onion+ compute link we spoke of above. In the SPUs, the developers have full control over the LocalStores and dedicated bandwidth. In Orbis, they may need mechanisms to make sure caches are not polluted easily. Similarly, bandwidth consumption is also something to manage.

There are things the SPUs can compute more efficiently than the CUs collectively. The reverse is also true.

EDIT: Oh and in Orbis, some specialized tasks have been farmed out to specialized hardware units so that the CPU is more "free" to do important work. These were all handled by the SPUs.

 

[Averagejoe]

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...

Well but that is ok remember the 7990 alone is $900 dollars the PS4 will not cost you half of that probably,and that is without counting a Sandy Bridge or a Haswells.

 

[arijoytunir]

the headset jack features stereo sound

http://www.gamesindustry.biz/articl...-simple-experience-for-developers-and-players

Does this means the in game audio can be heard on headphones ?

 

[whome0]

http://www.gamesindustry.biz/articl...-simple-experience-for-developers-and-players
Does this means the in game audio can be heard on headphones ?

100% sure it does stream chat+game audios on headset, anything else would be a big (software) engineering failure.

"The system is capable of streaming 32Khz sound to the controllers' speakers for up to 2 players, but that reduces to 16Khz when 3 or more players are hooked up."

What do people think about this, do system is able to stream dedicated gameaudio+(shared online chat) on each headset? Living room 2-player game and each can only hear an environment audio around own pixel hero?

 

[yewyew]

The system is also set up to run graphics and computational code synchronously, without suspending one to run the other. Norden says that Sony has worked to carefully balance the two processors to provide maximum graphics power of 1.843 teraFLOPS at an 800Mhz clock speed while still leaving enough room for computational tasks. The GPU will also be able to run arbitrary code, allowing developers to run hundreds or thousands of parallelized tasks with full access to the system's 8GB of unified memory.

http://arstechnica.com/gaming/2013/...4s-hardware-power-controller-features-at-gdc/

 

[Rangers]

Am I smelling 14+4 again? It's clear as mud...

What does he mean "two processors"? The CPU and GPU, or two "sections" of the GPU? Or is the writer enough of a luddite to think compute and graphics are separate processors within a normal GPU?

Also, wow

Speaking of memory, Norden hyped up the 8GB of GDDR5 RAM in the system as the type of memory that's currently usually found only on high-end graphics cards. Calling the RAM "expensive" and "exotic," Norden stressed that you "can't buy this [RAM] for 50 bucks... that's why high-end graphics cards cost as much as they do."

Sony has been forthcoming with PS4 details, it's good stuff, hope MS follows suit. If only they could do the same for NPD....

 

[ZiGgY]

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

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

Wonder if they'll reveal the clock speed of the CPU cores...1.6Ghz stock or the rumoured 2Ghz speed bump.

It would be nice to know.

 

[onQ]

http://arstechnica.com/gaming/2013/...4s-hardware-power-controller-features-at-gdc/

Sounds familiar.

Am I smelling 14+4 again? It's clear as mud...

What does he mean "two processors"? The CPU and GPU, or two "sections" of the GPU? Or is the writer enough of a luddite to think compute and graphics are separate processors within a normal GPU?

Nope it says clear as Day that it will have 1.84TFLOPS for graphics & still have room for computing.

 

[Silent_Buddha]

Nope it says clear as Day that it will have 1.84TFLOPS for graphics & still have room for computing.

Obviously as he specifically states "two" processors. The CPU will always be free to do computational tasks.

Regards,
SB

 

[Deleted member 13524]

100% sure it does steram chat+game audios on headset, anything else would be a big (software) engineering failure.

"The system is capable of streaming 32Khz sound to the controllers' speakers for up to 2 players, but that reduces to 16Khz when 3 or more players are hooked up."

What do people think about this, do system is able to stream dedicated gameaudio+(shared online chat) on each headset? Living room 2-player game and each can only hear an environment audio around own pixel hero?

I think this is wrong and what they're actually talking about the audio being sent to the console through a headset connected to Dualshock 4.

At least I hope so, because 32KHz (which in 16bit is what, a 48Kbps MP3 or 32Kbps AAC?) is pretty much useless for the game's sound IMO.

Bluetooth can do 2Mbps, so I think the PS4 could send up to 4 streams of 128Kbps AAC to 4 Dualshock 4 controllers, or an advanced for of Sony's own ATRAC.
For a single player / single controller, a 320Kbps sound stream should be doable.

This would translate into 48Khz sampling rate, at least.

 

[Averagejoe]

http://arstechnica.com/gaming/2013/...4s-hardware-power-controller-features-at-gdc/

Wow very interesting so the PS4 GPU will handle both at the same time without pone hurting the other,could this have anything to do with sony content swiching patent.?

 

[Gradthrawn]

I think this is wrong and what they're actually talking about the audio being sent to the console through a headset connected to Dualshock 4.

At least I hope so, because 32KHz (which in 16bit is what, a 48Kbps MP3 or 32Kbps AAC?) is pretty much useless for the game's sound IMO.

Bluetooth can do 2Mbps, so I think the PS4 could send up to 4 streams of 128Kbps AAC to 4 Dualshock 4 controllers, or an advanced for of Sony's own ATRAC.
For a single player / single controller, a 320Kbps sound stream should be doable.

This would translate into 48Khz sampling rate, at least.

This would be my guess as well, similar to the 360's controller headset if I remember correctly (haven't used it in a while), as the audio would have to share the bandwidth of the controller's BT connection unlike the separate wireless/BT headsets now that use a separate connection. However, I thought the BT "piconet" was limited to 7 active slave devices mated to the BT master controller? Has that been updated for BT 2.1?

 

[onQ]

Obviously as he specifically states "two" processors. The CPU will always be free to do computational tasks.

Regards,
SB

Seems to be a lot of confusion going on but there is most definitely something going on.

Additional hardware: GPU-like Compute module, some resources reserved by the OS


However, there's a fair amount of "secret sauce" in Orbis and we can disclose details on one of the more interesting additions. Paired up with the eight AMD cores, we find a bespoke GPU-like "Compute" module, designed to ease the burden on certain operations - physics calculations are a good example of traditional CPU work that are often hived off to GPU cores. We're assured that this is bespoke hardware that is not a part of the main graphics pipeline but we remain rather mystified by its standalone inclusion, bearing in mind Compute functions could be run off the main graphics cores and that devs could have the option to utilise that power for additional graphical grunt, if they so chose.

http://arstechnica.com/civis/viewtopic.php?f=22&t=1193497&start=440
Poster Blacken00100
"So, a couple of random things I've learned:

-It's not stock x86; there are eight very wide vector engines and some other changes. It's not going to be completely trivial to retarget to it, but it should shut up the morons who were hyperventilating at "OMG! 1.6 JIGGAHURTZ!".

-The memory structure is unified, but weird; it's not like the GPU can just grab arbitrary memory like some people were thinking (rather, it can, but it's slow). They're incorporating another type of shader that can basically read from a ring buffer (supplied in a streaming fashion by the CPU) and write to an output buffer. I don't have all the details, but it seems interesting.

　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.

The system is also set up to run graphics and computational code synchronously, without suspending one to run the other. Norden says that Sony has worked to carefully balance the two processors to provide maximum graphics power of 1.843 teraFLOPS at an 800Mhz clock speed while still leaving enough room for computational tasks. The GPU will also be able to run arbitrary code, allowing developers to run hundreds or thousands of parallelized tasks with full access to the system's 8GB of unified memory.

About 14 + 4 balance:

- 4 additional CUs (410 Gflops) “extra” ALU as resource for compute

- Minor boost if used for rendering

Dual Shader Engines:

 

[Scott_Arm]

GCN can do GPGPU and graphics processing at the same time because of the way the ACEs manage scheduling, but that doesn't mean you can exceed the theoretical peak of 1.84 TFlops. I believe the arstechnica article is probably incorrect.