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

[Shifty Geezer]

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

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

It's not what I'd call great journalism, leading to yet more wild, unrealistic speculation.

eg.:

Sony is building its CPU on what it's calling an extended DirectX 11.1+ feature set...

World's first DirectX featureset CPU? Or a complete muddling of ideas? The APU may be designed as DX11.1+ allowing the CPU cores to snoop GPU state and help with debugging, but the CPU itself is a completely different architecture to anything related to hardware DX compliance.

What does he mean "two processors"? The CPU and GPU, or two "sections" of the GPU?

Just the CPU+GPU...

The system (PS4 system, APU) is also set up to run graphics and computational code synchronously, without suspending one to run the other (like every other CPU+GPU combination in existence). Norden says that Sony has worked to carefully balance the two processors (CPU + GPU) to provide maximum graphics power of 1.843 teraFLOPS at an 800Mhz clock speed while still leaving enough room for computational tasks.

There is no more magical compute power on the GPU. It's 18 CUs, from Sony's own mouth in their reveal, running 1.8 TF maximum as that's how many ops that hardware can do at that clock. You cannot magically get additional compute power from the same resources. That's a peak figure, the very maximum hypothetical performance of the CUs. If spending 1.8 TF on graphics work, there's nothing left on the GPU to spend on non-graphics tasks.

Where Sony can potentially gain advantages is switching between graphics and compute workloads to maximise efficiency. If compute and graphics thread states (don't know what the APU terminology is) can be stored concurrently, the CUs could potentially switch from one to the other with very little overhead. We also have the data sharing between CPU and GPU which eliminates memory dependencies and slow main RAM accesses. But there's no magical extra hardware there. Posts like onQ's are making a mountain out of a molehill similar to misinterpretation of old physics processing remarks on Wii leading to belief in a physics processing unit by some. That poor Japanese translation is clearly talking about the CPU having 8 cores, 8 threads, for example.

The PS4 system is 8 single threaded x86 cores, connected to 18 CUs with a peak float throughput of 1.8 TF. Nothing said has changed that, or hinted at any change.

 

[onQ]

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.

I don't think anyone has said anything about exceeding the theoretical peak of 1.84 TFlops.


what I'm getting from it is that graphic rendering doesn't use GPGPU to it's full potential so sony made the APU so it can run compute without taking away from the graphic rendering.

 

[3dilettante]

I'm thinking they're pointing out one of the improvements recent architectures have over earlier GPUs, where if you go back to VLIW5 you'll find GPUs that had to context switch the GPU to juggle the workloads, which is a very expensive operation.
VLIW4 and to a much greater extent GCN and Orbis, can run different kernels from different threads at the same time, so the heavy contexts don't need to switch out.
There's still a tradeoff, but there's no massive drop because the GPU can't juggle.

 

[patsu]

Shifty,

Originally Posted by ArsTechnica
The system (PS4 system, APU) is also set up to run graphics and computational code synchronously, without suspending one to run the other (like every other CPU+GPU combination in existence). Norden says that Sony has worked to carefully balance the two processors (CPU + GPU) to provide maximum graphics power of 1.843 teraFLOPS at an 800Mhz clock speed while still leaving enough room for computational tasks.

They are talking about GCN here. On PC, GPU physics are typically asynchronous because of the latency in issuing commands to the GPU.

As for maximizing 1.8 TFLOPs, it's like what you mentioned, improve efficiency since based on previous discussions, the GPU may run @ say, 60% efficiency at some stage.

Due to the Onion(+) compute link, careful cache management, and copious number of buffers and queues, developers can squeeze in compute jobs at the right time, while running graphics jobs at the same time. It will be interesting to see various synchronization schemes between the jobs, and between the 2 processors. That's where some of the SPU + RSX co-operation experiences come in. It's also interesting to see how developers use the Onion(+) and Garlic bandwidth.

In Cell, they have dedicated bandwidth to separate pools. Here, they will need to manage them somewhat programmatically.

EDIT: Beaten

 

[patsu]

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.?

Which context switching patent is this ? Sony's or AMD's ?

 

[Cyan]

There is an interesting article mentioning that the Battlefield 4 demonstration was actually running on an AMD 7990, and they wonder if the PS4 is going to be able to match that.

I think it will, especially taking into account that DICE is going to create the game on PC hardware and next gen consoles first and downscale it for the PS3 and 360 versions.

http://www.mcvuk.com/news/read/rade...d-4-demo-will-ps4-be-able-to-match-it/0113264

 

[AlphaWolf]

PS4 won't be matching a 7990, but it won't have to because the console will probably be rendering at 720p with fxaa instead of 1600p with 4xMSAA.

 

[patsu]

Someone should profile a few representative GPUs and see how they behave in different games.

PS4 and 720 should be targeting 1080p @ 30fps right ? DICE's first effort should not be taken as their best.

 

[Cyan]

Someone should profile a few representative GPUs and see how they behave in different games.

PS4 and 720 should be targeting 1080 @ 30p right ? DICE's first effort should not be taken as their best.

Well, sebbi :smile: loves how BF3 runs at 60 fps on the PC, and judging from the leaks I see the PS4 as a quite capable machine which can run BF4 at 60 fps and 1080p. That would be awesome but it's all hearsay right now.

 

[Arwin]

Thank god I missed the Ars Technica article and could just skip ahead to the much clearer Eurogamer article

 

[pjbliverpool]

I thought it was already confirmed 720p @60fps?

PS4 sports roughly 1/4 the raw power of a 7990 so somethings got to give. And this early in the generation the console specific advantages will be minimal at best.

 

[Shifty Geezer]

Digital Foundry provides a much better record on the GDC presentation, including the full quote of a quote that's been confusing some:

"The cool thing about Compute on PlayStation 4 is that it runs completely simultaneous with graphics," Norden enthused. "So traditionally with OpenCL or other languages you have to suspend graphics to get good Compute performance. On PS4 you don't, it runs simultaneous with graphics. We've architected the system to take full advantage of Compute at the same time as graphics because we know that everyone wants maximum graphics performance."

They're talking about handling threads, or compute jobs, and the GPU's ability to run compute in parallel with graphics which is going to be using some CUs in compute and the rest for graphics. There's no running both across the full GPU simultaneously. Whether that's the old 14+4 idea or the scheduler managing jobs across CU clusters is unknown at this point, but I'd put money on the latter.

 

[onQ]

http://www.eurogamer.net/articles/digitalfoundry-inside-playstation-4

"Compute" architecture

What was intriguing was new data on how the PlayStation 4's 18-compute-unit AMD graphics core is utilized. Norden talked about "extremely carefully balanced" Compute architecture that allows GPU processing for tasks that usually run on the CPU. Sometimes, employing the massive parallelization of the graphics hardware better suits specific processing tasks.

"The point of Compute is to be able to take non-graphics code, run it on the GPU and get that data back," he said. "So DSP algorithms... post-processing, anything that's not necessarily graphics-based you can really accelerate with Compute. Compute also has access to the full amount of unified memory."

"The cool thing about Compute on PlayStation 4 is that it runs completely simultaneous with graphics," Norden enthused. "So traditionally with OpenCL or other languages you have to suspend graphics to get good Compute performance. On PS4 you don't, it runs simultaneous with graphics. We've architected the system to take full advantage of Compute at the same time as graphics because we know that everyone wants maximum graphics performance."

Leaked developer documentation suggests that 14 of the PS4's compute units are dedicated to rendering, with four allocated to Compute functions. The reveal of the hardware last month suggested otherwise, with all 18 operating in an apparently "unified" manner. However, running Compute and rendering simultaneously does suggest that each area has its own bespoke resources. It'll be interesting to see what solution Sony eventually takes here.

 

[Cyan]

I thought it was already confirmed 720p @60fps?

PS4 sports roughly 1/4 the raw power of a 7990 so somethings got to give. And this early in the generation the console specific advantages will be minimal at best.

This brings up an interesting point, but... Why not? I mean, PS4 is allegedly 8-10 times more powerful than the PS3, so if the PS3 can run BF3 at 720p and 30 fps, 1080p and 60fps seems feasible on the PS4.

If it took a significant effort then I'd ditch that possibility, especially for a launch title, but I have heard that the 720p/60fps thing was a rumour and the guy who started it admittedly said that he made it all up, iirc.

I can't confirm nor deny this, it's what I remember hearing the last time on that rumour going round.

 

[patsu]

Ok stupid question time. What fixed function units are in the Orbis GPU ?

This brings up an interesting point, but... Why not? I mean, PS4 is allegedly 8-10 times more powerful than the PS3, so if the PS3 can run BF3 at 720p and 30 fps, 1080p and 60fps seems feasible on the PS4.

If it took a significant effort then I'd ditch that possibility, especially for a launch title, but I have heard that the 720p/60fps thing was a rumour and the guy who started it admittedly said that he made it all up, iirc.

I can't confirm nor deny this, it's what I remember hearing the last time on that rumour going round.

Oh I see you just went ahead and pour oil over fire.

 

[Cyan]

http://www.eurogamer.net/articles/digitalfoundry-inside-playstation-4

Does this mean that they can use those 4 CUs in games featuring deferred renderers?

I say this because PS4 is UMA and it was a handicap on the X360 when it came to deferred renderers, which they solved using the eDRAM for the G-buffer.

I wonder if the same applies in the case of the PS4.

 

[babcat]

Could they have added extra compute units to every CU so that an individual CU can run graphics and compute?

 

[babcat]

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.?

This would be great. However, we need clear confirmation from Sony.

 

[3dilettante]

"Leaked developer documentation suggests that 14 of the PS4's compute units are dedicated to rendering, with four allocated to Compute functions. The reveal of the hardware last month suggested otherwise, with all 18 operating in an apparently "unified" manner. However, running Compute and rendering simultaneously does suggest that each area has its own bespoke resources. It'll be interesting to see what solution Sony eventually takes here. "

The author's conclusion is a reach, given the data he is using as justification. Running compute and graphics simultaneously suggests nothing about reserving resources.

 

[Averagejoe]

Which context switching patent is this ? Sony's or AMD's ?

Sony filled for one not so long ago.