By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Sony - Mark Cerny Explains The PS4's GPU - Can Perform Asynchronous Complex Processes

A couple of days back, Mark Cerny, the lead architect on the PlayStation 4, attended Gamelab 2013, a developer conference which took place in Barcelona. At the conference he spoke in great length about the PS4 and its set of features.

He also spoke about how the PS4′s GPU is not only limited to handling the graphics but also complex processes which is traditionally handled by the CPU.

“We have worked hard to ensure that the console has a rich feature set, which will allow it to grow over the years and support the overall evolution of gaming. And our work there on that rich feature was to focus on making sure that for those teams that were interested in investing time, the GPU could be used for far more conventional graphics. Principally, we enhanced the GPU to make the use of asynchronous fine grained compute practical on the platform. So the asynchronous refers to the GPU doing many tasks which is not directly related to graphics.”

He believes that the developers will be able to develop richer worlds if they are able to unlock the true potential of this feature set.

“Physics simulation, collision detection, ray casting for audio, decompression and the like. And these operations are fine grained meaning that there will be many small world simulation tasks running on the GPU simultaneously alongside rendering of the game scenes. So the concept is that as game developers learn to use these techniques later on in the console life cycle, we will see richer and even more interactive worlds.”

The presentation also had a slide which gave more details about the PS4 GPU and it’s capabilities:

As you can see from the image above, the PS4′s GPU can also handle some of the CPU based tasks like collision detection and simulation.

  • Decompression
  • Raycasting of Audio
  • Physics Simulation
  • Collision Detection
  • World Simulation

On Friday we had reported on why the Cell processor was not included in the PS4. You can read that report over here. Additionally, you should check out Mark Cerny’s brilliant 49 minute long presentation over here. Trust me, you won’t be disappointed.

http://gamingbolt.com/mark-cerny-explains-the-ps4s-gpu-can-perform-asynchronous-complex-processes



Around the Network

This man is inspirational to the god damn CORE.



The customization to 8 ACEs and 8 queued pipelines (instead 2:2 in HD 7000) is the responsible for thatI guess.



Just asking but is he really stating anything that isn't just part of the AMD GNC GPU. So far I believe we all know it has compute units which can offload CPU task to the GPU. This is nothing new about the system at least from reading the blurb.



Machiavellian said:
Just asking but is he really stating anything that isn't just part of the AMD GNC GPU. So far I believe we all know it has compute units which can offload CPU task to the GPU. This is nothing new about the system at least from reading the blurb.

GNC can't do that how PS4 GPU can... in GNC you have a lower queue pipeline and ACE... so if you want to perform both taks (compute and graphics) at the same time you need to wait one or other finish.

PS4 GPU added more 6 queue pipelines and 6 ACEs.. 8:8 compared to 2:2 on GNC... so if you want to execute compute and graphics at the same time you have units to make this happen.

That the two ACEs units in GNC.

PS4 have 8.



Around the Network

Just to add the PS4 GPU have the same numbers of ACEs and queue pipelines than what the first documentation of GNC 2.0 shows to have.



ethomaz said:
Machiavellian said:
Just asking but is he really stating anything that isn't just part of the AMD GNC GPU. So far I believe we all know it has compute units which can offload CPU task to the GPU. This is nothing new about the system at least from reading the blurb.

GNC can't do that how PS4 GPU can... in GNC you have a lower queue pipeline and ACE... so if you want to perform both taks (compute and graphics) at the same time you need to wait one or other finish.

PS4 GPU added more 6 queue pipelines and 6 ACEs.. 8:8 compared to 2:2 on GNC... so if you want to execute compute and graphics at the same time you have units to make this happen.

That the two ACEs units in GNC.

PS4 have 8.

Ok, thanks for clearing that up.  That should mean that the GPU can perform both graphics and Compute functions at the same time, how does this effect processing.  In other words, will graphics take a hit if you are using the compute units at the same time or vice versa.  I know in the past this was a problem for AMD chips.



Machiavellian said:

Ok, thanks for clearing that up.  That should mean that the GPU can perform both graphics and Compute functions at the same time, how does this effect processing.  In other words, will graphics take a hit if you are using the compute units at the same time or vice versa.  I know in the past this was a problem for AMD chips.

No hit at all.

GNC can do that too in a limited way... in GCN you have 2 schedules (ACE) so if they are full you can't send more asynchronous (Compute or Graphics) process to GPU... so you need to wait the tasks finish to send more taks to GPU.

I don't how many tasks a ACE can schedule to CPU at the same time but I will use 1:1 to make easy the example... so each ACE can schedule 1 task in GPU per time.

2 ACE = 2 tasks... so you can run 2 different traks in the GPU at the same time.
8 ACE = 8 tasks... so you can run 8 different tasks in the GPU at the same time.

Eg.

I have 4 different tasks to execute in a GPU with 18CUs (I'm using CUs to make more easy the example):

1. Graphic render (needs 10CUs)
2. Decompression (needs 1 CU)
3. Physics Simulation (needs 4 CUs)
4. Collision Detection (needs 2CUs)

So in GNC you can schedule only 2 tasks at the same time, so you will execute:

1. Graphic render (needs 10CUs)
2. Decompression (needs 1 CU)

You will have 7 CUs free but that can't be used... so you need to wait the task 1 or 2 to finish to start the 3.

But in PS4 GPU you have 8 tasks at the same time, so you will execute everyting at the same time:

1. Graphic render (needs 10CUs)
2. Decompression (needs 1 CU)
3. Physics Simulation (needs 4 CUs)
4. Collision Detection (needs 2CUs)

You will have 1 CUs free yet... so if you have another tasks that needs 1CU you can use it because you are only using 4 ACEs and have another 4 free.

What that means? More efficience... you can use all the resource of the GPU at the same time without need to wait the GPU tasks finish.

NOTE. All the numbers are created and just used to try to explain with examples... the ACE can schecule more than 1 task per unit... and there are more than 1 task for graphic render too.

The point is... with 8 ACEs you can use the idle units of GPU to do others things in parallel... you can put more tasks in parallel into GPU... if at same time you are using only 50% of the units in GPU with 2 ACEs then with 8 ACEs you can the other 50%.

Efficiency is the word here.

That's another feature that put the console GPU over the PC GPU for now... you can use more and better the power of the GPU... in console you can extract more from the GPU with less idle time.



Enjoying digesting all the technical aspects of the PS4 from the amazing Mark Cerny. Interesting that he talks about learning the GPU techniques later on in the console life cycle, and that the PS4 has a rich feature set that will allow it to grow.

This is great as the games will visually and in size improve over time and with that 7GB GDDR5 to tap into I can’t wait to see what the developers will create! It may be all about the games, but without improved hardware to power the dreams of game developers those dreams would die and so would the games!



Damn Mark Cerny is sexy!

no homo



Talal said:
I will permaban myself if the game releases in 2014.

in reference to KH3 release date