What is PS4's 14+4 CU thing all about? *spawn

Status
Not open for further replies.
The 14 + 4 thing came from a Sony slide (though not one meant for public consumption of course).

The slide never said 14+4. That was a simplification of "Hardware balanced to 14 CUs..4 extra CUs for GPGPU."

The problem with 14+4 is that it could be taken to indicate a hard partitioning of the resources or that there was something about the 4 CUs themselves that made them different from the rest. Sony's initial slide could also be taken that way, but VGLeaks' 14+4 was more misleading IMO.
 
The 14 + 4 thing came from a Sony slide (though not one meant for public consumption of course).
The origins of this most recent discussion are that very same slide.

Yes, diminishing returns and a knee in the performance curve for "graphics".
And in that slide, we have,
Hardware balanced at 14 CUs
4 additional CUs (410 GFlops) "extra" ALU as resource for compute
Minor boost if used for rendering
That's Sony's own words on their own slide, from 2012. So you can't begrudge people looking at that and thinking that 4 CUs aren't going to be contributing much to the graphics. Of course, that info wasn't on a slide meant for the general public, and there's zero context. As we know the rest of the architecture, we can decide for ourselves whether it's worth turning those 4 CUs towards graphics workloads or not. There's no denying the fact, though, that Sony said there was minimal benefit is used for rendering.

Niether Cerny nor the MS fellows are contradicting each other when they talk about "balance", so it's pretty amazing that the internet has managed to whip up a shit storm and hate campaign.
No-one in this thread (or the one that spawned it) was whipping up a shit storm. Airon was asking a question, perhaps oblivious to prior discussion. No need to drag the rest of the internet into it.
 
That slide is ridiculous, is like made in a world in which an AMD 7770 is the most powerful card in the world and so having 1,3 tflops is enough for your rendering engine.Tell that to Epic and their voxels or to Dice whose BF3 jumps from 30 to 45 fps at 1080p at ultra between a 7770 and a 7850. Or well tell that to Mantel...an API made to get more rendering from the ALUs.
Is marketing.Cerny want to sell us they have gone for a GPGPU monster, but in a world ruled by Titans and R9-290s this argument is so so weak....
 
Last edited by a moderator:
That slide is ridiculous, is like made in a world in which an AMD 7770 is the most powerful card in the world and so having 1,3 tflops is enough for your rendering engine.Tell that to Epic and their voxels or to Dice whose BF3 jumps from 30 to 45 fps at 1080p at ultra between a 7770 and a 7850. Or well tell that to Mantel...an API made to get more rendering from the ALUs.
Is marketing.Cerny want to sell us they have gone for a GPGPU monster, but in a world ruled by Titans and R9-290s this argument is so so weak....

Re-read ERP's post earlier in the thread. Try to understand it. Then reconsider what you've said here.
 
AMD exec in a vrzone interview said the mantle sdk will go public in a year or so, so we'll be able to test 7850s and isolate individual CU and test whether there really is a drop off in benefit going from 14 to 18 additional CU for rendering.

I really hope that Cerny designed the +4CU to be as (nearly) capable in the graphics rendering pipeline as the 14. I doubt many 3rd parties aren't going to make significant use of those ACE over the consoles lifetime, and only 1st party titles like Knack will. Knack seems like a fun good game, but Cerny's method of legitimizing the ACE-ification of the CU.
 
Last edited by a moderator:
Re-read ERP's post earlier in the thread. Try to understand it. Then reconsider what you've said here.

I have read it and i am sure that almost all devs will go full force with rendering in those ALUs except for if being true there are idle cycles, and in this case will be Sony first parties who will take care of them with GPGPU .
 
New influx of xbots new re telling of the 14+4 'myth'

Ps4 is 18 cu they are all identical and fully functional.

I don't see why it hurts feeling here that "balanced at 14" should be mentioned/queried.

How is it even a negative? There are still 18 CUs.
 
AMD exec in a vrzone interview said the mantle sdk will go public in a year or so, so we'll be able to test 7850s and isolate individual CU and test whether there really is a drop off in benefit going from 14 to 18 additional CU for rendering.

I really hope that Cerny designed the +4CU to be as (nearly) capable in the graphics rendering pipeline as the 14. I doubt many 3rd parties aren't going to make significant use of those ACE over the consoles lifetime, and only 1st party titles like Knack will. Knack seems like a fun good game, but Cerny's method of legitimizing the ACE-ification of the CU.

See Battlefield 3 benchs and how frame rate change from a 7770 to a 7850( or compare a 7770 with Bonaire better that is a 7770 with more ALU power ).More CUs impact performance.
 
I have a feeling that the reason people are concerned about the 14+4 is the implication that somehow the PS4 is no longer 50% more powerful graphically (but the reality is that it never was at any point). ;-)
 
I don't see why it hurts feeling here that "balanced at 14" should be mentioned/queried.

How is it even a negative? There are still 18 CUs.
It's negative to console warriors. The XBox contingent would say PS4 only has 14 CUs for graphics versus 12 in XB1, and could go as far to say the other 4 CUs are for audio and stuff.

As a technical discussion, the most relevant question is, "what are the limiting factors in the rest of the GPU where 14 CUs would be optimal and 18 CUs a little too much for optimal graphics?"
 
I take it to mean you can simply have 29% more shader/texture performance (a marginal improvement) or 400% more compute performance (measured in relation to the CPU). Obviously the latter has the potential to be more benefitial especially in scenarios where the rest of you system is bottlenecking the CU's and so they are not fully utilised anyway (as would be the case in cross platform games where xb1 is the lowest common denominator). However I dont think theres any doubt that in a shader/texture limited game the extra CUs should give a near linear performance increase.
 
I don't see why it hurts feeling here that "balanced at 14" should be mentioned/queried.

How is it even a negative? There are still 18 CUs.

Knee-jerk reactions from over-sensitive members. This information has been used as ammo in the fanboy wars in the past and that makes some members automatically get defensive when the subject comes up. I hope there are enough members still interested in nuanced, intelligent conversation to rise above the noise.
 
It's negative to console warriors. The XBox contingent would say PS4 only has 14 CUs for graphics versus 12 in XB1, and could go as far to say the other 4 CUs are for audio and stuff.

As a technical discussion, the most relevant question is, "what are the limiting factors in the rest of the GPU where 14 CUs would be optimal and 18 CUs a little too much for optimal graphics?"


Bandwidth?
 
A risk of derailing the already derailed discussion

However I don't think theres any doubt that in a shader/texture limited game the extra CUs should give a near linear performance increase.

Bottlenecks are massively misunderstood by most people on forums, and it annoys me how much it's bandied around almost as much as the use of the term bottlenecking. There is always some bottleneck in the system, and games generally are not bottlenecked by a single item for an entire frame, but at various points in a frame are bottlenecked by various different systems, I could be Vertex or possibly Fill limited when drawing Shadows, I might be Triangle setup limited when drawing my main character. A post processing step could be bandwidth limited or ALU limited, I could also be CPU limited (and it's surprisingly common for games) etc etc etc.

It's true that you see some games where one aspect is dominant, but it isn't generally the case.
The point of having more than one ring buffer is to allow compute jobs to be submitted in parallel with the graphics jobs, then when the ALU's would otherwise idle waiting on other systems they can do something. FWIW I'm unconvinced that having 64 queues in this case is much better than having 2 unless there are use cases, where you need to submit relatively low latency tasks while very long running tasks sit in other queues, I guess we'll see.
 
[/B]

Bandwidth?

That's probably one of the factors.

When thinking about the performance and the "balanced" statement, I think you have to evaluate the system as a whole and not just each component in isolation.


I have been wondering if perhaps the 18 CU's were a better match for the rumored 4 core 3.2 GHz Steamroller version of the PS4. (I think that was a rumor a year+ ago) and when the decision was made to go with Jaguars, they just kept the GPU intact even though it maybe limited a little by the CPU. If the Steamroller rumor had any validity that is.
 
A risk of derailing the already derailed discussion



Bottlenecks are massively misunderstood by most people on forums, and it annoys me how much it's bandied around almost as much as the use of the term bottlenecking. There is always some bottleneck in the system, and games generally are not bottlenecked by a single item for an entire frame, but at various points in a frame are bottlenecked by various different systems, I could be Vertex or possibly Fill limited when drawing Shadows, I might be Triangle setup limited when drawing my main character. A post processing step could be bandwidth limited or ALU limited, I could also be CPU limited (and it's surprisingly common for games) etc etc etc.

I think you just said what the Tech Fellow from MSFT said with EuroGamer.

"The goal of a 'balanced' system is by definition not to be consistently bottlenecked on any one area. In general with a balanced system there should rarely be a single bottleneck over the course of any given frame - parts of the frame can be fill-rate bound, other can be ALU bound, others can be fetch bound, others can be memory bound, others can be wave occupancy bound, others can be draw-setup bound, others can be state change bound, etc. To complicate matters further, the GPU bottlenecks can change within the course of a single draw call!"

Or are you actually Mr. Goossen?
 
I wish 14+4 was banned from the B3D lexicon. Its a vague figure that illicit deep discussion recycled many times over and mostly pulled from our asses.

LOL.
 
Status
Not open for further replies.
Back
Top