No, the 14+4 was specifically regarding rendering, not things like physics or GPGPU. Everyone agrees theres more raw computational power in the PS4 GPU, people are just just trying to make sense of the vgleaks stuff and even Cerny's comments about it not being 'perfectly round', does he have an agenda?
PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)
- Thread starter Love_In_Rio
- Start date
- Status
No, the 14+4 was specifically regarding rendering, not things like physics or GPGPU. Everyone agrees theres more raw computational power in the PS4 GPU, people are just just trying to make sense of the vgleaks stuff and even Cerny's comments about it not being 'perfectly round', does he have an agenda?
On-screen results in the PC space are capped by the lowest common denominator that the developers want to target. A GTX 780 can accomplish way more game at 1080p than a GTX 630 if targeted. A console with 18 CUs can accomplish more than a console with 14 CUs if targeted.
Especially someone like Mark Cerny.
I see a lot of people saying Cerny said it, but can't find an exact quote. Does anyone have an exact quote from Mark Cerny stating diminishing returns with each CU they added to the system?
Especially someone like Mark Cerny.
I see a lot of people saying Cerny said it, but can't find an exact quote. Does anyone have an exact quote from Mark Cerny stating diminishing returns with each CU they added to the system?
Here:
...I'm not seeing it.
Everything Cerny said so far seems to be based on the usefulness of these 18CUs, he thinks they will be useful for stuff other than shaders, and that 14 to 18 is "just a bit more ALU". He seems proud of the GPGPU capabilities of the PS4 and is trying to convince developers to spend the time to use that. It's like he's downplaying the difference between 14 and 18 for the graphics, because he strongly believes it makes a bigger difference for other things. However he also said devs will only start to tap then them for compute in 2-3 years, that implies most devs will still used 18CU for graphics for the first year of games.
Interestingly SOE have an MMO coming up with huge open areas and the terrain is all voxels and 100% destructible/constructible, some kind of photorealistic minecraft. I expect that to require a lot of GPGPU work to transform the voxels. All of a sudden the "balance" is completely difference from one game to another, and if it needs 6CUs to manage the voxels and physics, you still have a fair 12CU for the shaders, while on another console you'd be down to only 6CU for rendering.
Interestingly SOE have an MMO coming up with huge open areas and the terrain is all voxels and 100% destructible/constructible, some kind of photorealistic minecraft. I expect that to require a lot of GPGPU work to transform the voxels. All of a sudden the "balance" is completely difference from one game to another, and if it needs 6CUs to manage the voxels and physics, you still have a fair 12CU for the shaders, while on another console you'd be down to only 6CU for rendering.
The use of the word incentive goes against this theory. He's not trying to convince anyone (why would he need to convince a developer to do anything?) , he's saying the way the system is currently designed incentivizes not using all the CUs for graphics. Its an incentive to use the last 4 for GPGPU because the returns are diminishing if used for graphics due to the balance of the system. It lines up perfectly with the 14+4 from vgleaks, and its not a bad thing either.
Holy shitballs, WHY IS ANYBODY STILL TALKING ABOUT THE 14+4 NONSENSE?!
That's been put to bed. Repeatedly. Let it go. You are operating on a flawed assumption. It's been confirmed multiple times. I can't believe anybody is still beating this dead fucking horse.
If you actually read the VGLeaks article, it specifies that it is balanced for graphics at 14 and the other 4 only offer very small improvements to the graphics. Ppl love to cite the article as if it invented the 14+4 thing as two staved off separate arrays when it actually is quite clear when you put it up against Cerny's interviews.
Holy shitballs, WHY IS ANYBODY STILL TALKING ABOUT THE 14+4 NONSENSE?!
That's been put to bed. Repeatedly. Let it go. You are operating on a flawed assumption. It's been confirmed multiple times. I can't believe anybody is still beating this dead fucking horse.
All the info we have is that it is balanced at 14CU's for graphics. You are right that the initial reaction to the leaks somehow convinced ppl it might be 14 for graphics, 4 for compute only. THAT is what was put to bed. Alas, that's not what we are discussing now. We are simply noting that it sure seems as if the real setup really is designed for 14 spent on graphics rendering and 4 spent on compute, even if those 4 can work on rendering tasks, as it seems they are handicapped in some sense in that area somehow.
No need to freak out and try to quench discussion on the issue. Just read the thread and you will see that ppl are talking about something slightly different than what you are referring to.
This is ridiculous. As others have already said, the only bottleneck for more than 14 CU's is the code that you write. Sure you can code a game that perfectly balances 14CU's with 8 Jaguar cores and 176GB/s bandwidth but you can just as easily do the same with 18CU's, 8 Jaguar cores and 176GB/s bandwidth.
Cerny's angle is clearly that he believes there's a greater return for using ~22% of the GPU's resources on GPGPU than if those resources were used to incrementally increase the graphics by an additional ~29%.
There's no mysterious handicap that prevents all 18 CU's from being efficiently used for graphics. There's just (arguably) more effective work you can put ~22% of your CU's to. The same argument could possibly be made for the XB1 in the same percentages as well.
For fucks sake, no. They are all identical CUs.
Silent_Buddha
Legend
No, as Brit mentioned they are all the same.
If a developer wanted to use all 18 CUs for compute, they could do so.
If a developer wanted to use all 18 CUs for graphics related work, they could do so.
But he's suggesting that perhaps the best use is to use some number of them for compute and some number of them for graphics. Which is where I'm going to guess the whole misunderstanding about the 14+4 came from. It was likely just an example he threw out at some meeting as an example of what he meant.
But nothing about the 18 CUs screams out don't use more than 14 for graphics related work or [pick your choice] happens.
Regards,
SB
itsmydamnation
Veteran
This is so simple i cant believe this is still going on!!!!!!!
i did this table
now when you look at this table theirs nothing out of the ordinary with either the 14 CU PS4 or 18CU PS4. What you need to do is look at in game performance NORMALIZED FOR CLOCKS (particularly in the SKU's with aprox the same number of total units)! what do you find? you find that on current games the chips that have the lower ALU:MTU:ROP ratios have higher performance per unit.
When you look at it from that perspective his comments make perfect sense. A 14 CU PS4 gpu is more balanced. a 18CU PS4 is ALU heavy, just like bonnaire is compared to Pitcairn. Yet if we look forward 3-5 years with compute shaders and a like, ALU operations per pixel outputted is going to increase the PS4 configuration does look more "future proof", just like things like 1950XT.
But at the same time thats nothing compared to how ALU heavy Tahiti is. its going to be interesting looking at Southern Islands GPU performance pound for pound, if resolution/ triangle counts dont grow much Tahiti could look very good, but if tessellation and alike really kick in its low ROP to ALU count could really hurt.
i did this table
now when you look at this table theirs nothing out of the ordinary with either the 14 CU PS4 or 18CU PS4. What you need to do is look at in game performance NORMALIZED FOR CLOCKS (particularly in the SKU's with aprox the same number of total units)! what do you find? you find that on current games the chips that have the lower ALU:MTU:ROP ratios have higher performance per unit.
When you look at it from that perspective his comments make perfect sense. A 14 CU PS4 gpu is more balanced. a 18CU PS4 is ALU heavy, just like bonnaire is compared to Pitcairn. Yet if we look forward 3-5 years with compute shaders and a like, ALU operations per pixel outputted is going to increase the PS4 configuration does look more "future proof", just like things like 1950XT.
But at the same time thats nothing compared to how ALU heavy Tahiti is. its going to be interesting looking at Southern Islands GPU performance pound for pound, if resolution/ triangle counts dont grow much Tahiti could look very good, but if tessellation and alike really kick in its low ROP to ALU count could really hurt.
Maybe I should clarify. I do not hold stock in the 14+4 theory. I was getting the impression that the CU's were designed with compute in mind and since Cerny hinted at "extra ALU" that this was intended to increase performance of compute. But that extra ALU will be part of the design for all 18 CU's.
pMax
Regular
I cannot really understand how Cerny's words can be twisted so much, being so clear.
1) You have given as much ALUs as possible.
2) you have onion+ bus to do fast GPGPU interactions with the CPU without disrupting any existing shader execution.
We know AMD SI family needs many kernel in flight to work properly: and GPGPU can fulfill this requirement.
Cerny made even a simple example of what he meant: make the same job running as shader and GPGPU job, and run both.
1) You have given as much ALUs as possible.
2) you have onion+ bus to do fast GPGPU interactions with the CPU without disrupting any existing shader execution.
We know AMD SI family needs many kernel in flight to work properly: and GPGPU can fulfill this requirement.
Cerny made even a simple example of what he meant: make the same job running as shader and GPGPU job, and run both.
We can even go so far to imagine a realistic scenerio where this was demonstrated. "Here's a scene rendered using 14 CUs. Now we add another 4 CUs to the rendering. Notice the improvement? No, we don't much either. So how's about we use those CUs to do non-graphics stuff like this...(effect showing something amazing like, I dunno, fluid dynamic volumetric smoke interacting with the environment (which of course should count as graphics!)). Better? We think so too. Of course, if you want to spend those extra 4 CUs on graphics, that's fine, but I hope you feel an incentive to use them for something else with a more dramatic impact on the game."
I consider that explanation as uniting all the evidence and soundbites, so it's the one I'm most comfortable with.
Sadly they weren't at all clear. He used loose language with numerous connotations (what does "round" mean? Certainly not the geometric arrangement of the hardware, making it completely unclear) and that's why people are reading it different ways. Cerny provided an ink-blot of understanding, and people are seeing it in different ways, which is to be expected.
- Status
