• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Full PS4 specs, console and controller [Blu-ray drive confirmed]

spwolf

Member
Why, the PS4 seems alot more simplistic than the PS3.
Im willing to bet the AMD cpu is way cheaper than the Cell BE was, which had something like the 20% yields some months before it launched iirc. No extra processors for backwards compatibly. Blu ray drive is cheap as heck now.

cpu+gpu+ram should be very similar at the end... difference will be BD + PSU + Cooling
 

Crisco

Banned
The way they called it a "game server" at one point, made me wonder if they plan on releasing tiny/cheap-ish (~$100) remote play units so you could play on multiple TVs without having to move the PS4 around, similar to how DirecTV Genie + RVUs and whole home DVR works. Even better if the PS3 could work as one.
 
It's nice we got the 18CU figure and not 14+4.

1.8TFlops vs 1.2TFlops
8GB GDDR5 vs 8GB DDR3 + ESRAM
Used games vs ???
Not always online vs ???

Do we know for sure this is out? Even with the 14+4 set up the developer had the choice to use 4CU's for compute or rendering. Everything else was right on the leaks. I don't see them making a drastic change to the GPU architecture this late in the game...its not as simple as doubling the memory.

btw 14+4 was a good thing. it meant certain CU would be better at compute then others. I see it as a negative if this was taken out.
 
Where does it negate that rumour? Also, those four CUs can still be used for rendering.

no the official spec sheet clearly says a dev could choose to use all the cu for compute if they wish, there's no seperate,upgraded, degraded CU, they are all the same and the dev can choose how to divide the task.

edit: it's even quoted in the op

The GPU contains a unified array of 18 compute units, which collectively generate 1.84 Teraflops of processing power that can freely be applied to graphics, simulation tasks, or some mixture of the two.
8GB GDDR5 176GB/s unified
 

artist

Banned
Do we know for sure this is out? Even with the 14+4 set up the developer had the choice to use 4CU's for compute or rendering. Everything else was right on the leaks. I don't see them making a drastic change to the GPU architecture this late in the game...its not as simple as doubling the memory.

btw 14+4 was a good thing. it meant certain CU would be better at compute then others. I see it as a negative if this was taken out.
14+4 was sort of split arch. Even if the 4 units for compute were available for use in rendering, the approach would have been different rather than relying on the hardware scheduler in Pitcairn, basically that approach implied more (coding) work to extract power.

I always saw 14+4 as a negative over 18 cohesive CUs.
Where does it negate that rumour? Also, those four CUs can still be used for rendering.
The previous rumor was that the 4 CUs were separate for primarily for compute. I dont see any mentions of that in the documentation we've got so far.
 
Where does it negate that rumour? Also, those four CUs can still be used for rendering.

this.

no the official spec sheet clearly says a dev could choose to use all the cu for compute if they wish, there's no seperate,upgraded, degraded CU, they are all the same and the dev can choose how to divide the task.

no your wrong. The original leak(which was right on every single thing down to the decimal point, besides the RAM) said that the 4CU's would be superior for compute and the same as the others for rendering(but it was speculated that it may take more manual programming work to do so, verdict was still out on this). Each of the 4CU's had an additional ALU.
 

THE:MILKMAN

Member
this.



no your wrong. The original said that the 4CU's would be superior for compute and the same as the others for rendering(but it was speculated that it may take more manual programming work to do so, verdict was still out on this). Each of the 4CU's had an additional ALU.

It is weird though because VGleaks info is very recent (1.5 month old documents they claim) Could this be another late change in specs?
 

Izick

Member
Is anyone else kind of concerned (maybe not concerned, but something along those lines) about the fact that the touchpad is clickable? I mean if you accidentally press down a bit to hard when you use it on or a game or something I mean.
 
I always saw 14+4 as a negative over 18 cohesive CUs.

The previous rumor was that the 4 CUs were separate for primarily for compute. I dont see any mentions of that in the documentation we've got so far.

No, it was a positive. It kept the flexability and let the developers choose what they wanted to do, but made certain CU's more powerful for compute. Devs are going to use a portion of the CU's for gpgpu tasks anyways, why not have 4 of them on there that are superior at doing this than the normal CU's?

If the dev only need 2 CU's for compute, then they can do that, leaving 16 for rendering. It made the whole architecture more efficient and powerful, while still leaving the decision on how to use it all in the developers hands(this basically what the press release says).

It was definitely a positive IMO.

It is weird though because VGleaks info is very recent (1.5 month old documents they claim) Could this be another late change in specs?

No I don't think so. Changing the architecture and changing the manufacturing plan(is this what it would be called?) with AMD is much more complicated then doubling the RAM. I don't see it as possible, the 1.5 month old document claim further supports this. Unless somehow this 14+4 thing is a simple flip of the switch change, and it mainly done in software. But that's not the case because they said there was hardware differences. Each of the 4CU's had an additional ALU. I envisioned this set up would have an impact of how everything is set up on the DIE. I may be wrong, and maybe someone with more engineering knowledge can let us know. I really don't know how the CU's are set up on the die inside a GPU.

My take is this. Purely educated speculation based on the info and facts we have(the leaks, which have been proven very credible).

14+4 is still in. Technically there is 18CU's for whatever the developers want. The spec document therefor doesn't not lie or refute the fact there is the 14+4 set up. They just didn't mention it because it too complicated and is unneeded knowledge for the press. Keep in mind this is an official press release. NOT a dev document, which is what the other info came from.

Is anyone else kind of concerned (maybe not concerned, but something along those lines) about the fact that the touchpad is clickable? I mean if you accidentally press down a bit to hard when you use it on or a game or something I mean.

the same theory would apply to buttons or anything else you click.
 

artist

Banned
no your wrong. The original leak(which was right on every single thing down to the decimal point, besides the RAM) said that the 4CU's would be superior for compute and the same as the others for rendering(but it was speculated that it may take more manual programming work to do so, verdict was still out on this). Each of the 4CU's had an additional ALU.
"Minimal boost if used for rendering" - from VGLeaks.

Like I said, 14+4 is yet to be confirmed. The spec sheet we got so far makes no distinction as such.
No, it was a positive. It kept the flexability and let the developers choose what they wanted to do, but made certain CU's more powerful for compute. Devs are going to use a portion of the CU's for gpgpu tasks anyways, why not have 4 of them on there that are superior at doing this than the normal CU's?

If the dev only need 2 CU's for compute, then they can do that, leaving 16 for rendering. It made the whole architecture more efficient and powerful, while still leaving the decision on how to use it all in the developers hands(this basically what the press release says).

It was definitely a positive IMO.
CUs are already very good at compute. A fixed 14+4 is not as flexible as 16+2 or 14+4 or 12+6 scenario.

And this is one CU;
gcn-cuth01qyz.png


Imagine the 4 CUs with one extra ALU .. thats a whole 1.5% extra.

IMO, I'm not a fan of the rumored 14+4 approach. I'd like to see a pure 18 CU design only.
 
"Minimal boost if used for rendering" - from VGLeaks.
That word is up to interpretation. It does not discredit the fact there just as good as the other CU's when it comes to rendering. IMO all it does imply that there much superior when comes to compute.

Like I said, 14+4 is yet to be confirmed. The spec sheet we got so far makes no distinction as such.
Like I said before. The VGleaks info has been accurate to a T when it comes to the specs, except for the RAM info. Which it seems was a last minute decision. Also VGleaks reports that the document wasn't even two months old. "No distinction" does not mean it doesn't exist. Like I said before, this is a press release, not a dev document. The VGleaks info was from a dev doc. This type of information would of been unneeded and too complicated.

CUs are already very good at compute. A fixed 14+4 is not as flexible as 16+2 or 14+4 or 12+6 scenario.
What you mean fixed? The idea is they can be still used for both rendering and compute. with the 14+4 set up they could still use 16 CU's for rendering, or 17, or whatever they want.

IMO we can use the new spec press release to support the fact that the 14+4 set up is very flexible.
Imagine the 4 CUs with one extra ALU .. thats a whole 1.5% extra.
Good point. It doesn't seem like one extra ALU would do much. There may have been other information left out, as to why it would of been so much more superior when it comes to compute though.
 

artist

Banned
That word is up to interpretation. It does not discredit the fact there just as good as the other CU's when it comes to rendering. IMO all it does imply that there much superior when comes to compute.
Well it clearly says minimal boost for rendering ..

Like I said before. The VGleaks info has been accurate to a T when it comes to the specs, except for the RAM info. Which it seems was a last minute decision. Also VGleaks reports that the document wasn't even two months old. "No distinction" does not mean it doesn't exist. Like I said before, this is a press release, not a dev document. The VGleaks info was from a dev doc. This type of information would of been unneeded and too complicated.
I'm a big fan of VGLeaks and have defended them quite a bit here, so yeah I dont doubt their info and give them full due credit. But I agree the possibility of it still being a 14+4 approach and that we've not been given the full details in this press release.

What you mean fixed? The idea is they can be still used for both rendering and compute. with the 14+4 set up they could still use 16 CU's for rendering, or 17, or whatever they want.
14 normal CUs plus 2 "minimal boost for rendering" CUs with 2 CUs reserved for compute versus 16 normal CUs plus 2 CUs slightly inferior for compute.

IMO we can use the new spec press release to support the fact that the 14+4 set up is very flexible.
Lack of confirmation is holding me from believing it to be a 14+4 setup.
 

Minamu

Member
Can stress really affect the color of your hair? Some googling suggests it's only a myth?
Yes, I've recently found a disturbing amount of silver white hair strands on my head and I'm only 27 :(
 
Lack of confirmation is holding me from believing it to be a 14+4 setup.

http://www.scei.co.jp/corporate/release/pdf/130221a_e.pdf

I can't believe I read over this and didn't notice this in the original spec sheet.

The Graphics Processing Unit (GPU) has been enhanced in a number of ways,
principally to allow for easier use of the GPU for general purpose computing (GPGPU) such as
physics simulation.
The GPU contains a unified array of 18 compute units, which collectively
generate 1.84 Teraflops of processing power that can freely be applied to graphics, simulation
tasks, or some mixture of the two

what could that possibly mean, besides the inclusion of the 4CU's for compute??? I can't believe everyone read over this, including DF. There saying in very simple way, in laymen terms, that the gpu is enhanced for compute. If you look at my post up above I said that the whole 14+4 set up was to complicated to talk about in a press release, BUT THEY DID. Just put it very simply.

Does this deserve a new thread?
 
http://www.scei.co.jp/corporate/release/pdf/130221a_e.pdf

I can't believe I read over this and didn't notice this in the original spec sheet.



what could that possibly mean, besides the inclusion of the 4CU's for compute??? I can't believe everyone read over this, including DF. This saying in very simple way, in laymen terms, that the gpu is enhanced for compute. If you look at my post up above I said that the whole 14+4 set up was to complicated to talk about in a press release, BUT THEY DID. Just put it very simply.

Does this deserve a new thread?

it means they can be all used for computing task or whatever hybrid of the two that the dev chooses

14+4,12+6,10+8,7+11 etc.
 
it means they can be all used for computing task or whatever hybrid of the two that the dev chooses

14+4,12+6,10+8,7+11 etc.

no it specifically says "enhanced for compute." This means the gpu has been modified for compute.

There is to much coincidence between this official document and the VGleaks article that originally told us about the 4CU's for compute. Vgleaks spot on everything down to the decimial point(besides the ram).

Lets also not forget that it was DF that originally told us about a "compute module" in the PS4.

So it obvious that something is different about this GPU when it comes to compute. So therefor we have to go by what VGleaks said, which is that 4CU's are "separate" from the other 14 in some way and are enhance for compute.

What the press release clarifies is that its really flexible and the 4CU's definitely can be used for rendering, and its up to the devs to do what they wish.
 
no it specifically says "enhanced for compute." This means the gpu has been modified for compute.

There is to much coincidence between this official document and the VGleaks article that originally told us about the 4CU's for compute. Vgleaks spot on everything down to the decimial point(besides the ram).

Lets also not forget that it was DF that originally told us about a "compute module" in the PS4.

So it obvious that something is different about this GPU when it comes to compute. So therefor we have to go by what VGleaks said, which is that 4CU's are "separate" from the other 14 in some way and are enhance for compute.

What the press release clarifies is that its really flexible and the 4CU's definitely can be used for rendering, and its up to the devs to do what they wish.

eh?

The Graphics Processing Unit (GPU) has been enhanced in a number of ways,
principally to allow for easier use of the GPU for general purpose computing (GPGPU) such as
physics simulation. The GPU contains a unified array of 18 compute units, which collectively
generate 1.84 Teraflops of processing power that can freely be applied to graphics, simulation
tasks, or some mixture of the two

unified array(as in same thing) whose processing power can be used as a mixture of compute and render... I don't see what's not clear about it.
 

i-Lo

Member
no it specifically says "enhanced for compute." This means the gpu has been modified for compute.

There is to much coincidence between this official document and the VGleaks article that originally told us about the 4CU's for compute. Vgleaks spot on everything down to the decimial point(besides the ram).

Lets also not forget that it was DF that originally told us about a "compute module" in the PS4.

So it obvious that something is different about this GPU when it comes to compute. So therefor we have to go by what VGleaks said, which is that 4CU's are "separate" from the other 14 in some way and are enhance for compute.

What the press release clarifies is that its really flexible and the 4CU's definitely can be used for rendering, and its up to the devs to do what they wish.

I'm confused. I thought that ALL 18 CUs are individually customized enough to handle the tasks which were relegated to 4 CUs, in the earlier rumour...
 
I'm confused. I thought that ALL 18 CUs are individually customized enough to handle the tasks which were relegated to 4 CUs, in the earlier rumour...

that's exactly how i undertstand this line

The Graphics Processing Unit (GPU) has been enhanced in a number of ways,
principally to allow for easier use of the GPU for general purpose computing (GPGPU) such as
physics simulation. The GPU contains a unified array of 18 compute units, which collectively
generate 1.84 Teraflops of processing power that can freely be applied to graphics, simulation
tasks, or some mixture of the two
 
eh?



unified array(as in same thing) whose processing power can be used as a mixture of compute and render... I don't see what's not clear about it.

The Graphics Processing Unit (GPU) has been enhanced in a number of ways, principally to allow for easier use of the GPU for general purpose computing (GPGPU) such as physics simulation. The GPU contains a unified array of 18 compute units, which collectively generate 1.84 Teraflops of processing power that can freely be applied to graphics, simulation tasks, or some mixture of the two

So?? They flat out tell you this gpu was modified for compute.

Saying it has a unified array 18 compute unit which can be applied to rendering or compute doesn't change the fact that 4 of the CU's are enhanced for compute.

What this changes maybe is there not separate CU's. That the 14+4 means that this 4 are just enhanced in someway, but maybe not separate. A better way to write it would be, 18CU's(4 enhanced for compute).

I'm confused. I thought that ALL 18 CUs are individually customized enough to handle the tasks which were relegated to 4 CUs, in the earlier rumour...

No I don't think so. The official spec sheet specifically says that the GPU is enhanced/modified for compute. Combining this info with the previous VGLeaks info, one could come to the conclusion that 4 of the CU's are in fact different. This new info suggest there not seperate from the other 14 though. Hence why I think a better way to illustrate it would be "18CU's(4 enhanced for compute)" Not 14+4. It also confirms devs can do whatever they want with it.
 

i-Lo

Member
So?? They flat out tell you this gpu was modified for compute.

Saying it has a unified array 18 compute unit which can be applied to rendering or compute doesn't change the fact that 4 of the CU's are enhanced for compute.

What this changes maybe is there not separate CU's. That the 14+4 means that this 4 are just enhanced in someway. a better way to write it would be, 18CU's(4 enhanced for compute).

All we are saying is that sounds like it's not 4, rather all 18CUs that are individually programmable for GPGPU purposes.
 
But that's not what the spec sheet says, I understand what you are getting at but i think you are over-thinking this, to me it just says all the CUs have been enhanced for GPGPU task and the developer can choose to dedicate any of them to a different task

All we are saying is that sounds like it's not 4, rather all 18CUs that are individually programmable for GPGPU purposes.

I understand. Based of the VGLeaks document and the original DF article, I dont think so. VGleaks was pretty much exact on everything else. Theres just to much coincidence in this official spec sheet and the vgleak document.

1. We have official document that says the main enhancement/modification done to the GPU is for compute(what could that be? what could of they done?? Answer: see vgleaks article).

2. We had leaked dev document from VGleaks that said 4 of the CU's were enhanced for compute.

3. We had DF article write up saying that Orbis has "Special Sauce" and it was a compute module.

I just don't see how we can refute the VGLeaks article now based on this new information, in fact IMO it further supports it.

edit: Basically I would say you guys are right, IF we hadn't of had the VGLeaks's information. We simply would only know that gpu had been enhanced for compute in someway. VGleaks dev doc tells how they went about it.
 
So?? They flat out tell you this gpu was modified for compute.


No I don't think so. The official spec sheet specifically says that the GPU is enhanced/modified for compute. Combining this info with the previous VGLeaks info, one could come to the conclusion that 4 of the CU's are in fact different. This new info suggest there not seperate from the other 14 though. Hence why I think a better way to illustrate it would be "18CU's(4 enhanced for compute)" Not 14+4. It also confirms devs can do whatever they want with it.

But that's not what the spec sheet says, I understand what you are getting at but i think you are over-thinking this, to me it just says all the CUs have been enhanced for GPGPU task and the developer can choose to dedicate any of them to a different task plus I wouldn't take the leaks for confirmation as like we saw last night the details may have changed since the vgleaks docs
 
But that's not what the spec sheet says, I understand what you are getting at but i think you are over-thinking this, to me it just says all the CUs have been enhanced for GPGPU task and the developer can choose to dedicate any of them to a different task plus I wouldn't take the leaks for confirmation as like we saw last night the details may have changed since the vgleaks docs

If I had to venture a guess I'd think the 14+4 idea was probably an example scenario from the original documentation suggesting a mix that would produce a "good balance", and that has just been misconstrued through several layers of misunderstanding and poor wording to seem like a physical difference. GCN itself is an architecture specifically enhanced for Compute. Sony's PR could just mean that. HSA technologies like a fully unified memory space has specific advantages for compute. We have reason to believe both of those are in the PS design, so there's no good reason to go looking for a specially modified subset with added execution hardware (which itself should show up in the officially calculated GFlop number, but doesn't) when we could point to those as our answers.
 
If I had to venture a guess I'd think the 14+4 idea was probably an example scenario from the original documentation suggesting a mix that would produce a "good balance", and that has just been misconstrued through several layers of misunderstanding and poor wording to seem like a physical difference. GCN itself is an architecture specifically enhanced for Compute. Sony's PR could just mean that. HSA technologies like a fully unified memory space has specific advantages for compute. We have reason to believe both of those are in the PS design, so there's no good reason to go looking for a specially modified subset with added execution hardware (which itself should show up in the officially calculated GFlop number, but doesn't) when we could point to those as our answers.

Except that VGleaks just pretty much copy and pasted directly from the dev doc. Everything they said was spot on(once again except for the memory.)

They specifically says its a highly enhanced GPU, principally for compute. In the leak it talks about extra ALU's. Now if I'm not mistaken ALU's don't affect the flop number, only the total number of CU's do. The way flops are measured, leave out other things that could not be "accounted" for in the number. So anotherwards just because it doesn't change the flops, doesn't mean that it isn't customized or enhanced.

IMO Sony and AMD have customized and enhanced 4 of the CU's in the GPU(I dont think they did it to all 18 based on the leaks) to make the whole system much more efficient for compute tasks.
 

KidBeta

Junior Member
Except that VGleaks just pretty much copy and pasted directly from the dev doc. Everything they said was spot on(once again except for the memory.)

They specifically says its a highly enhanced GPU, principally for compute. In the leak it talks about extra ALU's. Now if I'm not mistaken ALU's don't affect the flop number, only the total number of CU's do. The way flops are measured, leave out other things that could not be "accounted" for in the number. So anotherwards just because it doesn't change the flops, doesn't mean that it isn't customized or enhanced.

IMO Sony and AMD have customized and enhanced 4 of the CU's in the GPU(I dont think they did it to all 18 based on the leaks) to make the whole system much more efficient for compute tasks.

ALU's is exactly how FLOPS are calculated whether or not they are enhanced in a way that does not increase them or not would probably decide FLOPS.

You can enhance a ALU (ways to enhance GPU ALU's do not come to mind) without affecting the FLOPS though.
 
ALU's is exactly how FLOPS are calculated whether or not they are enhanced in a way that does not increase them or not would probably decide FLOPS.

You can enhance a ALU (ways to enhance GPU ALU's do not come to mind) without affecting the FLOPS though.

I thought it was multiplied by how many CU's you have?(which inicidentely are supposed to contain a set amount of ALU's)
 

coldfoot

Banned
Guh? Isn't that a weird way to describe a GPU? Don't they normally mention the clockspeed?
You can deduce the clock speed from those figures by looking at existing AMD 78xx GPU's:
HD7850 is 16CU@860Mhz, and is advertised as a 1.76 TFLOP card.
Now multiply 1.76 by 18/16 and you get 1.98 TFLOP if the PS4 GPU ran at 860Mhz.
But we know the PS4 GPU is 1.84 TFLOP from Sony, so solve for x in the equality x/860 = 1.84/1.98 and you get x=799.2 Mhz, which means 800 MHz with rounding errors taken into account.
 
I read all about these specs,but one thing is bugging me.
What is better for the graphics 4gb gddr5 and a graphics card with 3TFLOPS,or how it is with 8gb gddr5 and 1.84TFLOPS?
I am very curios about this,so if anyone who knows about something like this can please answer me.
 
I was watching the PS meeting over again and one of the developers in the video talking about the hardware said there's no start up? That would be crazy if true. Consoles have had start up's since they first came out.
 

artist

Banned
http://www.scei.co.jp/corporate/release/pdf/130221a_e.pdf

I can't believe I read over this and didn't notice this in the original spec sheet.



what could that possibly mean, besides the inclusion of the 4CU's for compute??? I can't believe everyone read over this, including DF. There saying in very simple way, in laymen terms, that the gpu is enhanced for compute. If you look at my post up above I said that the whole 14+4 set up was to complicated to talk about in a press release, BUT THEY DID. Just put it very simply.

Does this deserve a new thread?
I think you are reading way too much into it.

It probably is referring to the fact that AMD's (GCN architecture) GPU is enhanced for compute.

I read all about these specs,but one thing is bugging me.
What is better for the graphics 4gb gddr5 and a graphics card with 3TFLOPS,or how it is with 8gb gddr5 and 1.84TFLOPS?
I am very curios about this,so if anyone who knows about something like this can please answer me.
Option #2 (3TFlops with 4GB GDDR5) will always win. Think of a 7950 with 3GB GDDR5 versus 7850 with 4GB GDDR5.

It'll also come at much higher cost, I'm not even sure if an APU of that monstrosity can even be manufactured before 20nm hits.
 
I think you are reading way too much into it.

It probably is referring to the fact that AMD's (GCN architecture) GPU is enhanced for compute.


Option #2 (3TFlops with 4GB GDDR5) will always win. Think of a 7950 with 3GB GDDR5 versus 7850 with 4GB GDDR5.

It'll also come at much higher cost, I'm not even sure if an APU of that monstrosity can even be manufactured before 20nm hits.

I don't understand how you can think that Sony/amd hasn't customized and enhanced this thing beyond the normal GCN architecture for compute. Listen to the way Mark Cerny talks about the GPU in the conference again. He says its something specific that they've done to the GPU, above and beyond the normal gpu architecture its based on. This blurb reads the same way as well, and back up what Cerny said.

I think its obvious that Sony has customized this GPU for compute based on what Cerny and this spec sheet says. The questions is: How? What did they do?

For that answer I think the first and obvious place to look is the VGLeaks leaked dev doc, that was pretty much a copy and past.
 

DBT85

Member
Has anyone asked if we will be able to swap hard drives ?

from GiantBombs interview with Yoshida. There were plenty of other "not ready to talk yet" replies.

http://www.giantbomb.com/articles/q-a-sony-computer-entertainment-s-shuhei-yoshida/1100-4584/

Hard Drive
We know the PS4 will have a local hard drive, but how upgradeable/interchangeable that hard drive might be is still a question mark. Yoshida's response? "We're not ready to talk about the exchangability of the hard drives, but it's our dream that people are filling up their hard drives on PS4. Because on PS3, not many people did."
 
Top Bottom