• 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.

Recent PS4 SDK update unlocked 7th CPU core for gaming

butman

Member
The secret sauce behind PS2 emulation, no doubt.

well.gif
 

RoboPlato

I'd be in the dick
Hopefully third parties get the whole core relatively soon.

I still want to know if the RAM reserve has opened up at all yet.
 

onQ123

Member
I'm afraid you're confusing Asynchronous Compute with running a full-blown OS.

A modern x86-64 processor has tons of features that a GPU doesn't have and it supports over 1000 different instructions. A GPU/SPU is specialized at certain tasks (mainly linear algebra/matrix multiplication). Its feature/instruction set is limited compared to a CPU and that's why they're able to process many Teraflops. It's a different design philosophy, depending on what you want to do with a limited transistor budget. A CPU is a jack of all trades and master at none, while the GPU/SPU is a specialized, streaming processor.


"Norden also revealed an extended Direct X 11.1 featureset, a new asynchronous compute architecture, "greatly expanded shaders", and features "beyond" Direct X 11, OpenGL 4.0, and OpenCL. Sony's Compute allows developers to run arbitrary code on the GPU instead of the CPU, which benefits fluids, particles and post processing. Compute has access to the system's full set of "very expensive, very exotic" 8GB 256-bit GDDR5 unified memory.


Shaders have nothing to do with traditional x86/CPU code.

Shaders are made to process graphics, physics and parallelizable stuff in general.


Look above you see where it says that it is beyond OpenCL?



Well, apart from the fact that it's just not very good at it, potentially, missing circuitry. I don't know enough to say for sure, but I wouldn't be at all surprised if the GPU simply lacked the transistors required to perform vital operations, because why would you waste silicon on functionality nobody will ever call? So maybe it can run the entire OS and maybe it can't, but what difference does it really make if it's a terrible idea to begin with?

That's what I was trying to get across in my hQ explanation, actually. It's dumb to force your decision-maker to waste all of its time crunching numbers, but it's equally stupid to make your number-cruncher try to make decisions. That's the advantage of GCN. Everything is designed around the idea of letting the appropriate chip handle the appropriate function, and letting the to chips work as independently from one another as possible. Running the entire OS on the GPU is just as dumb as running the entire OS on the CPU.

Having information like visibility magically update for the CPU is a huge fucking win, so we should be talking about stuff like that, or what devs are gonna run on this extra CPU core, instead of bickering about moot points like running the OS entirely on the GPU. Yes? :)




Fight!!


I'm not the one who brought GPU's running a OS I was just pointing out that it could.

Do you know what arbitrary mean? the GPU can run any code that's sent to it it's not restricted to any special instructions.
 

omonimo

Banned
I have the suspect you don't have understand at all what the quote talking about. You can't do all the CPU stuff with GPGPU. It talks of particle, fluid basically the physics stuff.
 
@onQ123

Apparently you still don't understand what parallelizable/streaming/SIMD means.

This discussion leads nowhere. Study some linear algebra first and then you'll understand why "fluids, particles and post processing" are parallelizable tasks, while others aren't.
 
Hopefully third parties get the whole core relatively soon.

I still want to know if the RAM reserve has opened up at all yet.

I imagine it will if it already has not. 3GB reserved seems kind of excesive. I am currently using only 2,4 on Win 10 with multiple tabs open and a full OS at my finger tips.
 

desu

Member
I'm not sure how feasible it is but, in light of this, I hope some devs patch their previously released titles to help even out framerates.

Will not happen, that means they would have to run through certification again and this will costs tons of money for no benefit (at least for the developer/publisher).
 
Where in the hell does this line of reasoning or idea come from?

Why would anyone want to use precious GPU resources to handle OS functions? Even if one could.
 

omonimo

Banned
I know the difference & that's only telling you what they are good at but there is nothing stopping the GPU in the PS4 from being able to run a OS.
I don't understand why you persist to substain something of so absurd. That's not what your quote talking about. Seriously.
Where in the hell does this line of reasoning or idea come from?

Why would anyone want to use precious GPU resources to handle OS functions? Even if one could.
Probably he believes to obtain more CPU resources with less sacrifices, using the 'more powerful' gpu. I imagine.
 
Where in the hell does this line of reasoning or idea come from?

Why would anyone want to use precious GPU resources to handle OS functions? Even if one could.



Because for some people, GPU computing is some magic that will solve all the problems, regardless of the fact that you still waste GPU ressources. Sure it's a great idea when your GPU isn't 100% sollicited and your CPU needs help... the thing is, in 99% cases, the PS4 GPU is already maxed out.
 
not happy about sony keeping the seventh core full power exclusive to ps4. come on sony let third part'ys go super saiyan as well.
 
I imagine it will if it already has not. 3GB reserved seems kind of excesive.
I note that using the media player doesn't kill the game, only suspends it, so that 3GB includes enough memory to run non-game apps. I also note that the bone reserves a similar amount, and it has a similar, OS + App + Game model. Note that I am not noting this in a system wars manner, just that it's a comparable device doing comparable things.
 

ElTorro

I wanted to dominate the living room. Then I took an ESRAM in the knee.
https://s-media-cache-ak0.pinimg.com/originals/96/d2/f7/96d2f7f1aa8250b3a1925521d3ae70fa.gif [img][/QUOTE]

Modern GPUs and GPGPU programming languages are turning complete. So, in principle, you could run a very simplistic "OS" on a GPU, ignoring hardware-specific issues like the lack of a ring protection model and (I assume) missing features in the areas of memory and interrupt management.

But it would be fucking slow and useless.
 

RoboPlato

I'd be in the dick
I imagine it will if it already has not. 3GB reserved seems kind of excesive. I am currently using only 2,4 on Win 10 with multiple tabs open and a full OS at my finger tips.
I've been complaining about these massive reserves since we heard about them. 3gb is absurd, especially with the featureset of the PS4's.

not happy about sony keeping the seventh core full power exclusive to ps4. come on sony let third part'ys go super saiyan as well.
I'm sure it will before long. First parties are likely "beta testing" the full core update.
 

onQ123

Member
I think the 7th core being free is only part of the story & what's being beta tested by 1st party devs is a API overhaul to take advantage of compute & it's HSA.

Mark Cerny to Edge

"Well, as a programmer myself, I wanted to be sure that there were lots of fun little features in the hardware that people could really explore in the later years of the console lifecycle. So when it came down to designing the feature set for years three and four of the console, that isn’t something where the development community could contribute as broadly because it had much more to do with personal vision about where GPGPU was going. I definitely did pitch it to a broad variety of teams and get their feedback, but there were many, many small and large decisions that had to be made in terms of actually creating the details of that hardware."


" There are definitely some features in the PlayStation 4 that start to get used broadly in the third or fourth year of the console lifecycle. It’s all about how the GPU and CPU can work together to do many tasks other than graphics, which is say that photorealism is a great target, but that world simulation is also important. This is underappreciated, but getting your audio right in a game and making sure that your character’s ears are really hearing what they should within the game universe takes a tremendous amount of processing power. And there’s a lot of features in the GPU to support asynchronous fine-grain computing. ‘Asynchronous’ is just saying it’s not directly related to graphics, ‘fine grain’ is just saying it’s a whole bunch of these running simultaneously on the GPU. So I think we’re going to see the benefits of that architecture around 2016 or so."


“With the PlayStation 4, it’s even such things as the share cores have a beautiful instruction set and can’t be programmed in assembly. If you were willing to invest the time to do that, you could do some incredibly efficient processing on the GPU for graphics or for anything else. But the timeframe for that kind of work would not be now. I don’t even think it would be three years from now.”

http://www.playstationlifestyle.net...und-2016-or-so-is-when-well-see-the-benefits/
 

onQ123

Member
But it's not.



Yes it is


"Sony's Compute allows developers to run arbitrary code on the GPU instead of the CPU"


“With the PlayStation 4, it’s even such things as the share cores have a beautiful instruction set and can’t be programmed in assembly. If you were willing to invest the time to do that, you could do some incredibly efficient processing on the GPU for graphics or for anything else. But the timeframe for that kind of work would not be now. I don’t even think it would be three years from now.”


AikcvIp.png


3mzcZhA.png
 

c0de

Member
You need a receipt to know that a general purpose processor can do general purpose processing?

Just to make it clear:
Every computing device that features at least
{add,sub}
combined with the boolean operators
{and,or,not}
can run arbitrary code, yes. Or perhaps let's say you can depict every boolean function with these. But OS is not only software but also relies on a lot of things like access to memory areas reported by BIOS, a dedicated boot loader, processor rings are also inherently needed and also OS's are not designed that way to boot off something different than a dedicated main processor.
For science it could be an interesting project but we won't see it on PS4, ever.
 

Blanquito

Member
As interesting as this is (and I hate to be that person), but can you make a new thread for this random topic or something? When I see that this thread is updated I kind of hope that it's talking about the 7th core and not about running an OS on the GPU or whatever.

:)
 

onQ123

Member
Just to make it clear:
Every computing device that features at least
{add,sub}
combined with the boolean operators
{and,or,not}
can run arbitrary code, yes. Or perhaps let's say you can depict every boolean function with these. But OS is not only software but also relies on a lot of things like access to memory areas reported by BIOS, a dedicated boot loader, processor rings are also inherently needed and also OS's are not designed that way to boot off something different than a dedicated main processor.
For science it could be an interesting project but we won't see it on PS4, ever.

I never said it would happen on PS4 or should happen on PS4, someone said it was impossible & I pointed out that it could.

As interesting as this is (and I hate to be that person), but can you make a new thread for this random topic or something? When I see that this thread is updated I kind of hope that it's talking about the 7th core and not about running an OS on the GPU or whatever.

:)


Sorry about that I'm done.
 

c0de

Member
I never said it would happen on PS4 or should happen on PS4, someone said it was impossible & I pointed out that it could.

Last one from me: but that doesn't have to do anything with PS4 or its GPU and all the stuff that has been posted in this thread. At all. It being impossible also depends on how you look at it. This can be right and wrong at the same time.
 

biteren

Member
EDIT: wrong thread.

a little on the subject, its neat they let devs have access, im hoping for more stable games
 
Yes it is

"Sony's Compute allows developers to run arbitrary code on the GPU instead of the CPU"

“With the PlayStation 4, it’s even such things as the share cores have a beautiful instruction set and can’t be programmed in assembly. If you were willing to invest the time to do that, you could do some incredibly efficient processing on the GPU for graphics or for anything else. But the timeframe for that kind of work would not be now. I don’t even think it would be three years from now.”

Yes, arbitrary code... assuming the instructions in the code CAN be processed by the GPU. GPUs aren't equipped to deal with running an ENTIRE OPERATING SYSTEM on them alone. OSs have a bunch of branching involved, and GPUs aren't made for that.
 

onQ123

Member
Yes, arbitrary code... assuming the instructions in the code CAN be processed by the GPU. GPUs aren't equipped to deal with running an ENTIRE OPERATING SYSTEM on them alone. OSs have a bunch of branching involved, and GPUs aren't made for that.



GPU

AMD R10x series GPU @ 800 MHz (Tahiti)
aggregate 10x RSX performance, 1.84 TFlops
DX”11.5″
cheap branching
18 compute units, focus on fine grained compute & tessellation


Read more at: http://www.vgleaks.com/world-exclusive-ps4-in-deep-first-specs

--------------------------------------------------

GPU+RAM+CPU = Beyond Fast!
- Plenty of power for a true Next Gen Game Experience
-  8 CPU cores
-  High polygon throughput
-  High pixel performance
- Efficient branching in
GPU Shaders
 

Locuza

Member
When you read "cheap branching" compare it to GPUs before GCN, don't think as a general rule or on the same level like CPUs.
 

LordOfChaos

Member
Lol, dude yes they were. Cell was a completely different beast, the PPE was used to send out as much task to the SPEs thereby increasing performance. There have been plenty of Naughty dawg and insomniac videos and articles about how they used Cell over the years. They were very proud of it, in any case the PS3 OS ran off the 7th SPE.

No need to be uppity; I think the implied question above was if the OS also used the PPE and had partial reservation, in addition to part of the OS running on an SPU.

If it /COULD/ run solely on an SPU is a different question, but it would be even slower and crappier than it already is. The SPUs were great at SIMD, to the expense of everything else, and that's not what an OS is sitting there doing all day. An OS with no L2 cache [it had local memory, but then you'd have to hand manage every bit of OS memory that should be there for performance] or branch prediction...Sounds nightmarish.
 
No need to be uppity; I think the implied question above was if the OS also used the PPE and had partial reservation, in addition to part of the OS running on an SPU.

If it /COULD/ run solely on an SPU is a different question, but it would be even slower and crappier than it already is. The SPUs were great at SIMD, to the expense of everything else, and that's not what an OS is sitting there doing all day. An OS with no L2 cache [it had local memory, but then you'd have to hand manage every bit of OS memory that should be there for performance] or branch prediction...Sounds nightmarish.
Exactly.

A dev said that it ran the OS and the game in separate soft threads (PPE 1.6 GHz + PPE 1.6 GHz*).

* "Well, as we understand it, the Xbox 360 CPU has two threads per core, and that threading is handled in a fairly primitive manner, with the processor effectively working as six 1.6GHz threads."

Source: http://www.eurogamer.net/articles/digitalfoundry-2015-vs-backwards-compatibility-on-xbox-one

Jaguar runs at 1.6 GHz, but we now have 8 cores and its microarchitecture is way more advanced (out-of-order execution/better branch prediction vs PPE in-order/poor branch prediction).

That's why I don't understand why people are so quick to dismiss Jaguar, while the Cell PPE is not that great either. The SPU part was its only saving grace, but as you said, that's limited to SIMD operations.

Maybe this could explain it: https://en.wikipedia.org/wiki/Megahertz_myth

I mean, if ND was able to run TLOU's NPC AI in PPE @ 1.6 GHz, then imagine what they can possibly do with 7 Jaguar cores @ 1.6 GHz.

TL;DR: Cell is not that great as people think it is (minus its SPU/flops grunt) and yet it served the PS3 for almost a decade. There's no reason to think that the PS4 won't be able to last until 2018-2020.
 

onQ123

Member
Exactly.

A dev said that it ran the OS and the game in separate soft threads (PPE 1.6 GHz + PPE 1.6 GHz*).

* "Well, as we understand it, the Xbox 360 CPU has two threads per core, and that threading is handled in a fairly primitive manner, with the processor effectively working as six 1.6GHz threads."

Source: http://www.eurogamer.net/articles/digitalfoundry-2015-vs-backwards-compatibility-on-xbox-one

Jaguar runs at 1.6 GHz, but we now have 8 cores and its microarchitecture is way more advanced (out-of-order execution/better branch prediction vs PPE in-order/poor branch prediction).

That's why I don't understand why people are so quick to dismiss Jaguar, while the Cell PPE is not that great either. The SPU part was its only saving grace, but as you said, that's limited to SIMD operations.

Maybe this could explain it: https://en.wikipedia.org/wiki/Megahertz_myth

I mean, if ND was able to run TLOU's NPC AI in PPE @ 1.6 GHz, then imagine what they can possibly do with 7 Jaguar cores @ 1.6 GHz.

TL;DR: Cell is not that great as people think it is (minus its SPU/flops grunt) and yet it served the PS3 for almost a decade. There's no reason to think that the PS4 won't be able to last until 2018-2020.

Sure 7 Jaguar cores is good enough in a console when a lot of work can be offloaded to the GPGPU but why would you minus the Cell's SPUs to say it wasn't so great?

449616PAlexisVaisseP11122.png




yUWQvs7.png


kVsbaQm.png


2690709-0579811166-i4AEa.jpg


http://gdcvault.com/play/1020939/Efficient-Usage-of-Compute-Shaders
 
Because a Pentium 4* from 2004 was able to beat it in most benchmarks (minus the SIMD/flops benchmarks, of course).

* Mind you, NetBurst wasn't exactly the most efficient x86 microarchitecture.

SIMD is not suitable for everything as you think it is. It's like having a super-smart kid who is really good at maths (linear algebra/matrix multiplication), but pretty much dumb at everything else (history, biology, literature etc.) That's what the Cell SPU/GPGPU is. It excels in one field, but it's useless in everything else.
 

Panajev2001a

GAF's Pleasant Genius
Because a Pentium 4* from 2004 was able to beat it in most benchmarks (minus the SIMD/flops benchmarks, of course).

* Mind you, NetBurst wasn't exactly the most efficient x86 microarchitecture.

SIMD is not suitable for everything as you think it is. It's like having a super-smart kid who is really good at maths (linear algebra/matrix multiplication), but pretty much dumb at everything else (history, biology, literature etc.) That's what the Cell SPU/GPGPU is. It excels in one field, but it's useless in everything else.

The problem with the setup was also that the PPU/PPX core had quite a few gotchas for developers (lack of efficient store to load forwarding leading to the much hated LHS stalls, micro coded instructions, etc...) that could in some cases actually allow an SPU to run the same general purpose code faster... not trivial to obtain especially on a wide variety of tasks, but interesting nonetheless. PS3 to PS4 means driving the frequency of the main general purpose core to half of what it was and losing the SPE's, but it means also going to 8 much more efficient and easier to code for general purpose cores (which do still pack capable AVX vector units).
 
Top Bottom