Support NeoGAF

[MORT1S]

I can understand xbone needing two cores for their 2 and a half OS setup but I'm perplexed as to why PS4 would need two as well.

Not sure. I think people are underestimating the OS requirements on PS4 for whatever reason.

sure, there is a ton of dedicated hardware to handle all of these little tasks... but what drives that? It really seems like that 14+4 thing was sort of a suggestion to possibly negate the two tied up cores, and not some hard lock.

This also jives with the Shadowfall slides that mentioned 6GB of usable RAM.

 

[Ryoku]

I finally understood where does that 196 GB/s value for the theoretical bandwidth peak comes from.

Thank you all for this informative thread!

Don't believe the 196GB/s value. It's wrong. Read the past page or so.

 

[pushBAK]

Gone are the days of complicated PS3 architecture.

This. This is why I'm excited for next-gen. Dev time will be reduced due to not needing to spend a ridiculous amount of time getting the engine running on the platform, leaving more time for the creative vision to be fleshed out and an overall better polished experience.

 

[Snubbers]

I thought the architecture was known for a long time now?

All seems as it was, Cerny's approach is very good indeed, and it's good to see devs not having many issues porting from PC..

This is one of those 'nothing to see here, move along' kind of situations!

 

[TheCloser]

Not sure. I think people are underestimating the OS requirements on PS4 for whatever reason.

sure, there is a ton of dedicated hardware to handle all of these little tasks... but what drives that? It really seems like that 14+4 thing was sort of a suggestion to possibly negate the two tied up cores, and not some hard lock.

This also jives with the Shadowfall slides that mentioned 6GB of usable RAM.

Firstly, the 14+4 has been debunked by Cerny. Secondly, the killzone slide only suggested that they were using 6 not that 6 was only available. People should really stop making assumptions.

 

[i-Lo]

I am assuming for back ground stuff, multi-tasking and social sharing. It does not matter any way multiplatform be around 6 cores anyways. Also Sony first party will be using compute so the extra core not going to be that big of a deal. The extra ACE units easily make up for this.

It is just surprising since I remember the rumour floating around about the OS may require only half of a single core.

Not sure. I think people are underestimating the OS requirements on PS4 for whatever reason.

sure, there is a ton of dedicated hardware to handle all of these little tasks... but what drives that? It really seems like that 14+4 thing was sort of a suggestion to possibly negate the two tied up cores, and not some hard lock.

This also jives with the Shadowfall slides that mentioned 6GB of usable RAM.

I do not recall that slide. Can you find it?

 

[RoboPlato]

I can understand xbone needing two cores for their 2 and a half OS setup but I'm perplexed as to why PS4 would need two as well.

I'm not sure either. I figured the 2 cores from the Killzone slides were just because it was an early version and would be improved later.

 

[Ryoku]

Not sure. I think people are underestimating the OS requirements on PS4 for whatever reason.

sure, there is a ton of dedicated hardware to handle all of these little tasks... but what drives that? It really seems like that 14+4 thing was sort of a suggestion to possibly negate the two tied up cores, and not some hard lock.

This also jives with the Shadowfall slides that mentioned 6GB of usable RAM.

The maximum amount of usable GDDR5 RAM in the PS4 is 7GB, I believe.

 

[Moral Panic]

I went ahead and made a "clearer" version of the diagram posted on the first page.

Hope it helps.

Fixed dem flops =D

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

 

[RoboPlato]

I do not recall that slide. Can you find it?

I don't remember that either. I'd like to see it. I only remember the slides saying they were using 4.5Gb of mixed RAM types.

 

[Guymelef]

Not sure. I think people are underestimating the OS requirements on PS4 for whatever reason.

sure, there is a ton of dedicated hardware to handle all of these little tasks... but what drives that? It really seems like that 14+4 thing was sort of a suggestion to possibly negate the two tied up cores, and not some hard lock.

This also jives with the Shadowfall slides that mentioned 6GB of usable RAM.

The Killzone tech presentation was with PS4@4GB.

 

[Eusis]

But I have heard it similar, it was pre-vita on Vita. I hope it should be better than this time.

I half expect that's why the Vita's even getting as much support as it is: at worst you can quickly port something and sell it DD only.

And I suspect more worked against the Vita: non-Nintendo handhelds usually struggling against them, the rise of mobile gaming, and just the thing costing $250 damn dollars. Remove at least ONE of the latter two factors and I imagine it'd be doing much more solidly.

 

[phosphor112]

I finally understood where does that 196 GB/s value for the theoretical bandwidth peak comes from.

Thank you all for this informative thread!

176 is the theoretical peak, not 196.

 

[qa_engineer]

hUMA. That is all.

 

[OldAsUrSock]

You know this wasn't anything special just confirming something that anyone with a little knowledge knew.

Sky is not purple

True, but now THE Cerny has declared it as such.

 

[Canis lupus]

The Crew was ported to PS4 in 6 months by 3 people?

 

[Locuza]

Secondly, the killzone slide only suggested that they were using 6 not that 6 was only available. People should really stop making assumptions.

Maybe i'm wrong, but i remember that GG said they using all avaible cores and the profiling tool showed 6 cores so at the moment it seems that 6 cores are the limit for developers.

 

[Perkel]

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

Onion bus is for the same memory that is used by both CPU and GPU.
There is no such a thing in PC.

 

[ChronotriggerJM]

Great thread, very clever use of hardware (be it Sony's extras or AMD's). From what I gather of it all, is that a developer could choose to be lazy, and just muscle it with the main bus (perhaps for some indy developers not planning on maxing out the console), or map out how they want to access the memory to significantly cut down on latency and make each component of the apu sing to it's capacity.

 

[benny_a]

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

No. But we already covered that in this thread. In short: hUMA.

 

[RoboPlato]

Maybe i'm wrong, but i remember that GG said they using all avaible cores and the profiling tool showed 6 cores so at the moment it seems that 6 cores are the limit for developers.

It said that there were only six cores available at that point in time, implying that there may be more later on. That said, if it hasn't changed by now I have my doubts that it'll improve.

 

[MORT1S]

The maximum amount of usable GDDR5 RAM in the PS4 is 7GB, I believe.

Do you have any confirmation? Shadowfall slides said otherwise.

It is just surprising since I remember the rumour floating around about the OS may require only half of a single core.



I do not recall that slide. Can you find it?

I will try to find it. There was a thread here, I believe.

Firstly, the 14+4 has been debunked by Cerny. Secondly, the killzone slide only suggested that they were using 6 not that 6 was only available. People should really stop making assumptions.

Relax...

I am only theorizing. Also, I only suggested that it was recommended to use the 14+4 for general use. Use all 18 if you can manage.

As far as the slide saying 6 was used, did it also not say 6 cores were in use?

 

[Nafai1123]

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

I believe the main difference between this and PC architecture is the fact that both the CPU and GPU can use the same data sets stored in the ram instead of copying from one allocation to the other. Should allow for much easier GPGPU function than a PC where the CPU is using system RAM and the GPU is using GPU RAM. CPU's don't require high bandwidth, which is why that Ubisoft dev said the 20GB bw is plenty.

 

[ChorusLindarr]

It is unified as it has a unified address space. What they are talking about is that you have to decide if your memory access commands are checked against the L1/L2 of the CPU caches (Onion) or not (Garlic). L1/L2 are essential for any CPU's management of latency, and latency-tolerant memory access from the GPU would corrupt them without needing them. Hence, the two pathways. As a developer you have to decide, if CPU or GPU access performance is most crucial for a given data set. In the former case, you would issue memory access commands via Onion, in the former via Garlic. If the GPU wants to read/write data relevant to the CPU for asynchronous compute, it can use Onion+ to use the CPUs L1/L2 caches and to, analogously, not corrupt its own internal caches.

It is not a contradiction to HSA but, on the contrary, a requirement.

ElTorro since you seem to be the only person in this entire thread that seems to be knowledgeable on the subject, might I ask you kindly help me clear something up in my mind.

The CPU 20GB/s bus to main memory, is this separate from the 176GB/s GPU peak read/write bandwidth on its own bus? Or is the figure actually 156GB/s GPU<-->GDDR5 and 20GB/s CPU<-->GDDR5?

From what I understand the GPU has full access to main ram at 176GB/s. And the CPU has its own separate bus to access the ram @ <20GB/s. If they are separate buses how can one be subtracted from the other?

 

[DieH@rd]

The Crew was ported to PS4 in 6 months by 3 people?

Only basic engine, just to get something shown on screen [game with 10fps preformance].

 

[Locuza]

The Killzone tech presentation was with PS4@4GB.

And was using 4,7 GB of Memory?

Page 6:
http://www.guerrilla-games.com/presentations/Valient_Killzone_Shadow_Fall_Demo_Postmortem.pdf

That said, if it hasn't changed by now I have my doubts that it'll improve.

/sign

 

[benny_a]

And was using 4,7 GB of Memory?

Page 6:
http://www.guerrilla-games.com/presentations/Valient_Killzone_Shadow_Fall_Demo_Postmortem.pdf

Yes, because it was running on a devkit that has more than 4GB of memory available.

I am only theorizing. Also, I only suggested that it was recommended to use the 14+4 for general use. Use all 18 if you can manage.

It isn't recommended to use 14+4. Cerny was specifically asked this question and denied it.

 

[mrklaw]

Really nice article, tons of details on PS4 arhitecture and SDK tools. Sony seems to be in very good position with PS4 deployment.


Article also confirmed two things - 2 CPU cores are dedicated to PS4 OS, and PSN download speed limit is [was] 12mbps.

Why would PSN download speed be limited? I that just when downloading in standby low-power mode? Or in the background to avoid hitting the HDD too much while playing a game? 12Mb/s is pretty bad if games are big. Hopefully all digital purchased games will have preload options.

 

[MORT1S]

I don't remember that either. I'd like to see it. I only remember the slides saying they were using 4.5Gb of mixed RAM types.

The Killzone tech presentation was with PS4@4GB.

Her is the thread, I believe.

http://m.neogaf.com/showthread.php?t=557670

I can't really check the videos on my cell, but if I am mis-remembering... You have my apologies

 

[Nafai1123]

Only basic engine, just to get something shown on screen [game with 10fps preformance].

That wasn't clarified. All they said is their first attempt was running at 10fps, but they've since resolved those problems.

At the Ubisoft E3 event, the PC version of The Crew was running at 30 frames per second, but the first working compilation of the PS4 codebase wasn't quite so hot, operating at around 10fps.

"We're getting near to the final state, we're not expecting huge performance changes, just finalisation of features. It's a lot more stable than it was early on. We haven't had to do any changes for a while."

Edit: Actually it was, basic porting took 6 months and it's running stable.

With the basic porting complete, the Ubisoft Reflections team is now ramping up its staff in order to complete the PS4 game ready for the Q1 2014 release, but the core engineering effort in moving The Crew across to PlayStation 4 was accomplished in six months with a team of just two to three people working on it. Overall, Reflections felt that the process of porting over the PC codebase was fairly simple and straightforward.

 

[RoboPlato]

And was using 4,7 GB of Memory?

Page 6:
http://www.guerrilla-games.com/presentations/Valient_Killzone_Shadow_Fall_Demo_Postmortem.pdf

It was a mix of DDR3 and GDDR5, which is why there was more than 4GB available.

 

[Ryoku]

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

On PC, the CPU pulls from its own, separate pool of memory (main system memory, or the usual DDR3). The GPU is left to its own memory (the VRAM) which it can utilize fully. So a GTX770 with GDDR5 RAM @224GB/s will have access to that full 224GB/s of bandwidth--theoretically speaking, of course.

The nice thing about the PS4 is the communication between the CPU and GPU. They're able to share the same address space. This helps with parallel processing between the CPU and GPU.

Wasn't there something about DX11.2 making it possible for the CPU and GPU to share the same address space or something?

 

[JJD]

This is kinda old news right?

And it's also the reason why I never understood all the latency problem when people talked about the PS4 and GDDR5.

 

[AgentP]

The cpu has to go through the gpu to acess ram. So its actually 176 minus 20 = 156 I belive

Leave to Major Nels.. eastman to spread FUD about the PS4,

The Onion and Garlic buses are for sharing data between the CPU and GPU and bypassing cache, they are not the main access for the unified memory.

 

[Beerman462]

This is kinda old news right?

And it's also the reason why I never understood all the latency problem when people talked about the PS4 and GDDR5.

The "latency problem" was a desperate attempt to find a flaw with PS4 architecture and somehow make the XB1 look better, because of super low latency.

 

[mrklaw]

Garlic and onion are good,but not as good as Paula Agnus and Denise

 

[phosphor112]

ElTorro since you seem to be the only person in this entire thread that seems to be knowledgeable on the subject, might I ask you kindly help me clear something up in my mind.

The CPU 20GB/s bus to main memory, is this separate from the 176GB/s GPU peak read/write bandwidth on its own bus? Or is the figure actually 156GB/s GPU<-->GDDR5 and 20GB/s CPU<-->GDDR5?

From what I understand the GPU has full access to main ram at 176GB/s. And the CPU has its own separate bus to access the ram @ <20GB/s. If they are separate buses how can one be subtracted from the other?

If I have 10 dollars, and I give you 1 dollar from my left hand, and I give my wife 9 dollars from the right hand, how many dollars are exchanged?

10.

Just because they are separate buses (in this case, hands) it doesn't mean I can somehow exceed the 176 (10 dollars) I have.

Keep in mind though, I can give all 10 dollars to my wife (all 176 for the GPU) but then then you won't get anything.

So there is a trade off. The more I give you, the less I give her. CPU has a hard cap though, at 20GB/s.

 

[LiquidMetal14]

Garlic and onion are good,but not as good as Paula Agnus and Denise

Now we're making strides!

 

[Locuza]

It was a mix of DDR3 and GDDR5, which is why there was more than 4GB available.

Any (official) evidence?

 

[Ryoku]

Do you have any confirmation? Shadowfall slides said otherwise.

There was a thread about it on GAF. I searched it up just now and it's still just a rumor, I think, that started off on PS4Daily. I don't know if that's a trusted source for news, but seeing as how WiiUDaily is not trustworthy, PS4Daily's rumors do seem questionable.

 

[RoboPlato]

Any (official) evidence?

Eurogamer, I believe, had a video of the presentation on lighting that GG did. The presenter said it.

 

[googleplex]

The cpu has to go through the gpu to acess ram. So its actually 176 minus 20 = 156 I belive

Oh god what? That's not true at all! Please stop it with the PS4 FUD.

 

[lyrick]

Other than the Onion buses, isn't this exactly like a PC? I thought there would be high bandwidth between memory and CPU, and we could see some new programming paradigms.

Yes, if by PC you limit your definition to Fusion (especially Jaguar) based SoC systems (Ultrabooks, laptops, all-in-ones).
Like the following:

Gaming PCs have completely separate memory pools and even can have even higher GPU to Video Memory bandwidth.

They operate how Ryoku described:

On PC, the CPU pulls from its own, separate pool of memory (main system memory, or the usual DDR3). The GPU is left to its own memory (the VRAM) which it can utilize fully. So a GTX770 with GDDR5 RAM @224GB/s will have access to that full 224GB/s of bandwidth--theoretically speaking, of course.

The nice thing about the PS4 is the communication between the CPU and GPU. They're able to share the same address space. This helps with parallel processing between the CPU and GPU.

Wasn't there something about DX11.2 making it possible for the CPU and GPU to share the same address space or something?

 

[MORT1S]

There was a thread about it on GAF. I searched it up just now and it's still just a rumor, I think, that started off on PS4Daily. I don't know if that's a trusted source for news, but seeing as how WiiUDaily is not trustworthy, PS4Daily's rumors do seem questionable.

Eh

I would imagine 2GB would not be a bad number to reserve.

I would rather assume 6, than 7 usable and hear better later.

 

[Rooster]

So the PS4 has more powerful hardware, an almost complete SDK and lower level access to hardware which should give higher performance for more effort.

Xbone has less powerful hardware, an SDK which still needs significant work and DirectX which provides ease of use at the expense of performance.

The Xbone supposedly has 2 games (Forza and Titanfall) already running on Xbone hardware at 1080p/60 and the PS4 doesn't have any IIRC (corrections welcome).

The PS4 games do seem to be graphically superior to me but not by a huge amount. Something isn't adding up.

 

[RoboPlato]

So the PS4 has more powerful hardware, an almost complete SDK and lower level access to hardware which should give higher performance for more effort.

Xbone has less powerful hardware, an SDK which still needs significant work and DirectX which provides ease of use at the expense of performance.

The Xbone supposedly has 2 games (Forza and Titanfall) already running on Xbone hardware at 1080p/60 and the PS4 doesn't have any IIRC (corrections welcome).

The PS4 games do seem to be graphically superior to me but not by a huge amount. Something isn't adding up.

PS4 has several 60fps games confirmed. BF4, CoD: Ghosts, Warframe, Blacklight, MGSV. It's just that the Sony's first party teams aren't targeting 60. That's all.

The XBO devkits also seem to be more final hardware, which means it can be exploited better.

 

[FlutterPuffs]

Do you have any confirmation? Shadowfall slides said otherwise.

From what we know the Shadowfall demo used 6 GB of RAM on the devkit (which had 8GB of RAM), that was planned to be further optimised for the previous 4GB GDDR5 RAM PS4 spec. It's changed now.

You're right that there's no official statement on how much RAM the PS4 OS will use, but it being 1GB is a pretty safe bet, considering that no developer has disputed it, or this guy right here:

http://67.227.255.239/forum/showthread.php?t=617901

 

[Lord Error]

Really interesting article, and somewhat worrying.
First of all, how on earth was thig game only running at 30FPS, on presumably some high spec PC? it's not that great looking at all.
Second, they said that initial PS4 build was running at just 10FPS which in itself is worrisome, when they didn't say what performance they achieved after all those optimizations that they've done.

 

[benny_a]

Yeah, initial port and boot up. The thing is only running at 10 fps. Let's scrap it. Back to the Sega Genesis that at least could do 60FPS, back when Cerny knew what he was doing.

 

[phosphor112]

Really interesting article, and somewhat worrying.
First of all, how on earth was thig game only running at 30FPS, on presumably some high spec PC? it's not that great looking at all.
Second, they said that initial PS4 build was running at just 10FPS which in itself is worrisome, when they didn't say what performance they achieved after all those optimizations that they've done.

....Sigh.