Do we know yet how much the PS4 is reserving for the OS? During the unveil it appeared to be similar (seamless transition in and out of games)
Edit: Ah just saw the post. It takes just 1GB
-
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.
DigitalFoundry: X1 memory performance improved for production console/ESRAM 192 GB/s)
- Thread starter ekim
- Start date
- Status
Edit: Ah just saw the post. It takes just 1GB
PR also means public relations so considering it's Leadbetter and the article is so masterfully written not only to hide the downclock but also to present the whole situation as a positive I'm inclined to believe that it's PR.
None of that nonsense has anything to do with what's written in this article.
phosphor112
Banned
Not just any increase.
An 88% increase on something that didn't get a clock increase.
I'm gonna bring back the term "Wizard Jizz."
Vestal
Gold Member
consumers no. Engineers yes
cyberheater
PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 PS4 Xbone PS4 PS4
Why don't games get 100% use of the GPU. Is it so you can snap etc...?
FranXico
Member
I doubt that they really are hired by MS, but Richie sure loves his XBox...
Proelite
Member
What's Leadbetter's slant?
Why would it be similar? PS4 isn't using 3 operating systems. Hell, we got our first peek at what OS it's using just a while ago. It's a variant of Free BSD 9. Knowing this, 1GB is actually plausible for the machine as it's a lightweight OS.
It has resources in reserve so it can switch to other tasks with ease. Sony's doing it via some chip they have in the PS4. Cerny talked about it in one interview.cyberheater said:
Man, I'm just now realizing that they've divulged a ton about that thing. Kind of frustrating to talk about the X1 when we don't have much in the way of official specs.
BlastProcessing
Member
Not a 3 gig deficit because PS4 is going to have some amount of RAM reserved also.
Also, I wouldn't say that Xbox One has 3GB reserved for the OS, I would say that it has 3GB reserved for the OS and any apps that are running. Things like Skype or Netflix or the TV stuff or Twitter or what have you. The idea is that that stuff can be running concurrent with games so you can quickly flip between them. Also if something like Illumiroom comes out in a few years, they could implement it without taking away resources from games.
Wizard Jizz = Secret Sauce?
...I don't think I want to eat Microsofts noodles anymore.
Like I said, the seamless switching in and out of games, I imagine that relies on the ram?
Acheteedo
Member
Richard Leadbetter is DF isn't he?
Manmademan
Member
So the actual news isn't "bandwidth increased!" but "downclock confirmed?"
oh man, this gen is a mess, and hasn't even started.
He is one of the people that write for DigitalFoundry, yeah.
That's the unconfirmed negative view on things. What seems to be confirmed is that they have the ability to read/write in the same cycle now which increases the overall bandwidth.
The article says that MS didn't tell them anything about a clock change. Where do you get from the article that the clock for the GPU has changed?
- Mentions SHAPE audio as if it's something everyone's going to use and has no counterpart in PS4 (It's not going to matter anyway because devs won't care, but PS4 has an audio chip and can spare a couple CU for sound processing if required).
- Perpetuates the DDR3 lower latency than GDDR5 myth without knowing the difference in memory controllers between the two APU's.
FranXico
Member
Considering that they were originally designing it for 4GB RAM, they aimed at a very low footprint.
A peak theoretical bandwith is a "best case" scenario so to speak - anything detrimental to this scenario isn't taken into consideration. The number only has academic value. And it should be 204,8GB/s in this case and not 192GB/s...
No
"they are making actual Xbox One titles and have not been informed of any hit to performance brought on by production challenges. To the best of their knowledge, 800MHz remains the clock speed of the graphics component of the processor."
There is nothing confirmed.
dragonelite
Member
Maybe the 102Gb/s for read and write was something that was 100% certain.
They could have know about the synchronous write but weren't it would make it in the final silicon.
It would be like sony saying to devs when they only knew 2gb of gddr5 was on the ps4 to use 8gb gddr5. You only tell devs what is certain guarilla also used the cpu to do sounds and only heard later there would be a dedicated audio block and they could use more ram.
This could be microsoft x1 luck that final silicon does perform by specification and they are adjusting drivers to offer these options.
Jinfash
needs 2 extra inches
I'm all for science and math but are we basing these calculations on a solid foundation of hardware and architecture knowledge or trial-and-error until something matches up? I mean, "engineers" realizing just now that a hardware has been bidirectional from the beginning sound embarrassing if true. Did MS know? Since they're the ones who started notifying developers lol.
Fwiw, I know jack shit about HW, but I enjoy asking questions.
Fwiw, I know jack shit about HW, but I enjoy asking questions.
hokagespit
Member
this
"Microsoft techs have found" it never gives a firm narrative for whom it is describing so carefully as to make you think its talking about sources outside of MS. hyperbole number fluff piece. before the specs are released. isnt it strange this occurs the same day they release a statement saying specs aren't everything?
I don't believe the DSP takes off the memory footprint of sound, only the CPU footprint of it.
I edited my post in response to someone else, but as I said there: The PS4 has hardware dedicated to that. Same with audio and encode/decode. I'll try to find the interview where Cerny says this was down to devote as many resources to gaming as possible.
Edit: Ah, here it is.
Gamasutra said:
Damn, I keep forgetting that they only had devkits with 4GB for gaming. So yeah, their OS choice (and rumored footprint) does make sense given that. They got lucky as fuck that Hynix was able to come up with higher density memory.
dragonelite
Member
Maybe its a feature they couldn't guarantee would make it in.
So they played save first till final silicon was done and the feature did make it in.
Kinda like how sony didn't starting mentioning 8gb of gddr5 maybe its all just pure fucking luck and some miss communication. Maybe engineer assumed it was common knowledge and the driver write didn't or that they got lucky. Shit happens even at that level of skill and knowledge especially in crunch time and launch window you don't want to take away promised performance from launch games better play save here.
even more downclocks confirmed?
did you forgot GG was working with an 4gb devkit or memory. And not a more recent devkit.
Are you deliberately ignoring everyone in this thread?
To me what you're saying has nothing to do with the solvable problem that the article presents. Peak or not what I am seeing is that every X clock cycle you get something extra. The leaps that people are taking to get the down clock number has nothing to do with the solvable problem I am picking up from the article.
I have little expertise in the making of hardware, but the only way I can make sense of a suddenly discovered bandwidth increase is that they now use both the rising and falling edge of the clock signal to transmit data, instead of just one of them. But I have no idea if that is possible without redesigning the memory controller.
I know you are trying to be facetious but I'll bite:
So in contrast to best-case scenarios this can't be solved as easily with math. Simply measuring perfomance while doing real-life tasks will do the trick though - as hinted in the quote...
Yeah, I never heard of that one. Smells like BS.
LMAO. Wake up & smell the coffee, bro.
blindrocket
Member
I love theoretic terms.
haterofpants
Neo Member
Sure it's theoretically possible, assuming you have a 100% hit rate on a 32MB cache. I'm guessing that's going to be more useful for GPGPU than your typical rendering shit. If your GPU has to access a bunch of detailed textures then I don't see how that 32MB is going to give you a consistent boost in overall bandwidth.
It certainly helps, but when you're trying to render a scene with 800MB of texture data, that 32MB cache isn't going to make up the difference. Again, those speeds are only relevant when you have a 100% hit rate there. Chances are you're NOT going to achieve that, and whenever you miss and pull from DDR3 that memory bandwidth is going to hurt.
It certainly helps, but when you're trying to render a scene with 800MB of texture data, that 32MB cache isn't going to make up the difference. Again, those speeds are only relevant when you have a 100% hit rate there. Chances are you're NOT going to achieve that, and whenever you miss and pull from DDR3 that memory bandwidth is going to hurt.
i dont think hes ignoring anyone, he states the truth that nothing in confirmed yet. Nor will it be till the actual console is in peoples hands.
dragonelite
Member
Maybe you can read 8bytes in 5 cycles instead of 8bytes in 8cycles. Give one command cycle wait 4 cycles and 8 bytes. instead 8 commands and for each command one byte.
Im no hardware man so yes i could be bullshitting.
Going by 192GB/s peak performance as stated by DF it would suggest that for 22 out of 25 cycles you'd get "something extra".
What do you think prevents the other 3 out of 25 cycles from getting "something extra" in a best case scenario (so forget pipeline flushes and the like...)?
Sorry, I misinterpreted your initial post.
Edit: Where did GG say that they hadn't head of the audio DSP until later on?
What you're saying makes no sense to me. Texture data will be stored in the DDR3. Its more than fast enough to handle high res textures no problem.
Actually we wouldn't be able to know this by opening it up and looking at it. Only devs working on it would
Your example is what I am thinking of.
Is dragonelite the only one who understands what I'm saying?
sweetvar26
Member
Do we have a conclusion for this thread yet or is the debate still going on?
Glorified G
Member
For some people, Math is hard
No conclusion.
I'm one of those people
I flicked through looking for Gifs to explain what this all meant and was disappointed.
qa_engineer
Member
No conclusions. Just clowds. Infinite clowds.
There are people who act like they know the in's and out's of the system and so they are taking some leaps to arrive at the down clock number. This what I have gotten out of this thread so far because people seriously aren't trying to figure out what the problem is that the article is presenting. They are jumping to conclusions and working towards a down clock amount when the article source stats that they haven't been told that their is a down clock.
Gemüsepizza
Member
This. What's more believable:
1. A downclock by 50MHz which would explain the numbers.
2. A bandwidth increase by 88% out of nowhere without clock increases.
Looking at what's written above, it should be clear who is taking some leaps here.
qa_engineer
Member
What are your own thoughts? I think this means absolutely nothing and this thread is huge. That Is all.
- Status