PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)

Status
Not open for further replies.
As later in the thread on neogaf says its just details on the bus's and how the devices access the ram.

Nothing negative.

I think it can be somewhat negative if DF is set to reveal the memory setup is more complicated than we're lead to believe.

It may not be the Armageddon GAF feared but it's probably not a good thing, rather a minorly bad one. But I guess we see today.
 
If PlayTV works on PS3 with only 512MB RAM (and roughly 56MB OS footprint), Sony should be able to make PS4 DVR work on similar or higher specs.

Apparently DVR is also done by hardware.
 
Last edited by a moderator:
I think it can be somewhat negative if DF is set to reveal the memory setup is more complicated than we're lead to believe.

It may not be the Armageddon GAF feared but it's probably not a good thing, rather a minorly bad one. But I guess we see today.

This is reaching pretty hard, the main goal of the machine was to be fast and easy to develop for, this flies straight in the face of that.

There is 0 evidence for this. The max complexity i can see is that you need to know what devices you want to access your data and you should know this already though so its not really that big a deal.
 
I think it can be somewhat negative if DF is set to reveal the memory setup is more complicated than we're lead to believe.

It may not be the Armageddon GAF feared but it's probably not a good thing, rather a minorly bad one. But I guess we see today.

At this point it doesn't even seem like he's talking about something that we don't already know about. My bet is the original source is just referring to his Spanish translation of the article about porting The Crew. Even that didn't really contain new information because we already knew there were multiple busses to memory with different characteristics, only lots of people apparently never realized it would be up to developers to decide when to use what. Which is to say even his original characterization of this as "memory drama" that would affect the relative performance of the PS4 vs the Xbox One is based on a faulty understanding of the subject.
 
At this point it doesn't even seem like he's talking about something that we don't already know about. My bet is the original source is just referring to his Spanish translation of the article about porting The Crew. Even that didn't really contain new information because we already knew there were multiple busses to memory with different characteristics, only lots of people apparently never realized it would be up to developers to decide when to use what. Which is to say even his original characterization of this as "memory drama" that would affect the relative performance of the PS4 vs the Xbox One is based on a faulty understanding of the subject.
It is pretty slow right now on the web with those summer vacations taking their tall on activity, I wonder if that is just plain click hunting... or a just a trollish post.
 
It's been Friday for a while without a peep from Eurogamer UK about anything new, and anyone can go look at Eurogamer.es and see the Digital Foundy section is well out of date. I don't think it was click hunting (after all, there was nothing to click on in the original tweet), it's just there's a large number of disgruntled Xbox enthusiasts out there desperate for any scrap of bad news about the PS4. Even the most pale shadow was enough to create a frenzy on GAF.
 
It's been Friday for a while without a peep from Eurogamer UK about anything new, and anyone can go look at Eurogamer.es and see the Digital Foundy section is well out of date. I don't think it was click hunting (after all, there was nothing to click on in the original tweet), it's just there's a large number of disgruntled Xbox enthusiasts out there desperate for any scrap of bad news about the PS4. Even the most pale shadow was enough to create a frenzy on GAF.
You might be right, could be another of those posts taken out of their context, with drama referring to "noise" existing prior the guy post. Anyway +50 pages on Gaf... :LOL:
 
...we already knew there were multiple busses to memory with different characteristics, only lots of people apparently never realized it would be up to developers to decide when to use what.
That's not how it was presented. We'd been told unified memory, that anything can be anywhere, and then we're told that you have to put stuff in the right parts of memory. The low level workings of memory access may be common knowledge to computer engineers, but you can't expect everyone else to know that you can't just ask in the CPU or GPU "fetch this data from RAM" and have it do it.
 
gaf has worked itself into a frenzy over...something...related to a eurogamer spanish editor tweets (so they speculate digital foundry)

http://www.neogaf.com/forum/showthread.php?t=634711




could be the onion/garlic bus stuff (most likely imo), or something more sinister? (gaf children wildly speculating everything up to a console delay)


"Calm down! It's just some info about how memory works and nothing more. Nothing serious! "

Welp, time to close the thread.


http://www.neogaf.com/forum/showthread.php?t=634711

It was nothing apparently..
 
That's not how it was presented. We'd been told unified memory, that anything can be anywhere, and then we're told that you have to put stuff in the right parts of memory. The low level workings of memory access may be common knowledge to computer engineers, but you can't expect everyone else to know that you can't just ask in the CPU or GPU "fetch this data from RAM" and have it do it.

Well, yeah, we're deep in the territory of people who don't understand how this stuff works trying to spin certain functions as a drawback, even if it may actually be an enhancement. Eurogamer kinda presented this stuff as a "problem" with PS4 development, when it sounds like to me that 3 guys porting a whole game in 6 months where the hardest part was discovering the best way to use each memory path is really promising. I'd love to see a corresponding article for the Xbox One port. I bet those folks would love it if their biggest challenge was choosing between the garlic, onion and onion+ busses.
 
That's not how it was presented. We'd been told unified memory, that anything can be anywhere, and then we're told that you have to put stuff in the right parts of memory. The low level workings of memory access may be common knowledge to computer engineers, but you can't expect everyone else to know that you can't just ask in the CPU or GPU "fetch this data from RAM" and have it do it.

My belief now is that it works like Llano, Onion behaving like CPU bus to memory with the AGP functions integrated, only that it will be faster than any AGP communication port ( or not, because is 20GB/s of bandwidth shared for both CPU only ops and GPU + CPU ops). And surely memory is divided in two regions, the addressable by the CPU northbridge and the addressable by the GPU memory controller ( via Garlic also called Radeon Graphics Bus in Llano ). It will be the same than Llano, neither HSA 2.0 nor full HSA...Unified Virtual Address, nothing of real Unified Memory Address.
 
Last edited by a moderator:
My belief now is that it works like Llano, Onion behaving like CPU bus to memory with the AGP functions integrated, only that it will be faster than any AGP communication port ( or not because is 20GB/s of bandwidth shared for both CPU only ops and GPU + CPU ops). And surely memory is divided in two regions, the addressable by the CPU northbridge and the addressable by the GPU memory port ( via Garlic or Radeon Graphics Bus ). The same, neither HSA 2.0 nor full HSA...Unified Virtual Address, nothing of real Unified Memory Address.

My understanding was that either device could access all of the memory space (aside from all the protect stuff such as the kernel, etc). Except there was a performance hit if you accessed something that was declared in the incorrect manner.

I think a good question to ask is, what kind of performance hit would happen on a machine that did not have such a option.
 
My understanding was that either device could access all of the memory space (aside from all the protect stuff such as the kernel, etc). Except there was a performance hit if you accessed something that was declared in the incorrect manner.

I think a good question to ask is, what kind of performance hit would happen on a machine that did not have such a option.

For example if your CPU has to look for GPU data the virtual address mechanism add many more cycles to the data retrieval than a real unified memory address space. To avoid that you have to put that GPU data in the Onion bus, and so, in the CPU memory space, but then you are limited to a 20GB/s of bandwidth, so better not have to put many GPU data there.
 
For example if your CPU has to look for GPU data the virtual address mechanism add many more cycles to the data retrieval than a real unified memory address space. To avoid that you have to put that GPU data in the Onion bus, and so, in the CPU memory space, but then you are limited to a 20GB/s of bandwidth, so better not have to put many GPU data there.

AFAICT, it's easiest to think of it in terms of caches:
- CPU-RAM (unnamed) uses CPU caches.
- GPU-RAM (garlic) uses GPU caches.
- GPU-RAM (onion) uses both GPU and CPU caches.
- GPU-RAM (onion+) uses CPU caches.

If you want to read a memory location, then you must either read it via the cache that writes to that memory location [or] flush the relevant cache before reading. (flushing is generally not a good idea)
 
For example if your CPU has to look for GPU data the virtual address mechanism add many more cycles to the data retrieval than a real unified memory address space. To avoid that you have to put that GPU data in the Onion bus, and so, in the CPU memory space, but then you are limited to a 20GB/s of bandwidth, so better not have to put many GPU data there.

My understanding previously, that the AMD's 3rd gen HSA systems would pound on the same memory addresses without penalties, thereby totally eliminating the need to copy data between memory spaces. But if you have to divide memory and not arbitrarily reach data by either GPU or CPU, and if working on the same data "heterogeneously" would require copy operation to fully utilize throughput, than that's sort of a let down.

Maybe it's just a simple question of reaching data through appropriate buses, and maybe the switch between onion/garlic on the same memory addresses doesn't really cause performance problems. Then it wouldn't be something to lose sleep over.

Edit: The cache explanation from Dumbo made some sense to me. The caches should be accounted for when talking about coherency. So any address space that CPU works on cannot be read faster than cpu's cache writing those results back, and that can't be faster than 20GB/sec. Have I understood correctly?
 
Last edited by a moderator:
Status
Not open for further replies.
Back
Top