LittleJohnny
Banned
Can and will be possible on both PS4 and Xbone, the difference between PS4 and Xbone is not as large as most people expect, and even more of a factor is that launch games are still being very conservative.
-
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
Can and will be possible on both PS4 and Xbone, the difference between PS4 and Xbone is not as large as most people expect, and even more of a factor is that launch games are still being very conservative.
qa_engineer
Member
Thats the speculation thats going on over at beyond3d. Sounds like a downgrade in clock. And... err... a Flux capacitor bi-directional bus and magic fairy dust.
LukasTaves
Member
They could if they were willing to sell a product with tremendous performance decrease for already released titles and that required all future titles to have optimizations that would only benefit their hardware.
That or if they developed magical drivers that would seamlessly for the developer do all the work required.
Interfectum
Member
Has MS actually released the specs of XB1 yet?
plagiarize
Banned
it's fairly simple. It's all about yields. You won't know what temperatures your chips can run stably at until after you're manufacturing them. Some will run stable at higher performance levels than others. Generally with something like this, it's more about how good your manufacturing process is.
Say Microsoft hoped 90% of the chips would be stable at 176 GB/sec, once they scaled up manufacturing they discovered that 90% of the chips were actually stable at 192 GB/sec.
http://www.anandtech.com/show/1689/2
I don't know man, but I think Anand knows more about this than you do.
It would be hilarious if Leadbetter inadvertently confirmed the downclock rumor at an article trying to dismiss it.
crazygambit
Member
Pretty much. If I'm limited in writing it doesn't really help me that there are available read cycles.
AgentP
Thinks mods influence posters politics. Promoted to QAnon Editor.
What does boredom have to do with the truth? Everyone has a bias, it is important to keep it in mind when you read a "technical" article that is about a company he seems invested in.
It is like reading gun control article written by a parent who's child was shot, it is an important fact to keep in mind.
Wow, some good news for MS..
Whilst peak bandwidth is meaningless, the MS architecture is not without it's own merits, and it's obvious from listening to insiders opinions that it does help improve system efficiency, not to the point anyone thinks it'll even match the PS4, but certainly way more then people who cling to #CU's + GDDR5 vs DDR3 as the only quantifier of performance think..
I always thought MS should have gone safe with things, but actually, with Intel now offering EDRAM as a efficiency boost (providing devs use it correctly), there does seem a little method in this asymmetrical madness...
Whilst peak bandwidth is meaningless, the MS architecture is not without it's own merits, and it's obvious from listening to insiders opinions that it does help improve system efficiency, not to the point anyone thinks it'll even match the PS4, but certainly way more then people who cling to #CU's + GDDR5 vs DDR3 as the only quantifier of performance think..
I always thought MS should have gone safe with things, but actually, with Intel now offering EDRAM as a efficiency boost (providing devs use it correctly), there does seem a little method in this asymmetrical madness...
dragonelite
Member
Or they changed the memory controller between the leaked docs and now?
Im not an hardware guy so what i now say could be bullshit.
plagiarize
Banned
What 'fact' do you have to say DF is biased towards MS?
Anyways, I never thought memory bandwidth was going to be an issue on the Xbone in the first place. The architecture was one we know to be quite workable in consoles. It still wasn't as good (or simple to code for) as the PS4s, but it wasn't crazy or bad. It's still likely true to say that the PS4 won't have any problems at all seeing a game ported to it that was designed around the memory architecture of the Xbone. This was not always the case when going from Xbox 360 to PS3.
Going from something designed around the PS4's memory architecture to the Xbones will be more work, but I don't think many games will be leading on PS4 and being ported to Xbone as an afterthought.
Can Crusher
Banned
It's funny that you are taking this personal when I'm just saying what almost all articles out there say.
Keep it up sport.
ChorusLindarr
Banned
This is a terrible article that is truly grasping at straws to try to paint the XB1 in a light where it might be considered (in some bizarre parallel universe) more capable than it actually is. Truly a low point for Digital Foundry, and I must admit that their recent articles regarding next-gen consoles have caused me to lose a lot of respect for them.
Very little written makes much sense.
I mean, why would the designers of the XB1 HW not know that the eSRAM is capable of both reading and writing to and from memory in certain peculiar cases?
Why does this only apply to production HW if it is merely a facet of the "unused cycles" and not anything like a change in clock speed in production HW (thus creating more cycles in a given time period)?
Why would this somehow present developers with a significant increase in memory performance across all operations, when in actual fact this concurrent read/write ability only occurs in very specific cases?
And why would this imply anything, either negative or affirmative, about the potential yield problems that MS may or may not be having with their APU? It's actually entirely irrelevant.
I must wonder who DF's sources are for this info.. I understand they have an even track record, but even if this is all true, and come down from a real update given to XB1 developers, DF's interpretation of it in this article is just abysmal. Really really bad.
Either way, this is nothing at all to be excited about. An increase in eSRAM bandwidth in the last thing that MS would need to close the gap between their console and PS4. Had this been about a change in main memory bandwidth (which is all but impossible with 256-bit DDR3) then it would have been much more interesting.
Very little written makes much sense.
I mean, why would the designers of the XB1 HW not know that the eSRAM is capable of both reading and writing to and from memory in certain peculiar cases?
Why does this only apply to production HW if it is merely a facet of the "unused cycles" and not anything like a change in clock speed in production HW (thus creating more cycles in a given time period)?
Why would this somehow present developers with a significant increase in memory performance across all operations, when in actual fact this concurrent read/write ability only occurs in very specific cases?
And why would this imply anything, either negative or affirmative, about the potential yield problems that MS may or may not be having with their APU? It's actually entirely irrelevant.
I must wonder who DF's sources are for this info.. I understand they have an even track record, but even if this is all true, and come down from a real update given to XB1 developers, DF's interpretation of it in this article is just abysmal. Really really bad.
Either way, this is nothing at all to be excited about. An increase in eSRAM bandwidth in the last thing that MS would need to close the gap between their console and PS4. Had this been about a change in main memory bandwidth (which is all but impossible with 256-bit DDR3) then it would have been much more interesting.
Kung Fu Grip
Banned
Yes they are. But people in this thread are acting like this one bandwidth improvement ( theoretical, final?) defines and changes everything.
DF just threw a slice of cheese into a cage full of hugry rats.
wow people believing this !!!
the 32mb esram is basically not even adequate for large algorithms and these days even a routine like moving a hand could take more than 32mb. If there are 10 routines with each wanting to access the esram ram pool, then the only way of acessing it would be when one process has been unloaded or killed. Xbone's ram's bandwidth for major ops would remain 68g/s . No way of fooling people with this 192g/s CLOUD LIKE number.
Cant believe they are combined the esram and ddr3 ram to fool people
the 32mb esram is basically not even adequate for large algorithms and these days even a routine like moving a hand could take more than 32mb. If there are 10 routines with each wanting to access the esram ram pool, then the only way of acessing it would be when one process has been unloaded or killed. Xbone's ram's bandwidth for major ops would remain 68g/s . No way of fooling people with this 192g/s CLOUD LIKE number.
Cant believe they are combined the esram and ddr3 ram to fool people
dragonelite
Member
They have been praising ps3 exclusives like hell.
Maybe with the vs articles they chose the 360 to many times?
famousmortimer
Banned
Not necessarily good news, it may be the result of a downclock. Going to be interesting to see how this plays out.
To your last paragraph though, MS did go the safe route. They knew from day one that they wanted 8gbs of ram. DDR3 was the only safe way to make that happen when they were designing the unit. Sony wanted unified memory with gddr5 and first targeted 2gbs... then 4gbs... and then right before the reveal found out that 8gbs was feasible. There is no way that Cerny and co could have seen this coming - affordable, mass produced 512mb sticks of gddr5 weren't supposed to be ready yet. So they shocked everyone, including Microsoft and first party devs, when they announced 8gbs. But the reason they got there was dumb fucking luck. Microsoft though they were going to have a memory advantage from day one and got unlucky.
DeadRockstar
Member
"I don't know how Microsoft managed to fit more bandwidths into a box the same size. Where did they put them?
I'm glad they have anyway because they are my favourite and now their console is more powerful than the PS4."
-This Thread
I'm glad they have anyway because they are my favourite and now their console is more powerful than the PS4."
-This Thread
AgentP
Thinks mods influence posters politics. Promoted to QAnon Editor.
Not DF, Leadbetter. His posting history as 'grandmaster' at B3D and his frequent editorializing in his technical articles at DF. IMO it is not even hidden, it is easy to see if you read what he writes. The one time he totally just gave up any illusion of objectivity was the FFXIII vs article, that was lol worthy he sounded like he was crying.
Not really my fault you took offence from a neutral statement and decided to be condescending.
uggh..
please, stop just looking at numbers and trying to guess through the rest.
Sony - unified memory runs at 176GB/s
MS - unified memory runs at 68GB/s. on-die 32MB eSRAM cache runs at 102GB/s.
MS is now saying that when combining read and write operations in the same clock cycle that can go up to 192GB/s. DF's info says that real world performance looks to be closer to 133GB/s. However we are AGAIN talking just about the 32MB eSRAM cache on die.. This does not mean that the XBONE's memory architecture now magically runs at 192GB/s, 133GB/s, or even 102GB/s. It means that where it works to pre-load data through the cache it will be read to the GPU at a faster bandwidth, but the unified system memory is still half as fast..
trying to compare the two directly next to each other is pointless. we probably truly won't even START to see the real differences in the architectures until 2015 or 2016 at the earliest.
Nothing detailed past basic amounts as far as I am aware, at least not officially to the general public, 8 Core CPU, 8GB RAM, 32MB eSRAM, and then typical WiFi, HDD, USB etc., there's no reason to think the leaked specs beforehand aren't the real ones seeing as everything else has lined up with them.
Jagerbizzle
Neo Member
What exactly are you suggesting then, that the esram is useless? If it can't be used for even a routine such as "moving a hand", what do you think it can be used for?
Please elaborate.
I think they are in kahoots with Adam Sessler also...
I swear every freaking story that has any kind of non condeming news for the XB1 is painted a shill or fluff piece.
Someone needs to change the year of the PS3 jpeg to who is in MS's pocket.
I'd like to hear what actually developers say about this
hydrophilic attack
Member
So it might actually be due to a downclock? I don't know how much of this stuff works, but that would be very unfortunate. Also, massive levels of fail on DF's part if they mistook a downclock for an upgrade.
But I guess this could mean downclock, but still better ESRAM capacity than previously thought?
Anyway, let's hope there is no downclock.
But I guess this could mean downclock, but still better ESRAM capacity than previously thought?
Anyway, let's hope there is no downclock.
LittleJohnny
Banned
Ding ding ding!
Everyone is focusing on the wrong thing, GDDR5 is just a different solution to make sure the system doesn't run into any bottlenecks. If both systems are balanced in a way that neither has any bottlenecking, having GDDR7 @8000Mhz adds nothing.
If a race car needs to get from point A to point B as fast as possible, and you have open highway with no traffic, then it's all up to the speed of the car (CPU/GPU) to make the difference, adding more lanes for the slower car wouldn't help at all. Xbox One's GPU has less shader cores so the GPU in the PS4 has a slight edge in performance, you will be getting better numbers but it's not that big of a difference, probably even less of a difference between Xbox and GameCube, the latter being compliant with an older generation of DX/OGL.
Anyone expecting a drastic difference between PS4 and X1 will be drastically disappointed
I don't think it's as daft as it seems... if you look at the memory timings, you see masses of clock cycles just 'waiting' in the various stages of the transaction, those are required for many reasons, one of those is stability.. I can imagine they'd have set the timings at design time to conservative values, only to find when they get final silicon that noise levels are better then expected, (or other factors affecting stability) and so can shave a few clock cycles here and there with the timings..
I'm basing this of currently working with ARMs talking to many different peripherals as memory mapped IO, where we design worse case timings until we get the final hardware and often find we can tighten up the timings be a reasonable margin..
ChorusLindarr
Banned
Nvm
funkystudent
Member
I am no expert but while the small amount of ESRAM seems like it will come in handy I dont get why people think it can magically make up the difference between DDR3 and GDDR5.
Jagerbizzle
Neo Member
Dude, what the you do you think devs are going to be using the extra ram for in the near term? >5 gig being used in anything other than a completely wasteful manner will not start happening until way way way later in this generation.
Calavera520
Member
Nah, each system has differing priorities. Xbone is targeting the dvr market primarily Sony is interested in the games market.
plagiarize
Banned
Fair enough. However, it's very common for a hardware manufacturer to not know what their performance is going to be due to yields. I was wrong on the specifics here, but I took the question as a general one. The notion that it shows some kind of problem or incompetence on Microsoft's part is nonsense. Hopefully that's clear.
It's worth stating as you did, though, that this instance has nothing to do with yields. My post was misleading, espescially to specifically mention the ram bandwidth.
Nizz
Member
I just subscribed to this thread. Lots of good info in here for someone like me who's not super knowledgeable on some stuff.
Can Crusher
Banned
Condescending? I specifically I said I didn't know, but it looked like you were wrong. And you pull one source to say that Anandtech knows more about it than I do?
Well no shit anandtech knows more than me, that's why I provided two sources for the information I knew. But because you are really in this, wearing your heart on your sleeve, you decided to be passive aggressive.
Anyway, it is what it is.
It's not one or the other.
MS's first party companies do not care about the "DVR" market and will work just as hard as Sony's first party.
For now I can see it being a draw, with Sony getting the edge a few years down the line
Sounds like MS are trying to hide a down clock with supposed good news, where under very specific circumstances they think they have observed some test results indicating the ability to read/write simultaneously.
Smoke and mirrors.
I don't believe anything this company says right now, until verified by third parties.
Smoke and mirrors.
I don't believe anything this company says right now, until verified by third parties.
Keep telling yourself that, Devs will find a way to use any option given to them.
LittleJohnny
Banned
They said it was a gaming system first before anything else. If you think they're lying that's another story, and if you're judging by the reveal, that's what the reveal was for, E3 focused purely on games. PS4 actually showed a lot of TV stuff at E3, doesn't mean that's their main focus.
Sandy Nipples
Banned
Oh fucking come on. You really think we should believe that right after news started spreading that the system was inferior to the PS4 because of that HD IR camera's costing and downclocking due to yield problems, all of a sudden some magical performance boost was discovered? Really? This is way too convenient.
Speculative article that is full of rumors + tech speek + GAF = hilarity
If this news is true... then this means that 33% slower GPU of Xbone will not be bottlenecked when accessing smallish 32MB of integrated ESRAM [PS4 dont have this problems, devs demanded easy architecture, and Sony provided] . Nothing has changed much on performance side, except life of high end game programmers will become little more easier when trying to optimize what to put in that small memory pool.
If this news is true... then this means that 33% slower GPU of Xbone will not be bottlenecked when accessing smallish 32MB of integrated ESRAM [PS4 dont have this problems, devs demanded easy architecture, and Sony provided] . Nothing has changed much on performance side, except life of high end game programmers will become little more easier when trying to optimize what to put in that small memory pool.
ChorusLindarr
Banned
LiquidMetal14
hide your water-based mammals
I wish someone from MSFT would just outline their platform like we've seen Mark Cerny.
One on hand we had all this info and big wording from MSFT.
On the other hand you have Mark Cerny making 2 speeches already. One in February and one just a day or two ago. There's no transparency in his words.
One on hand we had all this info and big wording from MSFT.
On the other hand you have Mark Cerny making 2 speeches already. One in February and one just a day or two ago. There's no transparency in his words.
Impeccable
Member
I wish I had the technical understanding to even participate in these conversations, all I know is that if the down clock thing turns out to be true... yikes.
- Status