Thank you, I never had a clear picture of the eSRAM situation until this thread.
eSRAM question
- Thread starter bGanci
- Start date
Thank you, I never had a clear picture of the eSRAM situation until this thread.
I feel like I am reading some sort of crazy fan fiction with this post.
I...can't tell if you're joking or not. Your name and tag suit you.
BruiserBear
Banned
It's amazing how many members of the gaming media still believe it might amount to something.
This right here.
It can offer some options to gameplay designers, and the way it spools up servers at a moments notice for games like Titanfall is great. But in terms of graphics, yeah, it's worthless.
SMOK3Y
Generous Member
gotta be taking the piss.... no lol
LordOfChaos
Member
Well actually the fastest bandwidth in the consoles will be the on-chip L1 cache, followed by L2. If PCs are any indication those might be just under 1TB/s.
But yes, in terms of outside memory the eSRAM is the fastest...But not by that much. 204GB/s in a 32MB pool, vs 170GB sustained for the whole 8GB. The proof is in the pudding, nearly all cross platform games are better on the PS4. I'm sure devs can make better use of that 32MB pool and optimize things further, but it doesn't make up for the discrepancy in shaders, ROPs, TMUs, etc. No amount of memory bandwidth ever could.
EDIT: Crap, were you pulling our legs? The sim city and diablo thing tipped it off. Damn it.
Head.spawn
Junior Member
DCDW
Member
Nope. That 204GB/s # is another one of Microsofts BS spins on their "we found holes in the ram process that allows it Read and Write at the same time". Fact is that it's still locked at 102 GB/s while doing both processes and even then if you count that as doing both processes at once (read and write) then you're essentially stating it's 16MBs of 204 GB/s since it doesn't magically allow double the amount of code to run through. 16 MBs of ram at 102 GB/s dedicated to write and reading doesn't do them a whole lot of good, considering 32 MBs is bottlenecking the system already, 16 MBs would just be a joke. At best it allows them some flexibility in micromanaging code but it's not in the ballpark of just outright claiming the ram is now 204GB/s.
BeforeU
Oft hope is born when all is forlorn.
BeforeU
Oft hope is born when all is forlorn.
The famous Microsoft Maths around the Xbox One's performance was incredible.
The claim that the 102GBs bandwidth was actually 204GB/s is like saying the speed limit of a 60MPH road is actually 120MPH, because it's 60MPH in both directions!
LukasTaves
Member
Exactly.
He has an apt username, as that's the facial expression I had when reading his post. It's like it's a copypasta from MisterXMedia.
LukasTaves
Member
It's not exactly 170GB/s sustained, 170 is the peak, and not sustained even if only the gpu is using it. And when the memory is serving the cpu and the gpu each GB/s the cpu consumes takes more than 1GB/s from the cpu.
That's one of the biggest advantages of a embedded memory: There's no contention from any one else: all the bandwidth is available to the gpu at all times (which doesn't mean it's sustained at 204 GB/s either).
Bending_Unit_22
Member
Heh, just need to add a cat paw going in the cat's mouth and perfection.
enanogrande
Member
Liabe Brave
Member
I recall the Sony employee DemonNite saying here on GAF that they'd worked with Hynix to try and ensure manufacturing matured in time. Unfortunately, I don't remember the actual wording, and I've been unable to find such a post. However, I am certain that this was said by some "insider", though it may have been someone much less reliable. Perhaps DemonNite can clear this up?
In any case, there is a source. The source may be wrong, but there's a reason besides fanboyism people are saying it.
LukasTaves
Member
None of that choices demand esram to work. How did any of that could possibly have forced the esram choice?
And more importantly, how is that more likely the chance that they had to chose that design to comply with those requirements than they simply were satisfied with 360's design and decided to improve it (which is exactly what they ended doing)?
If you said that running 3 VMs at the same time, having a non trivial die space dedicated to Kinect and voice recognition, etc were decisions forced down to the engineers I could agree with, but esram? Nah XD
I tend to dislike the sarcastic troll posts which are fashionable on gaf, as they are usually either too obvious or they give insufficient clues to deny Poe's law, but open_mouth's post was quite well done and had me going for a moment. The use of sim city/diablo 3 was nicely judged indeed.
open_mouth_
insert_foot_
This well thought out explanation was definitely not what I was expecting but thank you for that lol I was playing folks.. But if the Xbox cloud one powered by google fiber one becomes a reality then you'll know where you heard it first!
silver.tongue
Member
The "cloud" is a lie, man. It's basically a dedicated server.
360 and PS3 already had it.
Is this a parody? I feel like I've suffered brain damage just by reading it...
PHOENIXZERO
Member
Here, for everyone asking for it.
Here, for everyone asking for it.
From what I understand, this picture is simplified to a fault. Things aren't so simple.
The picture doesn't explain the situation, it just explains what a bottleneck is, which most people know. ESRAM is not entirely a bottleneck on all fronts.
The ESRAM is not the bottleneck though, that would be the DDR3.
You're right. That's what I mean.
The ESRAM is very, very fast. The DDR3 is "slow." (I mean, it's fast too, but not as fast as GDDR5 and far and a way slower than ESRAMN)
Liabe Brave
Member
Cost-effectiveness is the only possible reason they chose DDR3. We know this because the only two advantages DDR3 had over GDDR5 at the time of design were lower price, and bigger modules. But size can't be the reason, because BKilian confirmed that DDR3 was chosen before 8GB was the target.
Other target features Brad Grenz listed, like Kinect and multitasking, aren't directly the reason for eSRAM. However, they did drive the increase to 8GB (as also confirmed by BKilian). After that choice was made, RAM module size did become an issue, because the only possible approach at the time was DDR3. Fortunately, Microsoft had already decided to go that way for cost reasons. But DDR3 is not fast enough for modern gaming applications; using eSRAM (or eDRAM) therefore becomes a forced move. So while the console's features don't technically impel eSRAM, along with other facts they create a situation where no other choice will do.
The idea that eSRAM was chosen first isn't impossible, but it goes against all logic and I find it hard to credit, given how smart Microsoft's engineers are. Put simply, industrial design is always pointed toward a purpose. The story that eSRAM-plus-slow-main-RAM was chosen without any cognizance of what the machine would eventually be asked to do is incredible. "Well, it's like what we did last time" is one of the worst possible justifications for design. I can't imagine it was used here...leaving DDR3-first-eSRAM-later the likely route.
arhra
Member
Having a small pool of fast, on-chip scratchpad memory paired with a larger pool of slower memory is a design that's been used by at least one of the competing systems in every generation since the PS2 (and arguably, you could include the N64, although the scratchpad there was ludicrously tiny and only used for textures). It's not some weird outlier, it's a tried-and-tested design with a well-understood set of tradeoffs (many of which the XB1 design mitigates anyway, due to the added flexibility over previous designs), and which has worked extremely well in past hardware.
The PS2 had an eDRAM scratchpad used for texture storage and render targets. GameCube had an embedded framebuffer memory, 360 had the on-package eDRAM die.
Going with an improved extrapolation of those previous designs was a perfectly reasonable and understandable decision.
Brad Grenz
Member
Good point, they also had the option of using a completely crippled design where it was so bandwidth starved by its unified DDR3 bus it would barely be faster than a 360.
Here, for everyone asking for it.
its been revised
(eSRAM itself has a slightly slower bus than GDDR5)
Except if you wanted to go 1080P60 on modern deferred rendering engines.
http://www.redgamingtech.com/xbox-one-esram-720p-why-its-causing-a-resolution-bottleneck-analysis/
Pimpbaa
Member
The subpar GPU in the xb1 has more to do with those difficulties.
But what about the secret GPU?
I think the only controversy is that it was a bad decision. The eSRAM eats up the APU's transistor budget, which is why the GPU is so much weaker compared to the PS4s. If I recall correctly, the Xbone's APU is actually more expensive to manufacture then the PS4's.
I can't believe anyone still believe that "cloud" nonsense.
no cloud computing can increase graphical performance. what it can do is help CPU with less intensive computing tasks like AI, physic etc.
even if it help alot, any company can create it's own cloud including Sony
and for the DirectX 12 bringing "twice performance for multi-core system" is another BS from MS or their developers. the purpose DirectX 12 is to bring "bare-to-metal" console development to PC. so it won't help much with XB1 since currently they already using "bare-to-metal" way.
even for PC, the performance gain won't be "twice" as stated. try to look Battlefield 4 DX11 vs Mantle. it only helps with less powerful CPU.
stop believing anything you read immediately and do your own research next time.
no cloud computing can increase graphical performance. what it can do is help CPU with less intensive computing tasks like AI, physic etc.
even if it help alot, any company can create it's own cloud including Sony
and for the DirectX 12 bringing "twice performance for multi-core system" is another BS from MS or their developers. the purpose DirectX 12 is to bring "bare-to-metal" console development to PC. so it won't help much with XB1 since currently they already using "bare-to-metal" way.
even for PC, the performance gain won't be "twice" as stated. try to look Battlefield 4 DX11 vs Mantle. it only helps with less powerful CPU.
stop believing anything you read immediately and do your own research next time.
LordOfChaos
Member
I've heard this mentioned when talking about the 360 compared to the Wii U as well. Is the 360s 22GB/s main memory similarly improperly added up like this, and it's really half of that in both directions? How about the PS4s? And the Wii Us 12GB/s, is that in both directions, or the total aggregate like Microsofts 204?
And yes to the other people, sustained was the wrong word, I just meant through the whole memory pool rather than one smaller pool.
the "Cloud Demo" could be easily rendered/computed using GPU since the graphics in that demo isn't that good (it's like demo Physx using CPU vs nVidia GPU)
if "Cloud Power" is true, I guess the best way to use it is to create stand alone app for PC, connected with 100/1000 Mbps LAN. my i7 3770k will gladly lend it's power. no need to use slow internet anymore.
Exactly. All you need to do is look at the APU's of each console and you can see how much space the eSRAM eats up for such a small pool size. It was a terrible decision after the fact but it's not like MS expected to be able to get 8 gigs of GDDR5
So when your ISP experiences latency and high traffic your game runs at 3fps!
Its cool, but its not realistic. Games can only offload what isn't immediately nessesary or can be shifted back to the console if the net fails. Otherwise its straight to the title screen any time you have a hiccup.
plus why not just run that game on a azure server itself and send the much less bandwidth intensive video stream and data inputs?
its been revised
(eSRAM itself has a slightly slower bus than GDDR5)
correct.
For a simple explanation....its a "band-aid" to try and speed up its much slower 8GB DDR3 ram to keep up to the speed of PS4's 8GB GDDR5 ram but it's still much slower overall to PS4's approach of just using that sexy GDDR5 faster ram.
2-3 years down the road, both consoles may get process node shrinks.Exactly. All you need to do is look at the APU's of each console and you can see how much space the eSRAM eats up for such a small pool size. It was a terrible decision after the fact but it's not like MS expected to be able to get 8 gigs of GDDR5
Assuming GDDR5 memory vendors don't follow the same pace of savings pass-thru, X1 can get far more cost savings from this since the ESRAM will be shrunk as well (correct me if wrong). This gives MS room to put out a killer price or sit on nice margins.
Kind of a weird question, but can anyone clear up why the RAM speed is so important anyway? I know with PC gaming, it used to be the case (still is?) that RAM speed didn't really matter, it was like a 2 or 3 FPS difference at the most. Is the difference here because this RAM is also being used by the GPU? Does RAM speed matter more with GPU's than it does for normal system RAM?
OG_Original Gamer
Member
So then what comes first for the design team. Do they focus first on memory(i.e texture storage, a.i, sound, and other game data. Or is it the CPU and GPU, personally I don't see the eSRAM as a bandaid. Some of you make it sound like the eSRAM was a "Oh shit" moment. It's either it does what it was designed to do or not. There's benefits and negatives, I guess to some the negatives out weigh the benefits.
eSRAM is on the APU. Assuming both shrink, both will come with cost savings. Transistors are just transistors, and the Xbox One is just made of more of them with fewer dedicated to the GPU because of the eSRAM.
You should be comparing GDDR5 price drops to DDR3 price drops. GDDR5 has much more room to drop in cost as it gets adopted by more GPUs and devices while DDR3 won't be getting that much cheaper and could very likely get more expensive during the lifetime of the Xbone as fewer devices use it.
You can't perform math on data you don't have loaded somewhere the CPU or GPU can't access. RAM speed helps you get the data to where you can actually use it.
EmptySpace
Banned
esram is more expensive. gddr5 will lower its price way faster than esram, especially once gddr5 becomes standard.
Yeah, with the previous gen, you had separate dies for CPU and GPUs (and the eDRAM on the 360) and that left a lot of room for future shrinking as technology improved. Now that almost everything is already squeezed onto one die of both systems save for the main system RAM, there's not as much room for shrinking beyond the fabrication size.
What are the consoles at now? 28nm?
What are the consoles at now? 28nm?