They said in the article that the GPU clock "remains" at 800mhz, which would re-confirm the 1,23TF for a GCN-based GPU. The only other way to bump the raw FLOP-count would be to add more CUs which would imply a redesign of the APU.
-
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
They said in the article that the GPU clock "remains" at 800mhz, which would re-confirm the 1,23TF for a GCN-based GPU. The only other way to bump the raw FLOP-count would be to add more CUs which would imply a redesign of the APU.
no, not at all.. nowhere near that big of a difference. more than likely ALL games will run at 30fps targets (people hoping for 60fps in most games are going to be majorly disappointed) and eventually you'll see differences closer to the middle of this gen (i.e. when PS3 porting stopped sucking and differences were mostly down to framerate, resolution, vysnc, AA, etc)
"To the best of their knowledge"
SEGAvangelist
Member
Oh this will go another 10 pages at least.
cebri.one
Member
Tallshortman
Member
You're both basing your analyses on leaks, yours is just as flawed, not sure what you're trying to claim here.
dragonelite
Member
You say that like it wont matter.
Take a PS2 game like FF12.
Upscale it to 1080p, and bam. Textures come alive, you can see way better off in the distance, low aliasing. The game does become better to play
Clear
CliffyB's Cock Holster
I think its funny seeing Leadbetter touting the benefits of the SHAPE audio processor in Xbone after pointedly ignoring the superior sound output of the PS3 for an entire generation of Face-offs.
And sorry, but the whole article reeks of spin.
Any difference is going to show in precisely the same way that its been for the last generation: peaks of load will highlight bottlenecks in the system.
And sorry, but the whole article reeks of spin.
Digital Foundry said:
Any difference is going to show in precisely the same way that its been for the last generation: peaks of load will highlight bottlenecks in the system.
Which discussion? This thread is rather Jumping-to-conclusions-that-fit-my-agenda from both sides. It's embarrassing and people that are really familiar with the matter probably died already from laughing after reading this thread. I would love to see people at least outing themselves as devs (we don't need names or studios) when contributing to these threads. (If they do post here anyway) this way we would know, who is talking shit.
Upscaling does nothing to textures or even overall image quality, re-render in higher resolution and you get new information.
In case of emulators and FF12 you also get a lot better texture sampling quality on top of the resolution.
Also, good news for xbox.
LittleJohnny
Banned
wow what happened in here, the first couple of pages are people going crazy and saying ridiculous things.
Its good that they have managed to bump up the bandwidth of the esram on the XBO this could mean slightly improved performance if a developer was to take advantage of it. But dont be mistaken this is no paradigm shift. the esram still only makes up 32MB of the XBO memory and while increased bandwidth on this will help certain operations, not everything will benefit. People saying this vastly reduces the gap between the two consoles are delirious at best.
this is good! but don't let hyperbole grab you!
Its good that they have managed to bump up the bandwidth of the esram on the XBO this could mean slightly improved performance if a developer was to take advantage of it. But dont be mistaken this is no paradigm shift. the esram still only makes up 32MB of the XBO memory and while increased bandwidth on this will help certain operations, not everything will benefit. People saying this vastly reduces the gap between the two consoles are delirious at best.
this is good! but don't let hyperbole grab you!
In addition, The PS4 has a dedicated audio processor as well. I guess the only substantial difference is that the XBox' hardware's "fancy" name got public through leaked documents.
Call me simple, but how does "really fucking fast" access to 32MB or RAM make a big difference if you're, say, dealing with way more memory than 32MB? It seems like if you're reading/writing to memory outside of that 32MB, it'd be slow again once you're outside of the super-fast RAM that can now simultaneously read/write under the correct circumstances...
I just don't see how 32MB of even "scientifically proven absolutely instantaneous" RAM could speed up the rest of the memory access to the point that we could be talking comparable overall performance. I keep picturing a Ferrari trying to get through mid-town Manhattan and hitting every single red light, versus the bike messenger blowing through intersections like it ain't no thang. From light to light, that Ferrari is fast... but once it hits that, overall performance favors the guy that just keeps on going at a steadily fast pace (albeit in this analogy, the bike messenger/PS4 himself is really friggin' fast to begin with.)
Is it really "nearly as fast" access to 32MB of RAM on the Xbox One comparing to "still faster" access to all 8GB of RAM on the PS4 that people are saying should be comparable?
I just don't see how 32MB of even "scientifically proven absolutely instantaneous" RAM could speed up the rest of the memory access to the point that we could be talking comparable overall performance. I keep picturing a Ferrari trying to get through mid-town Manhattan and hitting every single red light, versus the bike messenger blowing through intersections like it ain't no thang. From light to light, that Ferrari is fast... but once it hits that, overall performance favors the guy that just keeps on going at a steadily fast pace (albeit in this analogy, the bike messenger/PS4 himself is really friggin' fast to begin with.)
Is it really "nearly as fast" access to 32MB of RAM on the Xbox One comparing to "still faster" access to all 8GB of RAM on the PS4 that people are saying should be comparable?
Phire Phox
Banned
DBZ power level?
From what I know - could be entirely wrong:
The ESRam is more or less handled as a cache - so while you are pumping memory from the ESRam to the GPU or vice versa you can fill the ESRam with data from the DRam. (Note that this is done in a split of milliseconds) so while the GPU is busy in one cycle you can "refill" the ESRam from the DRam to have this data accessible in the next cycle from the "faster" memory pool.
LittleJohnny
Banned
102GB/s was also confirmed, amirite?
I don't recall any speak of TFlops performance, link?
Insane Metal
Gold Member
Where's cboat? We need to know!!
Caayn
Member
Running speed =/= acces speed
For example the GDDR5 is faster but it has a higher latency. While the DDR3 is slower but has a lower latency.
102GB/s was also confirmed, amirite?
I don't recall any speak of TFlops performance, link?
They talked ops per cycle - from knowing the GCN architecture, you can deduct the number of CU and thus the Flops count. So unless there is something missing in that picture, the 1.2TFlops are 99% confirmed.
http://www.eurogamer.net/articles/digitalfoundry-spec-analysis-xbox-one
The FLOP-count of the PS4-GPU has been calculated the same way.
the cpu in both consoles is the same isnt it?
as I understand they are trying to make devs use GPGPU processing to offload task's to the gpu and leave the cpu to general game logic
No - this would be same for the PS4(rumor is, that both have the same CPU)so why should sony build a 1.8Tflops GPU if they could only use 1.2 of it?
Iced_Eagle
Member
Would you happen to have a link to that? I'm curious to know the math involved, and my google-fu isn't working.
*Edit* Nvm, I was beaten with the math. Interesting.
http://www.eurogamer.net/articles/digitalfoundry-spec-analysis-xbox-oneWould you happen to have a link to that? I'm curious to know the math involved, and my google-fu isn't working.
Not really. Latency is a result of the way the memory controller is designed to work, an memory controllers tend to work differently on CPUs and GPUs. In general, there is no inherent difference in latency between the two.
dragonelite
Member
Already posted in the first 3 pages or so
Here
So if we update the console power to DBZ level..
PS4:
XBone:
WiiU:
Wait for what?
All of this would end if one of the leakers would leak some test case numbers.
Technically, you are right, it could be on both platforms. However both designs are made with a bunch of assists that not as much needs to be thrown to the CPU.
What's that supposed to mean? Feel free to pm if you don't want to discuss this puplicly ;-)
graywolf323
Member
something about those numbers from Microsoft just seems off
really surprised Digital Foundry seems to have just taken them at face value
really surprised Digital Foundry seems to have just taken them at face value
Haha no. Either MS or Sony shill depending on what the numbers would show.
Caayn
Member
Yes there is. The faster the memory the more clocks it takes before it can respond.
102 * 2 != 192 - that's what's off...
Can Crusher
Banned
Maybe the console was designed by Seth Rogan. "Oh shit man.. look at this, didn't even notice it was there before"
Kung Fu Grip
Banned
This is what i was trying to say earlier.
I seriously don't get the hype for this. And now some are thinking the GPU will be better too.
Sephiroth_VII
Member
Okay, so reading all through this.
It looks like we're talking about an optimization, one of those things you can do on consoles, when you know exactly what hardware you'll be working with, and can take advantage of behavioral quirks.
It also looks like we're talking more along the lines of bursts of speed, rather than sustained speeds. That is helpful for some tasks, but between that, and the fact that this appears to only be possible on the ESRAM, means that this will only be of limited utility for game development.
The big problem I believe the X1 is going to have, are idle clock cycles while waiting for information stored on the DDR RAM. To be fair, we're probably not talking about a huge number of idle clock cycles, but they'll probably be somewhat more frequent on the X1 than on the PS4. Only so much data can be stored on the ESRAM after all, whenever the APU needs data that isn't on the ESRAM, there'll likely be a marginal delay resulting in an idle clock somewhere. Now, we're not talking about an immense bottleneck here, no system is going to be entirely free of the occasional idle clock cycle, the 360 certainly wasn't, and that did just fine. However, as a result of this, the X1 set up should be just a tiny bit less efficient, on top of slightly more difficult to develop for.
It's also worth noting, that it isn't altogether unlikely that a similar optimization will be found for the PS4 at some point. I cannot say how likely, as I do not know which component this optimization can be attributed to. As many components as the two share however, there's a decent chance that something similar can be achieved there.
Now of course, full disclaimer, I'm not actually an expert on such things. I've been apart of conversations like these long enough that I've picked up some knowledge of the subject. Others will have to confirm whether or not my read on on the situation is correct however.
It looks like we're talking about an optimization, one of those things you can do on consoles, when you know exactly what hardware you'll be working with, and can take advantage of behavioral quirks.
It also looks like we're talking more along the lines of bursts of speed, rather than sustained speeds. That is helpful for some tasks, but between that, and the fact that this appears to only be possible on the ESRAM, means that this will only be of limited utility for game development.
The big problem I believe the X1 is going to have, are idle clock cycles while waiting for information stored on the DDR RAM. To be fair, we're probably not talking about a huge number of idle clock cycles, but they'll probably be somewhat more frequent on the X1 than on the PS4. Only so much data can be stored on the ESRAM after all, whenever the APU needs data that isn't on the ESRAM, there'll likely be a marginal delay resulting in an idle clock somewhere. Now, we're not talking about an immense bottleneck here, no system is going to be entirely free of the occasional idle clock cycle, the 360 certainly wasn't, and that did just fine. However, as a result of this, the X1 set up should be just a tiny bit less efficient, on top of slightly more difficult to develop for.
It's also worth noting, that it isn't altogether unlikely that a similar optimization will be found for the PS4 at some point. I cannot say how likely, as I do not know which component this optimization can be attributed to. As many components as the two share however, there's a decent chance that something similar can be achieved there.
Now of course, full disclaimer, I'm not actually an expert on such things. I've been apart of conversations like these long enough that I've picked up some knowledge of the subject. Others will have to confirm whether or not my read on on the situation is correct however.
In a few hours? What's going to happen? (PM?)
Just seems like a slow CPU for gaming.
What are you talking about?
Crimson_Gold
Member
Please share
Interfectum
Member
How many TFLOPs would Fulgore need to function in real life?
If you compare them to PC gaming CPU, of course.
- Status