News & Rumors: Xbox One (codename Durango)

Status
Not open for further replies.
I would think the more concerning thing right now for MS is the fact that the difference between 720P vs 1080P is loser to 100%, much larger than the ALU deficit, and closer to the ROP deficit.

The engines in use for these early games aren't designed around the limited size of the ESRAM, and the prevalence of deferred renderers probably isn't helping that, which likely explains some of deficit, but I would be concerned that the virtualization of the GPU is introducing significant overhead, or the limited ROPs are an issue.

Having never worked on an XB1, I would assume that juggling the 100MB's of render targets most modern games use wouldn't be a huge problem, and that you should be able to get 80% of the way to optimum relatively easily, but maybe it's harder than I imagine, especially considering the launch timeframe.

It would be interesting to know why various tradeoffs were chosen and what buffers were put where, but I'd guess we'll never know.

Do you think there is any chance for whatever reason, they ended up putting their render targets on ddr3 thus the ROPs are being bandwidth starved, or do you think that even if they managed to fit the the most ROP intensive target fitting on esram they might still get fillrate issues?
 
However, if they are going to decrease the number of pixels on multiplatform games when creating the Xbox One version, I hope that developers take advantage of that and add more quality per pixel.

+1

I don't think we've maxed out pixel fidelity at 720p yet so I'm hoping they take advantage of the "lower resolution" if they must go that route.
 
Last edited by a moderator:
Even with the price 25%+ higher? That's the most noteworthy reason to stick to discs IMO; DD sells at RRP and discs can be got for much cheaper, especially if one's happy to wait. But also these consoles can only fit a dozen or so games. If you remove a game and then want to revisit it later, you'll have to redownload the thing if you don't have the disc to just pop in.

Yeah. I'd pay the extra to have the convenience as long as I had large enough bandwidth.
 
Do you think there is any chance for whatever reason, they ended up putting their render targets on ddr3 thus the ROPs are being bandwidth starved, or do you think that even if they managed to fit the the most ROP intensive target fitting on esram they might still get fillrate issues?
I'm not a dev but I was wary at the time DF issued its paper about how the ps4 and XO could compare based on PC hardware. I'm not sure I posted my opinion as I discard posts quite often and I remembered reading something here and I could not find the source/poster, no quote no post.

Anyway what I had in mind was something implying that you are not that often blending limited (that is part of the post I ciuld not find), in a more vague manner I wonder the circumstances under which the ROPS become heavily dependant on external bandwidth (vs cases when bandwidth amplification throught the dedicated cache does it work).
I don't know (so I ask for insights) but a hypothesis could be that actually whereas the Rops in the ps4 don't have enough andwidth to achieve their max throughput all the time , in practise they do quite often and when it comes to filling various render targets they indeed act twice as fast as 16 Rop even more properly fed aka in alot of case they are not bandwidth constrained.

Now I could not find the post back in time and may be it was a good thing, I won't say who I believed posted this as I could be wrong.

Either way may be they were dead set on locked 60fps and so decided to "over use" the edram trying to fit here RT that could have remained in the main ram at the cost of lesser perfs under stressful scenarii.
 
Last edited by a moderator:
I don't know (so I ask for insights) but a hypothesis could be that actually whereas the Rops in the ps4 don't have enough andwidth to achieve their max throughput all the time , in practise they do quite often and when it comes to filling various render targets they indeed act twice as fast as 16 Rop even more properly fed aka in alot of case they are not bandwidth constrained.


What I don't understand about the ROPs is basically what are the bandwidth requirements? (I know this has been discussed here before, but I can't seem to find it now). Whatn I look at a 7850 GPU, it has 32 ROPS with 152 GB/sec BW. While the 7970GHz Edition also has 32 ROP/s but with 288 GB/sec. So either the 7850 is BW bound on the ROP's, or the 7970 is ROP "compute" bound. Which is it? Or are they simply not that important once you're past some filll rate?
 
It would come down to what the ROPs are being tasked to do, and how much of the workload is dominated by it.

ROPs have their own caches which allow for effective bandwidth higher than external numbers would suggest. The leaked Durango docs actually mention optimizing for this case as well.
The number of ROPs, and by extension the caches doing the amplification, would figure more strongly the more a given operation can use them.
 
I'm not a dev but I was wary at the time DF issued its paper about how the ps4 and XO could compare based on PC hardware. I'm not sure I posted my opinion as I discard posts quite often and I remembered reading something here and I could not find the source/poster, no quote no post.

Anyway what I had in mind was something implying that you are not that often blending limited (that is part of the post I ciuld not find), in a more vague manner I wonder the circumstances under which the ROPS become heavily dependant on external bandwidth (vs cases when bandwidth amplification throught the dedicated cache does it work).
I don't know (so I ask for insights) but a hypothesis could be that actually whereas the Rops in the ps4 don't have enough andwidth to achieve their max throughput all the time , in practise they do quite often and when it comes to filling various render targets they indeed act twice as fast as 16 Rop even more properly fed aka in alot of case they are not bandwidth constrained.

Now I could not find the post back in time and may be it was a good thing, I won't say who I believed posted this as I could be wrong.

Either way may be they were dead set on locked 60fps and so decided to "over use" the edram trying to fit here RT that could have remained in the main ram at the cost of lesser perfs under stressful scenarii.

I understand you are saying, but assuming a scenario where there aren't many blending going on, wouldn't that mean you mostly draw stuff once and so you are not ROP limited either?

Other than blending, how else could you be severely ROP bound, in ways that you are not bandwidth bound? I would guess shadow maps and filling up all the render targets of a deferred render, but it's really just a guess XD
 
*AHEM* This is the Xbox One thread. This is not the Sony or PS4 thread. B3D still isn't allowing for Versus talk between the 2 next-gen consoles.

Don't force me to nuke the entire console forums from orbit because the rampant inability to stay on topic or follow current rules.
.
 
Other than blending, how else could you be severely ROP bound, in ways that you are not bandwidth bound? I would guess shadow maps and filling up all the render targets of a deferred render, but it's really just a guess XD

That's pretty much the issue I would expect, shadow maps and the G Buffer.

I should note that it's not unusual for a deferred renderer to have say 28 bytes/pixel in the G Buffer, that's 56MB's of data at 1080P right there. At 720P the G Buffer barely fits and that might be what's driving the resolution deficit.

There are ways to avoid buffers that big even without tiling, but you'd have to work with the 32MB constraint, and design around it which just may not have been possible in the launch time frame.

I should also note if the issue is just the G Buffer memory, that the additional slow downs people are seeing could be attributed to additional bandwidth limitations because engines are rendering other render targets to main memory, or just CPU issues, Sony's low level API has some significant work put into managing the overhead of constant updates that is probably not available in what ever DX11 variant X1 is running, and that could greatly reduce draw call overhead.
 
ERP, I thought that shadow maps would be hugely improved using Tiled Resources, which sound like a perfect fit for the eSRAM. The ROPs, I can see where you come from...

Even so, I don't get why they can't be used like in GPUs with excellent performance / price ratio, like the Bonaire, which comes close to superior GPUs -at least according to the paper numbers- that feature 32 ROPS vs 16 ROPs of the Bonaire.
 
News:

Edge recommends the Xbox One as the superior multimedia box.

http://www.edge-online.com/features/ps4-vs-xbox-one-why-microsofts-con​sole-is-the-superior-multimedia-box/

Talking of which, Xbox One is going to support streaming via DLNA

http://www.neogaf.com/forum/showthread.php?t=707051






5xkuDx3.jpg


The Play To option is something I am accustomed to use...

http://www.product-reviews.net/2013...e-review-lacks-depth-wont-influence-decision/

3dilettante, I will send you a brief pm about where I get the news and the reason of that mistake with the date.
 
That's pretty much the issue I would expect, shadow maps and the G Buffer.

I should note that it's not unusual for a deferred renderer to have say 28 bytes/pixel in the G Buffer, that's 56MB's of data at 1080P right there. At 720P the G Buffer barely fits and that might be what's driving the resolution deficit.

There are ways to avoid buffers that big even without tiling, but you'd have to work with the 32MB constraint, and design around it which just may not have been possible in the launch time frame.
I see... Now I'm curious to see what changed in the Cod engine compared to this gen... It used to be a forward render, right? So it should fit on esram with some AA even without any tiling... And from what I've seen, other than smoke grenades it's not all that heavy on blending either... Is there any way the CPU could be the bottleneck but reducing the resolution could actually increase the framerate?

ERP, I thought that shadow maps would be hugely improved using Tiled Resources, which sound like a perfect fit for the eSRAM. The ROPs, I can see where you come from...

It will probably improve, but I guess it's not automatic...
 
News:

Edge recommends the Xbox One as the superior multimedia box.

http://www.edge-online.com/features/ps4-vs-xbox-one-why-microsofts-con​sole-is-the-superior-multimedia-box/

Talking of which, Xbox One is going to support streaming via DLNA

http://www.neogaf.com/forum/showthread.php?t=707051






5xkuDx3.jpg


The Play To option is something I am accustomed to use...

http://www.product-reviews.net/2013...e-review-lacks-depth-wont-influence-decision/

3dilettante, I will send you a brief pm about where I get the news and the reason of that mistake with the date.

Where do they get DLNA from all that?
 
Where do they get DLNA from all that?
"Allow Play To " streaming. When you use the DLNA in Windows for instance, right-clicking on a file gives you the option to "Play To" (I mean playing that file to...another device) one of the devices in your Homegroup.

My TV accepts DLNA for instance and when you connect it to a Wi-Fi network for the first time, one of the options that appear afterwards let you use it as an open device other devices can detect and share content with. If you don't change that option the TV will remain invisible for other devices though.
 
"Allow Play To " streaming. When you use the DLNA in Windows for instance, right-clicking on a file gives you the option to "Play To" (I mean playing that file to...another device) one of the devices in your Homegroup.

My TV accepts DLNA for instance and when you connect it to a Wi-Fi network for the first time, one of the options that appear afterwards let you use it as an open device other devices can detect and share content with. If you don't change that option the TV will remain invisible for other devices though.

Maybe it is WiFi-Direct/Miracast option.
 
Having to right click files on a Windows PC to play them on your console isn't exactly what most people are looking for with DLNA support.
 
Having to right click files on a Windows PC to play them on your console isn't exactly what most people are looking for with DLNA support.

It is only a way to play DLNA content. My BluRay support "play to", but it support "play from" too (browsing sharing folders).

My Lumia 920 can "play to" Xbox 360, and it is nice because I can play what I want and I don't need to browse folders and folders.
 
I understand you are saying, but assuming a scenario where there aren't many blending going on, wouldn't that mean you mostly draw stuff once and so you are not ROP limited either?
Without proper occlusion culling you draw everything within FoV even if they are behind another objects.
 
Having to right click files on a Windows PC to play them on your console isn't exactly what most people are looking for with DLNA support.

Play To (renamed Play in 8.1) is also part of the Devices Charm in all versions of Windows 8. Means I can throw any media content that I've opened on my tablet to the tv with a swipe and two taps. Even if that content is not present on the tablet, since all variants of Windows 8 have support for connecting to homegroups and other network locations. It's extremely useful.
 
Status
Not open for further replies.
Back
Top