News & Rumors: Xbox One (codename Durango)

[function]

FWIW my expectation is that PS4 ought to have a performance advantage, but I wouldn't expect it to reflect the difference in CU counts. CU's are MASSIVELY underutilized on vertex heavy workloads and plenty of the frame will be ROP or bandwidth limited.

Wouldn't 1Bone have the advantage of a having fewer CUs for each triangle setup engine? Shouldn't that increase CU utilisation on vertex heavy workloads?

 

[Rangers]

I found the last Penello post (for now) on GAF apparently. If it's too versus, delete, but I thought one (bolded) passage very interesting.


http://www.neogaf.com/forum/showpost.php?p=80999997&postcount=1163

Basis is this:

1. Personally, I've seen both systems and games live at E3, Gamescom, and PAX. Played a little bit of PS4 games at PAX. There are great looking games on both systems. Ryse looks great, Forza is fantastic. I think Killzone looks awesome, and honestly have not been impressed with DriveClub (although it came a LONG way since E3). I really would like to see Infamous in person, but it hasn't been shown that I've seen. However, any difference between those games is subjective of course and subject to viewer bias, but looking at both systems it's hard to substantiate any claims around 40% - 50%

2. We're in the final stages of game and SW development now. We're working closely with 3rd parties as we approach launch. Reports we're hearing back from developers consistently are confirming that cross-plat games are running the same on both platforms.

I believe there is a point back near E3 where developers would have said their games were running better on PS4. I think Marc Whitten made the point that we'd just completed some driver work about a month ago just before Gamescom. And look at the frame-rate improvements in DR3 between Gamescom and PAX. We're making huge strides in our SW stability, and again, we have customized and balanced the system to reduce bottlenecks and optimize performance, in ways that aren't seen in the published specs.

As to the follow-up questions; I posted earlier I'm working with the engineers directly who I think (and you guys think) will be more credible than I am. I did say that I went to them for some supporting points which I posted here - so at this point it's probably best to let them speak on it. I'm not sure when, but more detail will show up.

It's late, and it's clear I'm doing more harm then good at this point. So best to let this rest from my seat and get back to you guys with the most credible source.

 

[Esrever]

Is there any "uncore" part of the gpu besides the CUs and ROPs and memory controller that would benefit from a overall clock increase? Things like geometry engines and such maybe and the command processor?

 

[Rangers]

Is there any "uncore" part of the gpu besides the CUs and ROPs and memory controller that would benefit from a overall clock increase? Things like geometry engines and such maybe and the command processor?

All of it, right?

 

[Esrever]

All of it, right?

I just wonder if having a slight 6% advantage there would mean anything.

 

[rokkerkory]

I found the last Penello post (for now) on GAF apparently. If it's too versus, delete, but I thought one (bolded) passage very interesting.


http://www.neogaf.com/forum/showpost.php?p=80999997&postcount=1163

Hmm I am a bit impressed he seems pretty straight forward and not wanting to spout anything that may cause harm. That's refreshing for MS which has had terrible PR since May.

Hopefully he will be able to provide some more details from the SMEs.

 

[3dcgi]

Going OT, but doesn't that somewhat suggest a return to discrete vertex shaders would be better silicon utilisation? Or can we adjust vertex shader workloads to do more stuff? Or can compute fill in the blanks?

Even if vertex work doesn't utilize all of the ALU horsepower it can still saturate memory bandwidth and for that it makes sense to keep things unified. Plus, if you have discrete vertex shaders that's area taken away from the main compute pool. Yes, compute and pixel work can fill in the blanks if there's work to be done. Async compute makes it easier to fill in the blanks.

Wouldn't 1Bone have the advantage of a having fewer CUs for each triangle setup engine? Shouldn't that increase CU utilisation on vertex heavy workloads?

That's not an advantage. Just not the disadvantage that the CU difference makes it seem for vertex heavy workloads.

 

[function]

That's not an advantage. Just not the disadvantage that the CU difference makes it seem for vertex heavy workloads.

Sorry, you've lost my dumb ass.

I wasn't referring to an absolute performance advantage relative to PS4 as both have 2 triangle setup units. I was talking about an advantage in terms of keeping CUs busy (see ERPs post I was replying to). Triangle setup seems to scale with CU count on larger GPUs, and I'm guessing that there is a balancing act where you decide on a typical workload for a GPU over it's competitive lifespan and set the CU to triangle setup/rasterisation rates accordingly (taking into account die area and power consumed etc).

It's interesting that MS have gone for so few CUs relative to triangle setup in an almost "2 x APU" way (while Sony have gone for more of a "discreet GPU" ratio).

Is this simply a quirk of how the systems were developed or does it hint at a difference in how they see GPU workloads evolving over the coming years?

 

[shredenvain]

Is there any "uncore" part of the gpu besides the CUs and ROPs and memory controller that would benefit from a overall clock increase? Things like geometry engines and such maybe and the command processor?

The geometry engines would gain a 6% polygon per second increase

 

[temesgen]

Not at all.

What exactly is the problem with his response?

 

[kots]

Well , in 2+ months , the truth will be out ... it' ll be soon and a lot could change , but we will have a general image where things are going .
Then we'll see if Penello was correct or everything was just PR .

 

[Johnny Awesome]

I personally determined my own point of reference at 20dbA, because that's what I get from my video projector above my head, I hear it if I concentrate on it, but it's quiet for me. I don't see the Xbone being much less than 30dbA at 1m when playing a AAA title. But I could be wrong.

You very well could be.

 

[Rangers]

Well , in 2+ months , the truth will be out ... it' ll be soon and a lot could change , but we will have a general image where things are going .
Then we'll see if Penello was correct or everything was just PR .

Yeah, we are in the weirdest time like I said. So much more will be known in just weeks, but right now all is excruciatingly quiet.

I hope we start getting major new infos (like leaked gameplay of games like BF4 actually running on next gen console) by late October. That's still over a month away, sadly.

 

[babybumb]

3rd party always like to say versions are running the same when thats not really true in digitalfoundry level inspection. They pretty much ignore resolution difference or FPS

Rockstar will easily claim Red Dead is "the same" on PS3/360

 

[Johnny Awesome]

Nevertheless, even if the performance difference in real world games was 30%, does it even really matter? I don't think so.

I'm pretty happy with the graphics performance of launch titles like Forza 5, Ryse, and DR3, from what we've seen and it can only get better from here. Last gen we went from Perfect Dark Zero to CoD: Ghost. It's night and day.

In the end, the worst case is that 3rd parties will make some resolution tweaks and minor adjustments and the games will be very close on Xbox One.

 

[XpiderMX]

3rd party always like to say versions are running the same when thats not really true in digitalfoundry level inspection. They pretty much ignore resolution difference or FPS

Rockstar will easily claim Red Dead is "the same" on PS3/360

Well, I guess it is not the same what 3rd party say to MS than what they said to media (web sites).

 

[BoardBonobo]

Nevertheless, even if the performance difference in real world games was 30%, does it even really matter? I don't think so.

I'm pretty happy with the graphics performance of launch titles like Forza 5, Ryse, and DR3, from what we've seen and it can only get better from here. Last gen we went from Perfect Dark Zero to CoD: Ghost. It's night and day.

In the end, the worst case is that 3rd parties will make some resolution tweaks and minor adjustments and the games will be very close on Xbox One.

But what if that 30% means that this is as good as it gets? Nothing will ever look better on this platform?

 

[Scott_Arm]

But what if that 30% means that this is as good as it gets? Nothing will ever look better on this platform?

Like last gen, software will always find new and better ways to do things. The games at the end of the gen are likely to look better than the games at the start. And by that logic, every new platform would have the same problem. Hardware doesn't get upgrades after launch. It's static. So if it was the opposite case, being 50% better, you could say nothing will ever look better on the platform just as easily.

 

[MrFox]

You very well could be.

I certainly could. But I try to guess a verifiable number, otherwise there's no fun in guessing.

30dbA is my guess, 25dbA wouldn't be surprising, 20dbA would be surprising to me, My expectations are similar for PS4. It's going to be interesting to see if the sound ramps up after hours of game play, or how much the difference between 20C and 30C temperature affects the fan speed (it was a huge difference with the PS3, for example).

...but I doubt anyone will ever make a proper measurement in a controlled environment, nobody ever did during PS360 generation. We'll get the usual internet posters with anecdotes ranging from a butterfly wing beat to a lawnmower. Psychoacoustics is a mysterious thing.

 

[Rangers]

Penello on Neogaf said he was going to get some engineers together and answer a lot of the tech questions/rebuttals to his tech posts/trolling from GAF.

So cant wait for that. Unfortunately he said it may take a "few weeks" and they havent decided what format (he thinks neogaf is a bad format, threads move too fast/his replies get lost).

A conspiracy theorist might think "few weeks" is to wait for Volcanic Islands NDA to end. More likely as someone posted, it's to get approval from various departments to talk about various infos.

Anyway it definitely appears Penello/Microsoft wants to do more tech "deep dives" as he calls them. Just a question of when they happen.

Albert Penello:

I was quickly corrected (both on the forum and from people at the office) for writing the wrong number.

The challenge with the NeoGAF format is that, because threads move so fast, posts disappear or get buried so people aren't reading everything. Maybe this part got lost:

I've stated - there is no possible way for one single person to know every detail about our platform. That means I need to go get the answers to the questions you guys ask sometimes. There's a lot I know first hand, and a lot I need to get updated on.

So people understand - I'm not dodging any of the follow-up. I actually stated the other night - there are a handful of people who asked some really legitimate follow-ups to understand what I posted. And I honestly said - I'm not the guy to answer at that level of detail. Out of respect for the people who are genuinely curious to learn how we derived those numbers, and to get the most technically accurate answers - the best course is to have the answers come from those engineers directly.

So we're working now on the best format to do that.

I still stand by what I stated (except for the aforementioned 204/218). In fact, some really interesting threads were going back and forth giving me even more excruciating detail behind those numbers based on the questions people asked.

I doubt it will take the format of an AMA, but I've collected a bunch of the follow-up questions. It may take a few weeks, but we'll be following-up on this for sure.