News & Rumors: Xbox One (codename Durango)

[zupallinere]

Somehow, I have a feeling that the Xbox One is one monster machine at efficiency.

Well I have heard of things like "it's a MONSTER when it comes to fill rate" but efficiency as a spec is kind of boring On the other hand it seems to be what MS is trying to accomplish, utilizing existing resources efficiently ( at least with directx code ). A worthy goal ( XB1 in general will only be crunching directx code )

Panello claims that they won't give "30%" to the competition and I am not to sure where he gets that number ( didn't digital foundry say it was more like 20% ? ). In any case I think we can translate to say that with little in the way of optimization for each console the XB1 will be closer than 30%. A reasonable claim based on what we know I guess. We don't know all the things that the NON-XB1 ( don't want to even JOKE about making a comparison post ) does to boost efficiency let alone what will be done with the extra resources available to it.

Being "more efficient" does give you the benefit of good performance early on rather than waiting for a dev to work out how to use "extra resources" and I am sure that tailoring their tool kit to the tool itself to maximize the performance early on in this console cycle is something that MS is working hard to do. I am also assuming that MS expects that the XB1 would the multi-platform target although that can be debated and maybe a bit of a fly in the ointment.

 

[eastmen]

Ms has a trick or to left but they will be small performance increases on par with the clock speed changes.

 

[brasnial]

Ms has a trick or to left but they will be small performance increases on par with the clock speed changes.

Since posting this morning I've done a bit of reading of what Microsoft has said about there tech .
It seems to me this time around Microsoft are all about maximizing performance of each and every part of there machine .
Now one thing I read was that the xenon in the Xbox and the Xbox in general was not that efficient at getting its true power onto the screen .
Others on here with more technical knowledge may no how efficient most gpus are in general .
So if seems to me they spent there engineering budget on making the parts they where using more efficient .
In theory would such a approach allow you to use cheaper and less power hungry parts to achieve the same or very near performance of a brute force approach where there's plenty of waste .
Does this line of thinking also help you to be profitable from the word go or sooner as I read a link on here saying Microsoft plan on turning a profit day one .
Questions I no but it seems Microsoft have used the word efficiency a hell of a lot when talking about the Xbox one .

 

[bagofsuck]

Look at the HotChips slides. Top of this page:

http://www.planet3dnow.de/cms/1538-...f-der-hot-chips-25-huma-sein-oder-nicht-sein/

...first slide, an overview of the SoC.

In the top left there's a label "15 special purpose processors offload CPU and GPU" and there's a labelled arrow connecting them to the memory system. Now, either one assumes the 15 custom processors are secret and not shown on the diagram, or one appreciates that they are included. So what 'processors' are shown. Well, for starters the "Audio Processors" label includes 3 or 4 processors according to the VGLeaks audio diagram (the actual SHAPE block is called an engine there, not a processor, but the other three components are named processors). We know we have 4 processors in the memory move units. These are all grouped under that "15 processors" label in the HotChips diagram. Along with them, we have an "AV out Rsz Cmpst" (resize composite) which'll be a processor we've already heard of. Then there are "Audio DMA", "PCIe", "AV in" and "Video Encode/Decode" in the same groups of diagram blocks that we may be uncertain of as they haven't been explicitly talked about.

The sensible interpretation is that these blocks, grouped with the known processors, are the custom processors - there's no other sane way to interpret the diagram. And that there answers your question. These 15 'processors' are ancillary blocks, doing housework. They aren't game changers. They mostly alleviate some percentage of work from the CPU and GPU for them to keep doing other things without being interrupted. There's probably no way to put a percentage figure on their contribution either. The functions are also often featured in other designs - other SoCs have graphics composite processing hardware, for example, and memory units - so it's not like they are 15 unique features that no other system has. The audio processors are special, and the rest may be tweaked to be better, or fairly standard fair.

But the most important point here is that there isn't any hidden hardware that's suddenly going to make XB1 far more powerful than we currently understand it to be. There's no secret second GPU. There's no amazing 15 processors as yet untalked about that deal with physics and AI and raytracing and voxel rendering and hardware sprites and copper-list display planes. The information is right in front of us, as presented by MS themselves to the public and to their developers.

they said these 15 co processors handle physics and graphics task

 

[3dilettante]

Other than that, it's not the first time I hear about Microsoft being under NDA for the time being til October and November. According to this news they are keeping some secrets under wraps. I consider you one of the geniuses here so any information could help to know what is going on.

I've seen reports of something like that, but I don't have much to go on about the veracity or motivation for it.
One possibility is that AMD's betting the farm on HSA, which means its internal resources and input from the semicustom design business have informed the implementation of its platform. There may be some hoops everyone has to jump through, since the console business and HSA involves contributions from some very large competitors.
I'd like to imagine that there's some kind of Mexican NDA standoff, with every party in the console and HSA space pointing their legal departments at each other's heads.

It may not be possible to disclose some elements of either console without disclosing information about products AMD is going to roll out in that time frame.
Both systems seem to be HSA or very near it, and elements of their consoles contributed to what AMD's HSA devices will be using.
A next-gen graphics card or APU announcement with HSA support may be the final all-clear. For PC products, Microsoft would also be involved since I think HSA needs some handling on the OS side, and Microsoft's OS is what is most relevant for AMD's own GPU/APU products.

Further explanation on how the hardware works might explain where are the rest of those 47MB of cache -Gipsel :smile2:exposed some pretty interesting information in that regard too, but it can also be sheer technical (something I know he loves) conjecture on his part 'cos maybe they aren't accounting for audio caches and stuff-. So perhaps we'll have to wait and see.

I believe they said storage, which isn't necessarily cache. Any array that holds information could get lumped into that. There is some storage not accounted for after adding up the eSRAM, known caches, and known register files. However, Microsoft didn't detail the architecture for every offload block or other potential design tweaks.
On a potentially related note, AMD has noted in its hUMA presentations that directories or probe filters may be used to prevent unnecessary work snooping every cache. One or more snoop filters capable of covering all caches would have a non-trivial amount of storage. If there is some kind of coherence checking on the eSRAM, that too could take up a fair amount of storage on its own.

Microsoft licenses the IP from AMD (so they can take to fab themselves, pay AMD royalty (time descending) and able to maintain cost control) and owns portions of the IP in use (such as DMEs)

Can you confirm this? A fair amount of discussion about the console deals has AMD selling the chips, and AMD's margins went down because the semicustom business became part of its financials, which licensing shouldn't do.

 

[XpiderMX]

they said these 15 co processors handle physics and graphics task

Are you sure they said "physics"?

 

[BoardBonobo]

I find it interesting that for all the bluster and faux attempts to imply power; Panello has basically confirmed that the PS4 is 30% more powerful as measured by MS themselves.

 

[BRiT]

You really think MS has a PS4 at this moment in time?

Both parties are going off the paper specs and all comparisons so far only take into account a single metric at a time. None of them seem to deal with the sum of all parts.

 

[BoardBonobo]

You really think MS has a PS4 at this moment in time?

Both parties are going off the paper specs and all comparisons so far only take into account a single metric at a time. None of them seem to deal with the sum of all parts.

It was Panello who's come out and made it absolute. And I bet that both of them have been playing with each others dev kits.

 

[Shifty Geezer]

they said these 15 co processors handle physics and graphics task

Link?

 

[ERP]

It was Panello who's come out and made it absolute. And I bet that both of them have been playing with each others dev kits.

I would be stunned if MS had a Sony dev kit or viceversa.

 

[DaveNagy]

Panello never "confirmed" a performance disparity. He's implying the exact opposite.

The "internet" believes that the PS4 will be 30% (or 40%, or 100%) more capable than the XB1. A.P. is asking, rhetorically I guess, why MS would ever put themselves in a hole that deep. He's clearly implying that they would not do that, for obvious competitive reasons, and in fact have not.

Talk is cheap. I'll keep withholding judgement. Perhaps there will be further shoes to drop.

 

[BoardBonobo]

I would be stunned if MS had a Sony dev kit or viceversa.

I would be too. But I wouldn't be surprised if they have access to dev kits via friendly studios. Even if it's not officially endorsed.

 

[warb]

Yes. He's suggesting there won't be the ~30% difference you get by simply comparing peak GPU FLOPS.

We'll find out soon enough.

 

[BRiT]

It was Panello who's come out and made it absolute. And I bet that both of them have been playing with each others dev kits.

People read way too much into the comments being said.

The only sane choice is "Both parties are going off the paper specs and all comparisons so far only take into account a single metric at a time. None of them seem to deal with the sum of all parts."

 

[BoardBonobo]

People read way too much into the comments being said.

The only sane choice is "Both parties are going off the paper specs and all comparisons so far only take into account a single metric at a time. None of them seem to deal with the sum of all parts."

Not necessarily the sanest option since it's measurable on paper. More like the safest option if you if you want it that way. ;-)

 

[XpiderMX]

Not necessarily the sanest option since it's measurable on paper. More like the safest option if you if you want it that way. ;-)

Some people (here in B3D) have explained that flops are not the only way to measure performance.

 

[Cjail]

@BoardBonobo

Sony has sent X devkits to X developers and I sincerely doubt they are not keeping track of them and that MS managed to get his hands on one of them thorough "unofficial" channels.

 

[zed]

I would be stunned if MS had a Sony dev kit or viceversa.

Not sure if serious.
Hint -it's a multi billion dollar industry and I assume the Dev kits have been in a lot of third party hands already

 

[golem]

Link?

Hes probably referring to the wording on top

I dont see anything specific about physics (and it seems most of the coprocessors are covered under the wording in the diagram)