Xbox One (Durango) Technical hardware investigation

[keenism]

It will be interesting to see if Microsoft is really willing to leave all that achievable clock speed on the table without taking advantage of it.

Assuming this "achievable clock" exists, you still have to remember...this isn't a discreet gpu. Further yet it's inside a console which has to function under a range of conditions(you wouldn't believe the amount of crap that accumulates over time in these things in some places) and placements reliably and consistently...(I think that's the goal this time? )

Realistically, more GPU speed = more heat/voltage/stress on other components = higher fail = money..

 

[Reiko]

Assuming this "achievable clock" exists, you still have to remember...this isn't a discreet gpu. Further yet it's inside a console which has to function under a range of conditions(you wouldn't believe the amount of crap that accumulates over time in these things in some places) and placements reliably and consistently...(I think that's the goal this time? )

Realistically, more GPU speed = more heat/voltage/stress on other components = higher fail = money..

MS likes to burn money if it reaches their end goal.

 

[SenjutsuSage]

800 mhz makes sense as it is half the clock of the CPU cores (1.6 Ghz).

Why does it need to be half the clock of the CPU cores? Does it have something to do with APUs or HSA?

Assuming this "achievable clock" exists, you still have to remember...this isn't a discreet gpu. Further yet it's inside a console which has to function under a range of conditions(you wouldn't believe the amount of crap that accumulates over time in these things in some places) and placements reliably and consistently...(I think that's the goal this time? )

Realistically, more GPU speed = more heat/voltage/stress on other components = higher fail = money..

It isn't like we are referring to an overclock. These are apparently safe clock levels for this GPU. It shouldn't lead to unreliability and failures. Also, do we know that Durango is using an APU?

 

[McHuj]

Why does it need to be half the clock of the CPU cores? Does it have something to do with APUs or HSA?

It doesn't have to be. And current AMD APU's don't follow that model, but I think power savings can be achieved if all the clocks can run from a common clock.

IMO, everything about Durango (to me at least) seems to be about finding the optimum point of performance/watt and performance/$$$ with both power consumption and cost being lower than the original 360.

I also think that the system is probably carefully balanced in terms of memory bandwidth, I'm not sure if over clocking the CPU or GPU would do much good if the memory bandwidth needs to be scaled up as well. According to spec, the memory looks to be 2.133 DDR3, could they (or would they need to) go to 2.4 DDR3 easily and cost effectively?

 

[SenjutsuSage]

Yea, they are clearly putting a high premium on finding the perfect balance between cost, performance and power consumption, without a doubt. Honestly, unless there's something I'm missing, they've found a pretty damn good balance. I still find it hard to think they wouldn't go for the higher gpu clock, though.

Perhaps, though, due to the choices they've made elsewhere, there might be even less of a reason for them to specifically need to count on the power savings from all clocks running from a common clock, thus allowing them to untether the gpu core clock, taking advantage of what a GPU within that range should be capable of.

 

[astrograd]

Are 12 or 14 Bonaire CUs at 1ghz inherently hotter than 18 Pitcairn CUs at 800mhz though? Point is we really dont know how hot either will be but i think we can safely assume that MS will not make an unreliable console, nor will they make one that sounds like a vacuum cleaner.

The question is if they weren't already planning on this chip at 1ghz, can they change it now? I think if Sony is able to double their memory so late in the game,which likely required them to rework their mainboard layout, then certainly MS can explore a more robust cooling solution (if such a change would even be necessary). Chances are they already looked at a number of possible cooling solutions throughout this process and its just a mixing and matching the parts they would now need.

I would imagine that they would plan ahead on raising the clock and then could go back on that if they felt they didn't need that fast a chip. Seems smarter to assume you'll need to go all out and pare back as necessary instead of putting yourself in a position to actually react.

 

[GrimThorne]

It's not a midrange 7000. There's a whole system that had to be designed around a long-term business plan, including evaluations of multiple designs. They had to come up with a solutions for substantial RAM without blowing the bank on GDDR5, which meant designing an eSRAM system and the DMEs and whatnot. They may also be designing the hardware with an eye on cross0device APIs for supporting Windows, so had to match Durango's design and ongoing API development to Surface and Windows 8 designs and APIs.

Sorry, meant to say 7700 series. And we're looking at the VGleaks specs right? That's either 7770 or 7790, both are mid-range performers. I hear what you're saying though, but it's still hard to imagine that Durango was the best that this team of engineers could come up with. It's just such obviously low hanging fruit.

 

[McHuj]

Sorry, meant to say 7700 series. And we're looking at the VGleaks specs right? That's either 7770 or 7790, both are mid-range performers. I hear what you're saying though, but it's still hard to imagine that Durango was the best that this team of engineers could come up with. It's just such obviously low hanging fruit.

I would say that you can't say that unless you know what criteria they were designing for. I think cost, manufacturability, power consumption, etc may have been more important than just raw performance.

 

[SenjutsuSage]

I would imagine that they would plan ahead on raising the clock and then could go back on that if they felt they didn't need that fast a chip. Seems smarter to assume you'll need to go all out and pare back as necessary instead of putting yourself in a position to actually react.

Exactly. It's pretty inconceivable that they would have boxed themselves in from the very start at some predetermined specification. More than likely, they had a number of plans that they've either altered somewhat or moved away from entirely since they've been planning for the next Xbox.

Obviously it's to be expected they're probably well past the point now where they have a favorite design, and that favorite design is probably more or less what we know now, but I fully expect that there are enough moving parts among their existing plans that, while I'm sure not wildly different from their current plans, have some meaningful differences and are fairly advanced, well thought out designs in their own right that if they did indeed arrive at the conclusion, that a change was absolutely, no way around it necessary, they would be prepared to make that change.

After all, we are talking about a highly qualified, no doubt hard working team of professionals that are paid to get things right and account for any and all possibilities. It isn't so difficult to imagine that these folks could have from their work with AMD, devised multiple paths forward.

It isn't their job to design the most expensive possible system they can make just because they are able to, or allowed to. They have to, at all costs, find a way around having to do such a thing. Their primary goal is giving the developers what they need and designing a system that can also meet Microsoft's own desired goals. If they can do that without going for the most powerful or expensive parts, then that's what they will do, and it's certainly what they should do.

After that it's up to Microsoft to user every resource available to them to make the console as desirable as possible from a software, features and services standpoint.

 

[Tap In]

Exactly. It's pretty inconceivable that they would have boxed themselves in from the very start at some predetermined specification. More than likely, they had a number of plans that they've either altered somewhat or moved away from entirely since they've been planning for the next Xbox.

Obviously it's to be expected they're probably well past the point now where they have a favorite design, and that favorite design is probably more or less what we know now, but I fully expect that there are enough moving parts among their existing plans that, while I'm sure not wildly different from their current plans, have some meaningful differences and are fairly advanced, well thought out designs in their own right that if they did indeed arrive at the conclusion, that a change was absolutely, no way around it necessary, they would be prepared to make that change.

After all, we are talking about a highly qualified, no doubt hard working team of professionals that are paid to get things right and account for any and all possibilities. It isn't so difficult to imagine that these folks could have from their work with AMD, devised multiple paths forward.

It isn't their job to design the most expensive possible system they can make just because they are able to, or allowed to. They have to, at all costs, find a way around having to do such a thing. Their primary goal is giving the developers what they need and designing a system that can also meet Microsoft's own desired goals. If they can do that without going for the most powerful or expensive parts, then that's what they will do, and it's certainly what they should do.

After that it's up to Microsoft to user every resource available to them to make the console as desirable as possible from a software, features and services standpoint.

exactly... it amazes me the sentiment that MS is "stupid" because they just didn't spend everything they had in the war chest to buy the biggest, hottest, baddest parts they could find to put in the new box.

instead of being brilliant for engineering it the smart way

 

[Betanumerical]

I would imagine that they would plan ahead on raising the clock and then could go back on that if they felt they didn't need that fast a chip. Seems smarter to assume you'll need to go all out and pare back as necessary instead of putting yourself in a position to actually react.

Most documents we have seen (all?) say they are targeting 800mhz, I see no reason to believe otherwise, why would they tell developers they are targeting 800mhz GPU clock if they are aiming for higher?.

 

[AlphaWolf]

Most documents we have seen (all?) say they are targeting 800mhz, I see no reason to believe otherwise, why would they tell developers they are targeting 800mhz GPU clock if they are aiming for higher?.

A lower boundary makes sense if you don't have final silicon and yield data. Moving it up won't set back projects like moving down would.

 

[Betanumerical]

A lower boundary makes sense if you don't have final silicon and yield data. Moving it up won't set back projects like moving down would.

Yes, but you would at least inform the developers you are aiming for higher, there is really no reason to leave them in the dark about trying to target higher clocks.

 

[SenjutsuSage]

Most documents we have seen (all?) say they are targeting 800mhz, I see no reason to believe otherwise, why would they tell developers they are targeting 800mhz GPU clock if they are aiming for higher?.

Don't get me wrong, it could very well be what they are targeting. However, even if they didn't tell developers that they intended on aiming higher, I don't see it as too problematic to the development process.

Microsoft may have simply provided devs with the 800MHZ parts because, due to yield related issues, it was the quickest possible way to get a close enough approximation of the final hardware into developer hands on time. I imagine it couldn't have been too easy producing GPUs with the ESRAM, so maybe it made some sense to keep clocks lower until they could get better yields.

If Microsoft ends up delivering a higher clocked GPU come final hardware, then it more or less eases its way into the pre-existing development process without very much chaos. It would be much worse, for example, to promise something much greater, convincing developers they're going to get it, and then suddenly pulling the rug out from underneath them by giving them something notably weaker, which I don't suspect happened here.

Yes, but you would at least inform the developers you are aiming for higher, there is really no reason to leave them in the dark about trying to target higher clocks.

If you think about it, who are we to really say they didn't notify developers of this? The rumored specs would still more or less be accurate, which is essentially what various site sources have been saying. Sony didn't exactly notify too many developers about 8GB of GDDR5, did they? Even some first parties apparently didn't know about it.

 

[AlphaWolf]

Yes, but you would at least inform the developers you are aiming for higher, there is really no reason to leave them in the dark about trying to target higher clocks.

I don't see why. Development would still need to proceed under the assumption of the lower boundary, so giving them maybes isn't really helpful.

 

[Squilliam]

It would be nice if Durango is clocked at 900/1800mhz respectively. It'd be nice for some single threaded performance and extracting some extra juice out of the GPU/eSRAM respectively. I think it is certainly doable with respect to the chip in question though it could be down to how 'close' the 20nm shrink is to being achieveable. If they can get the 20nm shrink within 12 months they may decide to take a little hit initially.

 

[SenjutsuSage]

It just makes way too much sense for them not to do it, but I guess we'll just have to see what they do.

 

[Nisaaru]

exactly... it amazes me the sentiment that MS is "stupid" because they just didn't spend everything they had in the war chest to buy the biggest, hottest, baddest parts they could find to put in the new box.

instead of being brilliant for engineering it the smart way

If I would decide for ESRAM instead of GDDR5 I would at least make the effort of getting the most out of it bandwidth wise. Sure that would mean more ROPs too but if you can't match high end GPUs CUs/Memory for TDP and price reasons you can at least try to reach higher peak bandwidth.

 

[AlphaWolf]

It just makes way too much sense for them not to do it, but I guess we'll just have to see what they do.

It only makes sense if yields and thermals allow it.

 

[Shifty Geezer]

...but it's still hard to imagine that Durango was the best that this team of engineers could come up with. It's just such obviously low hanging fruit.

You're comparing Durango to (desired) performance, assuming the intention of the engineers is Moare Powarrr. What if they've got the performance of a 7790 in 80% of the cost and 75% of the power consumption, coupled with a 50% in RAM costs by going with DDR3? Then their goals, which weren't low hanging, were achieved by good engineering.

The only way to determine the effectiveness of the engineers is to know what targets they were designing for, not what we want them to design. Same with Wii U - the engineers managed to get comparable performance to PS360 in a much smaller and cheaper package. We can grumble all we like about performance, but that's not due to the ineptitude of the engineers; that's a fault of the decision makers giving them low-power targets.