News & Rumors: Xbox One (codename Durango)

[3dilettante]

Sounds fishy to me. ICs last pretty long unless they get HELLA hot and with that huge fansink shown, I wouldn't think that would happen. I would assume this is BS, someone trying to respin that whole RRoD fiasco from years ago. Manufacturing techniques have improved since then, I'm fairly confident we won't see mass scale hardware death in just a few years' time in next gen.

For what it's worth, in some ways, modern manufacturing processes can be better characterized as being smaller, not better. Flash memory's rapid drop in write endurance is a highly visible example of trends that also influence other products built on silicon.
The last I saw discussion on this for mainline x86, many nodes ago, the goal was something like a decade at least, so the three year period and even the "better" five year rumor sound pretty bad. I'd hope this rumor is false because the "improved" time seems too short for a console generation.

However, it was far easier to stave off problems with electromigration, mechanical strain on structures like vias and insulating layers, chemical diffusion, and gradual shifts in transistor performance characteristics when everything was 8,16,32 times larger and the chemistry involved was only crazy, not unbelievably insane like it is now.

Even if metal doesn't migrate across a junction and short out the chip, shifts in transistor performance can slowly cause circuits to misbehave before outright failure. One example is the mis-specified transistor that slowly degraded in the original Sandy Bridge chipsets.
SRAM's longevity and the degradation in noise tolerance over time is a heavily researched topic.

 

[Brad Grenz]

He clearly had access to the Durango sdk documents, he just didn't have the background to properly interpret them.

 

[SenjutsuSage]

He clearly had access to the Durango sdk documents, he just didn't have the background to properly interpret them.

Ahh good point. That's likely it.

 

[(((interference)))]

Yeah aegies probably has contacts but most likely doesn't have the technical nous to evaluate technical info he's being given - hence him thinking Durango was more powerful that it really was.

I can personally vouch for SenjutsuSage's source, they are definitely senior enough and technical enough to be in a position to know.

However, one caveat applies; his source is a dev and it is unlikely but not completely improbable that these GAF insiders and aegies are hearing these rumours directly from someone within Xbox team where they have recently decided on a down clock and are going through the due process and specifics and are yet to inform devs.

This is not completely without precedent as the PS4's 8GB GDDR5 surprise was not known by even internal Sony devs like GG until the reveal - however some GAF insiders were in the loop before the reveal.

However, to me, the most likely scenario is that the clocks remain unchanged.

 

[Gradthrawn]

Hold on! What the hell is going on here?!? Tommy Lee Jones is one hell of a sexy guy. I'm reporting Grall's post to the mods immediately.

Heck, if you look like Tommy Lee it's already way too late for you!

 

[upnorthsox]

Every time I come into this thread

You guys are gonna ruin my nice smooth face.

And everytime I leave it I'm like:

 

[expletive]

However, to me, the most likely scenario is that the clocks remain unchanged.

Or that 1600/800 are the downclocked speeds

 

[SenjutsuSage]

Or that 1600/800 are the downclocked speeds

Well, if this is at all what's going on, then at least on the GPU side of things, we've sorta always viewed 800MHZ for a GCN class GPU in this performance range as a downclocked part, haven't we?

Both the 7770 and the 7790, for example, are both 1000MHZ GPU, so anything below that I guess would be a downclock. If that's the case, though, the latest rumors would have to somehow be based on the initial belief that Microsoft somehow once intended on running the Xbox One's GPU at 1000MHZ, or even 900MHZ?

 

[(((interference)))]

Or that 1600/800 are the downclocked speeds

Nope, 1600/800 was the target specs since the beginning of 2012 and the first Durango conference, at the very least, if not earlier. I think MS finalised the design around Nov 2011.

 

[ninzel]

I'm glad MS adressed the privacy issues regarding Kinect. I'm satisfied with their answers and will stay that way unless they give me a reason to doubt them.

 

[BRiT]

I moved the PRISM discussion off to a more appropriate location -- http://forum.beyond3d.com/showthread.php?t=63731

 

[BeyondTed]

Nope, 1600/800 was the target specs since the beginning of 2012 and the first Durango conference, at the very least, if not earlier. I think MS finalised the design around Nov 2011.

Finalizing a console design two years before shipping (Oct/Nov 2013 shipping?) does not sound right. That is just way too soon. The expected process would not be available then making it impossible to finalize the design then. I keep being told it is 28 nm and nothing else.

If that is true then I would say it lends credibility to the rumor of an earlier design which was finished and sent to certain developers but was later replaced with a higher spec design which isn't quite ready for production release yet. I think the rumor was silicon tape out date of Sept 2012 for that with the re-spec in April/May 2012. So it sounds like the first design was 40 nm and the new spec/new design is 28 nm... Makes sense as putting 5 billion transistors on 40 nm is going to yield and cost like a lead balloon. 40 nm would make the 768 rumor make some sense.

After all the anti-MS bashing it is rather hard to sustain that a design was finished in Nov 2011 but now they suddenly have clock and yield issues. They go through proving that stuff (yield, clocks, cost) long before finalizing a design.

 

[Bagel seed]

After all the anti-MS bashing it is rather hard to sustain that a design was finished in Nov 2011 but now they suddenly have clock and yield issues. They go through proving that stuff (yield, clocks, cost) long before finalizing a design.

And the more problems a chip might have the longer the test phase needs to be to get it sorted out with each respin. It's not unusual for this period to take at least a year's time to complete, from initial tape out to finalization. And that's without issues. Factor in manufacturing prep time during 2013 and the Nov 2011 date definitely fits the timeline for the design we see now, and to make a worldwide launch for the end of the year.

 

[AlphaWolf]

I doubt clocks would have been final till quite recent.

 

[BeyondTed]

And the more problems a chip might have the longer the test phase needs to be to get it sorted out with each respin. It's not unusual for this period to take at least a year's time to complete, from initial tape out to finalization. And that's without issues. Factor in manufacturing prep time during 2013 and the Nov 2011 date definitely fits the timeline for the design we see now, and to make a worldwide launch for the end of the year.

Right. To have finalized a design in Nov 2011 they would have started a couple years earlier. And no way work in 2009/2010 was in 28nm.

 

[BeyondTed]

I doubt clocks would have been final till quite recent.

Perhaps for the system as a whole but I don't know of any place that calls a chip design done until they have proven (with data on multiple batches) the yield, reliability, cost, clocks, power, etc.

Now stuff can show up later after you go through all the procedures but my point was you do go through a very thorough procedure regarding such items before calling it done.

 

[AlphaWolf]

Until you know yields, finalizing clocks could be bad. They would have an idea.

 

[Bagel seed]

Right. To have finalized a design in Nov 2011 they would have started a couple years earlier. And no way work in 2009/2010 was in 28nm.

Finalized as in design. Still needs to be tested. Everyone forgets this part.

 

[BeyondTed]

Until you know yields, finalizing clocks could be bad. They would have an idea.

Could be very bad. I assume they ran slow slow (corner lots) and what ever the particular company has codified as the appropriate number of lots over the appropriate period of time.

I suppose these days slow slow isn't the whole picture with the leakage of modern devices.

 

[BeyondTed]

Finalized as in design. Still needs to be tested. Everyone forgets this part.

In my opinion no design is done until it is tested.