(((interference)))
Veteran
Yup, MS eating the cost of defective chips.
- Status
Yup, MS eating the cost of defective chips.
I guess he's just ya know, an insider.
He also did not really specifically confirm anything about the downclock though. He just made another post you can infer from, but isn't specific.
Dunno if you've been reading my posts but GopherD also seemed wrong in a post about the PS4 core reservation, he said signs point to less less than one back in Feb and it now seems to be 2.
But I'm sure he's been right about stuff too.
No triple A source (IE just Matt, CBOAT, I guess) has said "yes, Durango was downclocked". In fact, it's starting to be kind of glaring that they're NOT saying it. Its only thuway and proelite with a side of Bruceleeroy (who supposedly is only well sourced on the Sony side). It sort of seems less likely to me as the hours go on now.
And yeah I'm well aware thuway had insider info in past, as I'm proud of saying thuway, you, and maybe even lil old me were probably the very first people putting out accurate Durango/Orbis specs anywhere on the net. I like to think maybe I'm a tiny bit of console history or something LOL.
But anyways, the point was it could be a Reiko situation. We dont know if thuway's sources for this downclock info, are the same sources that gave him good info in the past. They could be new, unreliable sources.
Yeah, somebody in there threw out for example a downclock to 750mhz. Still not good but palatable.
It's also interesting to me thuway started out saying he heard the same thing (downclock) about Sony, then he seemed to quickly sweep that under the rug, but I'm suspicious.
I guess it's possible both consoles receive mild downclocks imo.
Sony has revealed the flops spec which makes a downclock harder for them but I guess still doable. MS has also revealed some leading specs (200 GB bandwidth) which may be a bit more malleable.
What exactly was corroborated? Is it just the fact that a down clocking happened, or is it the actual amount?
It's easy to believe they clocked the GPU to 750Mhz, but much harder to believe 480 to 600. Also it's weird the down clock was specified in Flops instead of MHz.
I wonder if the info might have been lost in translation somewhere (it happened before). What if it was the memory that was down clocked 800Mhz or 933Mhz instead of 1066MHz? It's much more reasonable, and very easy to believe considering the state of DDR3-2133.
It's easy to believe they clocked the GPU to 750Mhz, but much harder to believe 480 to 600. Also it's weird the down clock was specified in Flops instead of MHz.
I wonder if the info might have been lost in translation somewhere (it happened before). What if it was the memory that was down clocked 800Mhz or 933Mhz instead of 1066MHz? It's much more reasonable, and very easy to believe considering the state of DDR3-2133.
With XBOX 360 MS took the role of IP licenser and the onus on production was on them, hence yield concerns were 100% on their shoulders. Is this the status quo now? We're dealing with x86 this gen and we all know that's tied in a rats nest of licensing; last time x86 was in a console MS bought an-off the-shelf part from Intel and all the onus on yield was the domain of the supplier.
If I'm reading Mr Baumann's post in a hopeful light, it seems like a possible downclock denial (though I know he cant comment on that).
It seems he's basically saying yields may not be MS problem, hence MS would have no reason to downclock. Any yield shortfall would only punish AMD and be "tough cookies"
That would have weird implications for the whole "fab anywhere" chip strategy though. I'm confused.
It seems he's basically saying yields may not be MS problem, hence MS would have no reason to downclock. Any yield shortfall would only punish AMD and be "tough cookies"
That would have weird implications for the whole "fab anywhere" chip strategy though. I'm confused.
(((interference)))
Veteran
This rumour, if true, might also explain why MS haven't revealed clock or TF numbers for XB1 yet - they're still in flux.
They did state that XB1 can only do 768 ops per clock vs PS4's 1152 ops/clock so it's not like they're staying away from giving out any numbers which make the competition look better.
They did state that XB1 can only do 768 ops per clock vs PS4's 1152 ops/clock so it's not like they're staying away from giving out any numbers which make the competition look better.
The general I think, valid, theory is that they're never going to reveal teraflops, since it's markedly lesser than PS4.
They could go to clocks I suppose perhaps without too much damage, but as I've pointed out, even Sony has technically not revealed any clocks (for the millionth time before somebody gets into this again, I know they revealed flops). They might fear if they say 800, and sony later comes out and says 900, they look bad. Any number of scenarios.
Perhaps they'd fear with clocks and the shaders count people could put 2 and 2 together. Also, sometimes I'm not 100% sure they plan all of this out. When the guy said 768 ops in the video, did he really plan out he was gonna say it or did it just kinda flow naturally? In such a way, maybe there's no 100% strict bans on saying clocks or flops numbers, etc, if the context is right and "relaxed" enough.
(((interference)))
Veteran
Yeah, I think so too - we're probably not going to get a TF number from them.
Do you know who Crazy Butt Train is?
This is the first I've ever heard of him, what has he divulged that gives him such top tier insider status?
Do you know who Crazy Butt Train is?
This is the first I've ever heard of him, what has he divulged that gives him such top tier insider status?
GrimThorne
Newcomer
Uh, that matter is quite debatable. Pointless, but very debatable.
But wasn't that video was taken down by MS?
http://www.neogaf.com/forum/showpost.php?p=974516&postcount=1
Maybe I'm wrong but I remember OXM rumors about lack of hard drive before 02-07-2005.
I dunno, that's just what I noticed skimming some threads.
Yeah. He writes all crazy and he is deffo a legit insider for years.
Hmm, cant remember offhand what he's divulged, I tend to have a terrible memory, but for example he has said a lot recently MS has a big exclusive with EA.
I figured it was the respawn game, but now I'm not so sure some gaffers think it might be mirrors edge 2.
Point is if he says it it's probably true.
He has been wrong once or twice though.
Bagel seed
Veteran
When you say eating the cost, you mean making more chips right? They'll want to have X number of chips to hit their worldwide launch targets. If yields are horrible then you'd have to assume they've already pushed the 'gimme all the chips you got' button several times before entertaining downclock options. Either that or completely miss the holiday launch of one or more territories which I'm guessing is out of the question.
Well, some "insiders", at NeoGAF, have said Microsoft paid to third party for not appear at Sony event, I guess MS can eat the cost of defective chips after all.
Jk.Themrphenix
Newcomer
Microsoft didn't downclock the gpu!They are getting bad yields,but talk is its on amd end!
There is no justifiable reason for randomly disabling features for individual chips manufactured for a console with a defined feature set. Selective removal of features as fundamental as the number of threads supported in hardware would be a baffling decision to apply to the Xbox One, not to mention that the vast majority of such feature tweaks by Intel frequently have more to do with market segmentation than physical need.
This is a memory pool alleged to have 102 GB/s of bandwidth. This could be two 64-byte connections, one in each direction.
You'll have to take into consideration how these many separate blocks tie into the eSRAM interface, and make trade-offs with respect to die area taken up by 33x duplicated decoders, amps, and bus connections.
The pressure to contain the growth and complexity of the large SRAM pool is the motivation I hypothesized could lead to reduced measures relative to the size of the storage.
The memory interconnect may be a separate topic that is being conflated into the general finger-pointing at the eSRAM.
This was already assumed as part of the design in handling faults. Parametric yields are another matter, since localized effects can make nearby spares share undesirable properties with the row we want to replace.
Are you talking about the quad core Jaguar modules? Any redundancy involves disabling within a cluster, not inserting additional cores or changing the quad-core arrangement. The area taken up by the L2 interface is already quite significant, and all indications so far are that there are two 4-core modules.
No eDRAM is offered for the likely foundry processes Durango would use, so this seems safely discountable. It would complicate manufacturing, is known to be a cost-adder, and has yield effects only IBM has considered palatable for the process geometry in question.
Also not likely at all.
For the same reasons, I'm not going to go too much in-depth with any of the others variation of DRAM mentioned.
Redundancy is an indirect means for trying to improve parametric yields. DRAM and FLASH work at very modest speeds, and they bin very heavily. The profile is much different for an SRAM memory pool running at GPU clocks for an APU. I wouldn't cite Intel as an argument for different results, since history has shown it to be very adept at implementing large (if still smaller than the eSRAM) caches and its processes are very advanced and far more focused on its architectures than AMD can be on a foundry process.
AMD is potentially on the hook for some or most of this, since they were paid to build much of the custom APU. I don't get how problems with a single physical design can be isolated to a Microsoft or AMD end. It's their baby.
Last edited by a moderator:
Well a more positive note
http://www.gamespot.com/news/forza-motorsport-5-will-run-in-1080p-at-60fps-6409332
http://www.gamespot.com/news/forza-motorsport-5-will-run-in-1080p-at-60fps-6409332
Best thing they could do is offer a premium $699 SKU with 5 TFLOPS of power...remember forwards compatibility is key.
News & Rumors: Xbox 720 aka Xbox 3 aka Xbox Next aka Durango aka Loop
Wouldn't the foundry be more to blame than AMD here? It's not as if AMD and GloFo are the same thing
I would just really hate to see bad yields for a few weeks/months in a rush to hit Holiday 2013 affect the performance of what is basically a decade-long box.
Wouldn't the foundry be more to blame than AMD here? It's not as if AMD and GloFo are the same thing
I would just really hate to see bad yields for a few weeks/months in a rush to hit Holiday 2013 affect the performance of what is basically a decade-long box.
- Status
Similar threads
- Sticky