-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
IF Sony's yields are worse than MS, which we have no evidence of, then it would be because the clock speed is so high. It's really not that surprising considering how insanely high the PS5 GPU is clocked. Higher clock speed = fewer chips that can remain stable at that speed. It's why some CPUs and GPUs can reach different maximum clock speeds even if they're the exact same model and brand.
the original Bloomberg report was 5-6 million units so they're still at roughly double the original production estimate
but yes y'all go on about how this is the worst news for struggling Sony
How reliable is this report, though?
It's unfathomable to me that Sony would go into next-gen with yields of 50% unless they knew it was only temporary and that they could eventually get yields up to 80-90% as usual. Clocks aren't the only thing that influence yield, size is also a very big factor. So I'd be interested in knowing if this is a problem due to clocks or just a problem with RDNA2 in general on the 7nm process.
If clocks ARE the reason, then why wasn't this known ages ago before Sony announced the specs of the PS5? Surely at that time they had done significant testing and early production runs to prove out the speeds they were going to target. A 50% yield for the duration of the generation is ATROCIOUS and not really viable, so I refuse to believe this is some sort of long-term issue. Sony isn't stupid. They should have just reduced the clocks IF that was a viable solution.
It's not even worth it at this point. Let them get their jollies here before Wednesday when they all will vanish all of a sudden.
Do MS even talk about the yields or amount of consoles in production?
Don't Sony only mention it because they're a hardware manufacturer.
With the yields issue, if that does mean potentially more defects (if true), then I think I may just wait for beginning of 2021 to snag a console for them to iron out any kinks.
I predicted the high GPU clock as potential problem back when Cermy had the tech talk. Not insurmountable or anything but potential road bump. I think this might be that bump and it's good to get it out of the way early.
Worse case scenario is to lower t he GPU clock officially to get the yields up. 2GHz might be enough maybe and won't affect perfofmance that much in the big picture...
Worse case scenario is to lower t he GPU clock officially to get the yields up. 2GHz might be enough maybe and won't affect perfofmance that much in the big picture...
Probably because Sony expected production even after this news is higher than Series production.
Sony: We want make 100, we reach 75.
MS: we wan make 50.
Something like this.
The only thing that will cause me to buy a PS5 at launch is to have full BC, which there is almost no chance. I am one of those who still has that small hope of it, but I am also realistic in my expectations.
So I most likely will not be buying a PS5 at launch, but don't envy those that do. Besides how difficult was to get a hold of the Switch, I remember the crazy shit that went on when the PS3 was released. I remember reading about at a Gamestop where someone was jumped right outside and had it stolen. Glad they caught the fucker. But I could see shit like this happening for the PS5.
This just goes to show you how unreliable these Bloomberg reports are, I'm not sure how much we can really trust their data.
While some people are here to console war or concern troll, the 50% yield news is concerning and that they are reportedly having trouble to reduce it is also troubling. It won't impact Sony at launch most likely but it will impact there ability to meet demand early next year if they fail to get that yield down significantly and will be something they would eagerly want to address to reduce further costs.
We don't know when they were having the yield issues. They were originally projecting a much lower production amount but have recently increased it. That suggests their yields have improved but not high enough to reach the 15mil number.
The original Bloomberg reporting was based on what Sony was telling partners at the time.
Poor yields != defects in retail units. Defective chips or chips that don't meet the requirements for being in the system don't get put in the system.
Read up on binning. I do this for a living.
If more of the chips are coming in on the slow side and you have to increase the voltage to meet the intended clock speed, you could end up blowing your power budget and the chip has to be tossed.
The process could be tweaked to help fix this in the short term (but "short term" is months). That is if TSMC would do this. I don't know. And in the long run, AMD could work on speed paths to fix the speed issues (now we are talking multiple quarters).
Fixable stuff. Just not in the launch window.
Not sure what you'd be gaining since consoles sold in early 2021 are part of this initial 11 million.
GPU clock. XsX GPU only needs to clock to 1.8GHz. 300ish MHz less than PS5.
Is the same source that everyone here swore by when it came to BOM price though.
Wide and slow prefab vs narrow and fast prefab is really the singular thing that sticks out based on what most of us know about production. In other words, clock speed and their abilities to sustain. We will likely never know the true reason but there is a reason that these devices don't typical clock high in retail devices. It was always a gamble with the small (in comparison to the Xbox Series X} SOC but had the chance to pay off dividends if it succeeded. Over time, it still might but now, it seems like they've run into a speed bump. The vents on the devkit and the size of the actual console, retrospectively lends itself to this report.
It's so crazy I'm inclined not to believe it's true. It's a lot to unpack.
I think more precision questions may be in order - is it 50% of what they expected? 50% of the total wafer sheet? Those may be very different numbers. The reason chips have CU's turned off is to try and increase the yields but honestly that number generally represents the expected loss. So if you have 2 disabled CU's on a, say 38CU (getting you 36 good CU's) - you're expecting 5%-6% failure rates (this is an oversimplification but you get the point). Launch yield issues may represent 10%, maybe 20% on the high end.
50% failures basically doubles the cost of the chip per box - which is the single most expensive part of the BOM.
My tendency when I hear a number like this is to not believe it. However, they are clocking that CU pretty high and depending on what you believe about how/when in the process that may have happened, it could have proven to be a problem in manufacturing.
I guess we'll find out more tomorrow!
If the article is true and the yields are slowly increasing without those major changes doesn't that suggest production line issues as well? Something that would improve over time regardless.
Yeah, if clocks are to blame for 50% yields....why wouldn't Sony just lower the clocks and bring them down into the realm of sanity?
That's why it's not really believable to me, unless these are somehow short term issues that should be resolved with time.
It's terribly cost inefficient and goes completely against Sony's hardware ethos. They've already lost the "Compute" power war, so getting down to 9TF isn't the end of the world in all honesty.
Is there something happening tomorrow that sheds more light on this?
Sony bumping production later on doesn't mean Bloomberg's report wasn't accurate.
I think we all believe price is coming tomorrow? Sony would have known about this for quite a while (before the leaks) and I assume it would impact the pricing strategy. Doubling the cost of the SOC would add over $100 to the BOM.
Maybe not. But if they come in with aggressive pricing tomorrow that would make me inclined not to believe a report like this.
Server rack space is also expensive. Why would they waste space handling fewer virtual consoles for cloud per server for what's likely less than $100 difference between a fully functional chip and a gimped one?
This to me sounds like the yields was 50% at one point of time, probably at the beginning, but has improved by now.
How many % is stable level, we don't know but I'd imagine 80-90% is the target. They could be anywhere between 51-80% for all we know.
it's possible that the yield issues are TSMC's problem and that the process isn't meeting the pre-production models. But there are other chips on this process (and not just Microsoft). They'd likely be seeing similar issues. But every chip has the possibility of stressing the process in different ways.
Penello gave a great breakdown of this exact topic in one of the old next-gen threads and explained it. He is in this thread right now so he can confirm it if he feels like it but I won't tag him to win an internet argument.
*puts on flame retardent suit*
This kinda gives credence to notion that PS5 had 9TF GPU originally like the leaked specs but up it to 10.28TF after learning about XSX being 12TF machine. Cranking clock and voltage over original design spec is never good for yields.
This kinda gives credence to notion that PS5 had 9TF GPU originally like the leaked specs but up it to 10.28TF after learning about XSX being 12TF machine. Cranking clock and voltage over original design spec is never good for yields.
Price coming Wednesday...still two days away :) But yeah, you're right, maybe price should shed some light on this. But it seems really silly again, if clocks are to blame, for Sony to accept such an abysmal yield...for every chip they are throwing away another one. They could have just went bigger a slower for less cost, again *IF* clocks are to blame and this isn't some new process issue in general. It's a colossal waste of money and would be one of the biggest hardware blunders ever.
I don't understand how Sony executives would accept such a scenario when they could have simply reduced clocks and improved yields.
To me it doesn't give credance because again, sarcrificing a huge amount of wasted cost just to increase from 9TF to 10.28TF is unimaginably stupid.
We'll, yeah. If you look at it, they will have 10mil consoles made by end of 2020. Then the plan seemed to be 5mil per quarter after that. The reduction from 15mil to 11mil means they only expect 1mil, from planned 5mil, in q1. That's a reduction of 80%.
Assuming an ASP of $449 (which splits the difference between all the pricing rumors) that's $1.8B in revenue just on console HW, and another $600M in games/accessories (assuming $150/per console). So it's not chump change, either.
Yes, that's what I would speculate as well. I'm just curious what the root cause is (clocks, or RDNA2, or new process, etc...maybe a combination of all three).
MS did reveal that this new process was a LOT more expensive, shouldn't that mostly be down to yields?
The
The wafers themselves are usually a fixed cost right? Do companies ever have pricing agreements where they can get a discount for overly defective wafers? Seems like the only way to have a low yield and prices the article is suggesting.
Yeah. You would think they would have showed a much larger impact on their financial forecasts if that was the case though.
Maybe the yields aren't abysmal. Not good but not disastrous maybe. If it was abysmL, you will see lowering of clocks for sure. We had GPU clocks tweak down before with GameCube going from 200MHz to 162MHz few months before launch IIRC. Im sure that was yields too. They upped the CPU clock from 400 to 488MHz to compensate.
That's true, assuming it was a known issue then. I haven't really kept up.