It was obvious he was completely wrong some months back because Micron themselves said they expect a large upswing in financials at the end of Q2 (i.e May) due to bulk sales of their new graphics memory.
same thing with the "Nvidia has to be 6 months behind AMD because TSMC is 6 months ate", when the delay TS<MC had was irrelevant to the launch of Pascal because Nvidia weren't trying to release the 1080 back in Q3 2015 which would have been problematic with TSMC's delay. TSMC's 16mm FF entered volume production Q4 last year so a Q3 2016 product is obvious not affected.
Lets stop rewriting history shall we. First with the Nvidia are 6 months behind thing. I said, Nvidia are
likely to be 3-6 months behind AMD based on all information at THE TIME. Second, why has GP100 launched but not in mass volume and not in mass volume until next year if 16ff+ entered production volume in Q4 last year... how can one thing effect the other because you're insisting they are unconnected.
Here is a hint, yields improve over time, Nvidia was unable to produce GK100 and subsequently GK110 at high enough yields to enter the desktop market till LONG after GK104 was available.... why? Why did GK104 come so long after smaller 28nm parts had been made in mobile? Because they are connected. If production of a process starts in Q4 2014, making a higher yield 300mm^2 part takes a little time and making a higher yield 600mm^2 takes a lot longer. A delay to the start of a process being deployed is a delay to EVERY stage of that process. If yields hit 70% 6 months after the process deploys, delaying the process delays the yields hitting 70% by the same amount.
Second, Micron stated several products would lead to increased demand including 8Gb GDDR5(non x) which is going to be on the massive majority of graphics card. Which chips will bring in more revenue, the non X chips available on say 7-8 graphics cards(2-3 of each Polaris, 1070, maybe a further cut down 1070 eventually) or the X version available on a single card in the lowest volume being the highest cost card?
Then again what were the rumours being talked about at the time, what was my reasoning? Half the people(you included) were talking about rumours for a Titan or 1080 release
IN APRIL. Micron said gddr5x production would start at the end of Q2, their talk about Q2 predictions also said gddr5x in production by the end of Q2... it went into production two weeks early. You lot were banging on about how Nvidia might launch gddr5x products LONG before JUNE and we were pointing out that Micron the whole time were saying summer before production starts.
Then we have the little fact that we have no idea how many 1080s will be available. I 'went quiet' because as usual people went insane and insist they know everything. Every indication is still that launch in June volume will be small and we won't see wider spread availability till a bit later.
I also months ago said what could easily happen is Nvidia launch early get a few cards out while AMD have massive volume launching around the same point and that relatively speaking comparing launches by comparing availability is fairly important.
Think about it like this, if Fermi 'launched' the first time with 1000 samples then didn't have any cards for 6 months, which was entirely possible, would that have meant Nvidia were ahead of AMDs 5870?
Cards come back 3-6 months in general before launch, almost always a 'normal' launch involves getting chips back, deciding they are good and then waiting 6-8 weeks to get production going and build up a supply for launch. There are many ways around that, costly ways. You can do hot lots, which is basically where you pay a premium and the fab will essentially jump your chips to the front of the queue. Consider that production takes now 3 months(roughly) at 16nm and that has potentially 50-100 separate stages, cooking, cooling, etching, other processes. You've got thousands of wafers inside a fab and the perfect timing for every step isn't going to happen so you can cut down time to produce if you pay more to get to the front of the line at every step throughout.
Another thing you can do is risk production, that is where instead of doing a few test wafers, testing the chips then if everything is okay you order full production which has the aforementioned full production cycle to go through before you even get chips back. Risk production is where you would maybe put 200-1000 wafers(which could cost in the range of 2-10million) in at the same time as the test chips. If the test chips come back needing a respin/fix then you have 200-1000 completely worthless wafers, if the chips come back fine then you get 200-1000x number of chips working per wafer for a 'launch' but then you get a 3 month delay to more chips starting to come out of the fab.
I suspect that we'll see a risk production lot go for sale in June/July, have extremely low availability for the next 2-3 months and see mass availability towards the end of august or sept. This could also be the case for ggdr5x, it could be Nvidia got as many sampling chips as possible to match to say 5000 1080s and we don't see more till late August when both Nvidia and Micron has had a chance to bump up volume.
Effectively anyone can short cut a relatively small number of cards to market months early, it costs more, the risks are massive and I suggested it as a possibility months ago.
But ignore what was said at the time, ignore that I didn't even claim Pascal wouldn't launch in April, ignore that I was attempting solely to disprove completely obviously BS stories from certain sites and made absolutely no claims about when Pascal would come out.
Something certain people chose to ignore is talking about the validity of one rumour isn't making a claim to the opposite. At the time I went to great length to explain I wasn't saying Pascal couldn't launch in April, but that the rumour the April launched was based off was completely bull. I also said I didn't remotely think Nvidia would launch in April and yes, every single sign pointed to Nvidia launching much later than AMD.
. It involves stripping the cooler off,throwing away the pcb part and chip and retaining the cooler ti use on a pascal card. It's true I tell ya
