Ya, that's why we should take AMD/NV marketing slides with a grain of salt but it took 3 pages until someone called them out.
I am shocked that AMD has actually gained mobile dGPU market share and it has nearly 35% market share in the mobile segment vs. just 23.6% on the desktop. Who is even buying laptops with AMD's dGPUs in them now? This is very surprising to me that their mobile dGPU division is doing miles better than their desktop cards because their desktop cards are actually very competitive in the $100-400 segments but their mobile dGPUs are not at all. I would have guessed that by now nV has > 90% mobile dGPU market share and I was way wrong.
Looks to me like the data is pretty close, actually. Just shifted over as one represents sell in and the other sell out.
One data set is the units sold to partners, the "sell in". Partners are shipped batches all at once and they will then have to be sold to the consumer. They will adjust depending on how well they sell.
Isn't it strange how close the data is, if you shift it over a quarter. This is because the other one is trying to represent the sell out figures. But this one is actually not a real number, it is not the amount of units sold to customers. The Mercury sell out figures are not actual units but an attempt to represent real data by averaging. It's a 4 quarter average by volume in attempt to smooth out heavy swings that would occur if a partner stocked up majorly with anticipation but then find the product failed to sell as well as they thought. In the end, it doesn't matter much because the partner will end up ordering less next time and the data will reflect that in the next quarter. Mercury's attempt to smooth out by a 4 quarter average is not more accurate and nvidia's data "not to be trusted", as you are trying to make it out to be. You just have to understand what each one is. Mercury's averaging method could be great accept it has one major flaw, if AMDs marketshare really does plummet, their average of the last four quarters is actually boosting up this drop off and smoothing it over. The Mercury's method only works if there are no real large swings in market share. So their attempt to smooth out possible swings from large orders a partner may make isn't the correct way to do it, it's just a 4 quarter average.
Using the actual numbers shipped to partners is real data. The draw back is that those units are just filling channels and stock on shelves, they haven't been sold to the consumer yet. The beautiful thing is that if these units don't sell, the partner will not order. So the data sorts itself out, when the partner doesn't order again. The boost up from the order will either actually be sold or it will be met with a dip when they aren't shipped another large quantity. But the data will reflect that. The numbers are real and it all sorts itself out. You just may see more bumps if a product fails to sell as expected. But unless these partners are totally incompetent, the bumps won't all that large.
Mercury average of 4 quarters can smooth out those bumps but it's not actual sell data for that quarter. It's an average of the last 4 quarters. The worst possible scenario in a 4 quarter average is a sudden and fast drop off. Then their data doesn't reflect this, it simply isn't a good method when a company is loosing market share fast. You just won't see it.
That's why the data doesn't look as bad, it's not because nvidia is lying or cheating. It is because Mercury is giving you the a 4 quarter average and 2 of those quarters AMD had close to 40% market share.
I hope you are capable to see this, the Mercury method is not the better way in this case. It is actually more flawed than looking at partner shipments. And if nvidia partners over ordered, then next quarter it will sort itself out. There will be a corresponding dip.
You should also know that the sell in data, the shipments to the partners, those would have happened. Your insistence that the 300series has no impact is just wrong here, because partners order and fill the channels long before you can buy one. And this is also the reason why you are wrong about the 980 and 970 surge, the market share data does reflect maxwell just as expected. The partners would have ordered and filled the channels long before the actual cards launched. So saying that maxwell didn't cause the market share shift last yr in Q3 is just wrong. It's clear you don't/didn't understand how this works. It I am pretty sure you will ignore the facts and just carry on.