Q4 2016 GPU shipments totalled 100 million, says JPR

[HEXUS]

AMD enjoyed the biggest unit gain during the quarter (by a smidgeon).

Read more.

[excalibur1814]

...and we're all idiots for buying them. When a company, such as nVidia, make 2 BILLION profit they really are over pricing their parts. We lap it up and then return to spend EVEN more cash. But, obviously, that's business.

P.s. I own an eBay purchased 980.

[kalniel]

...and we're all idiots for buying them. When a company, such as nVidia, make 2 BILLION profit they really are over pricing their parts. We lap it up and then return to spend EVEN more cash. But, obviously, that's business.

P.s. I own an eBay purchased 980.

It's good that you help other people to afford new cards by buying their old ones off them. I don't think people are idiots for buying cards though - the leap in speeds we see each generation is quite impressive. Only a few years back the power you get in something costing under £200 today used to cost closer to £1000.

[scaryjim]

... Only a few years back the power you get in something costing under £200 today used to cost closer to £1000.

Case in point, an RX 480 4GB - released in June 2016 for $199 - is generally a little faster than a GTX 780, released in May 2013 for $649. The 1080 Founder's Edition, which launched at $699, is at least twice as fast as a GTX 780, and in some cases gets close to three times as fast.

Unlike CPUs, which have had relatively small generational performance increments recently, GPUs have continued to increase performance massively (remember, 780 - 1080 is actually only 2 generations).

EDIT: in fact, the 8GB RX 480 is pretty much level with a 780 Ti, which launched 6 months later for $699...

[Ozaron]

...and we're all idiots for buying them. When a company, such as nVidia, make 2 BILLION profit they really are over pricing their parts. We lap it up and then return to spend EVEN more cash. But, obviously, that's business.

P.s. I own an eBay purchased 980.

In the last 8 years I leapt from HD5450 1GB DDR3, to GTX 460 SE 1GB GDDR5, to R9 Fury 4GB HBM. 80 to 3584 streams. I paid just over 3x as much for the last card as the previous two but there is no doubting I got more than 3x the relative performance!

Do we judge over pricing based on how much they ask, compared to material & development cost? Or compared to what they're worth to us? Material cost of spring water is almost 0, but we still pay up to a couple £ per bottle.

[CAPTAIN_ALLCAPS]

...and we're all idiots for buying them. When a company, such as nVidia, make 2 BILLION profit they really are over pricing their parts. We lap it up and then return to spend EVEN more cash. But, obviously, that's business.

P.s. I own an eBay purchased 980.

Well nVidia are only printing money because their upper mid range is totally unopposed and selling for high end prices.

AMD's strategy was a mistake; if Vegas was going to be released in 2017 they should not have made Polaris so small.

[spacein_vader]

Well nVidia are only printing money because their upper mid range is totally unopposed and selling for high end prices.

AMD's strategy was a mistake; if Vegas was going to be released in 2017 they should not have made Polaris so small.

It's depressing when they've managed to redefine the market pricing so much that people refer to £230+ as upper mid range.

[CAPTAIN_ALLCAPS]

It's depressing when they've managed to redefine the market pricing so much that people refer to £230+ as upper mid range.

I'd say that upper mid range is a bit higher than that, and why not? Improving fab processes is difficult and expensive these days, and the chips are bigger. It is hardly as though fast graphics cards are a vanity item; they are much cheaper and much easier to enjoy at their fullest than fast cars or expensive furniture.

Sure consoles are cheaper overall, but things have advanced to the point where AMD will likely release an APU later this year that is as fast as an entire xbox one.

[scaryjim]

... AMD's strategy was a mistake ....

Really? You're commenting on a story that shows AMD's made more market share gains against both nvidia and Intel (is this their third consecutive quarter of increased share?) and you think that's an indication that AMD got it wrong?

They hit the low end first because they wanted market share. If your market share is slipping investors get twitchy and your financial position is also under question. By going low and increasing market share, you increase investor confidence, have a more stable financial footing, and are in a better position to release high end products.

And as I've mentioned in connection to Ryzen; AMD's lack of a competitive high end product last year doesn't matter to people buying a card this year. And no matter how many people bought GTX 1070s and 1080s last year, there will still be people looking to buy high end cards this year. All AMD have to do is make sure that when they do have a product available, it's a desirable alternative to nvidia's extant cards.

[CAPTAIN_ALLCAPS]

Really? You're commenting on a story that shows AMD's made more market share gains against both nvidia and Intel and you think that's an indication that AMD got it wrong?

They hit the low end first because they wanted market share. If your market share is slipping investors get twitchy and your financial position is also under question. By going low and increasing market share, you increase investor confidence, have a more stable financial footing, and are in a better position to release high end products.

And as I've mentioned in connection to Ryzen; AMD's lack of a competitive high end product last year doesn't matter to people buying a card this year. And no matter how many people bought GTX 1070s and 1080s last year, there will still be people looking to buy high end cards this year. All AMD have to do is make sure that when they do have a product available, it's a desirable alternative to nvidia's extant cards.

The gains were from a low starting point though, and don't forget those numbers include AMD APUs. The 12/2016 Steam hardware survey (the most recent with comprehensive data) put the combined market share of the 480, 470, and 460 below the 1060 or the 1070 or the 1080.

Having your mainstream GPU being outsold by your opponent's top end is hardly the sign of a winning strategy.

[scaryjim]

The gains were from a low starting point though, and don't forget those numbers include AMD APUs. ....

OK, let's just look at a) discrete GPUs, and b) actual shipments. If you look at the figures over the last few years, nvidia's shipments have been fairly static, for all they follow a seasonal pattern - lowest sales in Q2, highest sales in Q4, figures generally in the 8M - 10M units per quarter. If you look at AMD over the same period, their sales plateaued around 5M units a quarter through the whole of 2013, then dropped steadily to a nadir of 1.7M units in Q2 2015. Over the following year they rose steadily to 2.8M units in Q2 2016, then leapt to 3.8M units in Q3 2016. That's the same increase in one quarter than they'd managed in the whole of the previous year. Coincidence that it happened in the quarter after Polaris was released? I don't think so, personally. This quarter they've maintained a small increase in dGPU shipments (3% - 4% depending on the relative distribution of desktop and notebook graphics in the mix). They're now back to the kind of sales levels they were seeing in mid-2014, and with a trend firmly in the opposite direction. You can say what you want about the technology, but financially they made absolutely the right call.

... Having your mainstream GPU being outsold by your opponent's top end is hardly the sign of a winning strategy.

a) I wouldn't trust the steam survey to be even vaguely accurate, tbh.
b) In 2016 NVidia sold > 2.5 discrete GPUs for every 1 AMD sold. There are, quite simply a LOT more nvidia GPUs out there than AMD. So even taking the steam survey with a pinch of salt it's not entirely surprising.

OTOH, in 2015 NVidia sold 4 discrete GPUs for every 1 AMD sold. So for whole year market share, AMD's gone from 20% to more like 30%. And what's more interesting: in 2015 nvidia sold ~ 35M GPUs and AMD sold around 9M. In 2016, nvidia sold around 35M GPUs and AMD sold around 13M. So nvidia have sold no more GPUs this year than last year, while AMD have sold around 50% more.

Selling 50% more GPUs this year than last year? I'd call that at least a successful strategy, if not a winning one...

[shaithis]

Selling 50% more GPUs this year than last year? I'd call that at least a successful strategy, if not a winning one...

Without knowing profit margins, its impossible to say. Selling at a loss isn't a winning strategy for instance....nor is helping your main rival look even stronger at the top end when you already had a perception problem there......that in essence is why they sell so few cards compared to nVidia and yet again they just helped nVidia cement that perception.

Lets not forget which end of the market makes the most profit per card.

[scaryjim]

... Lets not forget which end of the market makes the most profit per card.

I haven't, but if you're not making *any* profit, selling a lot of low-margin cards is better than hoping to sell a few high margin cards. AMD don't have the halo factor of having the majority market share, they don't have a huge R&D budget, and they clearly didn't have a high end card ready last year (partly, it appears, due to betting the high end on HBM2 then having to wait for volume production). They bet their bottom line on selling cheaper cards in volume, and it got them significantly more sales than they've managed for several years.

AMD's recent financials have been good; the loss they made last quarter was entirely due to writing down a one-time charge for varying their wafer agreement with GloFo, and without that they'd've been profitable with gross margins of 31%. So I think it's reasonable to read into that the fact that they made a profit on those increase sales.

[CAPTAIN_ALLCAPS]

OK, let's just look at a) discrete GPUs, and b) actual shipments. If you look at the figures over the last few years, nvidia's shipments have been fairly static, for all they follow a seasonal pattern - lowest sales in Q2, highest sales in Q4, figures generally in the 8M - 10M units per quarter. If you look at AMD over the same period, their sales plateaued around 5M units a quarter through the whole of 2013, then dropped steadily to a nadir of 1.7M units in Q2 2015. Over the following year they rose steadily to 2.8M units in Q2 2016, then leapt to 3.8M units in Q3 2016. That's the same increase in one quarter than they'd managed in the whole of the previous year. Coincidence that it happened in the quarter after Polaris was released? I don't think so, personally. This quarter they've maintained a small increase in dGPU shipments (3% - 4% depending on the relative distribution of desktop and notebook graphics in the mix). They're now back to the kind of sales levels they were seeing in mid-2014, and with a trend firmly in the opposite direction. You can say what you want about the technology, but financially they made absolutely the right call.



a) I wouldn't trust the steam survey to be even vaguely accurate, tbh.
b) In 2016 NVidia sold > 2.5 discrete GPUs for every 1 AMD sold. There are, quite simply a LOT more nvidia GPUs out there than AMD. So even taking the steam survey with a pinch of salt it's not entirely surprising.

OTOH, in 2015 NVidia sold 4 discrete GPUs for every 1 AMD sold. So for whole year market share, AMD's gone from 20% to more like 30%. And what's more interesting: in 2015 nvidia sold ~ 35M GPUs and AMD sold around 9M. In 2016, nvidia sold around 35M GPUs and AMD sold around 13M. So nvidia have sold no more GPUs this year than last year, while AMD have sold around 50% more.

Selling 50% more GPUs this year than last year? I'd call that at least a successful strategy, if not a winning one...

Well firstly, the Steam Hardware Survey might overstate the marketshare of older GPUs, but people who buy a new graphics card are going to be using it with steam and I severely doubt they are spoofing their video card descriptions whilst doing it. So I am confident that the Steam Hardware Survey's relative marketshare within concurrent generations is a good ballpark figure. And it isn't as though the gaps between the 480 and the 1060, 1070 or 1080 are small.


As for your sales figures, well those 5M units per quarter AMD shipped in 2013 are rather revealing since the cards on sale then were still the first generation of 28nm cards released in 2012. That was also the generation where AMD led with the excellent (and large) HD7950 and HD7970 - something they had to rebadge three or four times.

So despite a big jump in demand due to the length of time we have stagnated at 28nm, AMD managed fewer sales with smaller margins than they did with year old cards in 2013 even though they deployed a strategy that targeted mass sales.

AMD/nVidia is not AMD/Intel; AMD were and should still be competitive in the GPU space - and Polaris' performance is competitive per mm2 of die. So that sort of sales performance is not good enough given how competitive Polaris could have been. The Polaris 10 die - which struggles to tie with a 1060 - was simply too small when the true mass market looks to be between the 1060 and 1070 as gamers get older and richer. I'd say the decision to make Polaris so small was a bad strategy.


One more observation from the Steam figures - which is more illustrative than definitive given my earlier caveat. Another notable graphics series that still has a higher reported market share (1.17%) than the RX480's 0.47% is the HD7900 series, like my current HD7950 that I bought in 2013.

[Xlucine]

Remember fury, that narrowly lost to the 980ti at launch and was considered rubbish ever after? AMD know whether it's better to focus limited resources on the top end or bottom end, because they've done both now.

[scaryjim]

Well firstly, the Steam Hardware Survey might overstate the marketshare of older GPUs, but people who buy a new graphics card are going to be using it with steam and I severely doubt they are spoofing their video card descriptions whilst doing it. So I am confident that the Steam Hardware Survey's relative marketshare within concurrent generations is a good ballpark figure. ...

It's not spoofing or misreporting I'm worried about, it's sampling. The method is completely opaque, so we have no idea how it calculates any of its figures. As a result, I'd always take it with a pinch of salt. If Valve ever release details about sapling rates, assumptions, and the modelling they use to produce those figures, I'll start considering taking it seriously. Until then, I'll treat i the same as most browser-share figures - a useful finger in the wind, no more.

AMD/nVidia is not AMD/Intel; AMD were and should still be competitive in the GPU space ...

Here's the thing - in 2010 AMD were selling > 7M units a quarter. IN 2011 it was around 6.5M, in 2012 it was 6M, in 2013 it was 5M. The decline occurred over the course of five years, even when they had those very successful and highly regarded products you were talking about.

There's no point comparing their sales now to their sales in 2013, because they're not starting with the same market position. What they have done is halted the decline, begun a small recovery, and then - by targeting a particular market - won back significant shipments and market share in a very short time. I really don't get it if you honestly can't see that that's a huge success from where AMD started last year.