AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[tential]

Are they 'several months' ahead of cut-down GP104 / GP106 (whatever comes next)? Not so sure.

Also he says 'especially', not 'exclusively' notebooks and the mainstream market.

It's even more than that as well. Read the full context of the interview it's completely unclear exactly what he means. I don't think they're a couple months ahead of Gp106. I think any person downplaying Nvidia's business savvy in comparison to amd doesn't know what theyre talking about. Nvidia is more than prepared for Polaris with GP104.
Like I said before, being first to market with $200 gpus means nothing. The $200 market doesn't rush to buy gpus on day 1.

Amd needed a high end win to trickle down their low-end.

The only thing amd gets with Polaris is apple and they already had apple.

 

[maddie]

Fantasy is believing that somehow everything is fine when the scraps of evidence and rumor point to it being the opposite.

I'll just reply to this as I see it as the key.

Unreasoned rumour is a waste of time as far as I'm concerned. We have too many here who post the most ridiculous things to incite a nasty response.

For my part, I see no evidence, just some saying if I was in charge, I would do things differently. One is free to do so, but that means nothing to AMD. To say however that if they don't do as you recommend, means something is wrong is fantasy, in my opinion. In 3 weeks time we'll see what happens. Who knows, you might be right.

 

[Armsdealer]

AMD has to be more aggressive on price because they're not targeting the top of the market with Polaris. ... Etc

RussianSensation has many times pointed out that amd share losses have been primarily in the mainstream and mobile. Amd themselves at Computex indicated their goal is to take share. There is nothing going on from this perspective that doesn't make sense. Most people buy cheap cards. Rx480 is a logical entrant to help solve amd's problems. Whether it succeeds or fails remains to be seen, but my bet is that nvidia isn't interested in creating a price war with amd in the mass market, so it should do well. The gpu market is moving toward duopoly pricing.

 

[Silverforce11]

Like I said before, being first to market with $200 gpus means nothing. The $200 market doesn't rush to buy gpus on day 1.

Amd needed a high end win to trickle down their low-end.

The only thing amd gets with Polaris is apple and they already had apple.

It's pretty simple. AMD wants to regain marketshare.

You can't majorly swing marketshare with high-end priced GPUs. Unobtainium to the masses..

 

[tential]

I think I understand your confusion.

He says.

"We believe we’re several months ahead of this transition, especially for the notebook and the mainstream market."

Transition in this case is the introduction of 14/16 nm generation to replace the 28nm one. Not as you might be thinking, first to 14/16 nm. When the replacement of all present GPUs have taken place, the transition will be complete.

Understand now. It makes sense, as it means that AMD expects to have the full range earlier than Nvidia especially for the notebook and mainstream. One can assume that Vega might be also on track to beat Nvidia big die to market.

I believe that may be completely wrong. When you read the transcript he says that but the line of talking is a lot more general and isn't specific about being first to market. Its more about bringing a suite of things to the consumer.

It's so much more confusing when you look at the comment in its actual context

 

[tential]

It's pretty simple. AMD wants to regain marketshare.

You can't majorly swing marketshare with high-end priced GPUs. Unobtainium to the masses..

Having a 1-3 month lead in a mainstream market more than happy to wait doesnt seem like the best way to get marketshare.

If amd had focused on the 100-250 market I'd be more inclined to agree their strategy may work but without any strong product coming out now at the 150 mark I just don't know.

 

[littleg]

I'd actually look at it the other way here. AMD is executing the launch as they originally planned (and, bizarrely for them, doing it right). They're sticking to their announced schedule, taking time to build inventories and ensure they have enough chips for a decent presence on shop shelves as well as their design wins.

For me it's Nvidia that have panicked a bit and rushed their release out early. You can see from the extremely low availability that they haven't had time to build inventory and are now scrambling to get enough chips made to show some stock in retail and supply their board partners.

If they'd released with a 480(X?) on the market for $199 - $250, inevitably comparisons would be drawn between the cards even though they're at different price points. The narrative would be '40% more performance for 250% of the price' and poor value comparisons with 480 CF.

By releasing early they've ensured that the comparisons in the main are between the 1080 and a $550 Fury X and 980ti which still makes it poor value but not as bad as it would seem if Polaris was already on the market in volume.

All of this is supposition of course as we simply don't have any real news, whether good or bad, on production and yields and whatever else. One thing's for sure, it's going to be a very interesting month!

I'm hoping to build a new PC at the end of the year / early next year and a Zen / Vega combo would be nice.

 

[Armsdealer]

Having a 1-3 month lead in a mainstream market more than happy to wait doesnt seem like the best way to get marketshare.

The fundamental issue with everyone's comments is that they assume amd and nvidia are still playing the same marketshare game they've been playing for the last ten years. This is a rational thing to do in an industry that isn't perceived to have much growth, but what everyone is missing is in the last three months nvidia's share price is up 125% after going nowhere for 15 years. Hint: It's not up because they're about to take more share from amd. It's up because the gpu business is about to see more growth due to VR, driverless cars, and AI. What that means is nvidia's #1 goal shifts from using its profits to take take marketshare via an ongoing pseudo-pricewar with amd to trying to use its profits to prevent other entrants (eg intel) from entering the market as well as investing cash to expand to correctly create products to capture those new markets. It also becomes rational for them to form a quid pro quo truce with amd (duopoly pricing) so they can successfully milk the opportunity in the expanding market while not being undercut by a competitor.

This concept is business 101 and kind of described here:
http://www.quickmba.com/strategy/porter.shtml

I'm going to leave it at that before this goes off-topic.

 

[Ma_Deuce]

Having a 1-3 month lead in a mainstream market more than happy to wait doesnt seem like the best way to get marketshare.

What exactly is making the "mainstream market" more than happy to wait? Also, what is at the end of your 1-3 month rainbow?

 

[JDG1980]

Are they 'several months' ahead of cut-down GP104 / GP106 (whatever comes next)? Not so sure.

I suspect they probably are. Everything we've seen so far on GP104 points to Nvidia being supply-constrained, and it will take time to change that. I still think it was originally supposed to be a Q3 release and got rushed out because Nvidia wanted to beat AMD to market on FinFET. This is workable for a product that currently beats everything else on the market and can be sold at gouging rates, but GP106 needs to be cheaply mass produced.

 

[Vesku]

Having a 1-3 month lead in a mainstream market more than happy to wait doesnt seem like the best way to get marketshare.

If amd had focused on the 100-250 market I'd be more inclined to agree their strategy may work but without any strong product coming out now at the 150 mark I just don't know.

Judging by 960 2GB sales and the amount of people wanting to know when and where they can buy "Founders Edition" 1070s I don't think most GPU buyers put off purchases by months. If something hits their personal balance of performance and cost, or marketing/reviews convinces them it's time to upgrade, they buy it.

 

[JDG1980]

I think that if yields were better, AMD would be talking about their $300+ 480X, but they couldn't get enough chips that had enough active CUs at a high enough clock speed to sell a $300 chip. I think Polaris 10 has the capacity to compete with the 1070, but not when fabbed on Global Foundries 14 nm node.

Even if GloFo is having yield problems on 14LPP now, that won't last forever. They do eventually get their nodes right, even if it takes them longer than some of the other fabs. My guess is that if yields, validation, or some other factor makes full Polaris 10 unviable for now, that will change in the next 4 months or so.

Some part of me thinks Polaris was designed as a low power efficiency part because AMD had no confidence in Global Foundries to allow them to make high performance silicon.

No evidence of that. As far as we know, Vega (either a Q4 2016 or Q1 2017 release depending on what leaks you believe) is still going to be on GloFo. Most of the heavy lifting on 14LPP was done by Samsung anyway; GloFo is basically tasked with implementation of an already engineered process.

 

[maddie]

Having a 1-3 month lead in a mainstream market more than happy to wait doesnt seem like the best way to get marketshare.

If amd had focused on the 100-250 market I'd be more inclined to agree their strategy may work but without any strong product coming out now at the 150 mark I just don't know.

From this I'm assuming that you don't think P11 cards are being released soon?

If yes, then that's a big assumption. Information being scarce, we can't assume anything with any certainty.

 

[renderstate]

What's the advantage of gaining marketshare for the sake of it (i.e. by making little to no profit) ?

 

[Armsdealer]

What's the advantage of gaining marketshare for the sake of it (i.e. by making little to no profit) ?

Let total fixed costs (chip design) = A
Let fab costs per chip = B
Let number of units sold = n


Cost to make a single unit = B + A/n

AMD is running a loss in GPU because of the A/n term because n is too small

 

[ZGR]

What's the advantage of gaining marketshare for the sake of it (i.e. by making little to no profit) ?

http://www.quickmba.com/marketing/market-share/

There's a lot of advantages to gaining hold of the GPU market share even if it makes little profit. I am not sure how much money AMD is making from a $200 GPU, but it certainly isn't "little to no profit". Even if the profit is miniscule, selling lots of GPU's will offset that.

Gaining hold of the market share also lets AMD leverage exclusive features, improve brand image, and increase their influence on key OEM's.

 

[sirmo]

What's the advantage of gaining marketshare for the sake of it (i.e. by making little to no profit) ?

You're missing the point. rx480 was always going to be a $200-250 card. It replaces Tonga.

AMD has said why they are making a product for this segment. In order to increase the TAM as they put it. Basically to lower the barrier of entry into PC segment. These are folks who have been playing their Xbox or PS4 and have heard of how awesome CS:GO is on PC perhaps but couldn't afford to buy a fast GPU. Or people who want to upgrade their Dell or whatever store bought computer to be VR ready.

Now they can.

There are also other advantages to a higher volume product.

- price of components goes down. VRAM, voltage regulators, cooling fans, jelly bean electronic components all become cheaper when they are ordered in larger quantities, similar is true for wafers, production can go longer without setup.

- ecosystem.. more gaming systems powered by AMD gives AMD the ability to influence development. They already control consoles, this gives them mainstream PC.

- inertia. Business folks will tell you, but inertia is a really important aspect of doing business. Folks who buy rx480 are more likely to buy another AMD card in the future.

A disruptive price of $199 could very well make them more profit in the long run than if they marketed that same product at $299 despite the better margins.

When you look at the position AMD is in, changing up their release pipeline to mainstream first is a brilliant move imo.

 

[Slaughterem]

Originally Posted by Mopetar

Explain to me in what world it makes logical sense for steady rumors to suggest that AMD is months ahead of Nvidia on 14 nm only for them to launch a month later even though they're launching a smaller die which by all counts should mean better yields as well as more supply in general when they've likely got the entirety of GF's 14 nm production capacity to themselves whereas NV has to contend with several other companies buying supply from TSMC.

It takes 90 to 120 days to complete a wafer on 14nm. Do you go the route of a paper launch in a high volume mainstream segment or do you stock up inventory to meet the expected demand? And yes it appears that they are months ahead in this segment especially for laptops and OEM desktops, even if NV rushes out 1060 card that will not be able to compete with the 480.

As I said, either they're being really coy because they've got a big surprise with a 480X that is uncut and clocks much higher, allowing it to trade blows with the 1070, or they've got a 480 that's still a great value at $200, but is being held back by issues at GF getting their 14 nm process working. GP104 is almost half again as large as Polaris 10, but Nvidia is selling uncut dies, so where's the full Polaris 10? I've said that AMD could very well be sandbagging to throw off Nvidia, but the other side of that is the process has issues that need to be sorted out.

This is a new architecture, do you maybe think that the amount of shaders per cu is different? Maybe 1 cu is not 4 X 16 ALU? Maybe 36 cu does not equal 2304 shaders. Maybe the patent they received makes them have 2 X 16 ALU with 2 X 8, 2 X 4, 2 X 2, and 2 highspeed scalers. maybe its 2160 shaders with 72 high speed scalers capable of running 4 threads in 4 clock cycles which would give them 2448 effective shaders?

Fantasy is believing that somehow everything is fine when the scraps of evidence and rumor point to it being the opposite.

From what we know about the 480 I think it's a great value. Adored had a video that did some math to conclude that the 480 probably only uses a little under 100W based on some claims that AMD had made during their presentation. Assuming that's true, it suggests that card has some definite OC potential simply due to the additional power that can be throw at it, but it could be a case of an immature process limiting the true potential of Polaris 10 at this point if it doesn't scale particularly well beyond the stock levels.

With a new arch how can you set expectations as compared to prior arch? They already indicated that it is up to 2.8 X the performance per watt of the prior gcn arch. 1.7 coming from the shrink and 1.1 from the new arch. They are not driving performance to compete against NV they are driving performance per watt and performance per dollar.

 

[maddie]

What's the advantage of gaining marketshare for the sake of it (i.e. by making little to no profit) ?

By posting this question in this thread I'm assuming you are making the reader think that AMD is making little to no profit. At least you appear to admit they will be gaining marketshare.

I see several posters have answered you directly already on your specific question.

Now, I would like to ask one.

How do you know or where can you show how much, or how little profit is being made at these prices and what are the volumes expected?

Very interested in your answer.

 

[nurturedhate]

By posting this question in this thread I'm assuming you are making the reader think that AMD is making little to no profit. At least you appear to admit they will be gaining marketshare.

I see several posters have answered you directly already on your specific question.

Now, I would like to ask one.

How do you know or where can you show how much, or how little profit is being made at these prices and what are the volumes expected?

Very interested in your answer.

Go ask a 3rd party marketing firm, you'll probably get a similar answer.

 

[sirmo]

Go ask a 3rd party marketing firm, you'll probably get a similar answer.

what would a marketing firm know about the economics of producing GPUs?

 

[topmounter]

what would a marketing firm know about the economics of producing GPUs?

Hopefully that is his point

 

[Mikeduffy]

Rumoured GP106, 1280 Cores, 192bit bus, 6GB GDDR5, 1600MHz base clock, 110Watts

I'm assuming most are in agreement that GP106 will be max around 55% of GP104 at 1080P.

Anyhow, I don't think GP106 will outshine the 480X, but I do think that the hoopla around single-pass stereo could create in image issue for AMD.

Is there any way for Polaris' New Geometry Engine to perform an equivalent single-pass stereo? Is this feature dependant on specialized hardware? Can existing AMD hardware perform this?

 

[Armsdealer]

How do you know or where can you show how much, or how little profit is being made at these prices and what are the volumes expected?

Very interested in your answer.

https://semiaccurate.com/forums/showthread.php?t=8757&page=157

Good discussion on this here. Broadly speaking, they estimate AMD is making around $30-$40 in gross profit per RX480. In order to cover the $80-$100MM in R&D expenses that it costs to make a 14nm part

http://semiengineering.com/10nm-versus-7nm/

you need to sell 2-3MM units before making a dime in real profit ($80MM / $40). This is why market share losses are so painful. Nobody loses money on the silicon, they probably almost always make ~40% of the sale price (not $199 mind you, but $199 - retail - distribution - PCB - RAM = ~$80-$100), but the painful part is that engineers are expensive, and the smaller these processes get, the more expensive they are to design. Again, I don't know the exact numbers, because they aren't broken out in the financials, and this is a very broad back of the envelope calculation, but it should serve to illustrate the general idea.

Keep in mind, this does not take into consideration the fact that a lot of the R&D for PC GPU is shared by semi-custom / embedded business (console), but the gross profit margins there are very low according to AMD's 10-K

" Gross margin, as a percentage of net revenue for 2015, was 27% compared to 33% in 2014. Gross margin in 2015 was adversely impacted by an inventory write-down of $65 million, which was primarily the result of lower anticipated demand for older-generation APUs and a technology node transition charge of $33 million. The impact of the write-down accounted for approximately two gross margin percentage points and the technology node transition charge accounted for approximately one gross margin percentage point. Gross margin in 2015 was also adversely impacted by a lower proportion of revenue from Computing and Graphics segment due to lower sales which has a higher average gross margin than our Enterprise, Embedded and Semi-Custom segment and due to lower game console royalties."

from
http://www.sec.gov/Archives/edgar/data/2488/000000248816000111/amd-12262015x10k.htm

Also note all of these numbers are *before* administrative & marketing expenses, so that will also need to be deducted before you get to any sort of operating profit number. That number is another ~50% of R&D company wide. Maybe bump up that 2-3MM unit number to 3-4MM units to be conservative.

 

[boozzer]

http://www.quickmba.com/marketing/market-share/

There's a lot of advantages to gaining hold of the GPU market share even if it makes little profit. I am not sure how much money AMD is making from a $200 GPU, but it certainly isn't "little to no profit". Even if the profit is miniscule, selling lots of GPU's will offset that.

Gaining hold of the market share also lets AMD leverage exclusive features, improve brand image, and increase their influence on key OEM's.

a time saving tip, you need to ignore poster. :thumbsup: