The rumors about the product schedule are false -- AMD

[Final8ty]

Recent AMD hire Roy Taylor, Vice President of Global Channel Sales, has asked AMD to break the yearly update cadence they began with Evergreen, under the Sea Islands series of products. His reasoning? AMD has the best graphics products but people don't think so.
View Image
AMD Radeon Graphics 2013

In a recent AMD Press Briefing, AMD gathered their team to address the recent controversy stirred by their own information release - that the AMD Radeon series will by 'stable throughout 2013'. What they meant by that was actually 'a top to bottom stack of new products in Q4'. Nice to have some clarity. So, is this Sea Islands? Nope, Sea Island is the codename for the codenames that make up the products of 2013. Yeah. AMD know we love codenames, so ...

Mobility

As previously reported, the new AMD Radeon HD 8000M series is codenamed Solar System. Thus far, the Mars GPU has launched - back in November 2012, in fact - and availability is ramping. This GPU doesn't feature a new architecture; AMD is sticking with GCN for the foreseeable future. Given how long VLIW5 lasted after being introduced in 2006's R600, that's historically consistent and we expect that previously seen tweaks and improvements will be reflected in future cards. Mars is a new configuration of GCN cores and design and more new 8000M series products will be released this year, with new configurations of GCN. HSA feature upgrades will be driven in the APU products.

Desktop

At CES, AMD announced the AMD Radeon HD 8000 OEM series, which are exactly the same as the current desktop HD 7000 series. This was done to placate the OEM customers who love new numbers every year, and they get a new card the retail channel didn't, based on the Oland GPU. The desktop side didn't get this treatment; in fact AMD will be launching multiple new Radeon 7000 series cards before the end of 1H13. This is not limited to Oland, as one of AMD's repeated points was that they will sustain the price/performance leadership they currently own now, throughout 2013. This kinda hints at an official AMD product in the retail channel using the New Zealand design, although AMD claim Powercolor's Devil 13 and ASUS ARES II as their own, as AMD countered discussion of NVIDIA's upcoming titanic single GPU card.

Sea Islands

Sea Islands is now the codename for the graphics updates happening in 2013, so this encompasses Solar System mobility GPUs and the OEM rebranding of the desktop 7000 series to being 8000 series. It also includes new HD 7000 desktop GPUs, some of which may be new chips - but not new architecture - and others that are the result of binning, power management (i.e. turbo) and clock speed increases. Charlie Demerjian at SemiAcurrate published details back in January of the AMD Radeon HD 8950, details of which he reiterated in the call; it's a clock speed bump with better thermals but still a Tahiti Pro and the card is production ready, has been for a while. So why isn't it launching in March of this year like we originally heard rumored? This goes back to Roy Taylor's comment; he asked them not to do so until AMD's cards are recognized as the best in the business. The timing here doesn't really line up - it looks more like Project WIN forced a re-examination of priorities and an adjustment to what codenames mean, and what products are branded. With NVIDIA's 7-series GeForce cards likely following the pattern set with Fermi, the GK114 GTX 780 expected to launch in March, the performance delta that AMD has to overcome is around 30% above the 680. Is that something a 15% faster Tahiti XT 2 can beat?

ATI vs NVIDIA

A big statement in the briefing was that AMD wants to bring back the rivalry, the 'we're better than you' fight that will have comment moderators everywhere groaning. AMD believes they have the best line up of GPUs, and that should be evident everywhere to everyone. When asked to clarify what best means, I was referred to AMD's many wins for Gaming Evolved, where titles and drivers are optimized for triple-A titles before launch. Offered the opportunity to comment on smoothness instead of raw performance being the metric people are using to gauge the 'win', AMD reiterated their commitment to drivers as a key part of their ongoing initiative for their products, and specifically looking to smoothness concerns.

New Cards

AMD will finish out 2013 with a new top to bottom line of cards, replacing the HD 7000 series. It's not clear at this time if these will be branded the AMD Radeon HD 8000 series of it they will be something else for 2014, back aligned with the OEM series. Codenames like Volcanic Islands and Pirate Islands went undisclosed, but 2014 is the target for HSA integrated GPUs. 2014 is also the target for GlobalFoundries 20nm process and is expected to be the cutover from TSMC to GloFo, as AMD keeps it 'in the family (of investors)'.

Clarity

The call left many of the participants expressing confusion or bemusement, but did in fact go a long way to define what AMD is doing right now, and set a marker for when new cards will be available. In the sense that we don't have details on new architectures, products, speeds it was a disappointment but that's hardly surprising - AMD claims the HD 7000 sales are still rising, with no peak yet in sight thanks in part to AMD's Never Settle and Never Settle Reloaded campaigns, although we're mindful that they didn't disclose the markets in which sales are ramping. For gaming enthusiasts, AMD is keeping quiet on what's coming but that there is more to come from the HD 7000 series; AMD is not done yet and ended the call with a strong statement - 'We believe we will maintain leadership.'

http://www.rage3d.com/articles/amd_there_is_no_sea_islands/

 

[zlatan]

It doesn't make sense to me, they have a decent product why sit on it?

I think the real questions are not answered. I spend a lot of time to analyze both Kepler and GCN, and the results are pretty mixed, just like what I'm expected. But with branchy codes the GCN is far more powerful. With the shader tracer API I think Microsoft did a good job, at least I'm amazed how easy to debug a complex shader code. This leads to some questions, especially when we see a "top secret DX11 effect" reference in the Tomb Raider slide. So, what is cheaper? Creating new effects for AAA games, which are designed to run well or run only on the GCN architecture, or simply launching some new graphics cards. Did AMD realized that they have a really revolutionary architecture, and also devs want to work them (because the console design wins? ... or just because NVIDIA is now more like a smartphone company?)?
There are many situations where the Kepler can't match the GCN architecture, and now devs have the tools to write more complex, more branchy shaders. It is a logical step for AMD to write some fully new effects, implement them in the games, and that's all. Even the existing cards can easily win againts Kepler in this scenario.

 

[RussianSensation]

Well if they saying anything otherwise people would hold out and wait.

Your theory still doesn't make sense. Majority of consumers would not wait 9 months for new GPUs even if AMD announced HD8000 series launching exactly in Q4 2013. If someone wants an upgrade for 2013 games, they are buying in the 1st half of this year. Games like Bioshock Infinite, Crysis 3 and Starcraft 2 HoTS will create waves of upgraders who won't be waiting until fall 2013 for a mere 20-30% potential bump in performance. Clearly AMD doesn't have GCN 2.0 series waiting in the wings in the form ready to go because they have hundreds of millions of HD7000 cards. HD7950-7970 are selling out at places like Newegg, Amazon, Superbiiz and Scan.co.UK. You can't have 9 months of inventory overstock and "sell out/out of stock signs".

Is AMD going to give up for a while? Sure reads like it.

Ya, the 0.3% of the PC gaming market they'll let NV have that with the Titan.

 

[dangerman1337]

http://www.rage3d.com/articles/amd_there_is_no_sea_islands/

I think this is quite beliveable and probably means that it'll be like Nvidia's 200>400 series (though hopefully nothing like Thermi ;p) but put out "Sea Island" GPUs as 7900 series (maybe 7980 and a cut down 7960 for example?) and similar.

 

[RussianSensation]

And yet it's a dual role card, what kind of penetration did AMD make with workstations? Couldn't have been too deep.

Precisely why it makes no sense for AMD to design a 520mm2 chip since they won't sell many in the gaming market, nor in the workstation/professional HPC markets.

Yet still Nvidia made great profits selling a 520mm2 40nm die for $370.

Looking at NV's gross margins during Fermi and Kepler eras, the price of 28nm wafers must have gone up. This aligns with NV's price increases per mm2 of die, $499 for 294mm2 680 and $899 for a 520mm2 GK110, up from 520-526mm2 of GTX480/580 at $499. Their gross margins barely changed. This suggests 28nm wafers are way more expensive. Therefore, AMD could not possibly launch a 520mm2 28nm die at $370-550.

I think you greatly over estimate the cost to fab these chips. R&D is costly, wafers were costly, the chips themselves in bulk amount to a fraction of the price you pay at best buy.

I wasn't talking about manufacturing them, but the total cost to get the chip from hypothetical/on paper form to retail. For AMD the financial benefit is not there to make 500-520mm2 28nm chips. That would force them raise prices far above $499 level at which point the sales are minimal and the Gross profits (Gross margin per chip x volume of chips sold) are too small to justify the investment. AMD also has never manufactured a GPU this size. Their GPU strategy is not aligned with this because they don't use large die chips to service 2-3 distinct market segments like NV does successfully.

Nvidia isn't throwing up record profits because chips are costly, even the difference between GF114 and GF110 was trivial compared to their retail prices.

That statement supports even more what I said to you. Large chips on 28nm wafers are more expensive to make this generation than in the previous ones. Therefore, AMD couldn't possibly be able to sell a $369-499 500-550mm2 graphics chip. Going beyond that price level would kill your volumes, ultimately making such a product a pure marketing exercise. It seems AMD thinks selling more cheaper chips in larger volumes with game bundles would be more profitable than spending hundreds of millions to hire even more engineers to design a gargantuan 550mm2 chip for gaming alone. AMD's transistor density is also higher than NV's which means their larger chip would be even hotter and use more power than NV's equivalent. Sounds like an exercise in inefficient use of human capital and financial resources mismanagement. AMD needs R&D money for next key battle which will be 20nm against Maxwell. Even when AMD released HD7000 series low-end and mid-range cards 5-6 months before GTX660 and below even dropped, it hardly made a difference. Your suggestion that launching HD8000 series would somehow change AMD's financial position doesn't support what even happened when AMD had 28nm chips going against 40nm hotter slower Fermis. People either still bought NV's Fermis or kept waiting 6+ months for Kepler <$300 parts. If that's the case might as well pour your resources towards your next key battle - 20nm - and just coast for the remaining 12 months of HD7000 desktop, with more focus on mobile side where HD7000M lost a lot of market share. This is exactly what AMD is doing. Further, it could very be the case that they need 20nm to get much more performance at a reasonable level of power consumption. HD7970Ghz isn't exactly using 150W of power.

Nvidia just hit over 60% market share, but their product is inferior. How long do you think Nvidia is going ride this train? Seems like the engine is running out of steam, and you can't make a faster 680. I'm positive Nvidia isn't interested in selling "Flagship" moniker cards for sub $350, which is where they need to be to stem "this tide"

They could launch the rumoured GK114. But as you said NV is already doing well despite selling slower more expensive chips with less extravagant game bundles. That means even if HD8000 series launch, it doesn't mean anything because consumers may just as well continue buying GK114. NV also has no incentive to lower prices if their fanbase keeps buying GTX680 for $440+. Why would they lower the price to $350? HD7970Ghz is 10% faster at 1600P. Even if GK114 is 30% faster than 680, it would only be 18% faster than the HD7970Ghz. AMD can just lower the price of HD7970Ghz from $430 to $369 and GK114 is $499. That would be the same situation as GTX580 vs. HD6970. If HD7970Ghz was selling for minimum of $499, then it would be different.

I think Nvidia will move in with a refresh, why not? They can, they have money they don't need to sit around twiddling their thumbs playing a close race when they can simply walk away with it. AMD has made it quite clear their only method to counter now is price reduction and expensive game bundles. If I had to guess their new plan is to move bulk at reduced prices, because they've lost so much market share.

Most of the market share they lost was in the mobile space, not desktop one. Q3 2012 AMD lost just 2% market share on the desktop but 17% in the mobile. If you were in charge of AMD you'd throw X times more resources at the mobile HD7000M/8000M side, not desktop one. Notebook GPU market is also growing more than the desktop one. Since HD7000 desktop cards have very strong price/performance, performance leadership and AAA game bundles, the focus should be on the mobile side. When NV responds with GTX600 refresh, AMD can reassess. Right now it's much better that they keep working on their drivers since many new consumers are buying HD7000 cards. Memory management rewrite on HD7000 series should be #1 priority since it impacts MILLIONS of existing owners and decisions of potential HD7000 buyers whether to go HD7000 or GTX600. Releasing HD8000 out while not fixing the driver could risk repeating the poor 3-4 months roll-out of HD7000 series if GK114 outperforms HD8000 due to still immature HD8000 driver.

Never said it did, or didn't, only that not having it and waiting reminds me of their FX days (before the bad FX days). Sitting around on a design in this industry get's you smoked, FX vs Conroe smoked. Instead of "being predators" and going after the market (which admittedly they tried, to wit they lost tons of market share) they've regressed into "sheep" mode at the first sign of adversity.

What gets you smoked is letting 90% of your key business lines (CPU/APU/server) evaporate on you. No point sitting there hypothesizing about wasting millions of dollars trying to combat $900 marketing chips like the Titan. AMD was primarily a CPU/APU/server business. The ATI graphics business was purchased to advance their CPU-->APU competitiveness in 5-10 years, not to make billions of dollars in discrete GPUs. AMD was always thinking about competing with Intel first. The reason AMD is where it is today has little to do with their desktop graphics. Even if AMD doubled or tripled their desktop GPU profits, it wouldn't save the firm unless they turn around their CPU/APU/server business lines or find new growth opportunities in smartphones/tablets, etc. The reason AMD is so focused on graphics now is because it's the only strong product line they have that's not losing hundreds of millions of dollars every quarter. It's better to talk about your graphics segment than to talk about how you are getting killed everywhere else. That doesn't mean the discrete GPU graphics division is some huge money maker for them - it isn't.

 

[BallaTheFeared]

AMD Not Competing with Intel Anymore, Goes Mobile

Do we change that to Nvidia now, is that what you're saying?

AMD&#8217;s market share in desktop computers in the quarter decline from 40.7% in Q2 to 35.7%, while Nvidia&#8217;s rose from 59.3% to 64.3%. In notebook computers, AMD&#8217;s share fell more dramatically, from 44.8% to 34.2%, while Nvidia&#8217;s share rose from 55.2% to 65.8%.

I think some of your numbers are off, but I agree on other points like driver issues holding back current hardware.

 

[tviceman]

Ya, the 0.3% of the PC gaming market they'll let NV have that with the Titan.

I think the bigger issue is that if Nvidia is respinning GK104 / GK106 / GK107 to increase yields and (to a much minor degree) performance, they will be squeezing AMD up and down it's product lineup. AMD is already selling a bigger die with more ram and (generally) a more complex PCB for less money than Nvidia's comparable products in the same tier. The role of die sizes and prices have completely been reversed since 40nm, Nvidia has much more lateral pricing room right now than AMD does. IF a GK114 yields high enough that Nvidia only bins 3 skus for it (instead of the 4 they are doing now), they can effectively get away with selling it at a slightly lower price but with the same margins. Lets be realistic and say GK114 ends up 5-7% faster than GK104 at the same TDP and comes in at $399. It's neither a radical price drop nor is it a huge increase in performance, but it's a little bit of both and it's also got a new flashy series moniker tied to it.

My whole point? If AMD had such a difficult time selling it's cards through most of 2012, it will be just as difficult (if not more) in the face of refreshed products.

 

[RussianSensation]

I think the bigger issue is that if Nvidia is respinning GK104 / GK106 / GK107 to increase yields and (to a much minor degree) performance, they will be squeezing AMD up and down it's product lineup. AMD is already selling a bigger die with more ram and (generally) a more complex PCB for less money than Nvidia's comparable products in the same tier.

It makes no difference:

1) Even when AMD had the market all to itself with HD5000 for 6+ months, it hardly changed AMD's overall financial profitability;
2) Even when AMD had the sub $350 market all to itself with 28nm HD7000 parts (until NV launched GTX660Ti and below 6 months late), it hardly changed AMD's overall financial profitability;
3) AMD is a dying CPU/APU/server business. Discrete graphics cannot save the firm. It is financially/mathematically impossible. The only way AMD can survive if it turns around the rest of the company.

AMD is not a graphics card company, but a CPU/APU/server business. It only now seems on the surface that AMD is a "GPU company" because everything else they have is dying. You do not go from $2.9 billion in Gross Profits from 2010-2011 to $1.2 Billion because of graphics.

The best analogy is you have an ice cream store that sells ice cream and some ancillary desserts. 90% of your business has been successful because of ice cream (CPU/servers). You also sold cakes/chocolates (GPUs) on the side for people who don't like/aren't interested in ice cream. Due to changing competitive forces and market trends, suddenly consumers either don't want your ice cream because they now want fruit (tablets/smartphones) or they like the competitor's ice cream more. Even though you still make some $ selling cakes/chocolates, your ice cream store is becoming irrelevant and it's just a matter of time before you run out of $ unless you fix your ice cream or start selling fruit or something else like alcohol.

 

[SirPauly]

It makes no difference:

1) Even when AMD had the market all to itself with HD5000 for 6+ months, it hardly changed AMD's overall financial profitability;
2) Even when AMD had the sub $350 market all to itself with 28nm HD7000 parts (until NV launched GTX660Ti and below 6 months late), it hardly changed AMD's overall financial profitability;
3) AMD is a dying CPU/APU/server business. Discrete graphics cannot save the firm. It is financially/mathematically impossible. The only way AMD can survive if it turns around the rest of the company.

AMD is not a graphics card company, but a CPU/APU/server business. It only now seems on the surface that AMD is a "GPU company" because everything else they have is dying. You do not go from $2.9 billion in Gross Profits from 2010-2011 to $1.2 Billion because of graphics.

The best analogy is you have an ice cream store that sells ice cream and some ancillary desserts. 90% of your business has been successful because of ice cream (CPU/servers). You also sold cakes/chocolates (GPUs) on the side for people who don't like/aren't interested in ice cream. Due to changing competitive forces and market trends, suddenly consumers either don't want your ice cream because they now want fruit (tablets/smartphones) or they like the competitor's ice cream more. Even though you still make some $ selling cakes/chocolates, your ice cream store is becoming irrelevant and it's just a matter of time before you run out of $ unless you fix your ice cream or start selling fruit or something else like alcohol.

However, their graphic and software prowess may help redefine the company. For example with nVidia's graphic and software prowess helped redefine their company with growth from Telsa and SOC designs.

 

[RussianSensation]

However, their graphic and software prowess may help redefine the company. For example with nVidia's graphic and software prowess helped redefine their company with growth from Telsa and SOC designs.

NV capitalized on the first mover advantage when they pushed CUDA. NV created the HPC market from scratch more or less. Now that the majority of corporate clients are using CUDA and getting NV's technical support, the switching costs, risk/reward to a competing brand is too high. It would take so many years to crack into that market and make serious $, AMD would never survive that long given its current situation and existing market trends in the CPU/APU/server markets. If Steamroller and whatever is right after and next 2-3 generations of APUs fail, Jaguar/Kabini, etc. do not take off, AMD's ability to grow its IP revenue stream doesn't increase substantially (outside of PS4/720/Wii U design wins), if they don't break into the server market with some new products (SeaMicro, etc.), or come up with some smartphone/tablet SOC home run, or some other new product, it's probably game over. HD8000-9000 is not going to do much to save them. The overall profits in graphics are just too little to make a difference for AMD.

Let me put this in perspective. NV's Net Incomes in 2012-2013 were $581-563 million. In Q3-Q4 2012, AMD as a whole lost $630 million despite a profitable GPU business. That means even if NV DISAPPEARED off the planet and AMD took over all of the remaining NV market share to give AMD 100% market share in every desktop/professional/mobile GPU segment of NV, AMD would still be Net Cash Flow Negative for the entire year. This is why AMD's Graphics division cannot save them. It is not mathematically possible unless PS4/720/Wii U fail and people start buying millions of graphics cards by shifting to PC gaming. Also, NV can't just disappear....

You say AMD's graphics can define them but their APUs already have class leading graphics performance. It's not helping. Even if AMD's APUs were 2x faster than Haswell, it wouldn't make any difference because for people buying basic i3-i5s, they don't care about graphics performance. Gamers are buying laptops with discrete GPUs. The entire APU strategy is flawed if AMD only focuses on APU's graphics performance. If the GPU inside the APU could perform common CPU-related tasks most consumers use, only then they will start to care to trade off higher power consumption of AMD's APUs for Intel's CPUs (and that trade-off is not going to change due to Intel's 1.5+ manufacturing node advantage). Right now most consumers would rather have a faster CPU with lower power consumption rather than a faster graphics card in their APU at the cost of CPU performance and higher power consumption/lower battery life.

 

[SirPauly]

Will see what Rory Read can do over time!

 

[RussianSensation]

Will see what Rory Read can do over time!

AMD's management knows the answer is not videogame graphics, but using graphics cards for general purpose compute to aid their CPUs to make them more attractive for general PC usage/experience. This has always been the main reason for buying ATI but AMD failed to execute on this strategy in a timely manner. AMD's old management saw the freight train coming light years away and panicked picking up AMD at $5.4 billion thinking they could execute on HSA fast enough before their CPU division collapsed. Their greatest fear is coming true because the CPU business is dying faster than they can execute on HSA, while the overall GPU business is not profitable enough to sustain the company. Even if they execute HSA on the hardware side, they are going to have to spend millions of dollars to get developers to make programs for Windows that can actually benefit from this.

 

[BallaTheFeared]

Will see what Rory Read can do over time!

This goes back to Roy Taylor's comment; he asked them not to do so until AMD's cards are recognized as the best in the business.

Oh god...

 

[RussianSensation]

Oh god...

How clever would it be if Roy is NV's trojan horse?

This statement in the Rage3D article makes no sense: "With NVIDIA's 7-series GeForce cards likely following the pattern set with Fermi, the GK114 GTX 780 expected to launch in March, the performance delta that AMD has to overcome is around 30% above the 680."

580 was only 18-20% faster than 480. Right now the 680 trails 7970Ghz. Is the author assuming that GK114 will be 40-45%+ faster than a 680 and that AMD's memory management re-write will bring a 0% performance increase for GCN cards?

 

[SirPauly]

AMD's management knows the answer is not videogame graphics, but using graphics cards for general purpose compute to aid their CPUs to make them more attractive for general PC usage/experience. This has always been the main reason for buying ATI but AMD failed to execute on this strategy in a timely manner. AMD's old management saw the freight train coming light years away and panicked picking up AMD at $5.4 billion thinking they could execute on HSA fast enough before their CPU division collapsed. Their greatest fear is coming true because the CPU business is dying faster than they can execute on HSA, while the overall GPU business is not profitable enough to sustain the company. Even if they execute HSA on the hardware side, they are going to have to spend millions of dollars to get developers to make programs for Windows that can actually benefit from this.

I don't have the answers but maybe Hector Ruiz can offer much more insight with his new book -- Slingshot.

Former AMD Chief&#8217;s Book Describes Fight Against Intel

http://blogs.wsj.com/digits/2013/02/14/former-amd-chiefs-book-describes-fight-against-intel/

 

[SirPauly]

You say AMD's graphics can define them

Offered graphics and software may help redefine them. APU's maturing, server, professional sku's, SOC designs and their engineering experience with software!

 

[3DVagabond]

I think the only way AMD can turn around anything in the pro market is with super computers. Here, if AMD can have the fastest super computer they can establish a foothold.

As I've said before, nVidia is just too entrenched in graphics workstations for AMD to have any chance without a gigantic over investment.

 

[f1sherman]

Quote from Titan thread:

All they've really said is that new desktop cards released will be HD 7***. He distinctly does not say that Tahiti will remain their high end product He also said he can say more in a couple of weeks. What will change in a week?

Console announcements? Not sure why that would matter to the discussion about discrete graphics.

When after that STABLE THROUGHOUT 2013 slide, and after their weird official tweet, AMD announced press event,
promising to squash confusion and quickly clarify 2013 Radeon plans,
I thought to myself :hmm: "o-ho this is going to be funny"
And boy did they deliver.

Now we have press believing everything from No Sea Island, HSA DoA, not knowing which part is which, to imminent top-to-bottom release of GCN2.

Some of ideas on beyond3d on how to squash confusion:

Which suggests we should follow the advice Dave intends to give to the publisher of this document, and talk about graphics IP levels too instead of those non-descriptive codenames.

I propose that on these forums and for the sake of clarity, everything from Oland to Tahiti be referred to as GCN 1.0 (or GCN for short), and the upcoming changes described in the paper above be considered part of GCN 1.1.

When the big refresh comes at the end of the year, we might call that GCN 2.0. I know that no one died and made me king of the codenames, but since AMD is not communicating much about the micro-architecture, and just about any codename so far has been mostly confusing, I think this would be for the best. At least we might start understanding what the hell we're even talking about.
http://forum.beyond3d.com/showpost.php?p=1708140&postcount=4306

Remember how simple it looked back when OEMs got their HD 8000...

 

[RussianSensation]

What I understand is the next iteration of GCN 2.0 should be 20nm Volcanic Islands. The designation of GCN 2.0 for HD8000 (aka 28nm HD7000 refresh) was made up by through rumours, just like rumours made up that HD8970 was supposed to launch in 1H of 2013.

Whatever the case may be, it's too hard to get excited about 15-20% faster HD7970GE at $499. No existing HD7970GE/GTX680 would waste their $ upgrading to that. They are going to be waiting for 20nm parts or a cut-down GK110.

 

[boxleitnerb]

You honestly believe 20nm will be ready in Q4 2013? I wouldn't hold my breath.
15% increase is nice, why not do it? There are always people with older cards willing to upgrade. And AMD surely wants to charge $549 again for 350-400mm2. The only way to do this is to release new products soon instead of waiting a year for 20nm which at that time may be even more expensive and broken in terms of yields as 28nm was a year ago.

 

[blastingcap]

AMD has had a ton of layoffs, including at the seniormost levels, plus the CPU side has been bleeding tons of money as well. It may be the case that AMD doesn't feel like it's worth it to issue a mildly faster refresh on 28nm and has elected to "swing for the fences" with a an ace 20nm product. Not that either product would save the company but at least a much faster 20nm has a chance of greater impact, imho.

 

[f1sherman]

From our favorite troll:


And so the vast echo chamber of ignorance that people call internet news went in to overdrive, and the frothing began. There was no story, and nearly 100% of the information was published weeks or months prior. Did that stop the idiots from running around in circles hyperventilating until they passed out from exhaustion? Take a guess.

AMD having to do a few hasty press conference calls last Friday should tell you that these nitwits didn’t pass out from lack of oxygen soon enough, and that the public noticed too. Not that the unwashed masses did anything radical like, I don’t know, possibly the barest minimum search for what was really going on. Nope, they just read something on a site and took it for truth, then went off on the ‘injustice’ that was taking place. There are several brutal wars going on right now that get less indignation from the technical ‘in’ crowd. I should get into the bridge selling business soon, there seems to he a large and ripe potential customer base of nitwits with net access to milk. I meant the readers, not the ‘press’, they seem far less capable of filling out a purchase form on a web page based on this sorry episode.

To make matters worse, although the presenters on the AMD call had a clear message and described the CI line directly and plainly, there were two problems. First is that the parts that anyone cares about, basically the highest end of this buff and polish job, haven’t launched so there were enough, “I can’t tell you about that yet” answers to annoy even those who were borderline ODing on Qualuudes. That was the good part, the second problem was that the line itself is essentially crap that breaks over a decade of tradition and is segmented as clearly as mud.

http://semiaccurate.com/2013/02/18/...ounding-amds-sea-islands-ci-end/#.USJsJjthFol

 

[RussianSensation]

You honestly believe 20nm will be ready in Q4 2013? I wouldn't hold my breath.

No, I personally think 20nm won't be ready until 2014. HD7970GE reference card peaking at 238W at TPU is an indication AMD needs a full redesign of Tahiti XT perhaps around Pitcairn, or wait for 28nm node to mature even more before they can net a tangible performance increase.

15% increase is nice, why not do it? There are always people with older cards willing to upgrade. And AMD surely wants to charge $549 again for 350-400mm2.

Remember HD5870 --> HD6970 was difference since they fixed 5000's poor tessellation performance and added 2GB of VRAM standard. HD7970 does not have weaknesses in either of these areas that urgently require a refresh like was the case for HD5870. HD7970GE is faster at 1600P and in multi-monitor resolutions, it costs less, overclocks well and has good game bundles. AMD needs to shift its focus to mobile discrete GPUs, not desktop because their desktop line-up is extremely competitive and their notebook one is not. When you have limited resources, you have to pick and choose how to allocate them effectively.

FC3 bundle cost $4 million. Even if it costs $15 million for the new bundle, it's cheaper than spending $100-200 million to design a new HD8970 chip given AMD's finances. Most of the GPU sales occur in the $100-350 price levels. $550 HD8970 would not do anything for AMD because a $549 HD7970 didn't either. It makes more sense for them to allocate their resources to the most pertinent areas this time. I obviously would have liked to see GK114 vs. HD8970 but if AMD has no $ to launch such a card or thinks it's more financially advantageous to sell cheaper HD7950-7970 in slower quarters for GPUs (Q1-Q2), then one can see some of their thought process.

Looking at the excellent Kepler architecture, AMD's Graphics Division will need as many engineering and financial resources as possible to any shot at a decent showing on 20nm. 2013 is a year riddled with console ports and January 2013 was the highest monthly sales of HD7900 cards. It's essentially a year no one cares about knowing that the battle between 2014-2015 will be A LOT more critical. There is no pressure at all to launch HD8970 for $549 under these scenarios. When 2014 comes and NV unleashes what is likely an amazing Maxwell architecture, AMD better be ready or their graphics division is toast. There is almost nothing NV can do in 2013 that will suddenly make HD7000 a complete turd since at current prices HD7950-7970 are selling, even if GK114 is 20-30% faster than HD7970GE, AMD could just drop the price to $299 for 7970 and $369 for 7970GE and neutralize GTX600 refresh completely. OTOH, if Maxwell is 30-50% faster because it's such a superior architecture, then AMD will be toast. Conservation and allocation of resources strategically for 2014-2015 is more important.

AMD has had a ton of layoffs, including at the seniormost levels, plus the CPU side has been bleeding tons of money as well. It may be the case that AMD doesn't feel like it's worth it to issue a mildly faster refresh on 28nm and has elected to "swing for the fences" with a an ace 20nm product. Not that either product would save the company but at least a much faster 20nm has a chance of greater impact, imho.

I agree. If AMD doesn't have a competitive 20nm GPU series, they will be losing $ in both GPU and CPU segments. Right now their HD7000 series is making them $ and NV has worse price/performance on the desktop at nearly every level. AMD doesn't have to do anything until GTX600 refreshes show up, which we don't even know if they will until Q3-Q4 2013.

 

[boxleitnerb]

Well, with a refresh they could raise prices at all price points again, not just in the highend segment.
Any numbers on Q4 2012 GPU market share yet? Apparently they lost some share again, but I cannot read the article, it isn't public. Just judging by the headline:
http://www.jonpeddie.com/
AMD was at 35.7% in Q3, they have a lot of opportunity to regain market share.

 

[RussianSensation]

Any numbers on Q4 2012 GPU market share yet? AMD was at 35.7% in Q3, they have a lot of opportunity to regain market share.

8 out of 10 HD7950s and 6 out of 12 HD7970s are sold out at Superbiiz

Replacing HD7950 at $260-300 with an HD8950 at $450 and HD7970 at $360-400 with HD8970 at $550 for a mere 15-20% bump in performance would destroy any chances of selling a high volume of cards because in 2013 people are going to be less interested in spending that much $ to play console ports. In 2012 people paid that because 28nm brought a much larger increase in performance. Even then NV's own sales didn't take off until Q3 when they introduced $299 and lower desktop cards. They even said Q3 was unusually high quarter as a result of new Kepler product roll-out (which happened to be sub-$300 desktop cards). If BF4 came out in Q1 2013, it might have been different. Most people who like buying cutting edge hardware have already done so last year. Some will upgrade to the Titan (<1%). That means more likely than not the consumers who have held out for 2013 upgrades are those running older generation cards that have been waiting for price drops on HD7000/GTX600 series because they couldn't afford them last year. Raising prices now for a small bump in performance defeats the entire purpose of capturing those consumers. I am not saying I am right, but 2013 looks like one of the least eventful years for GPU upgrades because hugely anticipated games like Starcraft 2 HoTS, Company of Heroes 2 do not require $550 GPUs. After people were very disappointed with Crysis 2's campaign, many will be waiting to see if the game is actually good before upgrading to $500+ GPUs for it.

Games like DMC, Aliens Colonial Marines and Dead Space 3 were all console ports. Those games aren't going to get people with GTX200-500/HD5000-6000 cards to upgrade. There is nothing there besides Crysis 3 and Metro LL that's enticing for the majority of gamers outside of the hardcore enthusiasts on forums like ours. And with PS4's possible announcement in 2 days, and Xbox 720 expected to launch by end of the year, many PC gamers will consider whether to upgrade their console/media center or get a new GPU too. Don't forget that many PC gamers also have consoles and there isn't enough $ for many to do both a GPU and a next gen console upgrade in the same year.