wccftech:AMD Fiji XT Leaked For The Third Time On Zauba – 20nm Looks Promising

[james1701]

I think AMD really needs at least a soft launch soon to maintain their customer base.

Sorry, no coffee yet.

 

[Cloudfire777]

I thought the same thing. SoC doesn't indicate a GPU that will go on a PCIe card to me.

As far as I know, 20nm HP process was ditched from GlobalFoundries. 20nm process from TSMC is just for SOC`s and is not suitable for graphic cards.

Both AMD and Nvidia are most likely working on engineering graphic cards for 28nm (Nvidia 28nm from TSMC. AMD 28nm from GloFo) and is on the drawing board with graphic cards for 16nm FinFET from TSMC which will roll out in 2016.

The upcoming GPUs from AMD will most likely be on 28nm from GloFo with a much more efficient architecture than current GCN 1.x. Like Maxwell.
If the preview from Chiphell is any true, its perfect indicator of just that.

 

[cmdrdredd]

As far as I know, 20nm HP process was ditched from GlobalFoundries. 20nm process from TSMC is just for SOC`s and is not suitable for graphic cards.

Both AMD and Nvidia are most likely working on engineering graphic cards for 28nm (Nvidia 28nm from TSMC. AMD 28nm from GloFo) and is on the drawing board with graphic cards for 16nm FinFET from TSMC which will roll out in 2016.

The upcoming GPUs from AMD will most likely be on 28nm from GloFo with a much more efficient architecture than current GCN 1.x. Like Maxwell.
If the preview from Chiphell is any true, its perfect indicator of just that.

Yep that's what I figured. It's nice too that the new Maxwell cards can shut fans off even with certain gaming conditions that don't tax the cards enough for them to heat up. I'm sure AMD's next GPU will be similar in thermals as well as power consumption numbers.

 

[railven]

I think AMD really needs at least a soft launch soon to maintain their customer base.

Sorry, no coffee yet.

AMD needs to hit it out of the park if they want me back as a customer. I still feel a little dirty using this GTX 780

That stock cooler on Hawaii XT really made me not want to even bother with them until their next product cycle.

 

[Bateluer]

I don`t understand WCCFTech`s logic.
Tegra Maxwell is made with TSMC 20nm SOC process. Therefor R9 390X is confirmed with 20nm?

What?

Nvidia's Maxwell was designed 'mobile first'. Tegra X1 is 20nm, so their logic follows that a chip family designed for mobile would be amenable to 20nm. There's more links in their article, but yes, 20nm 300 series is just as likely if not more likely than a 28nm GloFlo part.

 

[Noctifer616]

Nvidia's Maxwell was designed 'mobile first'. Tegra X1 is 20nm, so their logic follows that a chip family designed for mobile would be amenable to 20nm. There's more links in their article, but yes, 20nm 300 series is just as likely if not more likely than a 28nm GloFlo part.

No. If there is no 20nm high performance process you won't see dGPUs at 20nm. You can't make a 200+ watt GPU on a process that is designed for low power parts.

That's why Apple and other mobile vendors get their parts first, not because they just buy all the wafers, they also only have the low power process only ready at that point.

If Tegra was a 100 Watt SoC it wouldn't be on 20nm right now because there is no process (as far as we know) that would support that kind of chip.

 

[Skurge]

Yep that's what I figured. It's nice too that the new Maxwell cards can shut fans off even with certain gaming conditions that don't tax the cards enough for them to heat up. I'm sure AMD's next GPU will be similar in thermals as well as power consumption numbers.

My vapor x does that with 2 of the 3 fans. You never believe how big of a deal it is until you have it. Would be great if all next gen cards had something similar, especially at idle.

 

[RussianSensation]

I think you a little confused. How outdated do you believe the Hawaii skus are? A 290X still offers performance near the GTX 970/980 while costing hundreds of dollars less. Seems like it was a solid design to me, lackluster reference cooler not withstanding. You really, really need to stop repeating this. Its flat wrong, no matter how many times you repeat it. The 290s have hit 200 repeatedly now, and for that price, its a steal. 970-ish performance for ~150 dollars less. 290X is usually ~250-300 dollars less than a 980 while offering 90% of its performance.

I have been recommending 290/290X for price/performance for many months, and well before 970/980 launched. That doesn't mean that AMD is not in a desperate need of a newer, more efficient architecture with some cool new features.

My view that 290/290X are outdated stems from these facts:

1. In the eyes of the consumer, a card that performs similarly but uses much more power is not preferred. GM204 has superior perf/watt here.

2. In the eyes of the consumer, a newer architecture brings more advanced features which are added bonuses. GM204 has HDMI 2.0 and MFAA, and despite questionable significant for 970/980 level of performance the DX12 checkbox. NV also had DSR earlier than AMD with Hawaii.

3. Major price drops on 290/290X and still most gamers picking up a new card now going for 970/980 show that in the eyes of the majority, while 290/290X can hang with 970/980, as an overall package the Hawaii cards are outdated.

4. The ability of many 970/980 cards to run at idle with no fans is a huge bonus. This enhances their green image, along with much lower quoted TDP. While more experienced gamers know that TDP does not always equate to real world power usage, the less informed gamer does equate TDP to actual power usage.

5. AMD still hasn't fixed their idle power usage when using multiple monitors. This is a pretty big deal for high-end enthusiasts running multiple monitors whether for productivity or gaming.

6. AMD's inability to have working CF in some recent titles like FC4 is a big determinant to their value proposition. Whether or not it's AMD's fault is almost irrelevant since the end user cannot use CF. They don't really care whose fault it is per say because at the end of the day the actual user experience matters. Small things like GeForce experience and more consistent SLI profiles are heavily favouring NV's SLI over AMD's CF, despite CF often showing superior frame time latency, when it works. The negative PR associated with poor frame times of Hawaii is definitely sitting in the back of many NV users' minds, making them a lot more reluctant to consider jumping ship to AMD for a multiple card setup.

And I'm sure the 20nm 390X will be a nice bump from the 290X, especially with the hybrid Hydra cooler it'll likely launch with.

Whether or not it will be 20nm and use hybrid cooler, I don't know. But I agree with you that 390X should be a nice bump from 290X or AMD wouldn't have waited so long to launch 390X. I don't recall any AMD generation where they bumped performance less than 30%.

The reference cooler was the only real negative with the Hawaii GPU after all.

I would say the high power usage, even higher than Tahiti XT was a big negative since performance/watt is all the rage nowadays. Perf/watt and higher power usage also cost AMD design wins in the mobile dGPU space. Don't forget that there are more mobile dGPUs sold now than desktop parts so winning there is more important than winning on the desktop when it comes to sales and profitability. AMD hasn't been competitive in the mobile dGPU sales as far as design wins all the way back to Pitcairn. Essentially everything starting with 7970M and below was dead in the water, with low worldwide availability and in many cases pairing 7970M with anemic AMD APUs/CPUs in laptops -- a horrible strategy.

7970 to 7970Ghz wasn't much of a refresh either, slight OC on the 7970. You can do that on any 290X with a non-reference cooler. My Asus DCU2 card comes with a comparable OC from the factory.

Right, but the difference is AMD did it and it was factory warrantied overclock on 7970Ghz. Asus Matrix 7970 had 1100mhz clocks, a whopping 19% higher than the original 7970. AMD didn't even attempt a 10% faster 290X despite 1 year passing. 7970Ghz was a quick refresh that came in June 2012 and provided after-market 680-beating performance. If AMD had not released 7970 1Ghz editions, things would have been a lot tougher from a PR perspective.

With 7970Ghz cards clocked at 1.05-1.1Ghz, NV could not claim the flagship card lead until Titan dropped. OTOH, 290/290X launched way later than 780 and NV trumped 290X with 780Ti. 290/290X's momentum was subdued due to high prices as a result of mining and by fall 970/980 launched. Essentially for most of 290/290X's life, it was overshadows by NV's products.

It actually would have been very beneficial for AMD to release a 10-15% faster Hawaii around 970/980 launch if they knew they couldn't get 300 series out by early 2015. A lot of us thought 300 series might come out Q1 2015 but recent rumours indicate a later launch.

The 'green' movement in IT pisses me off sometimes. 3 generations of a product line and performance has remained static.

I fully respect performance/watt from a point of view that with better perf/watt, you can get a lot more performance as you keep scaling TDP. Also, you can fit much faster products in thinner form factors (GM204 in the mobile is an excellent product). However, when people who bought GTX200 and Fermi generations now suddenly start hyping up performance/watt as the greatest thing and make it sound like 70-100W of power differences on the desktop are like having a volcano inside the chassis, I just shake my head. Considering most of these same gamers never abandoning NV when HD4000-6000 series was more efficient, there is an obvious double standard. But what can you do....

Reading your comments in various threads, it seems AMD's primary problem isn't technology, but rather marketing. You've cut the performance of the 290/29)X drastically in your head, putting the 970/980 on some kind of pedestal.

It's because the average consumer isn't going to look at reviews of after-market 290 cards. They are going to see hot, loud and slow performing reference 290/290X cards in reviews and obviously 970/980 look amazing next to those turds. I fully realize that in real world benches, an after-market 290 and an after-market 970 are very close, which is even more impressive given the much better price/performance of the 290. However, the average consumer clearly doesn't see it that way which is why we have seen NV continue to gain market share with slow and underperforming 750Ti, but perf/watt sold it. I expect when AMD/NV release Q4 2015 results, NV will have gained a large amount of market share due to GM204. Also, keep in mind NV didn't even release the GM200 flagship cards which means NV hasn't revealed its trump cards yet. This puts additional pressure on 390/390X to deliver well beyond 980's level of performance.

In reality, the Hawaii parts cost substantially less than the Maxwell skus while offering performance thats only slightly behind.

It's mostly true for the US market, but not as much for the worldwide market. In Canada, 290/290X are a lot more expensive and I presume in many parts of the world, there isn't a large premium for 970. Certainly I don't see many places having crazy sales like $200-225 after-market 290 that the US enjoys.

Because AMD's cards are not viewed as favourably as NV's due to higher perceived noise levels and higher real world power usage, with less advanced features (no TXAA, MFAA, HDMI 2.0, full DX12 'box checkmark'), AMD is forced to sell a 438mm2 chip on a 512-bit PCB board for much lower prices than they would have liked. This is probably wiping out a lot of AMD's profit margin and I can't see anything good coming out of AMD's Q4 2014 earnings report this month. I actually expect the GPU division to perform very poorly, with the only bright spot being console APU sales.

I loathe to suggest a company focus more on their PR and marketing, but in AMD's case, I think its very important given the level of falsehoods and blatant misinformation that people like yourself keep repeating.

I don't think I am providing some misleading information when I state that for most gamers the 970/980 are far more appealing than 290/290X cards. Based on various forums, a lot of gamers are upgrading to these cards and seeing various 970/980 SKUs out of stock on Newegg/Amazon is probably a good indication that they are selling better than 290/290Xs. AMD does need to better market its products and that starts with having a good reference cooler and sending after-market factory pre-overclocked cards to reviewers in the same way NV does.

to this day gcn is the most efficient compute architecture in the world despite the chip beeing only 438mm² small. this is not too shabby.

http://fireuser.com/blog/announceme...er_w_amd_firepro_s9150_gpu_ranking_1_in_gree/

Even if that were true, compute comprises a niche segment of the PC gaming community and an even smaller one for laptop users.

I think you went too harsh on 290X here, probably because of the lack of personal experience with the card. Fully dynamic clock engine alone is a huge improvement, i still enjoy how stutter free and power efficient the card performs at variable clocks. And raw rendering power is still enough to run most games in 2.5k as long as you tune some overly costly things (generally any option that spikes GPU usage by 30-50% vs previous level of quality, good example - ultra lighting in MGS GZ at night) down.

290X is a good improvement over 7970/7970Ghz but less impressive when considering what NV accomplished in terms of perf/watt with mid-range GM204 vs. GK104. AMD cannot simply make a larger chip with the same underlying architecture as 290X or the power usage would be unacceptable. Hence why I make comments that AMD needs a major redesign of the GCN architecture. Thankfully they have already addressed some key areas of weakness with 285 as a testbed. Just imagine how would 290/290X look if NV had launched GM200 with 250-275W of power usage? Essentially we have AMD's current gen high-end competing with NV's next gen mid-range and losing to a 980, sometimes by 20-25% too. AMD cannot just aim to match 970/980 but they need to outright beat them in price/performance and absolute performance because sooner or later NV will unleash GM200. NV can also drop prices on GM204 and re-spin it by Summer 2015. It's not like 980 will remain at $550 for another year.

 

[monstercameron]

snip

Interesting points and theories! Where do you think all the mindshare went? For a few months during the mining craze, amd absolutely blitzed the news. What happened? it couldn't just have been lack of availability.

 

[RussianSensation]

Interesting points and theories! Where do you think all the mindshare went? For a few months during the mining craze, amd absolutely blitzed the news. What happened? it couldn't just have been lack of availability.

AMD's reference cooler went from bad to worse due to increased GPU power usage when moving from 5870-R9 290X, while the cooler itself and the fan largely remained very similar. HD7900 series created poor image of early high prices, hot and loud reference cards, poor initial launch drivers, and then frame times issues for CF, exacerbated by Kepler beating GCN in BF4 for 10+ months. That set off a wave of users jumping to NV. Additionally, inferior perf/watt result and AMD being late for mobile dGPU bids meant that AMD basically forfeited mobile dGPU from the time Tahiti/Pitcairn launched. Riding off this negative image and lack of design wins into late hot and power hungry reference Hawaii and late and underperforming R9 285 didn't help. Then high prices and supply issues due to mining only made it worse for R9 200 series.

Once a gamer switches brands, it's hard for them to switch back if their experience is satisfactory. That's why it's a lot more costly for companies to attract new customers than retain existing ones. In some industries, it's actually 6-7 times more costly to acquire a new customer than to retain an existing one:

http://www.androidauthority.com/qualcomm-4k-display-smartphones-568785/

AMD's negative perceived image with its CPU division, poor drivers (whether justified or not), and hot, loud and power hungry cards (based on launch and reference card reviews) have resulted in a snowball of PC gamers lacking confidence in their products and shifting to NV. Since then NV has been charging a premium for perf/watt and gaining market share simultaneously. AMD launching late also hurts them A LOT more because the types of consumers who tend to buy AMD GPUs aren't particularly brand loyal as they are price/performance conscious. As a result, if NV releases a better price/performing part, most AMD owners won't wait 6-12 months for AMD's response, counter to most NV users.

 

[railven]

AMD's reference cooler went from bad to worse due to increased GPU power usage when moving from 5870-R9 290X, while the cooler itself and the fan largely remained very similar. HD7900 series created poor image of early high prices, hot and loud reference cards, poor initial launch drivers, and then frame times issues for CF, exacerbated by Kepler beating GCN in BF4 for 10+ months. That set off a wave of users jumping to NV. Additionally, inferior perf/watt result and AMD being late for mobile dGPU bids meant that AMD basically forfeited mobile dGPU from the time Tahiti/Pitcairn launched. Riding off this negative image and lack of design wins into late hot and power hungry reference Hawaii and late and underperforming R9 285 didn't help. Then high prices and supply issues due to mining only made it worse for R9 200 series.

Once a gamer switches brands, it's hard for them to switch back if their experience is satisfactory. That's why it's a lot more costly for companies to attract new customers than retain existing ones. In some industries, it's actually 6-7 times more costly to acquire a new customer than to retain an existing one:

http://www.androidauthority.com/qualcomm-4k-display-smartphones-568785/

AMD's negative perceived image with its CPU division, poor drivers (whether justified or not), and hot, loud and power hungry cards (based on launch and reference card reviews) have resulted in a snowball of PC gamers lacking confidence in their products and shifting to NV. Since then NV has been charging a premium for perf/watt and gaining market share simultaneously. AMD launching late also hurts them A LOT more because the types of consumers who tend to buy AMD GPUs aren't particularly brand loyal as they are price/performance conscious. As a result, if NV releases a better price/performing part, most AMD owners won't wait 6-12 months for AMD's response, counter to most NV users.

This post sums up my switch to a GTX 780 for my primary rig last year. I got the HD 7970 when the price was still >$500. I defended it, then of course the "mid" range GeForce 680 basically destroyed it. Solution, juice it, but them temps.

Fast forward to abysmal CFX performance, and I wear red-tinted glasses (least I did.) Sold that off quickly thinking the R9-series would be awesome. Then that heat/power, yet again, and worse, that cooler, and then the constant "wait for custom coolers" or "just water cool it." Yeah, because that's how you satisfy your user base, sell them subpar and tell them to deal with it or tell them to wait for something better. Ended up getting a GTX 780 Lightning for $450 brand new. Re-TIM'd the card and enjoying <55C at 1120/3500 clocks. AMD needs to do a lot to bring me back.

 

[RussianSensation]

Then that heat/power, yet again, and worse, that cooler, and then the constant "wait for custom coolers" or "just water cool it."

Lack of custom coolers at launch, poor availability and mining price gouging killed all the positive price/performance PR the R9 290 had against the 780. The irony though is today an after-market 290 beats the 780 in both performance and VRAM and you can easily buy affordable after-market cool and quiet cards. However, if you have had a great experience with NV's 780, you'll feel less compelled to go back to AMD since you feel they let you down and deep down you feel after you have supported them for so many years, they under-delivered. What makes less sense to me is brand devotion/attachment because even NV had massive turds with GF5 and 7 but I still bought Fermi since it had more VRAM, huge overclocking headroom and excellent game bundles with EVGA. I actually assess AMD/NV cards irrespective who won the previous generation -- a "memory blackout" strategy if you will. I guess for me the end user experience is largely similar making it extremely difficult for either of these brands to get me to become brand loyal to their brand. For example, if I am choosing between GM200 and 390X or some other GPUs in 2016, I am less likely to care that 680/970 beat 7970/290 in perf/watt, but what I will remember is NV basically threw Kepler under the bus by stopping to optimize its drivers too soon. I'll also be very cognizant of the VRAM amounts.

 

[Sunaiac]

AMD needs to do a lot to bring me back.

It's probably hard to "bring back" someone who feels he's done a better deal with a 450€ driver-abandoned 3G card that just had to be re TIM'd and is 10% slower than a 350€ 4G card as silent and cool with same quality coolers without need for anything to do and that's still top tier through actual driver update.

But well, the "where in the charts fell my card" 680s destroyed (as in was 6% faster than) the "wow, 3yo card still in top lines of charts, just wow" 7970s at launch, and everyone puts 450€ each year in new cards, so...

 

[railven]

It's probably hard to "bring back" someone who feels he's done a better deal with a 450€ driver-abandoned 3G card that just had to be re TIM'd and is 10% slower than a 350€ 4G card as silent and cool with same quality coolers without need for anything to do and that's still top tier through actual driver update.

Dat AMD defense force! For the record, when I bought my Lightning the cheapest R9 290 was $450 with stock cooler. Hey, remember that Bitming craze that sucked up supply thus made what was available cost more!

But, yerp, I lost out buying my card

 

[RussianSensation]

But well, the "where in the charts fell my card" 680s destroyed (as in was 6% faster than) the "wow, 3yo card still in top lines of charts, just wow" 7970s at launch, and everyone puts 450&#8364; each year in new cards, so...

The irony is that AMD never sold a reference 7970 Ghz edition and retail, and as early as June 22, 2012, AMD had recaptured the performance crown, even against after-market 680s. After-market 7970Ghz also cost less than 680 2GB, $100 less than 680 4GB and came with better game bundles than the 680. Since BF4 was a big game at that time, and people continued to focus on reference card reviews only, NV never lost momentum with Kepler.

http://techreport.com/review/23150/amd-radeon-hd-7970-ghz-edition/11

While one can say that 290/290X were late, AMD never threw GCN under the bus while 680/770 2GB are now on near life support. So it is a little bit of odd to point out how AMD was late with R9 200 series but ignore the very poor performance of GM204 and even GK110 in recent titles against Tahiti and Hawaii. The poor frame times with FCAT were investigated by many sites at the time after NV providing the tools (after they lost the performance crown), but ironically, the same sites that made this a big deal never noticed the poor frame times of Fermi. Go figure. Perhaps the most shocking/mind-blowing part to me is that NV continued to gain market share in the last 5 years despite Bitcoin mining making AMD cards free. Essentially the majority of the market decided to pay hundreds or thousands of dollars for GPUs in the last 5 years. That tells me AMD will probably continue to have a niche 20-30% dGPU market share, with mos to that userbase comprised of price/performance buyers and enthusiasts who unlock/overclock to get the best value. Unless AMD can get the performance crown for 3-4 generations in a row, I don't see them going back to 50/50 market share with NV for a LONG time.

Knowing all of this, with R9 300 series AMD needs to address the reference cooler, have custom versions of cards on day 1, send after-market pre-factory overclocked cards to reviewers, make sure they have enough stock for discrete and mobile GPU, make sure their drivers are working well almost right away, and undercut NV at all price segments while delivering more performance too.

I think GM200's 6GB of VRAM could potentially hurt 390/390X's 4GB of HBM though. I am not sure how AMD will address this through marketing.

 

[Black Octagon]

Yawn...wish we had something more solid about the specs and release date.

Things have been so stale in recent years. I often think about upgrading my 2500K and 7970, but performance has simply not improved enough.

I am really hoping that both Skylake and FijiXT/Pascal deliver substantial improvements for gamers...and that they hit shelves at around the same time

 

[RussianSensation]

Pascal should be a 16nm FinFET design though slated for 2016. I don't see how it will be out in time to compete with R9 390. Skylake-K seems to be MIA. I haven't read anything concrete on that chip coming out before Q3 2015.

 

[Black Octagon]

Q3 is a time I could wait for. I expect Windows 10 to launch around then, and for the monitor wars (FreeSync vs. G-Sync, etc.) to have led to a verdict. If Fiji XT is out a few months before then, it means I can avoid first adopters prices and get a nice after-market one. My one regret with my 7970 was getting a vanilla one during the first weeks, when I could have gotten a custom Ghz. edition later on

 

[RussianSensation]

Q3 is a time I could wait for. I expect Windows 10 to launch around then, and for the monitor wars (FreeSync vs. G-Sync, etc.) to have led to a verdict. If Fiji XT is out a few months before then, it means I can avoid first adopters prices and get a nice after-market one. My one regret with my 7970 was getting a vanilla one during the first weeks, when I could have gotten a custom Ghz. edition later on

One note -- the rumours point to Fiji XT being an R9 380 class card, with Bermuda XT being the flagship R9 290X successor.

 

[Black Octagon]

Oh you're right, thanks. Gotta start paying closer attention. From a cursory Google search it seems that no-one is speculating yet about the Bermuda XT release date, right?

 

[Silverforce11]

I think GM200's 6GB of VRAM could potentially hurt 390/390X's 4GB of HBM though. I am not sure how AMD will address this through marketing.

You think?

You can bet on it if NV has more vram, their reviewer's guide will consist of 4K + 8x MSAA. "Post AA is trash, real gamers use MSAA!!"

 

[RussianSensation]

Oh you're right, thanks. Gotta start paying closer attention. From a cursory Google search it seems that no-one is speculating yet about the Bermuda XT release date, right?

Honestly, it's pure guessing game. Some rumours said Q1 2015, some said Spring 2015, some said Summer 2015. I've never seen AMD having such secretary around its GPU launch. I might even be wrong about Fiji XT vs. Bermuda XT, which just goes to show how vague the rumours are.

The only thing I can find that seems to be recurring regarding 390X is this leak from SiSoftware Sandra database screenshots:

4096 SPs (GCN 1.2 or GCN 2.0?)
256 TMUs
64 Compute Units
4096-bit memory bus (1024-bit x 4 stack HBM) with memory running at 1.25Ghz, resulting with 640 GB/sec memory bandwidth
4GB of VRAM due to 1st gen HBM
http://www.youtube.com/watch?v=71xRFtupZ5k

There is some variation of the above specs with 390X as Bermuda XTX, with 4224 SPs, 264 Texture units, 96 ROPs, but these are earlier specs and no screenshot / database ever captured them. So I wouldn't trust these as much.

If either of these specs is true, 980 is basically toast, which is to be expected considering it's a mid-range next gen card through and through.

You think?

You can bet on it if NV has more vram, their reviewer's guide will consist of 4K + 8x MSAA. "Post AA is trash, real gamers use MSAA!!"

^_^ Ya, and all those NV loyalists who bought VRAM gimped 570 1.28GB, 580 1.5GB, 670/680/770 2GB and 780 3GB would start hyping up extra VRAM as more future-proof. You don't have to be Nostradamus to predict that.

What's going to be more interesting is AMD might have stronger 4K performance but less VRAM, resulting in no clear winner with either GM200/390X as either solution would entail a compromise. Then it would come down to which card overlcocks better while using the least amount of power in overvolted/OC states.

I also expect a lot of negative comments regarding a Hybrid WC system, implying that AMD "needed" WC to keep its cards within operating parameters, instead of embracing Hybrid WC as the superior cooling solution as proven by 295X2 vs. Titan Z (like carbon fiber chassis + aluminum/magnesium exotic materials used in high end sports cars). Don't forget if GM200 uses 50-60W less power at load, a whole mountain will be built out of that too.

NV has its GW as a secret weapon too, which a lot more financial resources to throw at developers. I don't see NV losing the performance crown unless AMD pulls a 9700 Pro moment.

 

[Ken145]

290X is a good improvement over 7970/7970Ghz but less impressive when considering what NV accomplished in terms of perf/watt with mid-range GM204 vs. GK104. AMD cannot simply make a larger chip with the same underlying architecture as 290X or the power usage would be unacceptable. Hence why I make comments that AMD needs a major redesign of the GCN architecture. Thankfully they have already addressed some key areas of weakness with 285 as a testbed. Just imagine how would 290/290X look if NV had launched GM200 with 250-275W of power usage? Essentially we have AMD's current gen high-end competing with NV's next gen mid-range and losing to a 980, sometimes by 20-25% too. AMD cannot just aim to match 970/980 but they need to outright beat them in price/performance and absolute performance because sooner or later NV will unleash GM200. NV can also drop prices on GM204 and re-spin it by Summer 2015. It's not like 980 will remain at $550 for another year.

I think they don't need a major GCN redesign, but rather do some gaming specific bumps (ala Tonga's improved Tesselator, Fast Conservative Rasterization etc.) and grow further in compute superiority. They simply can not compete with gk104/gm204-style designs directly without making a fully dedicated big gaming chip that is not designed for DP. Correct me if i'm wrong, but even GK110 has fully DP-dedicated CUs which can be disabled for a efficient gaming GPU, while AMD's design has fat DP ALUs in each CU. This worked well for AMD untill they stucked on 28nm and HBM recieved a massive delay from original 2012 plans. Hovewer designing a stipped-down gaming chip is simply a false root for AMD. Efficiency can still be brought to more than reasonable levels with things like HBM, color compression, aggressive adaptive voltage control alongside with already existing Dynamic clock engine etc. What they should focus on is HSA/Compute features, push game developers to use compute based effects that work well across PS4 and GCN cards, the same way Nvidia pushes tesselation based effects with Gameworks.

 

[Dribble]

They do need a major re-design, there performance/watt is just killing them. It forced them out the mobile market years ago (as mobile kepler was still significantly better then anything AMD has), and now it's forcing them out of the desktop market - the one least effected by power requirements.

They can't get ahead by using a better process node as nvidia will be able to use exactly the same one. The real question is can their hardware team manage it after all the cutbacks? The answer is unlikely - nvidia has more R+D, better focus, and is probably years ahead of AMD in design and is still pushing hard.

 

[BlockheadBrown]

I just wish we'd get something concrete sooner rather than later. Last I read it was "early 2015". I guess March would still be "early" by their definitions. I wouldn't be surprised if this was a 1H2015 release vs. 1Q2015. Frankly, a 3Q or 2H wouldn't be surprising at this point either. Call it apathy. I had a r9 290 and resold it. I'm hesitant on the nv9x0 series.