Not too much of a surprise, the GTX 9X0 cards have been selling like hot cakes, the balance of performance, low power consumption, cool running and near silent operation make them a winner.

FluffmeisterNot too much of a surprise, the GTX 9X0 cards have been selling like hot cakes, the balance of performance, low power consumption, cool running and near silent operation make them a winner.

Just wait until TSMC actually starts producing GPUs on its new 20nm process. I think it's only a winner in the short term.

AMD is going further and further down.... Feel so sorry for those guys. Honestly.

It's such a sad, but interesting affair.

We know people want strong GPUs; be it integrated in a mobile device with increasingly higher rez screen, apu for a sff/laptop, or discrete desktop part for gaming, especially as even low-power arm cpus are reaching the general purpose tipping point that x86 hit around the time of Conroe or so. It's unfortunate when you see companies like Qualcomm (ironically with adreno) and nvidia with K1 release products that are well-regarded, especially given amd's mission statement since 2006. Even Intel had 4k decoding before AMD...it's just sad, given they used to be the pioneers in so many areas over the years that attracted customers for one reason or another. Such a long way from avivo and the AIWs (not to mention core gpu features).

While ofc AMD is still in a (forced) holding/restructuring pattern until 14nm/2016ish, and I'm sure I'm not alone in hoping they can shock the world with products at that point...the last couple years of generally stale products have devastated them; it really is starting to bite them when their competition (intel in apus, nvidia in gpus) have or will have lower-end parts competing with the strengths of their flagships...conceivably even before some launch. Say what you will about their finances, but that never should have been allowed to happen when there are so many ripe markets to plunder with their specific strengths and IP. Dirk has been gone for FOUR YEARS, and it is beyond the blame of anyone of that time period. It's one thing to readjust focus to where they can be profitable, it's another to allow your competition to seriously start to nullify your purpose of existence in the meantime. Everyone has got to have an angle; at least then a niche will bite (more cpu or gpu performance discrete or apu, best price/perf per market, perf/watt, arm, x86, unique or first to a feature-set). AMD has the potential for any of them and is close to having none, while the areas that have held them back from further innovation (bandwidth and process tech) are being solved by the companies that used to so far behind them in such matters it didn't even use to be a concern*. When you're a company that has largely played second fiddle in your market(s), the last thing you want people to do is forget your name or what you do uniquely/best...because it takes a lot to make them come back.

I'm not saying this because I bought one (which coming from a big supporter of amd I do think is something), but when you see a product like the 970 versus the 290x, that's really quite shameful...and I've already expressed my reservations about Fiji (ram size for efficient bw with HBM, power/size/cooling requirements, inability to likely compete with anything above GM204 while probably being less efficient). When you see intel, the laughing stock of hardware graphics, surpass your video capabilities (and soon integrated graphics capabilities as the edram Iris pros float into more markets with broadwell and skylake), it's inexcusable.

*Intel/nvidia largely solved bandwidth issues with on/off die cache and unit optimization, while amd has been flops-heavy for years (and should have been in the forefront about solving the issue). AMD used to be first to a new process (compared to nvidia) by a large margin, but are having difficulty innovating without it (one could argue contrary to Maxwell).

TLDR:

I'm not saying anything anyone doesn't already know, and surely amd will live to see another day (with better products or a dead-shot aim on specific markets), but damn...this is the start of what's going to be a truly rough year...and I think it was avoidable on multiple counts. I'm not saying we may have been talking about a better cpu than intel, nor a more efficient gpu than maxwell...maybe it could have been a competitor to Tegra, or a faster time-to-market for off-die buffer than HBM, or something else completely that was better/unique. But instead, it's nothing...just hopes (and further anticipation of having) to catch up...with the sales numbers starting to show it...and that bums me right out.

alwaysstsEven Intel had 4k decoding before AMD

You say that like Intel doesn't have resources to do R&D. Intel is a vastly larger company than AMD with a lot more revenue to support that R&D, it's important to remember that. It sucks for AMD, but they did it to themselves IMHO. We'll see how TSMC's 20nm process changes things up, although AMD has been waiting way too long to release new GPUs, same deal with CPUs. Intel and nVidia keep churning out new products where AMD looks like they're sitting still in comparison.

If the rumoured GM200 is really 50% faster than Titan Black, that'd equate to a 50-60% (or more, hexus.net/tech/reviews/graphics/69561-gigabyte-geforce-gtx-titan-black/?page=4) performance increase over 290X.

AMD would need to make a massive perf leap over the 290X to compete. This is all of course, rumour. The stacked memory technology is a positive step but it's not yet known how much that will help with the new architecture in terms of overall performance improvement.

All that said, I think a GM200 >500mm2 won't sip power but it'll be a perf/watt increase, instead of an absolute reduction.

I would dearly like to see the AMD 'rabbit out of a hat trick' - we need it moving forward with 4k adoption. They need it for their finances.

the54thvoidIf the rumoured GM200 is really 50% faster than Titan Black, that'd equate to a 50-60% (or more, hexus.net/tech/reviews/graphics/69561-gigabyte-geforce-gtx-titan-black/?page=4) performance increase over 290X.

AMD would need to make a massive perf leap over the 290X to compete. This is all of course, rumour. The stacked memory technology is a positive step but it's not yet known how much that will help with the new architecture in terms of overall performance improvement.

All that said, I think a GM200 >500mm2 won't sip power but it'll be a perf/watt increase, instead of an absolute reduction.

I would dearly like to see the AMD 'rabbit out of a hat trick' - we need it moving forward with 4k adoption. They need it for their finances.

Fiji will have 4096 stream processors according to rumors. If they are clocked at 1Ghz they should be aboout 45% faster than an R9 290X. If they also tweaked them to be more efficient per clock then the R9 390X will easily be 50% faster than an R9 290X. If they also manage to bump the clock speed to 1100mhz then it should be about 60+% faster than an R9 290X so AMD is fine for 20nm.

Last quarter market share total for Intel/AMD/NVIDIA is 100% and this quarter it's 123.3% ... :pimp:

Yep. They gotta do the ol rabbit outta the hat trick and that rabbit better have some long fangs and a vicious attitude. I feel this next series is, like, determining their future. They impress us or forever fall behind nvidia.

ChampYep. They gotta do the ol rabbit outta the hat trick and that rabbit better have some long fangs and a vicious attitude. I feel this next series is, like, determining their future. They impress us or forever fall behind nvidia.

Allow me to preface this with the fact that 90% of my GPUs are AMD based cards.

Haven't AMD been behind Nvidea for the last several years? They've spent the last decade being one step behind Nvidea, but generally offering a much better value proposition. I'm loathe to pay the Nvidea tax, and the mild loss of performance in going with AMD is generally not an issue for the games I play. The 970/980 are eating into AMD's sales because they aren't really an Nvidea move. They're priced reasonably, relatively silent, and more importantly they've got a substantial performance lead over the AMD offerings. Even with lower pricing on the high-end AMD cards, I'd be very hard pressed to recommend a 290x over a 970.



AMD has been on the down-swing since introducing their new architecture. It was a bold experiment, that unfortunately did not pan out. AMD went whole-hog on the Bulldozer, and has wound-up paying the price in the GPU department. None of this is easy to say, but it's the truth. Hopefully the new AMD GPUs are released soon, and showcase substantial performance gains. As of right now, AMD can't really bring anything to the table except price cuts. Such a poor showing is depressing, given AMD has demonstrated so much more potential.

lilhasselhofferHopefully the new AMD GPUs are released soon, and showcase substantial performance gains.

I would like to see it on TSMC's new 20nm process.

lilhasselhofferAMD has been on the down-swing since introducing their new architecture. It was a bold experiment, that unfortunately did not pan out. AMD went whole-hog on the Bulldozer, and has wound-up paying the price in the GPU department. None of this is easy to say, but it's the truth. Hopefully the new AMD GPUs are released soon, and showcase substantial performance gains. As of right now, AMD can't really bring anything to the table except price cuts. Such a poor showing is depressing, given AMD has demonstrated so much more potential.

Agreed. I think the only thing that has been working out well for AMD recently are their APUs, not just for AM1 and FM2(+), but the embedded ones as well like what's in the Xbox and Playstation.

In all seriousness though, I see there being very little wrong with the GPU architecture in general. I think GCN is pretty solid. The problem is that they're not making incremental improvements regularly like nVidia and Intel have been. They ride out a card for a long time then overhaul it. I think the risk that AMD runs when they do this is that when you have a company like nVidia that releases new cards more often, it adds incentive to people to get the latest technology and right now if you look at AMD, the 200-series Radeons are, relatively speaking, old. The 900-series GTX cards are brand new, hot off the press.

With that said, I think that AMD's CPU shortcomings influenced the rate at which new GPUs have been released which is a shame because that hurts sales.

With that said, I still don't see myself buying a new GPU to replace my 6870s until we start seeing some 20nm GPUs.

AMD's problem lies not with their flagship 290 cards. It's everything below them that cant compete!!! The biggest problem of all being that is where the money is.

OEMS are still the major issue in this whole argument. Without them you cannot drive enough of your products into user hands because most users are not building from scratch not to mention the fact that many users only buy some type of mobile device (Laptop, Tablet, etc) which there are not enough OEMS on AMD's side to keep that going.

buildzoidFiji will have 4096 stream processors according to rumors. If they are clocked at 1Ghz they should be aboout 45% faster than an R9 290X. If they also tweaked them to be more efficient per clock then the R9 390X will easily be 50% faster than an R9 290X. If they also manage to bump the clock speed to 1100mhz then it should be about 60+% faster than an R9 290X so AMD is fine for 20nm.

You forgot the most important parameter. If the rumors for 4 HBM chips that will give a 4096bit data bus and 640GB/sec bandwidth are true, then both gpus and APUs might see a good performance improvement before 2016 when Zen will comes.

badtaylorxAMD's problem lies not with their flagship 290 cards. It's everything below them that cant compete!!! The biggest problem of all being that is where the money is.

The opposite. Nvidia can't touch them in the low and mid range. For example 750 and 750ti have a huge advantage in efficiency but they are behind in performance compared to 260 and 265. In general 240, 250/X, 260/X, 265, 270/X, 280 and possibly even 285 and 280X are mostly better choices compared with Nvidia's offerings at the same price if we just look them with the performance/dollar parameter in mind. Now if we start thinking about brand recognition and other stuff like PhysX, CUDA, Gameworks and stuff, or power efficiency, then Nvidia cards do gain extra value, but of course not extra frames per second.
AMD loses in the high end and that affects it's brand strength against Nvidia's in the whole market. People will buy Nvidia hardware or Intel hardware even if AMD's offering is better, because Nvidia and Intel are stronger brands. When you don't know what to buy, you just buy the one with the stronger brand on it.

john_The opposite. Nvidia can't touch them in the low and mid range. For example 750 and 750ti have a huge advantage in efficiency but they are behind in performance compared to 260 and 265. In general 240, 250/X, 260/X, 265, 270/X, 280 and possibly even 285 and 280X are mostly better choices compared with Nvidia's offerings at the same price if we just look them with the performance/dollar parameter in mind. Now if we start thinking about brand recognition and other stuff like PhysX, CUDA, Gameworks and stuff, or power efficiency, then Nvidia cards do gain extra value, but of course not extra frames per second.
AMD loses in the high end and that affects it's brand strength against Nvidia's in the whole market. People will buy Nvidia hardware or Intel hardware even if AMD's offering is better, because Nvidia and Intel are stronger brands. When you don't know what to buy, you just buy the one with the stronger brand on it.

I think that APUs may have influenced this because AMD had to keep power consumption down to cram enough CPU and GPU on the same PCB. That design probably leaked into their low-end discrete cards. This is just more evidence (imho) that AMD has been investing more resources into APUs versus anything else.

Just how popular are high end cards anyway, if GTX 9xx is behind this? AFAIK most action (fiscally speaking) happens in the low-mid range, and there AMD are excellent.

AquinusI think that APUs may have influenced this because AMD had to keep power consumption down to cram enough CPU and GPU on the same PCB. That design probably leaked into their low-end discrete cards. This is just more evidence (imho) that AMD has been investing more resources into APUs versus anything else.

I agree they are going to work hard on APU's mostly because the mobile segment is so dominant in society right now and the needs of smaller areas in which a good CPU and GPU can sit without taking up much room is becoming the norm.

Also the GTX 9XX is not the reason for this shift in my opinion because its just a repeat of what happens over the years. Its the expansion of extra products like the Tegra line and a lot of new changes in the OEM world bringing this on.

FrickJust how popular are high end cards anyway, if GTX 9xx is behind this? AFAIK most action (fiscally speaking) happens in the low-mid range, and there AMD are excellent.

People read about Titan and 900 series, how great they are and how good they perform. Then because they don't want to spend that amount of money they go and buy a GT620 because it is from the same company that build that Titan of that 980 card they just read or heard about. It does write on the box that offers unprecedented features and speed anyway, so it must be like a little Titan or a little 980. Right?
That's how the hi end affects the low end sells.

AMD sells silicon chips. Not cards, not motherboards, not much else.

They have consistently failed to shrink the die size for performance increases in the last two cards, so engineering dollars went to make a better mud pie, silicon costs are the same.

They have failed to materialize a CPU that could compete in per clock performance, so have been forced to add more cores, and thus more silicon to get closer, higher cost.

Manufacturing hasn't improved at the rate they have needed and planned on, and this may be holding up production of plans for smaller chips, and without the R&D budget to allow for side by side development it halts their production improvement.

They need to go back and look at the R300 (9700 Pro/9800Pro/XT) days of ops. Heck r300 was even in peg form.

AquinusI would like to see it on TSMC's new 20nm process.

Agreed. I think the only thing that has been working out well for AMD recently are their APUs, not just for AM1 and FM2(+), but the embedded ones as well like what's in the Xbox and Playstation.

In all seriousness though, I see there being very little wrong with the GPU architecture in general. I think GCN is pretty solid. The problem is that they're not making incremental improvements regularly like nVidia and Intel have been. They ride out a card for a long time then overhaul it. I think the risk that AMD runs when they do this is that when you have a company like nVidia that releases new cards more often, it adds incentive to people to get the latest technology and right now if you look at AMD, the 200-series Radeons are, relatively speaking, old. The 900-series GTX cards are brand new, hot off the press.

With that said, I think that AMD's CPU shortcomings influenced the rate at which new GPUs have been released which is a shame because that hurts sales.

With that said, I still don't see myself buying a new GPU to replace my 6870s until we start seeing some 20nm GPUs.

badtaylorxAMD's problem lies not with their flagship 290 cards. It's everything below them that cant compete!!! The biggest problem of all being that is where the money is.

Are you talking about... today? AMD’s 270X, 280/285 & 280X are extremely competitive against the 760-770 price points! In all matters of perf/power etc. Nvidia is the one with gaping hole in the "most lucrative money” area 1080p. As nice as a 750ti is, it's not a card gamers can look at for mainstream 1080p performance. It is so strange the focus of everyone as narrowed in on the 970 price point (~20% at best), and seem to forget that when it comes to gaming what's like >60% market share, is the true bread and butter for both.

The strange part of JP's. numbers are all they're about Q2 (April-June) sales, and has nothing to do with 9X0 Maxwell products, but is perhaps a strong indicator of the 750Ti that had released in Feb. Remember in this quarter; April (first month) was when the "mining craze went bust", and AMD had perhaps not much volume in the channel. While AMD started to see pricing normalize they couldn't necessarily cut prices as they probably hadn't product to offer. I think a lot of this "doom-n-gloom" is just... Monday morning quarterback’n. What I think it indicates is AMD caught off guard by mining, never had the volume (wafer starts) to meet the demand, and AMD had some sales jump to Nvidia, as some folk seeing still high prices that hung around well too long after things went bust!

What happen in to past... is history, sure it has a bad feel but look at it in "context" and we see perhaps meaning. Oddly there's more "the sky is fall’n" going in here than was made of TSCM 16nm FF, just now passing risk production, which is perhaps put a bind in Nvidia's bottom line moving forward.

SteevoAMD sells silicon chips. Not cards, not motherboards, not much else.

They have consistently failed to shrink the die size for performance increases in the last two cards, so engineering dollars went to make a better mud pie, silicon costs are the same.

They have failed to materialize a CPU that could compete in per clock performance, so have been forced to add more cores, and thus more silicon to get closer, higher cost.

Manufacturing hasn't improved at the rate they have needed and planned on, and this may be holding up production of plans for smaller chips, and without the R&D budget to allow for side by side development it halts their production improvement.

As to GPU's that statement is true for AMD and Nvidia, while Nvidia went another round of chips (cards) without a shrink.

As to R&D, AMD has had less and most of the work we see now was spent three years ago (or longer) when AMD was really bad off probably spending money they didn't have and working from executive decision’s that where poorly put in place long before that.

john_People read about Titan and 900 series, how great they are and how good they perform. Then because they don't want to spend that amount of money they go and buy a GT620 because it is from the same company that build that Titan of that 980 card they just read or heard about. It does write on the box that offers unprecedented features and speed anyway, so it must be like a little Titan or a little 980. Right?
That's how the hi end affects the low end sells.

Exactly, brand recognition is something that will sell to people regardless of what they are getting. Its like buying a Ferrari because in the end of the day you will probably buy a Ferrari because its a Ferrari and not because of its top speed or what leather it uses. Your just going to buy it because its a Ferrari and you know its fast and all about being the best right?

SteevoAMD sells silicon chips. Not cards, not motherboards, not much else.

They have consistently failed to shrink the die size for performance increases in the last two cards, so engineering dollars went to make a better mud pie, silicon costs are the same.

They have failed to materialize a CPU that could compete in per clock performance, so have been forced to add more cores, and thus more silicon to get closer, higher cost.

Manufacturing hasn't improved at the rate they have needed and planned on, and this may be holding up production of plans for smaller chips, and without the R&D budget to allow for side by side development it halts their production improvement.

Well the CPU segment has basically pushed them off to the side for a lot of reasons including some scandalous which has pushed them out of many OEMS for years which accounts to losses of funds to develop these areas.

CasecutterAre you talking about... today? AMD’s 270X, 280/285 & 280X are extremely competitive against the 760-770 price points! In all matters of perf/power etc. Nvidia is the one with gaping hole in the most” lucrative money” area 1080p, as nice as a 750ti is it’s not a card gamers can look at for mainstream 1080p performance. It is so strange the focus of everyone as narrowed in on the 970 and seem to forget the what like >60% market share that’s the true bread and butter.

The strange part of JP's. numbers are all they're about Q2 (April-June) sales, and has nothing to do with 9X0 Maxwell products, but is perhaps a strong indicator of the 750Ti that had released in Feb. Remember in this quarter is April (first month) was when the "mining craze went bust", and AMD had perhaps not much volume in the channel. While AMD started to see pricing normalize they couldn't necessarily cut prices as they probably have product to offer. I think a lot of this "doom-n-gloom" is just... Monday morning quarterback’n. What I think it indicates is AMD caught off guard by mining, never had the volume (wafer starts) to meet the demand, and AMD had some sales jump to Nvidia, as some folk seeing still high prices that hung around well too long after things went bust!

What happen in to past... is history, sure it has a bad feel but look at it in "context" and we see perhaps meaning. Oddly there's more "the sky is fall’n" going in here than was made of TSCM 16nm FF, just now passing risk production. which is perhaps put a bind in Nvidia's bottom line moving forward.

I think no one predicted the mining craze would take off like it did and while it may have caused sales of AMD cards to skyrocket they were to people building mining machines instead of gaming rigs. They also could not control the whole up charge area that really saw them no extra money and actually hurt them more than anything in that area. History is full of things we never saw coming and honestly that could all change soon enough again as we may find another craze that causes something to happen with the market.

CasecutterAs to GPU's that statement is true for AMD and Nvidia, while Nvidia went another round of chips (cards) without a shrink.
As to R&D, AMD has had less and most of the work we see now was spent three years ago (or longer) when AMD was really bad off probably spending money they didn't have and working from executive decision’s that where poorly put in place long before that.

The worst decision they made with their chips in the situation they were forced into was the idea that multi-core was going to come out faster than it did. The problem was with the Intel compiler problem, the pushed out of the OEM segment by many manufacturers, and the fact they were already starting to get behind on single core performance decided the idea that they should put a focus on multiple cores over single cores. While developers of software were focusing on multithreading, they were only really focusing on 2-4 cores at most and were stuck not wanting to spend the time to go beyond that because most chips were just 2 or 4 cores and still are to this day. Developers were not going to waste time developing for it so while if you include using all the cores on an FX 8350 makes it much stronger than the i7 quad core (with hyper threading 8) you still suffer from the fact that each core in a core to core comparison the Intel is much better which caused all programs that only look to the first 2-4 cores to run way faster and better on which ever had the better first cores (Intel). The mistake was thinking that this would eventually catch on and jumping the gun way to early, they basically got obsessed that having more cores available would make software developers start working towards heavy threading which never came through in the end. Sure some games or software like Adobe can use pretty much every core you throw at it (or at least 8) but its so far and few between that its really does not help having extra cores that really are not benefitting you in any way.

I think we may see something change later on in this world but right now its going to be a very slow process of rebuilding and honestly the best course of action is to develop the APU and work towards their single core performance and getting into the OEM's machines over than trying to release hard core CPU's.

I feel bad for AMD. I wish they would just pull two rabbits out of their hat. A CPU that can compete with intel's i7's, and a GPU that can compete with nVidia's Maxwell cards.

AsourciousI feel bad for AMD. I wish they would just pull two vicious rabbits out of their hat. A CPU that can compete with intel's i7's, and a GPU that can compete with nVidia's Maxwell cards.

Well I think the CPU part is to farfetched at the moment because they have already decided how they are going to focus for the time being until they win back enough OEMS to gain some serious ground.

As for the GPU's, its all going to depend on the node it comes out on and the overall performance. The waiting game they played instead of just sticking to the basics like Nvidia did could pay off heavily if the 390X is power efficient enough and out performs the 980 using a 20nm die and the 3D stacked memory (HBM) then they will have stuck a hard foot in the ground on the GPU segment of once again. However if they stick to the 28nm and even with the stacked memory only perform on par or around the 980 even if the power consumption for all intensive purposes is about the same then they have made a huge error in waiting this long.