AMD CPU speculation... and expert conjecture

[juanrga]

AMD will use HBM on the 390 series to reduce power consumption and compensate for the lack of efficient architecture. Same reason why will use 20nm instead 28nm.

Why isn't the move from 28nm to 20nm considered an advance in architecture to a more efficient one in itself for AMD?
And why isn't HBM considered an advance in architecture also?

You seem to cherry-pick and negate AMD's achievements calling them "compensations" just to say NVIDIA has a better architecure and approach, but the truth is AMD is also advancing in architecure just in different areas than NVIDIA. Correct me if I'm wrong...

The move from 28nm to 20nm is an advance in the fabrication process. And replacing GDDR5 by HBM is an advance in the memory subsystem. In both cases the graphical (micro)architecture is the same GCN than now.

Nvidia did its homework and got an efficient architecture that allows them to compete from phones to supercomputers. AMD is trying a brute force approach to get the performance needed to match a Titan II and then using die shrink plus HBM to maintain power consumption under control and liquid cooling to maintain safe temperatures. This brute force approach will allow AMD to continue playing on a niche market (HEDT), but will move them apart from supercomputers, servers, tablets... As a consequence their dGPU division will continue loosing market share and money...

The recent launch of Nvidia’s Maxwell chips is no doubt limiting AMD’s high-end graphics ASPs. With all of that said Mr. Kumar considers the performance of AMD’s graphics business to be disappointing.

What do you think will happen when Nvidia makes the die shrink and adds HBM to its already efficient Maxwell architecture?

AMD needs a new GCN architecture to compete, but it will be not ready before 2016 and I have my doubts about what AMD will achieve because lacks the money

 

[blackkstar]

AMD will not remove compute from their mid range cards. AMD and Nvidia have vastly different approaches when it comes to their GPGPU market.

Nvidia wants to ensure that people spend the big money on the big GPGPU cards. They don't want people buying GTX 970 and then rendering crazy fast in mental ray. they want everyone using Mental Ray on expensive Quadros.

AMD wants everyone using GPGPU. It means more potential customers for HSA enabled software, which means developers are more likely to make it.

Nvidia waiving efficiency around is just a ruse. It's them trying to spin removing features from their GPUs as a positive thing. It's them saying "you can't use all these fancy CUDA features we pushed earlier to their full potential unless you buy the big card" and then saying "but it saves a few dollars a year in electricity!", when in fact it is far worse because you do GPGPU things on CPU instead which uses more power.

AMD wants HSA across cat cores to HPC. Neutering their product range doesn't align with their goals.

 

[juanrga]

AMD will not remove compute from their mid range cards. AMD and Nvidia have vastly different approaches when it comes to their GPGPU market.

Nvidia wants to ensure that people spend the big money on the big GPGPU cards. They don't want people buying GTX 970 and then rendering crazy fast in mental ray. they want everyone using Mental Ray on expensive Quadros.

AMD wants everyone using GPGPU. It means more potential customers for HSA enabled software, which means developers are more likely to make it.

Nvidia waiving efficiency around is just a ruse. It's them trying to spin removing features from their GPUs as a positive thing. It's them saying "you can't use all these fancy CUDA features we pushed earlier to their full potential unless you buy the big card" and then saying "but it saves a few dollars a year in electricity!", when in fact it is far worse because you do GPGPU things on CPU instead which uses more power.

AMD wants HSA across cat cores to HPC. Neutering their product range doesn't align with their goals.

AMD has crippled the DP performance of Hawai to 1/8 from the original 1/4 (Tahiti)

http://www.tomshardware.com/reviews/radeon-r9-290x-hawaii-review,3650.html

to force people to purchase the more expensive FirePro cards. AMD doesn't want people using cheap R9s beyond gaming and a pair of apps and forces people to "buy the big card".

Also the emphasis on efficiency for future architectures is not a Nvidia marketing strategy, but a well-known technical result from academic research. AMD doesn't have another option than follow Nvidia and develop an efficient architecture if want remain on business. The big "IF" is on if AMD has the resources needed to do it. I remain skeptic.

I don't know if "AMD wants HSA across cat cores to HPC" or not, but I know the official roadmap doesn't even mentions HSA for Carrizo-L

 

[jdwii]

AMD will not remove compute from their mid range cards. AMD and Nvidia have vastly different approaches when it comes to their GPGPU market.

Nvidia wants to ensure that people spend the big money on the big GPGPU cards. They don't want people buying GTX 970 and then rendering crazy fast in mental ray. they want everyone using Mental Ray on expensive Quadros.

AMD wants everyone using GPGPU. It means more potential customers for HSA enabled software, which means developers are more likely to make it.

Nvidia waiving efficiency around is just a ruse. It's them trying to spin removing features from their GPUs as a positive thing. It's them saying "you can't use all these fancy CUDA features we pushed earlier to their full potential unless you buy the big card" and then saying "but it saves a few dollars a year in electricity!", when in fact it is far worse because you do GPGPU things on CPU instead which uses more power.

AMD wants HSA across cat cores to HPC. Neutering their product range doesn't align with their goals.

No one would buy a r9 or 970 for that anyways. I still can't find any software I use for encoding that works well on cuda or opencl let alone HSA. I guess the majority still agree except in workstations or cad machines which generally use cuda and hate changing. Amd is simply using the add more transistors approach which is brute forcing and in chess that's the worst strategy something I find sad. But this does give them more time to make a update for their architecture. From the rumor it seems like the 390x will hold its own in the gaming market it will just use more power and heat something most gamers don't care about I'm the odd one out

Amd needs every transistor to be more efficient then ever for their APU series. Its not even worth discussing.

 

[cdrkf]

AMD will use HBM on the 390 series to reduce power consumption and compensate for the lack of efficient architecture. Same reason why will use 20nm instead 28nm.

Why isn't the move from 28nm to 20nm considered an advance in architecture to a more efficient one in itself for AMD?
And why isn't HBM considered an advance in architecture also?

You seem to cherry-pick and negate AMD's achievements calling them "compensations" just to say NVIDIA has a better architecure and approach, but the truth is AMD is also advancing in architecure just in different areas than NVIDIA. Correct me if I'm wrong...

The move from 28nm to 20nm is an advance in the fabrication process. And replacing GDDR5 by HBM is an advance in the memory subsystem. In both cases the graphical (micro)architecture is the same GCN than now.

Nvidia did its homework and got an efficient architecture that allows them to compete from phones to supercomputers. AMD is trying a brute force approach to get the performance needed to match a Titan II and then using die shrink plus HBM to maintain power consumption under control and liquid cooling to maintain safe temperatures. This brute force approach will allow AMD to continue playing on a niche market (HEDT), but will move them apart from supercomputers, servers, tablets... As a consequence their dGPU division will continue loosing market share and money...

The recent launch of Nvidia’s Maxwell chips is no doubt limiting AMD’s high-end graphics ASPs. With all of that said Mr. Kumar considers the performance of AMD’s graphics business to be disappointing.

What do you think will happen when Nvidia makes the die shrink and adds HBM to its already efficient Maxwell architecture?

AMD needs a new GCN architecture to compete, but it will be not ready before 2016 and I have my doubts about what AMD will achieve because lacks the money

Lol why are you so anti AMD all of a sudden?!

AMD co developed HBM, and have invested the time to get to the shrink first. *IF* we believe the leaks, then what they are launching is both more efficient *and* faster than what Nvidia is supposedly going to launch. The liquid cooling is simply the best solution AMD can offer for cooling atm (Nvidia has a top notch air cooler- efficiency / performance aren't really that related this is more down to noise and overclocking headroom).

I'll bet the 390X will still be a smaller die than Titan 2... Whatever you say about the 290 / 290X, the performance / die size is unrivalled- I think what happened to the efficiency is they were forced to push the clock speed way higher than initially anticipated due to the 780 / 780ti being somewhat faster than AMD was expecting. GCN has a sweet spot around 900mhz, above that it starts to get power hungry.

What I find *so* frustrating is that *if* the situation was reversed and nVidia were the ones about to release the faster, more efficient card (due to new memory and process tech) everyone would be praising. Yet as it's AMD the context is negative. Your entire 'dark' assessment of the situation hinges on the fact you believe AMD can't catch up on the efficiency of the underlying uArch, however I see no evidence to suggest they can't other than the fact nVidia have *finally* surpassed AMD's research budget for a very brief period of time after many many years of spending considerably less.

To be clear, any spending cuts now are not going to be visible in the near future, any new uArch is in development *years and years* before it's released. These cuts might hurt things in the 2020 time frame sure but what we're seeing now from all parties was in play well before.l

 

[jdwii]

That's nothing but made up by Amd fans and quite honestly I'm sick of hearing it. Nvidia gets more praise over marketing and they have more money to pay for that marketing. Not to mention everyone knows between the late 90's and mid 2000's Nvidia was on matched they had no competition so in a world with trust its big on their end.

I have plenty of Amd products a8 3520m laptop, 6950, 4770, 6670. I got my mom a a4 jaguar laptop not bad actually. My friend owns the 280x xfx GPU and 8320fx that I built for him. I built A10 5800k gaming build with 8gb of ram for 350$. 760k+250x gaming rig for 400$.
And I'm sorry but that doesn't mean I'll be a blind fanboy before I buy any GPU I read as much as I can on it. I make sure it runs quiet and cool. Sorry but I don't want water cooling I don't want a 212+ just to get a quiet CPU fan which is what I had to do while my friends i5 machine I built him runs quiet on his Intel heatsink.
Amd makes great products but since the 7970ghz GPU came out its been boring rehashing of older products or just moar cores. If the 390x uses hbm memory and is being built with 20nm and its 10-15% faster then the 980 while using the same amount of power it will be made fun of severally.
Juan's point and mime is that if Nvidia makes a ti product on 20nm with their efficient design were will amd be then? HBM matters more if you need that type of bandwidth which we do but we need the GPUs to be fast enough for it to.

Also Amd was the first to get gdddr5 memory and 40nm(first again with the 4770(which I own) loved that 105$ GPU! Kinda makes me feel a bit wishful for something like that again my 770 might live a short life with me plus amd is bringing a frame cap kinda like adaptive v-sync to their drivers something Nvidia had for quite some time in their drivers.
Anyways as stated again and again Amd needs a efficient design more then Nvidia since they have their APU series with horribly inefficient modules. Nvidia at least has arm CPUs allowing more headroom for a crappy power hungry design.

 

[Rafael Luik]

AMD will use HBM on the 390 series to reduce power consumption and compensate for the lack of efficient architecture. Same reason why will use 20nm instead 28nm.

Why isn't the move from 28nm to 20nm considered an advance in architecture to a more efficient one in itself for AMD?
And why isn't HBM considered an advance in architecture also?

You seem to cherry-pick and negate AMD's achievements calling them "compensations" just to say NVIDIA has a better architecure and approach, but the truth is AMD is also advancing in architecure just in different areas than NVIDIA. Correct me if I'm wrong...

The move from 28nm to 20nm is an advance in the fabrication process. And replacing GDDR5 by HBM is an advance in the memory subsystem. In both cases the graphical (micro)architecture is the same GCN than now.

Nvidia did its homework and got an efficient architecture that allows them to compete from phones to supercomputers. AMD is trying a brute force approach to get the performance needed to match a Titan II and then using die shrink plus HBM to maintain power consumption under control and liquid cooling to maintain safe temperatures. This brute force approach will allow AMD to continue playing on a niche market (HEDT), but will move them apart from supercomputers, servers, tablets... As a consequence their dGPU division will continue loosing market share and money...

What do you think will happen when Nvidia makes the die shrink and adds HBM to its already efficient Maxwell architecture?

AMD needs a new GCN architecture to compete, but it will be not ready before 2016 and I have my doubts about what AMD will achieve because lacks the money

But can NVIDIA suddenly make the die shrink and add HBM before 2016?
Do you get what I mean? They just invested in different places... I think you're picking what you want to call an "architecture" just for the sake of painting a brighter picture into NVIDIA. That's my impression.

I prefer to not comment on rumors, but it looks like AMD's perf/watt/cost is doing just fine whether they'll be using water cooling or not.

And let's not forget dGPUs won't exist anymore in 2020.

Amd needs a efficient design more then Nvidia since they have their APU series with horribly inefficient modules. Nvidia at least has arm CPUs allowing more headroom for a crappy power hungry design.

Can you show me a Tegra vs comparable AMD product comparison? I'm really curious to see it.

 

[cdrkf]

That's nothing but made up by Amd fans and quite honestly I'm sick of hearing it. Nvidia gets more praise over marketing and they have more money to pay for that marketing. Not to mention everyone knows between the late 90's and mid 2000's Nvidia was on matched they had no competition so in a world with trust its big on their end.

I have plenty of Amd products a8 3520m laptop, 6950, 4770, 6670. I got my mom a a4 jaguar laptop not bad actually. My friend owns the 280x xfx GPU and 8320fx that I built for him. I built A10 5800k gaming build with 8gb of ram for 350$. 760k+250x gaming rig for 400$.
And I'm sorry but that doesn't mean I'll be a blind fanboy before I buy any GPU I read as much as I can on it. I make sure it runs quiet and cool. Sorry but I don't want water cooling I don't want a 212+ just to get a quiet CPU fan which is what I had to do while my friends i5 machine I built him runs quiet on his Intel heatsink.
Amd makes great products but since the 7970ghz GPU came out its been boring rehashing of older products or just moar cores. If the 390x uses hbm memory and is being built with 20nm and its 10-15% faster then the 980 while using the same amount of power it will be made fun of severally.
Juan's point and mime is that if Nvidia makes a ti product on 20nm with their efficient design were will amd be then? HBM matters more if you need that type of bandwidth which we do but we need the GPUs to be fast enough for it to.

Also Amd was the first to get gdddr5 memory and 40nm(first again with the 4770(which I own) loved that 105$ GPU! Kinda makes me feel a bit wishful for something like that again my 770 might live a short life with me plus amd is bringing a frame cap kinda like adaptive v-sync to their drivers something Nvidia had for quite some time in their drivers.
Anyways as stated again and again Amd needs a efficient design more then Nvidia since they have their APU series with horribly inefficient modules. Nvidia at least has arm CPUs allowing more headroom for a crappy power hungry design.

*wow, just wow*

1: I have owned a mix of 3DfX, Nvidia, ATI, AMD and Intel products, I have *no* specific allegiance.

2: The general attitude about nVidia *doesn't* reflect reality. They released a card a tad faster than AMD's best and are suddenly 'years' ahead and AMD can't surpass them (cmon, really?). ATI / AMD have been neck and neck with nVidia for years. nVidia weren't ahead in the 90's then it was 3Dfx's time. nVidia had a short run where they were undisputed leaders from the Geforce 256 through to the Geforce 4 range. After that it was by no means a clean sweep.

3: I'm fed up with the idea that if AMD's next get cards aren't immeasurably faster *and* more efficient than the competing nVidia cards they have somehow failed, when nVidia are given the 'win' in the reverse situation.

4: If you actually read the leaks we're discussing then:
- the 380x is the card that is about 15% faster than a 980 whilst using a bit less power whilst using air cooling
- the 390x is significantly faster than the 980 and also faster than the equally theoretical Titan 2, uses less power than the latter (not quite as efficient as the 980 but neither is the Titan 2 according to same leak) whilst being equipped with water cooling.

So according to all the metrics you can come up with, the updated 3xx range is faster and more efficient than Maxwell, but apparently it's still somehow 'worse' than nVidia because, well you know, it's nVidia after all?

Final point, these are just leaks so shall we wait until something actually comes out eh? All I ask is that people treat AMD & nVidia with the same rules when looking at these. We are tech enthusiasts and should be capable of rising above the marketing and spin put out by *both* parties and look at the hardware underneath. What factors are *actually* important? What is *actually* being offered? You should be purchasing based on facts not the 'feeling' that one vendor is better than another.

 

[jdwii]

No because you know its built on a new 20nm which allows for adding more transistors with little to no work and HBM memory. Its like Intel making something on 14nm and screaming how much better it is then Amd parts at 32nm.
Who says Amd can't pass them again this is yet another endless rant about how everyone hates Amd. The true fanboys of amd want them to come out with a design that can compete with Nvidia using 28nm and still release it on 20nm with hbm memory.

It feels like piledriver on 22nm instead of a newer design on 22nm. Amd would have Nvidia by the neck if they could pull a maxwell like design but on a smaller node and with hbm memory it would probably be 30% faster. And if it didn't use water I'd probably buy it as soon as I can.

I actually think this has more to do with me being a big fanboy for Amd I want them to be ahead for awhile not slightly better don't you understand if nvidia shrinks their design they will meet amd in performance and maybe even beat them. Remember when the 8800 came out took Amd a couple of gens to beat it or at that time ati.

 

[esrever]

And I'm sorry but that doesn't mean I'll be a blind fanboy before I buy any GPU I read as much as I can on it. I make sure it runs quiet and cool. Sorry but I don't want water cooling I don't want a 212+ just to get a quiet CPU fan which is what I had to do while my friends i5 machine I built him runs quiet on his Intel heatsink.

Amd makes great products but since the 7970ghz GPU came out its been boring rehashing of older products or just moar cores. If the 390x uses hbm memory and is being built with 20nm and its 10-15% faster then the 980 while using the same amount of power it will be made fun of severally.
Juan's point and mime is that if Nvidia makes a ti product on 20nm with their efficient design were will amd be then? HBM matters more if you need that type of bandwidth which we do but we need the GPUs to be fast enough for it to.

Holy selective memory batman. I have no idea what you are doing but an AMD stock fan should cool the cpu under the same circumstances as an intel fan. I have built multiple computers with both and they are both bad, in the intel's case the fan is terrible and is not able to cool even their 77W cpus if you put a stress test level load on them. For AMD's case their fans are better but their CPUs use more power but they still run cooler than intel if both are using stock. Unless you mess with the settings or overclocked, they both will keep the cpu cool enough under normal load and neither of them are quiet.

You say AMD's been rehashing their old products? Go read a hawii tech sheet. They put more things into hawaii than nvidia did with gk110 that for sure. AMD actually launched more chips than nvidia in the same time frame. Their held the performance/watt crown and generally the price/performace crown for the majority of the generation. All the time nvidia has been touted as more power efficient just because people only select the top end cards to compare by people with selective memories.

I am not going to speculate on how efficient AMD's next platform will be but it is to be expected they will increase efficiency a decent amount. Generally speaking all they need should be competitive products at competitive prices but it seems all nvidia has to do is lie to people do get them to seem to perfect.

Nvidia's just been pretty much lying about power consumption of maxwel. Its efficiency is overblown by the absurdly low TDP nvidia printed on them. They actually use more power than their TDP in gaming tests, the first time any card has done this. The reviewers get 250W power target cards to do the testing on and to hit the TDP, the cards actually need to downclock. Sure they got a good efficiency boost but a 980's TDP should be over 220W if nvidia want to use AMD's standards for measuring TDP. Even when you compare power consumption of a 980 vs a 680, the 680 used less power but has a much higher TDP printed on it by nvidia and even the 680's TDP was under quoted compared to what AMD usually quote.

 

[esrever]

No because you know its built on a new 20nm which allows for adding more transistors with little to no work and HBM memory. Its like Intel making something on 14nm and screaming how much better it is then Amd parts at 32nm.
Who says Amd can't pass them again this is yet another endless rant about how everyone hates Amd. The true fanboys of amd want them to come out with a design that can compete with Nvidia using 28nm and still release it on 20nm with hbm memory.

It feels like piledriver on 22nm instead of a newer design on 22nm. Amd would have Nvidia by the neck if they could pull a maxwell like design but on a smaller node and with hbm memory it would probably be 30% faster. And if it didn't use water I'd probably buy it as soon as I can.

I actually think this has more to do with me being a big fanboy for Amd I want them to be ahead for awhile not slightly better don't you understand if nvidia shrinks their design they will meet amd in performance and maybe even beat them. Remember when the 8800 came out took Amd a couple of gens to beat it or at that time ati.

Your logic is terrible and you know nothing. Do you even know the electric characteristics of 20nm? If so show me how it works.

 

[jdwii]

It seems like when someone critizes Amd, fanboys get emotional and resort to horrible logic or even go as far as attacking another user or spreading false information that can easily be confirmed by looking at techpowerup reviews.

Read many reviews under new egg for the fx 8350 many claim the same as me through the stock heatsink away unlessm you want to hear a vacume like sound whenever you use the CPU heavily.

 

[jdwii]

But is that all they will ever be is fine? If they can get some design wins they will make lots of money and be more than just fine. Actually I'm not to sure but isn't Amd expensive in some parts of the world? On the build section at toms I here comments that anfx 6300 is close to a i5 in price yikes.

 

[Rafael Luik]

No because you know its built on a new 20nm which allows for adding more transistors with little to no work and HBM memory. Its like Intel making something on 14nm and screaming how much better it is then Amd parts at 32nm.
Who says Amd can't pass them again this is yet another endless rant about how everyone hates Amd. The true fanboys of amd want them to come out with a design that can compete with Nvidia using 28nm and still release it on 20nm with hbm memory.

Do you know how this discussion started right??
A claim that NVIDIA built better architecture so now they can beat AMD forever from now and make their dGPU unit go bankrupt, etc...

Then we say that AMD is also bringing better architecture, because they are probably doing a die shrink and bringing HBM, it makes them beat NVIDIA's current card offerings in performance and energy efficiency. But you can't accept it? WTF?

I do have a preference for AMD but I'm not illogical and I'm not ignoring either's advancement nor shouting "it's not faaaair" when someone does a die shrink.
You're illogical with that talk of "since they made a die shrink it's not fair". How about NVIDIA doing the die shrink too then?
They can't right now because, just like AMD didn't invest in the things Maxwell brought, they invested in a different place.
It's totally fair. Although for comparisons we have to consider their product launches are never aligned so each ~six months they switch position in their neck and neck race and we can't really compare these products.

It seems like when someone critizes Amd, fanboys get emotional and resort to horrible logic or even go as far as attacking another user or spreading false information that can easily be confirmed by looking at techpowerup reviews.

Read many reviews under new egg for the fx 8350 many claim the same as me through the stock heatsink away unlessm you want to hear a vacume like sound whenever you use the CPU heavily.

Yay let's discuss the future of graphics architecture by talking about a CPU from 2 years ago that's not even an APU.

 

[juanrga]

AMD will use HBM on the 390 series to reduce power consumption and compensate for the lack of efficient architecture. Same reason why will use 20nm instead 28nm.

Why isn't the move from 28nm to 20nm considered an advance in architecture to a more efficient one in itself for AMD?
And why isn't HBM considered an advance in architecture also?

You seem to cherry-pick and negate AMD's achievements calling them "compensations" just to say NVIDIA has a better architecure and approach, but the truth is AMD is also advancing in architecure just in different areas than NVIDIA. Correct me if I'm wrong...

The move from 28nm to 20nm is an advance in the fabrication process. And replacing GDDR5 by HBM is an advance in the memory subsystem. In both cases the graphical (micro)architecture is the same GCN than now.

Nvidia did its homework and got an efficient architecture that allows them to compete from phones to supercomputers. AMD is trying a brute force approach to get the performance needed to match a Titan II and then using die shrink plus HBM to maintain power consumption under control and liquid cooling to maintain safe temperatures. This brute force approach will allow AMD to continue playing on a niche market (HEDT), but will move them apart from supercomputers, servers, tablets... As a consequence their dGPU division will continue loosing market share and money...

What do you think will happen when Nvidia makes the die shrink and adds HBM to its already efficient Maxwell architecture?

AMD needs a new GCN architecture to compete, but it will be not ready before 2016 and I have my doubts about what AMD will achieve because lacks the money

But can NVIDIA suddenly make the die shrink and add HBM before 2016?
Do you get what I mean? They just invested in different places... I think you're picking what you want to call an "architecture" just for the sake of painting a brighter picture into NVIDIA. That's my impression.

I prefer to not comment on rumors, but it looks like AMD's perf/watt/cost is doing just fine whether they'll be using water cooling or not.

And let's not forget dGPUs won't exist anymore in 2020.

Nvidia decided to skip the 20nm node and move directly to 16FF. And the 16FF node will be not ready for 2015.

I am calling things by their name. Neither die shrinks nor changing the memory subsystem are changes in (micro)architecture. E.g. the PS4 uses GDDR5 instead the DDR3 used in Kabini laptops but the (micro)architecture is the same in both cases: Jaguar.

Amd needs a efficient design more then Nvidia since they have their APU series with horribly inefficient modules. Nvidia at least has arm CPUs allowing more headroom for a crappy power hungry design.

Can you show me a Tegra vs comparable AMD product comparison? I'm really curious to see it.

I didn't wrote that of above. You mixed my post with the post of someone else.

In any case, I agree with him/her. Nvidia has an efficient design enough to win tablets and phones designs, whereas AMD is avoiding phones as the plague and its tablet plans were canceled a pair of weeks ago...

I mentioned this before, but I will repeat because my point has been ignored: AMD brute force approach will work in a niche market: HEDT. The brute force approach is useless in phones, tablets, servers, supercomputers, laptops... and AMD cannot pay bills by selling a dozen of 390X to enthusiasts.

 

[juanrga]

amd is fine, console revenue still growing. apu market penetration still growing. firepro apu's continue to make headway in the overseas market yielding 184 GFLOPs of double precision performance for half the cost of traditional intel workstation. only in the usa do you hear amd is on the ropes and most of them are gamers with more money than sense. nvidia wins because it is consistent with there products, nothing magical and because of this they can demand a higher price. again amd is fine, it has access to the best fabs for current and future products, low margins and a growing mobile market ripe for carrizo, no i do not own amd stock

Let us see.

The former server head abandoned the ship. Currently market share has dropped to 3%, Seattle 16-core canceled, Warsaw not wining any important design (everyone prefer Xeons), Seattle 8-core being crushed by Atoms on one side and by X-Gene 1 on the other. And Berlin APU still missing (I suspect was a paper launch).

AMD impact is supercomputers is similar. About a 5% of top-500 list uses AMD CPUs. Last big customer (Cray) abandoned AMD by Intel and last month announced that will building ARM-based supercomputers as well. Guess what? Cray will use Cavium SoCs, because AMD doesn't have anything to compete.

Laptops: it is difficult to impossible to find some of the mobile Kaveri chips. Another paper launch.

Desktops: FX-9000 series was a fiasco. The 6000-8000 series have got a niche market thanks to aggressive pricing. This pricing has as result that the CPU division is loosing money. AMD uses financial engineering to hide this fact by reorganizing itself.

And desktop roadmap for 2015? Once mentioned Excavator, but latter that was canceled and recently the plans for a Excavator E/D did appear again. Who knows maybe tomorrow will disappear again.

Phones: AMD is absent here and avoids it as the plague because is far from competitive.

Tablet: Where are the promised design wins. Nowhere. Where is the Discovery project? Ups! it was canceled a pair of weeks ago.

Graphics: AMD has lost 22% market share in last count. The pro products are nowhere. They only got the Apple design because Nvidia is now a competitor of Apple in mobile. AMD is selling Firepro cards at one half the cost because otherwise nobody want them. But this will hurt financials and I don't see clear plans for future. Unsurprisingly AMD vicepresident has admitted that the GPU business division is in bad shape

Mr. Kumar considers the performance of AMD’s graphics business to be disappointing.

Add that the financial targets for the whole company were missed and that AMD did lose more money than expected and you will understand why the president abandoned the boat. Now AMD has a new president, but there are two problems: she lacks the experience and her first move at AMD has been a sound mistake. She has been denunciated by inversors by violating the internal rules of the company regarding her salary. Good start!!!!!

Add that AMD is broken into two, with only a 60% of employers approving Lisa Su management...

And AMD continues pivoting towards semicustom market to maintain itself afloat. But there is a problem. The console market is a low margin market and the next designs (K12 and Zen) are not the saviors that everyone expect. According to AMD the new designs won two semicustom designs and AMd expect to get 3 billions in revenue from those, during a period of 3 years. Pardon? 1 billion per year? That is peanuts. Currently Nvidia spends more than 1 billion per year only in R&D. And Intel spends 1 billion per quarter on incentives for mobile OEMs.

This seems to imply that K12/Zen are not so good and it seems that AMD only got the design wins thanks to aggressive pricing.

Mantle: Where are the dozens of games promised by Huddy for this year? Where is the promised Mantle for linux? Not many time ago someone tried to convince us that Ubuntu was full MANTLE. Bla bla bla bla bla bla. And when DX12 was here Mantle will vanish in the air.

HSA: Where are the half dozen of HSA applications promised? Nowhere because AMD had the smart idea of pushing HSA without the spec being finished and without hardware ready. Only Carrizo will be full HSA. Therefore, we could start seeing developers to consider HSA seriously by 2016/2017 or so.

It looks as everything is rosy. Isn't?

 

[juanrga]

As you say juan Bla Bla Bla, consoles may be lower margins but guess what amd does not have to do anything and revenue will continue it is called a back end revenue stream. And now Nintendo’s next console will use amd adding to that revenue stream. Server amd is doing bad but workstation is growing , I know for a fact amd firepro a320 our kicking the shit out of intel in pakistan and india. When you only look for bad news that is all you will find, Intel is subsidizing the mobile market a lot more than one billion a quarter, and is actually losing 1 billion a quarter net net in the mobile market. This is coming to a end as intel ceo Bryan Krzanich pointed out.

Do you really believe atom will be around in tablets without subsidies ? The expert say NO
"Atom shipment growth has come directly from Intel's liberal subsidies; shipments are expected to collapse as soon as Intel stops paying subsidies, as it plans to do next year" is everything rosy for amd some areas yes some NO, But as a whole amd is fine. You can stand on your soap box and proclaim amd is losing but amd is not subsidizing the market like intel and will be better off for it going forward.

Last point microsoft and amd have gotten a lot closer these past two years and the benefits of that closer relationship will pay huge dividends.

Consoles is what has maintained AMD affloat. But the revenue is too small to pay the bills of the whole company and that is why AMD is selling offices, canceling projects, reducing R&D to one half, cutting jobs, reducing marketing dept, delaying products, staying on cheap 28nm for a while... E.g., when you ask AMD about Mantle for linux and they say that are "still looking at the feasibility of the project", this is a fine way to say "sorry we don't have the money".

Those are facts, not rumors. Of course, if the Nintendo rumor is confirmed it will be a good notice, but very far from what AMD needs to solve its problems.

Of course, Atom doesn't have any future in tablets. I predicted here that Intel would fail on mobile pretension and I explained why, thus you are not saying me nothing new. The difference is that Intel is already planning to move to other markets, and of course has the money to try, whereas AMD simply has no backup plan except reducing costs, cutting jobs, canceling products... waiting to win another semicustom business to stay afloat another bunch of quarters.

 

[juanrga]

Compared to Q3-2013 the financial results for Q3-2014 show that revenue, operating income and net income have reduced. E.g. the net income reduced from $48M to $17M. Part of bad results are due to the Computing and Graphics division whose Operating Income reduced to $9M to -$17M. AMDF attributes this loss to decreased sales as compared to a year ago. The total debt at the end of this Q3-2014 is of $2.20 billion.

Evidently Lisa Su doesn't have another option than cancel projects and focus on those with higher priority. Moreover she ordered to cut costs by reducing workforce by about 7% which means around 710 people will be cut from the company by the end of Q4.

Expect further 10--15% decline in revenue for fourth quarter. But if you want say that AMD is fine, I can live with that.

About Intel atoms I said two things: first that Intel would fail due to x86 ISA disadvantage and second that Intel couldn't fight giants such as Qualcomm and Samsung.

TSMC is selling 20nm to Apple because Apple has the money to get early access to new nodes. I don't buy your "TSMC is set to produce amd 300 series gpu on 20nm"... the information that I have now is that AMD will be stuck to Glofo 28nm for the next gen GPUs.

 

[cdrkf]

Compared to Q3-2013 the financial results for Q3-2014 show that revenue, operating income and net income have reduced. E.g. the net income reduced from $48M to $17M. Part of bad results are due to the Computing and Graphics division whose Operating Income reduced to $9M to -$17M. AMDF attributes this loss to decreased sales as compared to a year ago. The total debt at the end of this Q3-2014 is of $2.20 billion.

Evidently Lisa Su doesn't have another option than cancel projects and focus on those with higher priority. Moreover she ordered to cut costs by reducing workforce by about 7% which means around 710 people will be cut from the company by the end of Q4.

Expect further 10--15% decline in revenue for fourth quarter. But if you want say that AMD is fine, I can live with that.

About Intel atoms I said two things: first that Intel would fail due to x86 ISA disadvantage and second that Intel couldn't fight giants such as Qualcomm and Samsung.

TSMC is selling 20nm to Apple because Apple has the money to get early access to new nodes. I don't buy your "TSMC is set to produce amd 300 series gpu on 20nm"... the information that I have now is that AMD will be stuck to Glofo 28nm for the next gen GPUs.

Ok, we get it AMD aren't in the best shape. Does that mean there is no hope? I don't think we're at that point yet.

Your earlier rant about how K12 / Zen are obviously rubbish due to the announced design wins is adding a lot of unknown factors together and jumping to a baseless conclusion. AMD won't have these designs available next year, so frankly the new design wins next year are unlikely to be related to K12 / Zen. The new wins are far more likely to be related to Skybridge which is more about the infrastructure than the processor design itself (i.e. the fact you'll have mix n' match ARM and x86 for the first time). The performance of these cores won't be anything revolutionary but then I think Skybridge is more of a proof of concept / building block much like how the earlier APU's have been slowly adding in HSA capabilities with each generation.

Also your comments about HSA are unfair- AMD are trying to get a new initiative off the ground and this is one area where the majority of the industry (excluding Intel and nvidia) is fully behind them on. I agree that Kaveri still lacks bits from the spec, but with something like this there is very much the old 'chicken and egg' situation. You need hardware to develop the software on, but you need software to justify the cost of developing the hardware. In AMD's case they've pretty much bet the company on the idea and have decided to try and kick the whole thing off by *providing* some hardware to get it going. The initial snippets shown about it are promising, if a few big names take it on and run with it then AMD could stand to do very well. The main thing is ensuring they stay afloat long enough to realise the potential from their investments, which is what the latest management have been all about.

What I would say is AMD need at least one of their new initiatives to take off and put them in the position of 'undisputed best' in a specific niche. The problem at the moment is they have very compelling products at lots of price points, however where they are good they are competitive rather than outright better (though as I've said before where dGPU are concerned I really don't agree with the opinion of nVidia being particularly ahead- their approaches differ but putting your research money into a die shrink + developing a new memory standard is no less worthy than taking the approach of optimising your design for the existing tech which is essentially what nVidia have done). If say Adobe got behing HSA in a big way, and an AMD APU was the *undisputed* fastest processor for graphics work then that would become a serious selling point for graphics workstations, and in turn start to built an 'AMD = best' reputation rather than the current 'AMD = value' which is the position they've been forced into.

As for K12 / Zen, if I'm being realistic then I seriously doubt they will surpass Intel in the way that Athlon 64 did a decade ago. I am however quietly confident that they will both:

A: Show significant per core performance improvement over the current Piledriver fx based chips (as Intel have shown there is *lots* of room for them the expand in that area).

B: Provide a decent efficiency improvement (if not outright performance boost) in areas that their current modular uArch is already strong in.

Being better isn't the same as 'saviour' or wiping the floor with Intel, however hopefully will be sufficient to keep them in the game on the CPU side of the equation (as they need a competent CPU core to make their 'semi custom' approach really fly).

 

[Rafael Luik]

I didn't wrote that of above. You mixed my post with the post of someone else.

In any case, I agree with him/her. Nvidia has an efficient design enough to win tablets and phones designs, whereas AMD is avoiding phones as the plague and its tablet plans were canceled a pair of weeks ago...

I mentioned this before, but I will repeat because my point has been ignored: AMD brute force approach will work in a niche market: HEDT. The brute force approach is useless in phones, tablets, servers, supercomputers, laptops... and AMD cannot pay bills by selling a dozen of 390X to enthusiasts.

What brute force? R9 285 energy efficiency gains = brute force?
The rumors showing performance and performance per watt of R9 300 series being both better than NVIDIA's current products?

I need more context to comment on the mobile market.

@ other posts
Did Intel or NVIDIA already released an APU with better graphics performance than AMD at the same price point and TDP? Take the A10-7800 as example.

 

[gamerk316]

amd is fine, console revenue still growing. apu market penetration still growing. firepro apu's continue to make headway in the overseas market yielding 184 GFLOPs of double precision performance for half the cost of traditional intel workstation. only in the usa do you hear amd is on the ropes and most of them are gamers with more money than sense. nvidia wins because it is consistent with there products, nothing magical and because of this they can demand a higher price. again amd is fine, it has access to the best fabs for current and future products, low margins and a growing mobile market ripe for carrizo, no i do not own amd stock

Let us see.

The former server head abandoned the ship. Currently market share has dropped to 3%, Seattle 16-core canceled, Warsaw not wining any important design (everyone prefer Xeons), Seattle 8-core being crushed by Atoms on one side and by X-Gene 1 on the other. And Berlin APU still missing (I suspect was a paper launch).

AMD impact is supercomputers is similar. About a 5% of top-500 list uses AMD CPUs. Last big customer (Cray) abandoned AMD by Intel and last month announced that will building ARM-based supercomputers as well. Guess what? Cray will use Cavium SoCs, because AMD doesn't have anything to compete.

Laptops: it is difficult to impossible to find some of the mobile Kaveri chips. Another paper launch.

Desktops: FX-9000 series was a fiasco. The 6000-8000 series have got a niche market thanks to aggressive pricing. This pricing has as result that the CPU division is loosing money. AMD uses financial engineering to hide this fact by reorganizing itself.

And desktop roadmap for 2015? Once mentioned Excavator, but latter that was canceled and recently the plans for a Excavator E/D did appear again. Who knows maybe tomorrow will disappear again.

Phones: AMD is absent here and avoids it as the plague because is far from competitive.

Tablet: Where are the promised design wins. Nowhere. Where is the Discovery project? Ups! it was canceled a pair of weeks ago.

Graphics: AMD has lost 22% market share in last count. The pro products are nowhere. They only got the Apple design because Nvidia is now a competitor of Apple in mobile. AMD is selling Firepro cards at one half the cost because otherwise nobody want them. But this will hurt financials and I don't see clear plans for future. Unsurprisingly AMD vicepresident has admitted that the GPU business division is in bad shape

Mr. Kumar considers the performance of AMD’s graphics business to be disappointing.

Add that the financial targets for the whole company were missed and that AMD did lose more money than expected and you will understand why the president abandoned the boat. Now AMD has a new president, but there are two problems: she lacks the experience and her first move at AMD has been a sound mistake. She has been denunciated by inversors by violating the internal rules of the company regarding her salary. Good start!!!!!

Add that AMD is broken into two, with only a 60% of employers approving Lisa Su management...

And AMD continues pivoting towards semicustom market to maintain itself afloat. But there is a problem. The console market is a low margin market and the next designs (K12 and Zen) are not the saviors that everyone expect. According to AMD the new designs won two semicustom designs and AMd expect to get 3 billions in revenue from those, during a period of 3 years. Pardon? 1 billion per year? That is peanuts. Currently Nvidia spends more than 1 billion per year only in R&D. And Intel spends 1 billion per quarter on incentives for mobile OEMs.

This seems to imply that K12/Zen are not so good and it seems that AMD only got the design wins thanks to aggressive pricing.

Mantle: Where are the dozens of games promised by Huddy for this year? Where is the promised Mantle for linux? Not many time ago someone tried to convince us that Ubuntu was full MANTLE. Bla bla bla bla bla bla. And when DX12 was here Mantle will vanish in the air.

HSA: Where are the half dozen of HSA applications promised? Nowhere because AMD had the smart idea of pushing HSA without the spec being finished and without hardware ready. Only Carrizo will be full HSA. Therefore, we could start seeing developers to consider HSA seriously by 2016/2017 or so.

It looks as everything is rosy. Isn't?

Looks who's suddenly come around to my line of thinking. Wasn't I the one who said years ago that console sales wouldn't generate much revenue? Or that AMD has missed it's chance at Mobile? Or its disappearing act in the lucrative server market? Or the lack of widespread Mantle adoption? And so on.

The problem AMD has is simple: They need people to realize there are alternatives to Intel and NVIDIA. They needed to win SIGNIFICANT OEM products to remind the masses there are alternatives. When's the last time you've seen an AMD advertisement on TV?

I said it when smartphones became a thing: AMD was in unique position to succeed because of the ATI merger. For reasons that still boggle the mind, they simply NEVER MADE AN ATTEMPT. Likewise, the high margin server market they basically ignored. And so on. Left with desktops, which isn't high margin at all.

So no matter how you spin it, AMD's long term future is not very bright.

 

[colinp]

It's never a good sign when a company is slashing headcount and R&D budgets. The signs point to AMD becoming a niche player, quietly carving out a small sector of the market but never again able to challenge the 500 pound gorillas in the market. Cyrix/Via anyone? And that's not a great outcome for consumers.

What can save AMD now? It'll need a competitive next gen GPU that wins OEM deals across the spectrum, with good margins. E.g. There's a reason why you can't buy a decent gaming laptop right now with an AMD part, and that's because nobody wants to buy an inferior product.

Next gen APU had also better be good, or at least good enough. But that'll be for nothing without some OEM wins. As I and others have pointed out, where are all the non ULV Kaveri laptops? Intel isn't afraid of putting a 47W chip into laptops and at that TDP, Kaveri is pretty good!

 

[jdwii]

Rafeal I'm on my nexus 7 and for some reason I can't find reply but this is a reply meant to be for you.
Your comment on the 285 is something I just wanted to check again and yes Juan if reading this I'll try my best to make sure its at stock

http://www.guru3d.com/articles_pages/amd_radeon_r9_285_review,5.html
Power consumption is basically the same as the 770 while being 10-15% slower here is a OC 285(so it even has a performance advantage) at techpowérup I chose this since it gives percentages and they review the card with a lot of games
http://www.techpowerup.com/mobile/reviews/Sapphire/R9_285_Dual-X_OC/25.html

So from the looks of it they caught up with nvidias 680 series in terms of performance per watt.

Also for others I would like them to look at my first URL I gave with power consumption as I stated time and time again the 280x(7970ghz) is less efficient then the 770 this site has both at stock. Also take a look at the 290x using as much power as a 690 and a 280x using slightly more powerpower then a 780 while being 20% slower(using techpowerups URL I gave on performance).

The thing is nothing here is subjective its charts with facts on them comparing different products. If some feel these results are false go to the reviewers themselves and tell them I just simply use other peoples work not my own testing. However I do CPU and GPU benchmarks on my own since my life is boring a lot of the time .

 

[con635]

@jdwii
Look at some of the newer reviews on tpu which include newer drivers and games, the 7970ghz comes around 10% of the 780 now, gcn deserves more credit its lasting fine, better than kepler. ppw is just one metric and because its the best for nvidia they market it well, is it just me or did no one care as much about it years ago? The performance and performance per dollar charts are healthy for amds old cards. The tpu ppw charts may change a bit if the dynamic frame rate control in future drivers works out also.
http://www.techpowerup.com/reviews/MSI/GTX_970_Gaming/27.html

 

[esrever]

People are so adamant about 1 metric. It really is overblown. If thats all you care about then good for you. Its not what everyone cares about. The difference isn't something you notice unless electricity is extremely expensive for you. Power consumption on a flagship GPU shouldn't matter that much to people but its all the nvidia fans are pushing these days. The 60+W difference loaded of a 290x vs a 970 is pretty substantial but the 20W difference between a 7970 and a 680 is not really much to write home about. Just gotta wait to see what comes out next.

Also JWII, please just stop beating this dead horse. I have already told you why AMD's top end cards use more power. Its blatantly obvious at this point that AMD isn't as efficient as nvidia at the top end. Whats the point of keep posting about it? Do you really need to justify your 770 purchase that much? Should I just continuously post about how nvidia cards have no longevity because their memory bandwidth/capacity limitations? Or how nvidia cards like the 680 are getting beaten by 7950s in the new games that are ported from consoles? What good is a 680's power consumption compared to a 7970 when a 7950 is probably just as powerful in the long run and uses less power most of the time?