CPU Whiz Jim Keller Leaves AMD

[Schmuckley]

inb4 anyone



Guess he Wizzed on outah there...

The open question is, What's after Zen ?

If it sucks..Not a thing.
I'm not too happy with AMD.
5 years of producing chips with worse performance than previous chips and they stuck with that?
Who makes these kind of decisions?
"Oh, It's good enough"
umm..Not in my eyes.
They've lost all server market pretty much.
Nobody wants to run a Bulldozer server;That architecture sucks.

 

[medi01]

If Zen is to hit shelves in second half of 2016, the project is complete-ish by now. (bugfixing etc)
He always did that, project complete => leave, so nothing alarming (yet).

I wonder how much of the 40% IPC improvement promise will Zen really deliver.

The question that people are asking regarding AMD's future is "will Zen be good?", but that's the wrong question. The correct question is "what comes after Zen?", and AMD doesn't have an answer for that, because they are terrible at long-term planning.

Meanwhile Intel is already ...

AMD's APUs that got them into both Xbox/PS4, who is catching up with Iris?
AMD's 64 bit instruction set vs I64?
AMD's "fuck long pipelines", vs, cough, Prescott?
AMD's DirectX 12 support?
HBM?
How long did it take them to respond with FreeSync? Why does it cost next to nothing (built-in into upscalers) and put no restrictions on ports (gsync => 1 display port and nothing else)?


And, FFS, you are mocking a company that manages to stay afloat vs competitors wich vastly more resources, please, at least get some clue.

 

[SaltyFish]

The i-series comprises several architectures. Every "Tick" is a new microarchitecture in Intel's "Tick-Tock" lifecycle.

I meant as a major leap; P4 -> Core 2 -> Core-i
Some people are still on first-gen Core-i because to the lack of major improvements.

 

[john_]

I meant as a major leap; P4 -> Core 2 -> Core-i
Some people are still on first-gen Core-i because to the lack of major improvements.

After the IMC integration, what else can we expect to be a game changer? I think there aren't many things that Intel or AMD can do. Pentium 4 was a disaster. Core 2 was nothing more than going back to Pentium III and improving it. i7 saw the integration of the IMC. What else can Intel do, to be considered a huge architectural change? I can't think anything other than HSA.

 

[SaltyFish]

After the IMC integration, what else can we expect to be a game changer? I think there aren't many things that Intel or AMD can do. Pentium 4 was a disaster. Core 2 was nothing more than going back to Pentium III and improving it. i7 saw the integration of the IMC. What else can Intel do, to be considered a huge architectural change? I can't think anything other than HSA.

If that's the case, and assuming Zen can match Core-i, then there really isn't much to do after Zen. Short of some major breakthrough, all AMD and Intel would be making are minor tuning adjustments to their respective architectures.

 

[john_]

Intel is already there. So, the only thing they can do right now is to continue improving iGPU to be more competitive against Nvidia and AMD and probably in a better position in case HSA becomes importand, and continue improving in efficiency lowering power consumption to be more competitive against ARM.

 

[HumanSmoke]

If Zen is to hit shelves in second half of 2016, the project is complete-ish by now. (bugfixing etc)

AMD themselves have said that Zen is sampling in 2016. Actual customer availability is now 2017.

 

[R-T-B]

If Zen is to hit shelves in second half of 2016, the project is complete-ish by now. (bugfixing etc)
He always did that, project complete => leave, so nothing alarming (yet).

I wonder how much of the 40% IPC improvement promise will Zen really deliver.



AMD's APUs that got them into both Xbox/PS4, who is catching up with Iris?
AMD's 64 bit instruction set vs I64?
AMD's "fuck long pipelines", vs, cough, Prescott?
AMD's DirectX 12 support?
HBM?
How long did it take them to respond with FreeSync? Why does it cost next to nothing (built-in into upscalers) and put no restrictions on ports (gsync => 1 display port and nothing else)?


And, FFS, you are mocking a company that manages to stay afloat vs competitors wich vastly more resources, please, at least get some clue.

Those are all short term successes. Yes, that's the hit list of AMD victories. Their long term planning has been terrible, is his point.

 

[medi01]

Those are all short term successes. Yes, that's the hit list of AMD victories. Their long term planning has been terrible, is his point.

That has little to do with what I said. That APU thing goes years back into ATI purchase. That's as long term as it gets.

 

[medi01]

AMD themselves have said that Zen is sampling in 2016. Actual customer availability is now 2017.

Yikes.
Although, no wonder, considering fabs are only about to go 16/14nm.

 

[OneMoar]

 

[R-T-B]

That has little to do with what I said. That APU thing goes years back into ATI purchase. That's as long term as it gets.

And tell me, how has that panned out from a profit perspective?

My main and only point is the company is sinking. I know they are better off alive but something needs to change to acomplish that.

 

[Blue-Knight]

The open question is, What's after Zen ?

The end.

 

[john_]

After Fable Legends reviews shows that AMD cards are more than competitive under DX12, I wouldn't say that AMD's end is near. They will be gaining market share in the next year and take back of the market share they lost after Maxwell release. The only way this can change is if Nvidia manages to enable Async compute in hardware - if we assume there is that hardware in Maxwell - or if they manage to create a software based solution that can really help their cards. But I don't really expect that. And I don't expect a Fermi DX12 driver either (there are half a dozen low end Nvidia cards based on Fermi still selling).

 

[HumanSmoke]

After Fable Legends reviews shows that AMD cards are more than competitive under DX12 (1), I wouldn't say that AMD's end is near (2). They will be gaining market share in the next year and take back of the market share they lost after Maxwell release (3). The only way this can change is if Nvidia manages to enable Async compute in hardware (4)- if we assume there is that hardware in Maxwell - or if they manage to create a software based solution that can really help their cards. But I don't really expect that (5). And I don't expect a Fermi DX12 driver either (6)(there are half a dozen low end Nvidia cards based on Fermi still selling).

I must admit, I am quoting you primarily for bookmarking reasons. I love pronouncements derived from emotion rather than historical precedent.
(1). What Fable Legends shows is that neither AMD's or Nvidia's architectures offers a home run. Games will need to be taken on a case by case basis depending upon what is implemented. From this analysis by Anandtech it looks like AMD's gains in asynch compute are offset by their lack of hardware-based conservative rasterization and voxelization.

Spoiler

(2) Neither would I, but this point isn't predicated upon how well or poorly the graphics division is doing. AMD's loans aren't due until the 2019-20 timeframe, and they aren't going to dig themselves out of that hole selling a few thousand units of enthusiast graphics cards. The bread and butter has to be the server space and a new workstation/server/HEDT platform capable of matching Intel. If I were you I'd worry less about the graphics side of things than Intel's Skylake-EP/-EX scaling up to 28 cores/56 threads, and ARM 48-64 core SoC's.
(3) Sounds like a tall order. For that to happen, AMD would need to sell 24.1 million cards more than Nvidia over the next four quarters. To put that into perspective, that means that AMD have to go from 18% (to Nvidia's 81%) of the market to 66% consistently starting from the quarter we are in at the moment. Just to add a little more perspective, AMD launched a whole top-to-bottom DX11 series in Q3 2009-Q1 2010 when Nvidia could not field a single DX11 capable card and relied upon a trickle of GTX 285/275's and the GTS 250 ( 9800GTX+ by another name), yet still conspired to lose out on market share ( AMD sold 22.51 million cards to Nvidia's 36.78 million over the three quarters before the GTX 480/470 appeared in the retail channel)
(4) see point 1. Also, the vast majority of graphics cards are sold to people who understand nothing of GPU technology, and of the tiny fraction of those who might actually frequent a tech site, they'll look at a bar graph and call it good.
(5) It has already been established that (1) Async compute isn't the be all and end all, and (2) For all the dire prophesy, AMD's architecture isn't destroying all comers in benchmarks. A GTX 980 Ti still compares well with a Fury X. The only real difference is price (excepting the Fury line), but since AMD has had that edge forever and still hasn't made any inroads into gaining market share, how vital is that?
(6) Nvidia's stance is that they'll provide a driver by the end of the year. I guess we'll find out if the first of your predictions will come true on 31st December. I'll keep updating this post to let you know how many cards AMD needs to sell to reach parity with volumes conceded after Q2 2014.

 

[john_]

.............

I have to admit that I NEVER bookmark someone's post. I am not that much of a hardcore fanboy to care about that. As for emotions, should I remind you in the past who was rushing to prove me wrong, only to realize that he didn't knew the facts? I haven't bookmark that. You are the only person reminding me about that post with your newer posts. But if I had to search for it today, it would take me some time.

1) What Fable Legents shows, is that AMD's cards are back in the game. More people will consider them as a future proof option, less will care if they are rebrands. Having a game like Fable Legents giving you a full analysis and showing that it doesn't favor any architecture specifically, makes it a really objective benchmark and probably representative of what we should expect. Many where considering AotS biased. They can't say that about FL, still 390X beats 980 and 390 beats 970. With DX11 things where much more in Nvidia's favor.

2) Money is money. There is no difference if it is money from CPUs or GPUs. It's money. Even if we are talking about 20-30 millions more every quarter, considering AMD's financial position, that difference can have a very positive impact on R&D expenses in the next 12 months. Now, if you want to see them becoming equal to Intel, until you say that their future is secured, then you will have to wait.

3) If in the last quarter the number of AIBs sold where less that 10 millions, how do you make that 24.1 millions cards number? And what exactly is your target here. By getting 66% of the market for 4 quarters, where is AMD going to land after 4 quarters? 20% market share? 40% market share? 60%? 80%? I think you are the one here with the tall order, just so you can make it look impossible.

4) We are in 2015. Almost everybody has access to internet or knows someone who do have access to internet. Many people ask others who DO know about GPUs. Or just register an account on a tech site, create a thread "What can I buy with that amount of money", get the answers, buy the card they where told them to buy and then never again visit that forum. People do look at graphs, also many just buy gigabytes. If things where as simple as you describe them, the market share would have been closer to 50-50 than 20-80. That 20-80 can happen ONLY if most people ASK before buying.

5) You need a full GM200 with it's 96 ROPs to have a victory over hi end AMD cards. Compare any GM204 with 64/56 ROPs against the 64 ROPs AMD hi end cards, and AMD cards win. I like how you say here that AMD had always the edge on price but that didn't help them enough. So can this be possible an indication that people DO ask about GPUs before buying?

6) The end of the year. The question is, which year. Don't answer that.

 

[HumanSmoke]

If in the last quarter the number of AIBs sold where less that 10 millions, how do you make that 24.1 millions cards number?

You said:

They will be gaining market share in the next year and take back of the market share they lost after Maxwell release.

Prior to Maxwell's release AMD were selling on average 5.1 million discrete cards per quarter. The last two quarters they have sold 2.54 million and 1.69 million. In order to regain their prior market position they would need to outsell Nvidia by a margin of 2:1...unless your thinking is that for AMD to regain market share in your eyes is to transiently hit 35-37% for a quarter to consider it job done. If that is your thinking then it is false economy. AMD's market share prior to Maxwell's launch was pretty steady- albeit in a slow decline, at 40.2% in Q2 2012 to 37.9% in Q2 2014, yet because of overall discrete market decline and AMD's lower ASP's, they have shipped fewer cards every quarter for the last two years and realized less and less profit.
A financial quarters market share taken in isolation means little, and history tells us that even with being first to DX11, and having virtually no opposition (including none at the enthusiast level aside from the GTX 295 for most of a year), and Nvidia's Fermi delay, AMD have never managed to claw back more than 10% market share before dropping back. The ONLY time ATI/AMD took serious market share from Nvidia was in 2004 and that had little to do with the cards and everything to do with ATI being able to field PCI-E interface cards when the Intel 915/925X chipsets launched.
How many times has the AMD revolution been imminent? and how many times has it come up short? You can argue the future all you like, but until it happens it is moot....just like the AMD revolution

We are in 2015. Almost everybody has access to internet or knows someone who do have access to internet. Many people ask others who DO know about GPUs. Or just register an account on a tech site, create a thread "What can I buy with that amount of money", get the answers, buy the card they where told them to buy and then never again visit that forum. People do look at graphs, also many just buy gigabytes. If things where as simple as you describe them, the market share would have been closer to 50-50 than 20-80. That 20-80 can happen ONLY if most people ASK before buying.

One huge flaw in your argument is that one company possesses top of the mind brand awareness among consumers, and one doesn't. Believe it or not, people had the internet in 2014, and 2013, and 2012...check if you don't believe me. AMD have had performance parity during almost the entire modern GPU era, including a period where they owned the fastest card (HD 5970), fastest single GPU card (HD 5870), best bang for buck card since the 8800GT ( HD 5850), Eyefinity, and full DirectX 11 compliance when its competitor had none. YET STILL LOST MARKET SHARE. Yet, for some reason - that you put down to people having internet access in 2015, AMD is going to double its market share because it has (at best) a slight edge in DX12 and a distinct disadvantage in DX11 (where a lot of titles will reside for some time). Yeah, OK. That's why I'm going to keep a cursory eye on your predictions - I love a good come from behind win.

I like how you say here that AMD had always the edge on price but that didn't help them enough. So can this be possible an indication that people DO ask about GPUs before buying?

Ask a hundred people whether Intel is better than AMD. Do you really think the split would accurately reflects the relative merits of the processors and platforms?
You seem fundamentally unaware of brand recognition and top of the mind awareness. AMD have plenty of adherents and advocates on tech forums, indulge in guerrilla/viral marketing, and have PR up to the eye teeth - yet they still suffer in image (and thus in sales), in part because they lack the profile, and in part because for every step forward they tend to shoot themselves in the foot taking the step. AMD's produces great hardware, but for every Fury launch and DX12 benchmark, AMD have a big Roy Taylor moment - and when sites start calling them on it, they and their more ardent fans pitch a full scale hissy fit. It really shouldn't be a surprise that company confidence leads to a stronger customer marketing perception (and customer perception equals reality in marketing - one look at Apple should suffice as an example) ....and quite frankly, AMD possesses none. Intel seldom indulge in public displays of whining and even Nvidia seemed to have learned after Jen-Hsun's "open a can of whoop ass" moment, yet AMD persist with the plucky underdog shtick - which just tells the average consumer that the company is giving itself a licence to fail.

Anyhow, we'll see how well in tune with the technology market we both are in due course. It is rather pointless arguing over future history.
/OUT

 

[cadaveca]

Ask a hundred people whether Intel is better than AMD.

40% will say "Who is AMD?"

30% will say "Intel".

28% will say "wtf did you just say?"

2% read forums.

Polled 452 people at school.

 

[john_]

@HumanSmoke
Obviously I don't expect them to go much higher than 25-30%+. We will have to see if that will continue with the FinFETs, or if Nvidia will manage to gain more market share. And it's not false economy. It is reality. You can keep asking from AMD to become Intel or Nvidia overnight, only so you can come to easy conclusions. Double standards and easy conclusions about AMD, are really common and boring to read. And don't make me look like someone who is struggling to pass his opinion. You started it remember? You write the big posts here.
The rest of your post is just your own conclusions and imaginations with the usual double standard approach. In many things you do have a point, but it is really easy to hit someone who is down. We can talk all day about Roy's stupidity, but he wasn't the one who wanted us to believe that in the same company enginnears do not use the internet and talk a different language than the marketing department. Of course Roy will go on the bookmarks list. Nvidia's excuses where swallowed instantly.
Now if you want to make this personal, start reading and writing Greek. Then I will be happy to lose some more time arguing with the wall you.

 

[rtwjunkie]

@HumanSmoke
Obviously I don't expect them to go much higher than 25-30%+. We will have to see if that will continue with the FinFETs, or if Nvidia will manage to gain more market share. And it's not false economy. It is reality. You can keep asking from AMD to become Intel or Nvidia overnight, only so you can come to easy conclusions. Double standards and easy conclusions about AMD, are really common and boring to read. And don't make me look like someone who is struggling to pass his opinion. You started it remember? You write the big posts here.
The rest of your post is just your own conclusions and imaginations with the usual double standard approach. In many things you do have a point, but it is really easy to hit someone who is down. We can talk all day about Roy's stupidity, but he wasn't the one who wanted us to believe that in the same company enginnears do not use the internet and talk a different language than the marketing department. Of course Roy will go on the bookmarks list. Nvidia's excuses where swallowed instantly.
Now if you want to make this personal, start reading and writing Greek. Then I will be happy to lose some more time arguing with the wall you.

@HumanSmoke is not "hitting them when they are down." He's just calling it like it is. Perception IS reality, just as he mentioned, and it's not just true in marketing. Name/Brand recognition is everything, and FAR more people at least know the name Intel or Nvidia (especially Intel) than AMD.

Very few average people have ever heard of AMD, and THEY are the ones who make or break a company, not the few of us with tech knoledge and interest. Obviously, that is a major problem in their strategy.

Until they fix that visibility image, it doesn't matter how good their products are.

 

[john_]

We saw Nvidia's brand recognition in the ARM platform. They failed miserably. They managed to find place for their Tegras in cars and a couple of their own products. We saw Intel's brand recognition also in tablets and smartphones, where they had to throw billions just so manufacturers use their chips.

I say that again and it's your choice to understand it or not. For the same things people in here accuse AMD, they will find plenty of excuses for Nvidia and Intel. And for every example they have to show where AMD fails, they will throw under the carpet 10 equal examples where Intel or Nvidia failed. It's really boring. But it is really good to see that drivers/performance/efficiency, as reasons for AMD failures, have been replaced with perception/name/brand recognition.

 

[rtwjunkie]

But it is really good to see that drivers/performance/efficiency, as reasons for AMD failures, have been replaced with perception/name/brand recognition.

As far as Joe Public is concerned, yes. And Joe Public is all that matters, not you or I or any of the infantesmally small percentage of people on tech sites.

No one is making excuses for anyone. We're just talking about reality. Until AMD can overcome the lack of name recognition, they will not recover their glory.

 

[cadaveca]

As far as Joe Public is concerned, yes. And Joe Public is all that matters, not you or I or any of the infantesmally small percentage of people on tech sites.

No one is making excuses for anyone. We're just talking about reality. Until AMD can overcome the lack of name recognition, they will not recover their glory.

I'd have to disagree. The public is unimportant, as they are not the major purchasers of PC hardware. Business is, and brand recognition is not as important in the corporate contracting world. For public purchases, sure, I'd agree with you, but discrete graphics cards sales by the public are a small part of the market, and that is really why cards that cost $100 are so popular. People aren't really broke so much as such data would like to represent. I kind of read that as John_'s point... What businesses look for and what you and I look for and what the general public look for are all completely different things.

 

[rtwjunkie]

I'd have to disagree. The public is unimportant, as they are not the major purchasers of PC hardware. Business is, and brand recognition is not as important in the corporate contracting world. For public purchases, sure, I'd agree with you, but discrete graphics cards sales by the public are a small part of the market, and that is really why cards that cost $100 are so popular. People aren't really broke so much as such data would like to represent. I kind of read that as John_'s point... What businesses look for and what you and I look for and what the general public look for are all completely different things.

Agreed, we all look for products differently, which is kind of the point I was getting at. What john_ seems to be attributing is tech site member knowledge, which isn't what the mojority of anything have, be it the public, or business.

Your business example is alot more believable than john_ attributing childish shenanigans to major companies.

It doesn't change though, that brand recognition is important for a company overall, just maybe not to the degree I attributed it.

 

[john_]

childish shenanigans

You had to downgrade the different opinion. Anyway, I wouldn't say that I wasn't expecting something like that.
I wasn't expecting any kind of comments about Nvidia and Intel in the smartphone markets either. You have two companies with, money, brand name recognition, hi technology that can give them the edge over the competition and they fail miserably. In the end one company is paying the manufacturers to use it's chips, the other end up suing the competition because it can not compete. But, let's forget about that. Let's talk about AMD. Shoot...