AMD mulling break, spinoff

[ShintaiDK]

I thought the future was fusion, not breaking it up or spinning it off?

The current business model isnt substainable financially. AMDs healthy part is the semicustom. The CPU side is a complete disaster and the GPU part is close to being completely starved to death. Some kind of breakup between CPU and GPU+Semicustom would at least leave a company that is viable for the time being.

But one thing is certain, something drastic needs to happen.

 

[nemesismk2]

this is bad news so guess AMD is worse off than i thought

 

[Fjodor2001]

The current business model isnt substainable financially. AMDs healthy part is the semicustom. The CPU side is a complete disaster and the GPU part is close to being completely starved to death. Some kind of breakup between CPU and GPU+Semicustom would at least leave a company that is viable for the time being.

But one thing is certain, something drastic needs to happen.

You're commenting on the current state of AMD. But since we're talking about the future, you need to take Fiji, HBM, Zen and 14 nm into account. Things will look very different then.

 

[ShintaiDK]

You're commenting on the current state of AMD. But since we're talking about the future, you need to take Fiji, HBM, Zen and 14 nm into account. Things will look very different then.

Fiji and HBM wont change anything at all. While the 300 series rebrand will just continue the marketshare loss. And Zen needs to be nothing less than a miracle. 14nm chips needs around 150 million $ just in chip design alone. And about 1.5 billion $ revenue to make them attractive.

AMDs turnaround for the better have been expected for around 10 years now. And it never happend. Always the promise f the next thing to save the company. But the company is weaker than it ever been before.

By the time 14nm chips may even be possible. AMD may have gone from ~2% x86 revenue to ~1%. And graphics marketshare from 22.5% to 10-15%. And the other companies isnt standing still.

 

[mrmt]

You're commenting on the current state of AMD. But since we're talking about the future, you need to take Fiji, HBM, Zen and 14 nm into account. Things will look very different then.

And why do you think things will change too much? AMD had performance parity with Nvidia before and they didn't gain much market share, HBM is something Nvidia can also use and Zen, well, we already had plenty of hyped AMD CPUs and none of them work. In fact, it is quite worrisome that their current cash cow, the cat family, will have no replacement at all.

In fact, by the looks of it I think Lisa will be trashed around Q3 or Q4 because of the failure in regaining share to Nvidia, not even semiaccurate believes it will happen.

 

[Wreckem]

And why do you think things will change too much? AMD had performance parity with Nvidia before and they didn't gain much market share, HBM is something Nvidia can also use and Zen, well, we already had plenty of hyped AMD CPUs and none of them work. In fact, it is quite worrisome that their current cash cow, the cat family, will have no replacement at all.

In fact, by the looks of it I think Lisa will be trashed around Q3 or Q4 because of the failure in regaining share to Nvidia, not even semiaccurate believes it will happen.

She has only been AMDs CEO for 9 months. Add on her brief stint as COO and its only been a year as a high level executive at AMD.

Honestly the market reaction was more because of the CEO than the product line. I doubt they axe her anytime soon.

 

[soccerballtux]

respins are expensive

 

[mrmt]

Honestly the market reaction was more because of the CEO than the product line. I doubt they axe her anytime soon.

Not axe her, but promise a come back in their GPU business and deliver a bunch of rebranded with a higher price tag won't be taken lightly by the market. The gap between what management promises and what they are delivering is still here.

 

[jpiniero]

There can be no spin off because there s nothing that could be spun off, AMD need both the GPU and CPU IPs since their products integrate boths in the same silicon.

There s no info at all in this fud article, must originate from some short sellers that were caught off guard by the recent stock surge...

I'll tell you what probably happened - AMD hired Goldman Sachs or someone similar to explore a spinoff or a sale of one of the units (to raise cash, since they obviously need it). They had to unwind their short position because they took them on as a client, hence the recent surge. The source of the report would be one of the employees of the investment bank and the IB wanted it known to drum up interest.

 

[Justinbaileyman]

I am not understanding what I am reading here.. So is this saying that there will be no more dedicated GPU's after this generation from AMD? Are they going belly up then?This is not good news..

 

[podspi]

It sounds to me like AMD really is betting the farm with Zen. I doubt they are planning on breaking up the company if their products going forward are a success, but if Zen flops it makes perfect sense to jettison the CPU business, hopefully firewall the GPU/SoC side away from AMD's huge amount of debt, and then sell it. Then whatever AMD's CPU business is called (maybe AMD?) either turns into the next Via or sells itself off, or continues to fight the good fight until finally declaring bankruptcy.

I don't necessarily see this as a bad sign, as much as management doing its due diligence. It also says a lot about how much is riding on Zen, imho. Personally I think their GPU-side of the business is fine (from a technical standpoint), but certainly it is not being managed as good as it could. It seems like AMD is pushing as hard as it can on perf/watt, which is their main issue on the GPU-side.

 

[NTMBK]

It sounds to me like AMD really is betting the farm with Zen. I doubt they are planning on breaking up the company if their products going forward are a success, but if Zen flops it makes perfect sense to jettison the CPU business, hopefully firewall the GPU/SoC side away from AMD's huge amount of debt, and then sell it. Then whatever AMD's CPU business is called (maybe AMD?) either turns into the next Via or sells itself off, or continues to fight the good fight until finally declaring bankruptcy.

I don't necessarily see this as a bad sign, as much as management doing its due diligence. It also says a lot about how much is riding on Zen, imho. Personally I think their GPU-side of the business is fine (from a technical standpoint), but certainly it is not being managed as good as it could. It seems like AMD is pushing as hard as it can on perf/watt, which is their main issue on the GPU-side.

Yup, this. They're getting a plan B ready in case Zen doesn't pan out. All the executives leap into the ATi lifeboat, leaving AMD to sink.

 

[Fjodor2001]

Fiji and HBM wont change anything at all. While the 300 series rebrand will just continue the marketshare loss. And Zen needs to be nothing less than a miracle. 14nm chips needs around 150 million $ just in chip design alone. And about 1.5 billion $ revenue to make them attractive.

AMDs turnaround for the better have been expected for around 10 years now. And it never happend. Always the promise f the next thing to save the company. But the company is weaker than it ever been before.

By the time 14nm chips may even be possible. AMD may have gone from ~2% x86 revenue to ~1%. And graphics marketshare from 22.5% to 10-15%. And the other companies isnt standing still.

The amount of FUD you are able to produce is astonishing as usual.

Observe this:

* AMD will be on 14 nm next year.
* Zen will be released next year, and is expected to reach about Haswell levels of IPC. It will also be 8-16 cores.
* Since Intel has not improved desktop performance much since Haswell (in fact since SB), Zen will be competitive with Intel's top CPUs, and likely also priced lower providing more "bang for the buck".
* Fiji & HBM is already presented and will be on the market within days. It beats anything nVidia has on the market and at a lower price too. nVidia will not have HBM products for a long time according to current plans.
* AMD has turned things around before, and can very well do so again.
* Your guesstimates of AMD's market share are based on the current status and extrapolates the future based on that, at the same time totally ignoring the bullets above and what AMD has in the pipeline.

 

[dark zero]

If AMD dies, VIA will leave x86 and the rest of the semicustom competence (Rokchip and Spreadtrum) might stop the implementation of the lame Atoms too... Leading that Intel takes the absolute monopoly of x86.

Also in GPU side AMD and Matrox dies together and definately Intel will nerf Nvidia hard despite the HBM. If Intel is smart enough they will buy off nVIDIA and implements ITANIUM again.

On the other side ARM and Open Power will recieve more boost since both are not as close as x86 and in Power case they are more powerful than the former.

In few words
X86 is the new Power Pc
ARM will be the new X86

Not trolling but seeing the situation, this might be the beginning off the enf of X86... Nothing is permanent. And x86 lived their time

 

[JDG1980]

In fact, by the looks of it I think Lisa will be trashed around Q3 or Q4 because of the failure in regaining share to Nvidia, not even semiaccurate believes it will happen.

That would be a foolish decision. As critical as I have been of the 300 series rebrands (and the failure to include HDMI 2.0 in Fury), this isn't Lisa Su's fault. The lead time on GPUs mean that these decisions were made before she ever stepped into the CEO's chair. The only real choice she had was to either put on a brave face and push forward with the existing lineup (which is what happened), or else cancel the rebrands entirely, launch only Fury, and just keep the 200 series until FinFET is ready. The latter decision might have saved some bad press, but it wouldn't have made much difference in the market share.

 

[Arachnotronic]

* AMD will be on 14 nm next year.

GloFo/Samsung "14nm" isn't likely to deliver the same kinds of performance/clock speeds that Intel's 14nm is likely to. Intel is very likely going to be on 10nm for its CPU+GPUs in late 2016/early 2017.

* Zen will be released next year, and is expected to reach about Haswell levels of IPC. It will also be 8-16 cores.

Haswell IPC I guess is nice, but at what clocks? 8 cores in 95W TDP doesn't suggest super high clocked cores.

* Since Intel has not improved desktop performance much since Haswell (in fact since SB), Zen will be competitive with Intel's top CPUs, and likely also priced lower providing more "bang for the buck".

Skylake should be a nice boost from Haswell, and it's coming this year. Also, Intel is the one with the cost structure edge here. With a 14nm process that will be quite mature by the time Summit Ridge rolls out, as well as a whole family of high core could Xeon processors that can drop into the LGA 2011 socket, it shouldn't be too hard for Intel to respond to any competitive threats from AMD here.

If Zen CPU w/ 8 cores @ 95W proves competitive, then what's to stop Intel from lowering prices on X99 and 6/8 core Broadwell-Es and then putting an unlocked 10 core+ Broadwell out as an extreme edition at the $999 price point?

You seem to think Intel has no options and/or is incompetent at understanding the competitive landscape, and I think that's probably not a wise assumption.

* Fiji & HBM is already presented and will be on the market within days. It beats anything nVidia has on the market and at a lower price too. nVidia will not have HBM products for a long time according to current plans.

Fiji should be nice, HBM is cool, but let's wait for third party benchmarks before claiming that AMD has gone ahead and "beaten" NVIDIA at anything.

* AMD has turned things around before, and can very well do so again.

It's up against much fiercer, better funded competition this time around than it was in the past. Turning things around only gets harder over time. Not saying it's impossible, but probably worth tempering enthusiasm.

* Your guesstimates of AMD's market share are based on the current status and extrapolates the future based on that, at the same time totally ignoring the bullets above and what AMD has in the pipeline.

You seem to be ignoring what the competition has in the pipeline.

 

[mrmt]

That would be a foolish decision. As critical as I have been of the 300 series rebrands (and the failure to include HDMI 2.0 in Fury), this isn't Lisa Su's fault. The lead time on GPUs mean that these decisions were made before she ever stepped into the CEO's chair. The only real choice she had was to either put on a brave face and push forward with the existing lineup (which is what happened), or else cancel the rebrands entirely, launch only Fury, and just keep the 200 series until FinFET is ready. The latter decision might have saved some bad press, but it wouldn't have made much difference in the market share.

I'm not saying that this is all Lisa Su's fault. Rory pretty much wrecked AMD R&D pipeline and we are just starting to see the results of his tenure. Basically he stopped all Derp designs and moved to Zen and K12, and hee seems to have stopped all GPU development too for something that is yet undisclosed.

But what she promised in the last Q&A is that AMD would regain share from Nvidia with their 300 series, and if their 300 series is nothing more that a rebranded 200 series with a higher price tag, how are they expecting to regain market share? Nobody forced her to promise anything, it was her own doing, and she will pay dearly in the next couple quarters.

 

[Sweepr]

The amount of FUD you are able to produce is astonishing as usual.

Observe this:

* Zen will be released next year, and is expected to reach about Haswell levels of IPC. It will also be 8-16 cores.
* Since Intel has not improved desktop performance much since Haswell (in fact since SB), Zen will be competitive with Intel's top CPUs, and likely also priced lower providing more "bang for the buck".

Just to make everything clear so we can laugh at your absurd expectations next year, are you expecting 8C/16T + Haswell IPC and high-clocks @ 95W TDP and below?
Reality will hit some people really hard as soon as the first performance numbers come out. They might finally understand that increasing ST performance @ legacy apps at equal or lower power is difficult by then (or whenever Zen refreshes with minor IPC bumps come out). Perhaps the number of arrogant forum users dissing the work of Intel engineers will be lower.

 

[Abwx]

GloFo/Samsung "14nm" isn't likely to deliver the same kinds of performance/clock speeds that Intel's 14nm is likely to.

Numbers are known, GF14nm lLPP (not Samsung s exactly..) has much better perf/watt at 2.41Ghz, the frequency for wich they published their tests datas, it should have no problem working at 3.5 , although not with the same perf/Watt difference.

 

[Arachnotronic]

Numbers are known, GF14nm lLPP (not Samsung s exactly..) has much better perf/watt at 2.41Ghz, the frequency for wich they published their tests datas, it should have no problem working at 3.5 , although not with the same perf/Watt difference.

Could you please provide a source? Thanks!

 

[DrMrLordX]

There can be no spin off because there s nothing that could be spun off, AMD need both the GPU and CPU IPs since their products integrate boths in the same silicon.

There s no info at all in this fud article, must originate from some short sellers that were caught off guard by the recent stock surge...

Well that, and also AMD hasn't actually done anything concrete. If you read the article, they've "thought about it" before, and they're "thinking about it" again. Doesn't mean they're gonna do it. At most, this planning seems to be preparation for a contingency in case Carrizo, Fiji, and/or Zen tank, bringing the company closer to bankruptcy.

Nobody seems to be thinking about whether Mubadala is going to like this kind of a change. Recently, AMD has been moving production of some products (notably graphics cards?) away from TSMC and towards GF, so the level of integration between AMD's CPU and GPU divisions is only increasing, as is GF's stranglehold over the company's future. I don't see Mubadala liking this kind of a move, especially since their primary investment came sort of as a bailout for AMD taking on all the debt that was necessary to acquire ATI in the first place.

 

[jhu]

Trying to escape debt? Jettison the competitive part(s) (small core + GPU) and allow the rest to implode?

It worked for GM.

 

[Fjodor2001]

Just to make everything clear so we can laugh at your absurd expectations next year, are you expecting 8C/16T + Haswell IPC and high-clocks @ 95W TDP and below?

What has been communicated by AMD so far for Zen is 8-16 cores, 40% above Excavator performance (i.e. Haswell level of IPC).

Do you have some better sources stating otherwise?

 

[JDG1980]

I'm not saying that this is all Lisa Su's fault. Rory pretty much wrecked AMD R&D pipeline and we are just starting to see the results of his tenure. Basically he stopped all Derp designs and moved to Zen and K12, and hee seems to have stopped all GPU development too for something that is yet undisclosed.

Flushing the Bulldozer R&D pipeline was the right move. That architecture is a failure and a dead end; every dollar spent on it is a dollar that can't be spent on Zen development, where AMD might have a chance of actually being competitive (or at least not 4-5 generations behind).

What happened with the GPUs was a massive miscalculation. I suspect AMD thought that 20nm would be viable sooner or later despite the delays, and were ready to go with slightly modified die-shrinks when that happened. What they failed to anticipate was that not only would 20nm be a complete bust for anything larger than cellphone/tablet SoCs, but that Nvidia had a backup plan including a whole new generation of GPUs on the 28nm process. Maxwell pretty obviously caught AMD flat-footed. I suspect that Tonga and Fiji were supposed to be 20nm and were hurriedly back-ported to 28HPM, and by the time this became apparent, no more 28nm GPUs could be made because they wouldn't be ready for market until 2016, when FinFET would be just around the corner.

But what she promised in the last Q&A is that AMD would regain share from Nvidia with their 300 series, and if their 300 series is nothing more that a rebranded 200 series with a higher price tag, how are they expecting to regain market share? Nobody forced her to promise anything, it was her own doing, and she will pay dearly in the next couple quarters.

The problem is that there was no way to give an honest answer without actually making the problem worse. What do you think would have happened if she had told the truth - "We're basically in a holding pattern for 2015, with just one major new GPU product and a bunch of refreshes; we expect to regain market share in 2016 when FinFET comes along." A quote like that would have made all the major tech news sites, and it likely would have been also included in the reviews of the rebrands when they came out. What little hope there was to increase sales would be completely gone.

 

[JDG1980]

If Zen CPU w/ 8 cores @ 95W proves competitive, then what's to stop Intel from lowering prices on X99 and 6/8 core Broadwell-Es and then putting an unlocked 10 core+ Broadwell out as an extreme edition at the $999 price point?

You seem to think Intel has no options and/or is incompetent at understanding the competitive landscape, and I think that's probably not a wise assumption.

Intel would rather lose some market share in desktops and servers than see their gross margins in those markets go below 60%. (Things are different in ultra-mobile, where Intel is willing to literally give their products away to gain market share.) I don't think Intel would respond with substantial price cuts until AMD's market share approaches 25% in the desktop and server markets, and maybe not even then.

That said, it does seem likely that with process improvements, Intel could bring 8-core further down the line on HEDT (say, the $599 price point instead of $999). That'll probably happen in another generation or two, maybe before Zen hits the market, maybe after.