AMD mulling break, spinoff

[dark zero]

Admit it guys... AMD finally died, but with that the traditional x86 market is on danger... and then just imagine a bunch of ET incomming shortly after that... it will kill the PC market, with nVIDIA in it. Only Intel will be left on that market. And MS will see that only Intel on that is not good...

That's why MS is moving to ARM, like Apple too.

 

[VirtualLarry]

Admit it guys... AMD finally died, but with that the traditional x86 market is on danger... and then just imagine a bunch of ET incomming shortly after that... it will kill the PC market, with nVIDIA in it. Only Intel will be left on that market. And MS will see that only Intel on that is not good...

That's why MS is moving to ARM, like Apple too.

I wouldn't count them out yet... but is the problem AMD, or the x86 market in general, being swamped by a sea of "mobile", powered by ARM? Is Intel going to succumb to what is ailing AMD?

 

[dark zero]

I wouldn't count them out yet... but is the problem AMD, or the x86 market in general, being swamped by a sea of "mobile", powered by ARM? Is Intel going to succumb to what is ailing AMD?

Intel won't die, but definately will see that brute force is not anything and they will commit the mistake to shrink the x86 cores.

Unless... Intel is trying to bring bigger changes and finally we might see the 128 bit instructions.

 

[VirtualLarry]

Intel won't die, but definately will see that brute force is not anything and they will commit the mistake to shrink the x86 cores.

Unless... Intel is trying to bring bigger changes and finally we might see the 128 bit instructions.

Huh? SKL-E/EP have 512-bit instructions, AVX512.

 

[DrMrLordX]

While that is technically true, I think dark zero was implying that they might be looking into a shift from a 64-bit to a 128-bit ISA. I don't forsee that anytime soon.

 

[ShintaiDK]

128bit ISA is pointless. All the performance benefits for even 256bit today is already implemented (And soon 512bit). So unless we reach 64bit memory addressing limit...

 

[ShintaiDK]

With that mentioned, I would like AMD to do the one thing Intel won't do:

Gives us the CPU cores instead of throwing in the extra iGPU at the low end.

4C/8T Zen with small iGPU is going to be more useful and versatile compared to 2C/4T Zen with medium size iGPU. (This especially considering AMD also sells dGPUs)

You cant be serious. AMD tried that route for how long now? And its been nothing but a complete disaster.

 

[Tsavo]

You cant be serious. AMD tried that route for how long now? And its been nothing but a complete disaster.

It wasn't the number of cores that was the problem, and you know that.

 

[Shehriazad]

http://hexus.net/business/news/corporate/84161-amd-company-split-rumours-officially-denied/

AMD valiantly denying it.

 

[ShintaiDK]

Nomatter if its true or not. The company has to deny it.

 

[Shehriazad]

If the sources had any viability they would step forward and actually deliver something worth our time.

And of course they would have LOOKED at the strategy...new leader in place...take a look at all options...and then shelf the stuff that obviously wouldn't work...like breaking up into multiple pieces...which CANNOT work since ALL business parts would suffer from it.

The very claim itself just seems ridiculous to me. Not to mention that they want to develop ONE CPU arch (Zen base) for ALL markets...how can that work if they all split off? Right...




Inb4 some anonymous Samsung spokesperson just said Samsung is gonna sell itself off to Intel.

 

[Fjodor2001]

To me it seems AMD wants to compete with Intel pretty much across the board.

Desktop PC chips? Yes.

x86 server processors for enterprise, networking, etc.? Yep.

Notebook PC chips? Yep.

Embedded? Yep.

ARM servers? AMD exclusive.

High performance computing? Yes.

AMD's FAD strategy seemed to be "we're going to play in all of the markets we played in before and go after some that we neglected, too!"

So when I see AMD spending ~$1B/yr on R&D trying to play in the same markets as a company that spends ~$11B+, it's really hard to shake off the feeling that AMD might be trying to do too much with too little.

You make it sound like those are completely different R&D areas. But in reality most of them could be bundled up into one. Server, desktop, laptop etc are more or less just different variants of the same building blocks. Once you have designed an iGPU, x86 core, and related on chip logic such as memory controller etc it does not take that much effort to create different SKUs for server, desktop and laptop markets.

And as for AMD's ARM CPUs, AMD already got the relevant ARM CPU core IP-blocks, so that does not take much effort either compared to designing a CPU from scratch.

Now compare this to all the areas that Intel spends its total R&D budget on:

Cable Modems

Products by technology: cable modems, set-top-box (STB), and video gateway solutions
Cable modems, STB, and video gateway solutions: DOCSIS* 2.0 data modem solutions
DOCSIS* 2.0 data modem solutions TNETC430436
DOCSIS* 2.0 data modem solutions TNETC46004602
DOCSIS* 2.0 voice and data gateway modem solutions TNETC540

Chipsets

Desktop
Laptop
Server
Workstation
Embedded
Consumer electronics

Consumer electronics

Media processing components
Demodulators and tuners
Widget Channel
Smart TV

Desktop

Intel® Graphics Technology
Intel® vPro™ technology
Processors
Boards
Chipsets
Adapters
Solid-State Drives (SSD)

Embedded and communications

Processors
Chipsets
Wireless adapters
Intel® Desktop Adapters
Gigabit server adapters
10 Gigabit server adapters
Intel® Ethernet controllers
Microcontrollers

Ethernet

Intel® Ethernet
Intel Ethernet Controllers
Gigabit Ethernet
10 Gigabit Ethernet

Health

Healthcare technologies and products

Internet Devices

Processors
Chipsets

Laptop

Intel® Graphics Technology
Intel® Centrino® processor technology
Intel® Centrino® with vPro™ technology
Processors
Chipsets
Wireless adapters
Internet Devices
Solid-State Drives (SSD)

Motherboards

Intel® Desktop boards
Server boards
Workstation boards

Processors

Desktop
Laptop
Server
Workstation
Internet Devices
Embedded and communications

Server

Processors
Chipsets
Systems
Boards
Adapters
Chassis
RAID controllers
Intel® Storage Systems
Solid-State Drives (SSD)

Software products

Intel® Software Products
Intel® Expressway Service Gateway

Storage and I/O

SAS controllers
Solid-State Drives (SSD)

Technical books

Programming
Computer system design
Network infrastructure design
Strategic technologies
IT best practices

Wireless products

Intel® Wi-Fi Products
Business-class Wi-Fi
Why Intel® Wireless
Intel® Mobile Communications Products
Intel® Wireless Display

Workstation

Processors
Chipsets
Boards
Solid-State Drives (SSD)

Heck, just Intel's process tech R&D and capex alone is likely much larger than their x86 uArch R&D budget. And AMD spends $0 on process tech R&D, since they do not have their own fabs and buy their chip wafers from foundries instead.

So to compare AMD's vs Intel's total R&D budget is completely irrelevant. You need to compare them area by area. E.g. how much R&D does Intel vs AMD spend on x86 uArch design.

 

[JDG1980]

Killing the CMT chips? Yes, it was the right move. But killing the cat family? that's dumb IMO.

Agreed. It's a shame to see the cat cores go, because they were a solid design. It would have made more sense for AMD to immediately kill the ARM-related projects. Unless AMD has a specific large-volume client (Nintendo?) that wants a custom ARM APU, it's hard to see how they make back their R&D budget on this stuff. Who is going to buy AMD ARM parts off-the-shelf? We've already seen Project Skybridge officially cancelled, and K12 pushed back to 2017. I think the writing is on the wall, and I would be surprised if K12 ever makes it to market.

AMD history is full of miscalculations like that. Funny that these "miscalculations" don't affect other companies as much as it does with AMD.

AMD is a smaller company with less room for error. NetBurst was a technical blunder roughly equivalent to Bulldozer (and for many of the same reasons), but Intel was able to muscle through by a combination of a much larger R&D budget (which allowed the old Pentium Pro design to continue development in parallel) and unethical marketing behaviors (which allowed the Pentium 4 to sell far more than it should have). Likewise, while AMD's recent 300-series rebrandings are a definite low point, Nvidia has had serious rough spots in the past as well - the first generation of Fermi was badly delayed (>6 months to do a silicon layer respin, at massive costs) and when it did come out, yields were so low that they couldn't release a fully-enabled part, and the power usage and heat output was way out of control. It took another full respin (not just a new stepping) to get Fermi working the way it was supposed to.

She said that about the CPU business in 2015 and the sky didn't fall. Today everyone wrote off their CPU stack but everyone already had anyway, so things didn't change too much from a commercial POV, but at least management gave a believable message in years, and they just throw it away by hyping the GPU business and underdelivering with the rebranded 300 series.

AMD isn't writing off the entire GPU business for 2015; the Fury cards are looking to be reasonably competitive (though we don't know the full story yet), and are expected to rack up some sales on the high end. It's the 300 series rebrands that are basically write-offs; they can't really expect sales of these to be anything more than minimal.

 

[JDG1980]

AMD also communicated that Bulldozer wouldn't suffer IPC loss.

Their 40% claim is shady as hell in my opinion.

"Communicated" means two very different things in these two cases.

With Bulldozer, you had an AMD marketing guy saying false stuff on forums. As you pointed out yourself in the past, the official slides for Bulldozer never said that IPC would remain constant; instead it implied otherwise, with talk of "knee-of-the-curve IPC" and talk of single-threaded performance "without significant loss" (implying there would be some minor loss).

With Zen, we've got official slides on the Financial Analyst Day touting improved IPC as a major feature of the new architecture. Of course we have to account for best-case scenarios; no one would be surprised if the actual IPC improvement in real-world applications turned out to be 35% instead of 40%. But AMD can't just issue blatant public lies to stockholders in this fashion without consequences. If Zen flops and its IPC improvements are significantly off predictions, then a lot of questions are going to be asked - and not just by the technical press, but by large investors and maybe even SEC regulators.

 

[mrmt]

They clearly couldn't afford the 3 billion loss from 2008 and were forced to spin off their fab just to stay alive. Added to all the debt created from the overpriced ATi acquisition.... They are a company that has proven to be cat-like with 9 lives.

I'm not sure what you are trying here. It's pretty clear that AMD balance sheet was big enough to withstand the 3 billion dollars write down and still survive, as the company had both an open debt market and assets to shed.

AMD today cannot get another 3 billion write down because it doesn't even have assets worth 3 billion to write down, AMD is basically a much smaller, leaner company than it was in 2008, so another 3 billion write down is an impossibility, and not because the company is healthier than in the past.

 

[ShintaiDK]

"Communicated" means two very different things in these two cases.

With Bulldozer, you had an AMD marketing guy saying false stuff on forums. As you pointed out yourself in the past, the official slides for Bulldozer never said that IPC would remain constant; instead it implied otherwise, with talk of "knee-of-the-curve IPC" and talk of single-threaded performance "without significant loss" (implying there would be some minor loss).

With Zen, we've got official slides on the Financial Analyst Day touting improved IPC as a major feature of the new architecture. Of course we have to account for best-case scenarios; no one would be surprised if the actual IPC improvement in real-world applications turned out to be 35% instead of 40%. But AMD can't just issue blatant public lies to stockholders in this fashion without consequences. If Zen flops and its IPC improvements are significantly off predictions, then a lot of questions are going to be asked - and not just by the technical press, but by large investors and maybe even SEC regulators.

Even in that case, remember Phenom and 50% faster than Core 2? Something AMD had pubished and then removed again on their site.

AMD got away with lying many times before. Even insider trading. And it doesnt look like Lisa Su will last long in the chair. She is already having her fair share of the burden. Remember the lawsuits from the stockhodlers because she couldnt even follow AMDs own rules on stock options awards?

 

[frozentundra123456]

"Communicated" means two very different things in these two cases.

With Bulldozer, you had an AMD marketing guy saying false stuff on forums. As you pointed out yourself in the past, the official slides for Bulldozer never said that IPC would remain constant; instead it implied otherwise, with talk of "knee-of-the-curve IPC" and talk of single-threaded performance "without significant loss" (implying there would be some minor loss).

With Zen, we've got official slides on the Financial Analyst Day touting improved IPC as a major feature of the new architecture. Of course we have to account for best-case scenarios; no one would be surprised if the actual IPC improvement in real-world applications turned out to be 35% instead of 40%. But AMD can't just issue blatant public lies to stockholders in this fashion without consequences. If Zen flops and its IPC improvements are significantly off predictions, then a lot of questions are going to be asked - and not just by the technical press, but by large investors and maybe even SEC regulators.

They carefully hedged those 40% IPC predictions by stating it was based on theoretical architecture analysis, or some such. So IMO, that is pretty much a meaningless figure, or at least absolute best case scenario. Not to mention they did not say what the clockspeeds will be. Personally, I think they will be lucky to reach Sandy Bridge single thread performance (clockspeed plus IPC). That still will make Zen much more competitive, but without an IGP initially, it will have very limited appeal in the consumer market, and have to compete with very efficient Intel servers with a well established support system.

As for the spin-off, I just see it as due diligence covering all possibilities, and seriously doubt it will happen. The only way is if they were somehow able to spin off ATI for some operating cash yet still keep access to the gpu IP for consoles and APUs.

 

[ShintaiDK]

Agreed. It's a shame to see the cat cores go, because they were a solid design. It would have made more sense for AMD to immediately kill the ARM-related projects. Unless AMD has a specific large-volume client (Nintendo?) that wants a custom ARM APU, it's hard to see how they make back their R&D budget on this stuff. Who is going to buy AMD ARM parts off-the-shelf? We've already seen Project Skybridge officially cancelled, and K12 pushed back to 2017. I think the writing is on the wall, and I would be surprised if K12 ever makes it to market.

The K12 project is one of the most mindbogling projects they ever made. And hopefully they cancel is sooner than later. I cant see any customers for it, and even worse. I cant see it compete with A72 etc. It seems yet again that someone had a meeting and thought something would be a good idea without doing the research.

Their current projects cant even repay themselves anymore.

 

[mrmt]

You make it sound like those are completely different R&D areas. But in reality most of them could be bundled up into one. Server, desktop, laptop etc are more or less just different variants of the same building blocks. Once you have designed an iGPU, x86 core, and related on chip logic such as memory controller etc it does not take that much effort to create different SKUs for server, desktop and laptop markets.

Servers aren't. The interconnect is far heavier than anything you'll find in the desktop, so is the memory controller, plus the validation process is a different matter, entirely.

 

[DrMrLordX]

128bit ISA is pointless. All the performance benefits for even 256bit today is already implemented (And soon 512bit). So unless we reach 64bit memory addressing limit...

I don't think we'll reach the 64-bit addressing limit anytime soon. It's good for . . . what, ~18 exabytes?

 

[dark zero]

I don't think we'll reach the 64-bit addressing limit anytime soon. It's good for . . . what, ~18 exabytes?

With the new material used for processors (graphene), 64 bits will be history shorter than expected.

 

[DrMrLordX]

. . . right.

 

[mrmt]

AMD isn't writing off the entire GPU business for 2015; the Fury cards are looking to be reasonably competitive (though we don't know the full story yet), and are expected to rack up some sales on the high end. It's the 300 series rebrands that are basically write-offs; they can't really expect sales of these to be anything more than minimal.

Fury won't change the picture on the notebook market, which is where Nvidia is mopping the floor with AMD GPUs. Basically three years being spanked on that market wasn't enough for AMD to change its course, and on top of that Fury won't come for the high volumes SKUs. Basically Fury is a lame duck for all that matters, it will be incapable of challenge Nvidia domination on the GPU market on the brackets that matter.

 

[Fjodor2001]

Servers aren't. The interconnect is far heavier than anything you'll find in the desktop, so is the memory controller, plus the validation process is a different matter, entirely.

Sure, but that's still very minor compared to the cost of developing all the other building blocks, i.e. x86 CPU core, iGPU, memory controller, and all other logic on the CPU die (i.e. the stuff you already have paid R&D for if you have designed a desktop APU anyway).

 

[cbn]

You cant be serious. AMD tried that route for how long now? And its been nothing but a complete disaster.

If AMD gets to Sandy Bridge level of IPC, I think they have a chance of being competitive.

And with regard to core count, remember AMD also has to compete with older Intel processors like the i7-2600 on desktop.

P.S. Also remember Vishera (AMD's current 8 thread processor) is a big expensive die without iGPU. Whereas a 4C/8T Zen with small iGPU would be much smaller and have the versatility to be used in a wider range of devices.