I wonder if this has anything to do with the very cheap Sandy Bridge line just launched by Intel. $300 for a chip that competes with $1200 chips is pretty tough to beat.
I assume you're referring to the ultra low voltage sandy bridge parts parts. AMD's Ontario and Llano, coming out later this year, will be very competitive with these parts. Not to mention cheaper to produce.
> A source familiar with the matter said the board viewed Meyer as somebody who stabilized AMD and helped turn it around, but wanted to look elsewhere for an executive to accelerate the company's growth.
These really are two different skills. A CEO that can turn a company around may not have the right skillset to put it into a sustained growth.
AMD's myopic view of only competing against Intel is also another problem. There's no real reason they also shouldn't be fabbing PowerPCs and ARM chips in bulk. At least in order to extend the life of older fab facilities.
> There's no real reason they also shouldn't be fabbing PowerPCs and ARM chips in bulk. At least in order to extend the life of older fab facilities.
I believe the reason is money. Margins on x86 products are _way_ higher than those on ARM products. AMD doesn't want to get off the x86 gravy train and invest on a product with many more competitors and much lower margins.
At this point, AMD doesn't own any fabs. They spun that part of the business off as Global Foundries, which then bought Chartered for good measure and is now a close partner with ARM: They've even abandoned the tradition of making a full wafer of SRAM to test and demonstrate new processes in favor of making test wafers of ARM SoCs. (Apparently SRAM cells don't do a very good job of exposing the kinds of problems that will crop up when trying to make CPUs.)
Indeed, SRAM layouts are very regular (it's a lot of the same), so the lithography process is easier to get right. This is also why new, less-nm processes will first be used to produce Flash/RAM and only later for complex CPU logic.
It's a shame. AMD always just seems to be a little behind the next big thing. With the exception of the K8 they've just been in perpetual catch-up mode. They finally have a pretty good mobile x86 platform when the rest of the industry seems to have decided ultra-low-power ARM is the way to go. They developed a pretty good high end server platform then the industry realized virtualization can save a ton of money on hardware. Poor AMD.
K7. I remember when motherboard makers were shipping wares in unmarked cardboard boxes like you got porn from behind the counter. (In a brown paper baggie). AMD absolutely throttled Intel in 1999 with the K7 and later Athlons.
"I remember when motherboard makers were shipping wares in unmarked cardboard boxes like you got porn from behind the counter. "
Ah, the beginning of Intel's antitrust troubles.
Those chips ran hot. I remember 40 degree celcius (104 F) summers with no AC, when I would compile the 2.4 version of linux kernel on my PC with its case removed. Those heat conducting pastes solved the problem eventually.
I once watched a brand new K7 burn before my eyes. I had just installed it and wanted to power it up briefly before securing the heat sink and cooler. I had done this many times with Intel chips. Well, the AMD took exactly two seconds to go up in smoke and this broke college student was out a couple of hundred dollars.
No offense, but that's about the worst thing you can do to a modern processor. The only reason the Intel didn't go up in smoke was because of proper on-die thermal protection. You still should never boot a modern processor without the heat sink in place. The temperature will almost instantly exceed the thermal design limits of the materials. AMD's only poor decision was not to provide good thermal protection in their CPUs. Running the processor without a heat sink was your own poor decision.
All chips of that generation ran hot. The clock speed arms race assured it. The AMD chips of similar generations generally ran cooler than the Pentium chips because AMD was squeezing more power out of fewer clock cycles, and therefore lower clock rates/power/heat dissipation/etc.
Let's not forget the Thunderbird, a fantastic AMD success. Also, AMD is not going out of business... might be a little too soon for you to start delivering the eulogy.
Maybe they need someone more aggressive to push the company forward again instead of trying to play catch-up.
> They developed a pretty good high end server platform then the industry realized virtualization can save a ton of money on hardware. Poor AMD.
There is still need for actual hardware on which to run all those VMs. One cannot just dump a 1000 loaded VMs on a quad core Xeon server and expect them to work as if there are 1000 real dual core machine. Sometimes VMs save on hardware but sometime just the cost of centralizing and simplifying the management of various legacy OSes and applications in a smaller (but not necessarily cheaper hardware-wise) data-center is a good strategy. So I imagine there will still be a good market for servers.
Yes they have to compete with ARMs on other end of the spectrum and that's a problem, and I am still not sure if buying ATI was that great of an investment.
Assuming virtualization makes physical servers more efficient, it should actually increase the demand for physical servers according to Jevons paradox: http://en.wikipedia.org/wiki/Jevons_paradox
Exactly, we just spent about $10K on a bunch of 6 core i7 machines to serve as VM hosts. Once we had the machines we quickly found enough VMs to fill them up -- VMs we didn't plan on having before getting all this hardware. And that was a clearly a couple of thousand dollars in Intel's pocket.
While in recent times this might be true, I seem to recall AMD also beat Intel to 1Ghz, and I was not paying that much attention at the time, but I think they came with multi-core cpus around the same time if not a bit earlier than Intel.
Fusion also seems to be slightly ahead of the curve, but I follow all this stuff less and less so I'm not that familiar with the details of how exactly it compares to SandiBridge.
This is for kingkilr, I accidentally downvoted you. Damn, my sloppy clumsy fingers. Really sorry. Could anyone upvote him to compensate please. And erm could pg increase the gap between the up and down arrows.
Was also going to comment that AMD beat Intel to the 64bit instraction set and it was Intel who had to play ctachup.
>I think they came with multi-core cpus around the same time if not a bit earlier than Intel.
My recollection is that Intel made it to the consumer market first, but did so in a kind of hacky manner. The original Pentium D was actually two dies on a single package, whereas the Athlon X2 was a multicore package with coherent cache from the start.
I believe AMD did beat Intel to the server market, but this is all from memory, so I can't say for certain.
For a long time AMD were the "indie" manufacturer in the microprocessor world - and prior to about '06 had vastly superior chips. I hung out on a forum dedicated to hardware at about that time, and the paradigm shift that Intel's Core 2 chip caused was insane.
Within literally weeks AMD was old news and Intel was king (this was after a couple of years of outright dislike of Intel as the "big corporate").
AMD beat Intel to 1GHz, X64(proper) and Multi-core(proper). But Intel produced a brilliant and quality line of chips in Core 2, which has left AMD basically floundering for nearly 4 years now.
I think that's fine. In a few years, with luck, AMD will produce a killer chip that leaves Intel scrabbling. Such competition is good :)
Do you (or the people voting you up) realize that the K8 processor mentioned was exactly the first AMD64 processor? Your comment seems quite redundant.
> AMD always just seems to be a little behind the next big thing.
Not really. They were the first to vectorize - with 3DNow! They were the first x86 with the on-chip memory controllers on the Opteron, which Intel were able to copy only 5 odd years later with Nehalem. They were the first to 1 GHz. They were the first (by a small margin, but still first) to build a CPU+GPU chip. They're also going to be first to build an x86 with an FMAC instruction with Bulldozer. These are just a few things off the top of my head, I'm pretty sure I could come up with a few more given some time.
The real problems AMD has are (1) Intel has way too much money because of which they can afford to screw up royally like they did with the P4 and now Larrabee and still not suffer any serious consequences and (2) Intel has always been ahead of them in the technology scaling game.
Good list of "firsts", even though you forgot some: first to implement x86-64/AMD64 which Intel copied 2-3 years later, first to implement inter-processor links (HyperTransport), first to ship 12+-core processors (Opteron 6100 series).
As an underdog, a company has no choice but to be the first to implement many features in order to even remotely hope to stay alive. Even then, it's often not sufficient because there may be a lot of inertia, so a small technical advantage may not be enough to convince customers from switching from the market leader to the underdog.
Also, correct me if I'm wrong, but AMD's GPUs have been almost neck-and-neck with NVIDIA's for quite a number of years, since before they were AMD's --- although in the last couple of years AMD's have fallen behind quite a bit. I have the impression that Intel's GPUs are dramatically inferior.
With the Radeon 5000 family of GPUs AMD really did well for themselves. Nvidia didn't have an answer for that until they delivered Fermi 6 months later. And what you say about Intel's GPUs is true, they are pretty much useless for any sort of advanced 3d graphics.
While NVidia has had the most powerful single chip for the last while, AMD has had the most powerful board and until the last half year or so generally beat NVidia on performance/price to. Before the 4000 series NVidia had the lead, and before that I hadn't been paying attention.
One area NVidia has held the lead in for the last few years in GPGPU, but thats a pretty small market.
37 comments
[ 3.0 ms ] story [ 112 ms ] threadThese really are two different skills. A CEO that can turn a company around may not have the right skillset to put it into a sustained growth.
AMD's myopic view of only competing against Intel is also another problem. There's no real reason they also shouldn't be fabbing PowerPCs and ARM chips in bulk. At least in order to extend the life of older fab facilities.
I believe the reason is money. Margins on x86 products are _way_ higher than those on ARM products. AMD doesn't want to get off the x86 gravy train and invest on a product with many more competitors and much lower margins.
Maybe they need someone more aggressive to push the company forward again instead of trying to play catch-up.
> They developed a pretty good high end server platform then the industry realized virtualization can save a ton of money on hardware. Poor AMD.
There is still need for actual hardware on which to run all those VMs. One cannot just dump a 1000 loaded VMs on a quad core Xeon server and expect them to work as if there are 1000 real dual core machine. Sometimes VMs save on hardware but sometime just the cost of centralizing and simplifying the management of various legacy OSes and applications in a smaller (but not necessarily cheaper hardware-wise) data-center is a good strategy. So I imagine there will still be a good market for servers.
Yes they have to compete with ARMs on other end of the spectrum and that's a problem, and I am still not sure if buying ATI was that great of an investment.
Fusion also seems to be slightly ahead of the curve, but I follow all this stuff less and less so I'm not that familiar with the details of how exactly it compares to SandiBridge.
Was also going to comment that AMD beat Intel to the 64bit instraction set and it was Intel who had to play ctachup.
My recollection is that Intel made it to the consumer market first, but did so in a kind of hacky manner. The original Pentium D was actually two dies on a single package, whereas the Athlon X2 was a multicore package with coherent cache from the start.
I believe AMD did beat Intel to the server market, but this is all from memory, so I can't say for certain.
Within literally weeks AMD was old news and Intel was king (this was after a couple of years of outright dislike of Intel as the "big corporate").
AMD beat Intel to 1GHz, X64(proper) and Multi-core(proper). But Intel produced a brilliant and quality line of chips in Core 2, which has left AMD basically floundering for nearly 4 years now.
I think that's fine. In a few years, with luck, AMD will produce a killer chip that leaves Intel scrabbling. Such competition is good :)
Not really. They were the first to vectorize - with 3DNow! They were the first x86 with the on-chip memory controllers on the Opteron, which Intel were able to copy only 5 odd years later with Nehalem. They were the first to 1 GHz. They were the first (by a small margin, but still first) to build a CPU+GPU chip. They're also going to be first to build an x86 with an FMAC instruction with Bulldozer. These are just a few things off the top of my head, I'm pretty sure I could come up with a few more given some time.
The real problems AMD has are (1) Intel has way too much money because of which they can afford to screw up royally like they did with the P4 and now Larrabee and still not suffer any serious consequences and (2) Intel has always been ahead of them in the technology scaling game.
The other good thing about 3dNow was that it was the first attempt at cleaning up Intel's x87 mess.
As an underdog, a company has no choice but to be the first to implement many features in order to even remotely hope to stay alive. Even then, it's often not sufficient because there may be a lot of inertia, so a small technical advantage may not be enough to convince customers from switching from the market leader to the underdog.
One area NVidia has held the lead in for the last few years in GPGPU, but thats a pretty small market.