Slashdot Mirror
AMD Withdraws From High-Density Server Business
An anonymous reader sends word that AMD has pulled out of the market for high-density servers. "AMD has pulled out of the market for high-density servers, reversing a strategy it embarked on three years ago with its acquisition of SeaMicro. AMD delivered the news Thursday as it announced financial results for the quarter. Its revenue slumped 26 percent from this time last year to $1.03 billion, and its net loss increased to $180 million, the company said. AMD paid $334 million to buy SeaMicro, which developed a new type of high-density server aimed at large-scale cloud and Internet service providers."
Looks like they're focusing on ARM chips:
"AMD still sees growth potential in the server market, but not from selling complete systems. It's returned its focus to x86 chips and to the development of its first ARM server processor, code-named Seattle."
8 core 64 bit ARM chips with GPU built in are fairly common and 10 core chips already announced (Mediatek), with 16-48 core vaguely hinted at for servers by other vendors. So if AMD plan on entering the ARM processor market they'd better get something special out and fast, and be prepared to stick at it and upgrade it and take the initial losses. Because they're unlikely to win companies over first time till they're confident AMD are in it for the long run and won't leave them hanging without a supplier.
On the other hand they could focus on x86 chips where Intel is already deep discounting at the low end, and likely will have to do that all the way up the range to compete.
AMD face a tough time either way.
Actually AMD defined the 64 bit extensions to the 32 bit x86 architecture, and Intel had to follow and is letting the Itanic sink.
Of course Intel does not even remotely admit it (and even Linus ranted on this fact in a mail a decade or so ago), but they have still not come over the NIH syndrome that it caused them.
This said, amd64 (as Debian calls it) could and should have been better designed.
The trouble is that Intel really needs some competition, this will take years, but it may come now that they put MBA at their head. Their lead on process technology won't last forever once we hit fundamental physical limits. Also they have a tendency to forget that they are where they are thanks to IBM: if IBM had chosen another architecture 34 years ago, Intel would be an also ran. Unfortunately the rumor is that IBM selected the 8086 crapitecture (segemented addressing) because it was so weird that it would never come to compete with their own high-end proprietary products.
AMD has played a losing strategy for as long as I have can remember. It is sad, but I remember my first few PCs were all AMD machines. I bought AMD on principle, and because they were price/performance leaders. They were even outright leaders for a while, but failed to capitalise on that. I think, however, that the whole Sledgehammer/Clawhammer phase has ultimately ruined them. Obviously, those processors were streets ahead of the Intel offerings at the time, but it was always a long term losing strategy, in particular if they were depending on selling CPUs to make money. Their obsession with OEM deals also hurt them.
AMD could have done one of a few things, in my opinion, to reinvent themselves.
- They could have become a whole-hog PC builder, using their own chips and pricing their laptops and desktops accordingly.
- When Android happened, AMD, without as much baggage as Intel, could have produced an Android phone and Android tablets, and gone to market with that, using their chip making expertise to develop offerings that would have been more competitive than Qualcomm, Samsung etc.
AMD was obsessed with being a mini Intel, which was never going to work out for them.
AMD should have taken a page out of Apple's playbook. At best, they might be taken over by a Chinese company, otherwise they are doomed to irrelevance.
Segmented addressing is actually a clever idea for 8-bit (8080/Z80) software compatibility, it's definitely easier that bank switching on 8-bit CPUs (2600 carts, mappers on NES carts, C64 and CoCo RAM/ROM swap, etc.) IBM probably designed the PC as a generic, but superior machine to run 8-bit CP/M software, and compete against Apple ][. Too bad they did not consider improving their '70s 5100 portable computer.
I am writing my own (multi-threaded) software and recently I had a chance to do a test run on an intel i7 processor (8-core, 2.67GHz) to compare it with my old Athlon II X4 (3GHz). Both programs compiled with the same version of GCC (4.6.1), both compiled with -O3 optimization. Running 8 threads on the Intel machine was only marginally faster than running 4 threads on the old Athlon. The threads were independent, so no threads were inactive while waiting for something else to finish.
Where Intel have the lead is in the compiler business. Back in 2003 or so they released their ICC 8.0 for free for Linux users. I was writing only single-threaded software at the time, and simply re-compiling it with ICC made it run about 5 times faster than the version compiled with GCC 2.96. And that was on a 2GHz Athlon XP.
What AMD have done right is the integration of the CPU and GPU allowing them to gobble up the console market. However, their bet that all developers will jump on the heterogeneous computing bandwagon did not pan out. But with HSA 1.0 coming up their lead will be too large and neither Nvidia not Intel will have a competitor ready for the next console refresh. All that Nvidia will do is to continue to pay game developers to optimize their engines for GeForce cards, and refuse to optimize for Radeon. AMD's resources are so limited that they will be forced to have a desktop version of their console processor, and maybe an ARM core for good measure.
Exiting the "dense server" are makes perfect sense, as the market is very limited. Running across many small cores is hard and developers will avoid it. It is the same story as taking advantage of the GPU, which also provides many simple cores.
So no, they are not dead, they are simply adapting to market realities and accept that they made a mistake when they jumped in the dense server bandwagon. Unlike Intel, who even now refuse to let go of the Itanium.