AMD: What Went Wrong? 497
Barence writes "In 2006, AMD could seemingly do no wrong. Its processors were the fastest in the PC market, annual revenue was up a record 91%, expansion into the graphics game had begun with the high-profile acquisition of ATI, and it was making exciting plans for a future where it looked like it could 'smash Intel's chip monopoly' for good. Now the company is fighting for its very survival. How did AMD end up surrendering such a advantageous position – and was it given an unfair shove on the way down? This article has plotted AMD's decline, including the botched processor launches, the anti-competitive attacks from Intel and years of boardroom unrest."
It's not so much AMD failed (Score:4, Informative)
Re:It's not so much AMD failed (Score:5, Informative)
Intel just succeeded HUGE. They few years of AMD dominance were more a result of intels missteps.
Bingo. The P4 was a dead end, Intel were betting on Itanium for the 64-bit market, and AMD just kept on building better x86 chips.
Once Intel realised they were falling behind, they dropped their brain-dead policies and pushed out better chips than AMD's.
Forgetting Intel tactics? (Score:5, Informative)
Re:Hello? It's a Monopoly! (Score:0, Informative)
The article is pretty explicit about how AMD dug its own grave. I don't think blaming an Intel monopoly is all that convincing.
Re:Products (Score:3, Informative)
It's really simple--Intel made better products.
BS. Intel was fashionable, that's all. AMD has always been at least neck and neck with Intel if they weren't ahead of them, despite all of Intel's dirty tricks. I've been very satisfied with AMD ever since my 486DX3-100, and my Sempron and Turion based boxes just build upon that. Was Intel smart enough to buy ATI, or is it more familiar to associate them with nVidia? Need we mention Itanium? Who sold 64 bit CPUs first?
Re:Products (Score:5, Informative)
>The problem was AMD didn't bet hard enough on ramping up clock speeds.
Clock speeds have barely moved lately.
>Hertz to Hertz, AMD makes a better processor.
Not since the Core2, and even less competitive in the "i" era
Re:Products (Score:2, Informative)
If it were true.
Our testing of everything up to quadcores says that clock for clock, AMD made mincemeat out of Intel. In notebooks, where there can be much different chipsets, AMD lagged with their ATI chipsets. Intel with nVidia combos ate their lunch. Of course, there was that little problem with nVidia's chips cratering, but we can overlook that for now.
Re:Products (Score:5, Informative)
You're doing something wrong, either it's overclocked, the heatsink isn't seated properly or you have no thermal compound, or your case sucks pretty badly for airflow.
I have a high end i5, running on the stock heatsink, at stock speed, in a 10 year old case with crap airflow and can do renders quite happily with no issues whatsoever.
Response from the accused (Score:0, Informative)
I'm getting sick of responding to these. I am not any of the 14 accounts you claim I am. Your past trolling has already gotten me modded down so harshly that my karma is -1, so I've stopped regularly posting. Now you're getting modded up by other people for this shit? I'm taking advantage of the new flagging feature and reporting your post. This is organized corruption of the moderation system.
Re:Products (Score:5, Informative)
You should come out of 2002 sometime.
Intel Compilers still backstabbing AMD (Score:5, Informative)
The Intel compiler, widely regarded as the best compiler available for x86, still produces code designed to make Intel chips look better than any others.
http://www.agner.org/optimize/blog/read.php?i=49 [agner.org]
That page was posted three years ago. Scroll to the bottom, and read the latest additions to the discussion there: "New Intel compiler version - still the same!"
http://www.agner.org/optimize/blog/read.php?i=49#179 [agner.org]
This makes it difficult to be sure how much better Intel chips really are than AMD chips. When the Intel chip scores higher on a benchmark, and the benchmark includes Photoshop, was the Intel chip actually better or was Photoshop compiled with the Intel compiler?
Sadly, I think Intel chips really are better now; given that Intel is leapfrogging past AMD on process technology, they have major advantages so their chips ought to be better.
But I still buy AMD. Yeah, I'm giving up some increment of performance... but the chips these days are so fast, I can survive on only 90% performance or whatever. And I prefer to avoid doing business with a company that continues to sell a compiler that sabotages performance on competitor's chips.
Personally, I would love to see AMD sell a line of processors that return "GenuineIntel" for the CPU ID, and thus run Intel compiler code at full speed. When Intel sues them, they can argue that this is necessary for full compatibility with the code produced by Intel's own C compiler. (Yeah, I know. It will never happen. It's a fun daydream but that's all.)
Even if AMD doesn't have the top performing chips, they continue to score very well on price/performance, and the performance is good enough for me. And they are less evil than Intel. So I remain an AMD customer.
steveha
Re:Products (Score:4, Informative)
This is hugely incorrect. A Bulldozer module contains two, physically separate integer cores. It shares only the decoder, L2 cache, and FP coprocessor (which can handle loads from both processors simultaneously, unless they're AVX instructions, but it still executes them fast enough anyway). Hyper-threading uses the same core for two threads, sharing all of the execution resources - it's much less efficient in terms of performance per thread.
Why is AMD doing this? By separating out the FP unit, they can replace it with a GPU-based SIMD unit, allowing them to _vastly_ increase their FP processing ability. Heavy workloads are either already FP-based, or shifting to FP. Once they get this going, I'd expect to see a four-core module built around a variant of the Graphics Core Next compute unit - four sets of 512-bit wide SIMD units, allowing truly enormous FP throughput* with a large amount of integer capacity, while having all the benefits of a large, shared cache. The only problem with this design is the decoder throughput, and I'm certain that AMD is aware of this.
*I wouldn't be surprised to see an instruction set for 512-, 1024-, and 2048-bit wide SIMD instructions, to take full advantage of the FP performance with fewer integer threads. There are also out-of-order advantages to having so many FP ports available, even when dealing with narrower instructions. Or they might use a cut-down version - one 512 bit unit - instead, presumably with a two-core module.
Re:Intel Compilers still backstabbing AMD (Score:2, Informative)
The part about the legal fine print being put in a GIF file just to make it harder to discover through search engines is truly special.
http://www.agner.org/optimize/blog/read.php?i=49#184 [agner.org]
http://software.intel.com/sites/products/web2010/prod-images/opt-notice-en_080411.gif [intel.com]
Re:Not for everyone, wonderful for some (Score:5, Informative)
A large but trivial example...
How about a small but perfect example: running more than one program at once.
I'm not sure where this idea that multiple cores isn't useful came from, but it's dead wrong. Many individual programs might not be written to take advantage of parallel processing, but the OS will quite handily dispatch different programs to different cores.
Re:Products (Score:5, Informative)
It's not all about the heat. We're pretty close to the speed of light being an issue. Electrical signals travel slightly below light speed, so for a 3GHz chip, your signal could make it approximately 10cm, round trip, in one cycle. Which means we may be able to get to ten or twenty GHz inside chips but when it comes to memory access, that's going to then take tens of clock cycles. And that also assumes a 1cm square chip, all straight paths, and zero latency in the transistors. My guess is that even if you made a diamond substrate, you wouldn't be able to get it too much faster than existing chips.
Re:Products (Score:4, Informative)
Tell that to my two servers with two 8 core Magny Cours 6128s per machine. Linux+KVM+fast RAID on these machines equals lots of responsive virtual machines at a price point way below what Intel could deliver. Is virtualisation a niche market? Really?
Comment removed (Score:4, Informative)