ParisLemon » AMD Is Going 3 Cores: This Is the New Shaving Razor Arms Race

AMD Is Going 3 Cores: This Is the New Shaving Razor Arms Race

Does anyone else think the microprocessor race between Intel and AMD is turning into something very similar to the razor race between Schick and Gillette? AMD is now said to be moving towards releasing a 3-core processor, which would of course trump Intel’s dual-core processors. You just know Intel is going to announce a quad-core processor any day now (no, the Mac Pro Quad-Core machines don’t really count, that’s just 2 dual-core chips).

What I want to know is if having all of these cores will actually make sense? Is anyone going to be writing programs that utilize them or are they just going to go to waste? It’s just like how the razor companies jumped to 2-blades, the other jumped to 3-blades, so the other went to four blades, only to be taken down by 5-blades. Does anyone actually need a 5-blade razor? I mean the very thought of it and the whole blade “arms-race” was so ridiculous that The Onion did a (classic) parody piece on it and then a year later Gillette actually came out with exactly what The Onion invented just to mock, a 5-blade razor!

To me, all of this one-upsmanship is just a pure marketing ploy.

[photo by flickr user altemark]

Post by MG Siegler on 09/15/07
Comments

Anonymous

"no, the Mac Pro Quad-Core machines don't really count, that's just 2 dual-core chips."

We can argue about this until we're blue in the face. For instance, IBM's POWER6 is a dual-core design but can be packaged in a dual or quad chip module, like its predecessors. Would you say that an 8-core POWER6 system doesn't really count cause that's just four dual-core chips in a multichip module? I think that a native design is more advanced and more elegant, but both approaches have merits and this is just a design tradeoff between the manufacturing process, fabrication yields, costs, etc.

At the moment Intel is combining a pair of dual-core processors to produce a quad, it works as intended if your apps are multithreaded and can take advantage of the four cores. Intel was struggling against AMD's Opteron in the small but lucrative multi-processor server space, even Dell began selling AMD quad-core servers last year. Intel had to do something fast and chose the short time-to-market solution to regain lost market share. The company has shipped over 1 million quad-core Xeons since last November, the first Intel quad-core processor was a year ahead of AMD.

Intel also combined two single-core dies in a single package when it introduced its first dual-core processors, nowadays Intel is shipping native dual-core chips. There is clearly a tradeoff between time-to-market and a more advanced native design. It also depends on the manufacturing process technology of the time and the die size of the chip.

Going monolithic wouldn't be an option for IBM server chips, the die size of a dual-core POWER6 manufactured on a 65nm process is already 341mm2. A native quad-core POWER6 would be huuuge, very difficult to manufacture without defects and thus too expensive, even for servers. On the other hand, the monolithic approach is starting to make sense, albeit at relatively high cost (decisions, decisions…), for the Cell processor developed by IBM, Toshiba and Sony and shipping in the PS3. Cell is an asymmetric design, it has more cores but is actually way smaller than POWER6 (1 PPE plus 8 SPEs all in 235 mm2 on a 90nm process) and will be smaller and less expensive to produce on a 65nm process. It has eight little SPEs, but note that one is disabled to increase fabrication yields. (PowerPoint presentation about the POWER line and the Cell BE)

Sun's UltraSPARC T1 has up to eight cores on a single die but each core is really simple (no out-of-order execution, a single floating point unit shared between all 8 cores, CPU not capable of SMP), thus the die area is 379mm2 for 8 cores on 90nm, on the same process it would be smaller than the POWER6. The chip was running at 1.2 GHz at launch with a very short 6 stages pipeline, compared to 4.7 GHz for IBM's POWER6. Those processors are not optimized for the same workload and the tradeoff is completely different. (Sun presentation)

AMD's tri-core processor is basically a quad-core CPU with one of the cores disabled. Unlike Intel, the four cores of the AMD processor are integrated onto a single piece of silicon, and it's really easy to disable one of the cores, "just a fuse to blow" according to the Inq. As I said, a large quad-core processor on a single die is more difficult to manufacture. If the fourth core is defective, AMD can still sell this CPU as a three-core part with core #4 disabled. The tri-core processor will be less expensive than a quad-core CPU, this is a nice mid-range option for customers and it gives AMD more pricing flexibility.

Three cores will consume less power and produce less heat, the tri-core processor can be clocked a bit higher than the otherwise identical quad-core CPU, like 2.2 GHz instead of 2 GHz. Depending on his apps and workload, each customer will chose the best processor. For a scientist, a 2 GHz quad-core Phenom CPU may be more suited for the job at hand, and for the moment there are chances that a hard-core gamer will prefer a more affordable processor with two or three cores, because of the higher clock speed.

BTB

And now Gillette also has a Phenom?