A couple of weeks back, my co-writer, kram, wrote a column entitled “It’s AMD’s Spotlight.” In it he noted AMD’s recent dominance in terms of benchmarks, heat dissipation and general popularity, and said that AM2 is just one more victory for them. AMD has been in a great position for the past year or two, but that’s about to change with the release of Intel’s new Core 2 Duo line, code-named Conroe on the desktop front and Merom on the laptop front.
I have to say that I expected more of AM2. It was pretty heavily hyped, but it turned out when all the dust had settled and the benchmarks were out that it actually had inferior performance to Socket 939 unless the very expensive DDR2-800 was used. This was a bit of a disappointment to me, as DDR400 has been climbing in price compared to DDR2 recently and I was looking forward to a good alternative. But AM2 is, in my opinion, a bust for AMD. Granted, it’s not really an inferior product in the sense that the AMD K6 line was inferior to the Pentium II and Pentium III, but I think it will turn out to be the next Athlon XP – playing second fiddle to Intel’s chips.
Why do I say this? It’s not like the Pentium 4 and Pentium D are particularly impressive. In fact, Intel has given up on being competitive with them on performance and is now selling them at a loss to maintain sales supremacy. But in a month or two Conroe and Merom are going to hit the market – and they’ll take the computer world by storm.
Tom’s Hardware Guide recently had a chance to benchmark a new Core 2 Duo 2.66 GHz at Intel’s facilities. They pitted it against the 2.8 GHz AMD Athlon 64 FX-62 in a similar setup to the Intel system. The FX-62 is AMD’s top-end chip and costs over $1000; the Core 2 Duo 2.66 will be the top of Intel’s regular range when it comes out, but at $530 will be much less than the FX-62. Furthermore, there will be a Core 2 Extreme running at a heady 2.93 GHz on a 1066 MHz (4×266) front-side bus. Still, even when overclocked, the FX-62 was bested by the Core 2 Duo in 13 out of 18 benchmarks Tom’s ran. When running at non-overclocked speeds, the FX-62 was eating Conroe’s dust in 16 out of 18 benchmarks, although a few were fairly close. Yes, this was far from a controlled laboratory experiment because the Intel systems could have been running any number of tweaks that the AMD systems weren’t, but the results are impressive nonetheless considering that the 2.66 GHz Core 2 Duo is not going to be the top-end model.
Why is Conroe so impressive? Well, ironically it’s using old technology; it replaces Netburst, which was introduced in 2001, with the P6 architecture, which dates back to the Pentium Pro of 1996. The architecture has received a number of updates since then, with an increased pipeline length, multiple die shrinks, and the Netburst-style quad-pumped bus, but it’s still basically the same. However, the P6 architecture has consistently demonstrated a high level of efficiency compared to Netburst. The 1.4 GHz Tualatin PIIIs were the performance kings compared to the inefficient Williamette Pentium 4s up close to until the P4s broke the 2 GHz mark, despite using a much slower system bus/RAM combination. Furthermore, when the Banias Pentium-M (arguably the genesis of Conroe in terms of the architecture updates that went into it) came onto the scene, a 1.8 GHz model was very close in performance to a 2.4 GHz P4-M.
Conroe further improves the performance equation by continuing to use a feature introduced with the Yonah chips (the original Core Duos about which I wrote a column several months back) – shared L2 cache. In Conroe’s case, the cache is a whopping 4 MB, and can be accessed by both cores (the Pentium D and Athlon 64 x2 both had separate L2 cache for each core and thus had to access data cross-cache across the system bus). This eliminates the benefit of the Athlon 64 X2 chips over the Pentium-D, namely that the X2s could use the faster HyperTransport link instead of the front-side bus that Intel dual-cores used to communicate between cores. The combination of the efficient P6 architecture and the advent of shared L2 cache makes for a killer chip.
However, not all is well in the world of Intel. There is talk that Conroe will be unavailable in the retail channel for a while because the OEMs will have scooped up most of the first run for use in their systems. This may put somewhat of a damper on hardware geeks’ enthusiasm for Conroe and cause them to build AM2 systems instead of waiting. But overall, the Core 2 Duo seems like a winner.
