Let's look at the players in some detail. On the one hand we have Intel with the Pentium line of indeterminable numbering, and AMD with the line of now meaningless numbering. The year looks to start on a low in the nomenclature category, and get lower as it progresses. Whoopee! Either way, we are going to start out the year with a Pentium 4 3.46GHz/1066/4MB on the Intel side, and an Athlon X2 at 2.6GHz.
In late Q1 or early Q2, we will see AMD update its line to the 90nm F-Step (FS) cores. These will bring two major things to the desktop, DDR2 and lower power consumption. It does add a lot more besides, a new socket, faster HT, improved cores, and other more server oriented features, but the two listed are the most important for the desktop. We'll leave server parts for a different day.
This means that the faster more power efficient desktop AMD parts will increase their lead, probably about five per cent for the DDR2, and a clock bump from Q1's 2.6GHz to 2.8GHz. It will do this at about half the real world power draw of the Intel equivalent parts. It isn't a fair fight, AMD will stomp Intel until mid-year, badly. In addition to the FS parts, there will be continued development on E-Step, so all the current owners won't be left out in the cold.
When the Merom based Conroe comes on line in Q3, early Q3 if the recent roadies are to be believed, then Intel will be on the road to redemption. But here is where the 'if' comes in, this is the one place where Intel has a serious potential for delays. We don't believe Conroe will be delayed, the early chips are doing quite well, but the company's track record on execution quite frankly sucks.
Conroe will be be a new core built from the ground up. Basically it is a four issue wide 65nm Pentium M inspired chip. You can read about the details here, and not much has changed since then. The Achilles Heel of the chip, or heels as is the case here, are the front side bus (FSB) and the Integrated Memory Controller (IMC), or lack thereof. The FSB, 1333 for Conroe, is a bit creaky old, and slow. Add in the lack of an IMC, and memory latency becomes a sore spot. Both should be worlds ahead of the Pentium 4, but not up to the performance of the AMD A64.
Intel's position on the IMC is that the concept is good on the server front, but the lack of flexibility makes it unsuitable for the desktop. If Intel held the performance crown or did not force chipset changes vastly more often than AMD, we might buy that argument, but the current lineup of desktop chips makes this quite a stretch to believe.
On the up side, the IPC (instructions per clock) of the Conroe will be about 30 per cent higher than the current Pentium M parts. Those current parts will give an Athlon 64 a run for its money, so with increased efficiency, Conroe should be quite a fearsome competitor. Add in projected clocks of 2.66GHz at launch, 3.0 by the end of the year, and AMD had better not screw up. There is a fallback plan for a 1066MHz FSB, but that would leave a lot of performance on the table. If a Conroe is about equivalent to a 50 per cent higher clocked Pentium 4, then a 2.66GHz part would need around the equivalent memory bandwidth of a 4GHz Pentium 4. Two cores at that speed would saturate a 1066FSB in no time, and even a 1333MHz bus could be restrictive. If Conroe is better than that, the bus becomes even more of a problem, and as clocks ramp, FSBs won't. This leaves Intel with a scaling problem. Basically, the core is too good for the platform.
Intel also has a habit of way underdelivering on clocks of late. Yonah was originally slated to be 2.5GHz +/- one bin, 2.33-2.67GHz. It is launching at 2.16GHz with 2.33GHz to follow as soon as volume for that bin ramps. If it overpromises and underdelivers - which is becoming a Chipzilla tradition - we could be in for a 2.33GHz Conroe, which wouldn't exactly set the world on fire versus a 3.0GHz A64 X2. A 3.0 at launch however would be untouchable.
With Conroe, barring massive screwups, Intel will retake the lead, but not by huge margins. The platform is way overdue for an upgrade, and that isn't on the cards. With it, it would be game over, without it, Intel still will have a decent lead.
In Q3, AMD will start what by all accounts will be a syphilitic trickle of 65nm parts, ramping to full production in 2007. This is slow and somewhat late, but not fatally so for AMD. It is also AMD's big chance to screw up in an epic way. If AMD blows this move, Intel will take the performance lead, run away and hide with it. If AMD does it right, Intel will still have a decent lead.
Personally, I am not hearing good things about AMD's 65nm move though, so I would put better odds on AMD screwing up than Intel, but being an optimist, I hope neither does. I like fair fights. The 65nm parts will pretty much be a dumb shrink of the 90nm FS parts, so don't look for a performance leap. Clock for clock, they will pretty much tie their predecessors. AMD of late has been following a core upgrade then a dumb shrink pattern. What 65nm will buy AMD is increased clocks and decreased power consumption.
By the end of Q3, we will most likely have a Conroe 2.66 vs an X2 3.0, both at 65nm. If a Pentium M is about five per cent slower than an A64 clock for clock, then Conroe's will be at about 125% of an X2's speed. FS will add maybe five per cent from core efficiencies and DDR2 another five per cent to the AMD numbers. This would put AMD at 3.0 * 110% or about 3.3 Inq Nebulous Units (INU) in performance, Conroe will be at 2.66 * 1.25%, 3.325 INU. In Q4, clocks will go to 3.2 and 3.0 for X2 and Conroe respectively, and the INU count will jump to around 3.52 vs 3.75.
In reality, we would expect Conroe to show a little more of an advantage than that on release, but the X2 will scale better due to a vastly superior platform. A slim lead at the beginning of Q3 will see a widening gap during Q4. The K8 is on its third process, and that's basically the end of the line for the core, Conroe is the first of a new line. Over 2007, Intel should ramp faster than AMD increasing the lead a bit, but platform deficiencies will hamper it, making it a win, but not a clean kill. 2007 will bring K8L vs Penryn, so the game starts all over.
Moral, the first half of 2006 is all AMD, no question there. Intel can screw up on the core launch in mid year, AMD can screw up on the 65nm process. AMD will tweak its core a bit, Intel will launch a new one. AMD's vast wattage lead will be eroded but most likely not outright beaten if you take the north bridge into account. The second half will open with Intel having a minor lead, and it will stretch out during Q4. My, how things change. µ
Sign up for INQbot – a weekly roundup of the best from the INQ