Where is AMD's Intel-killer Barcelona?

Back in May when speculation on AMD's "Barcelona" quad-core CPU was still running rampant, I wrote that "AMD Barcelona can potentially smash Intel" if high clock speeds at sufficient yields could be produced in a timely manner.  AMD needed to hit 2.5 GHz to reclaim the performance crown and release the product on time in July to give it some breathing room before Intel strikes back with its new 45nm "Penryn" processor at the end of the year.  Now that it has been confirmed that AMD will miss their original Barcelona release date by at least 2 months and the best clock speed they can do is 2.0 GHz, the future is looking grim for AMD.  With shocking news like this it's no wonder AMD decided to announce Barcelona's release schedule and clock speeds on iPhone Friday.

Based on AMD's own claims, the quad-core CPU codenamed "Barcelona" will have a 20% clock-for-clock advantage on Integer performance over Intel's "Clovertown" quad-core CPU.  [UPDATE 7/4/2007 - The 20% clock-for-clock advantage claimed by AMD is actually a 1% clock-for-clock advantage in the dual-socket 8-core Server space.  Intel holds a much stronger clock-for-clock position with fewer core tests since that doesn't put as much a strain on their shared memory controller and FSB architecture.]  The problem is that the Intel's 3.0 GHz quad-core CPUs have a 50% clock speed advantage so a 20% [UPDATE 7/4/2007 - actually a 1% clock-for-clock advantage] clock-for-clock advantage doesn't come close to beating Intel.  At this rate the only thing that Barcelona is going to smash are AMD's existing dual-core CPUs and last year's Intel dual-core products.  By the time AMD releases the Barcelona product in September, Intel would have had nearly a full year's lead in quad-core CPUs and it's no wonder AMD has lost nearly all of its market share gains in just the last year.

Based on the paper AMD released at the ISSCC in March, it came to light that AMD's Barcelona CPU had electrical capabilities that permitted a maximum clock speed of 2.8 GHz.  At the height of the speculation on Barcelona in May, Charlie Demerjian reported that AMD was "dancing in the aisles" in Austin Texas because early indicators for Barcelona was so good that it would hit a whopping 2.9 GHz for a desktop part and 2.8 GHz for a Server part.  That post fueled a lot of stories and cheers for AMD watchers but in hindsight it was couldn't have been more wrong.

I asked CPU analyst David Kanter (Real World Technologies) what could be the problem behind AMD's failure to deliver an Intel killer and Kanter came up with the following three possibilities.

1. Manufacturing cannot get the frequency high enough
2. There are speed paths that prevent higher clocking
3. You can hit high frequencies, but not within the TDP

Kanter noted that all those statements have an implied "with good yields" at the end.  Manufacturing problems shouldn't be surprising since that has always been an Achilles heel for AMD.  As for the second possibility, Kanter explained that "Speed paths" mean that certain parts of the CPU can't handle higher clock speeds even though the rest of the CPU can.  Then there's the possibility that manufacturing can hit the desired clock rates but with unacceptable amounts of leakage.

These problems may all just be temporary and AMD will eventually iron out the bumps and produce high quantities of higher clocked Barcelona CPUs, but time isn't on AMD's side when Intel's tick-tock cadence keeps ticking away.  If AMD manages to release higher clocked Barcelona CPUs closer to the end of the year, Intel will raise the bar even higher when it releases the Penryn processor.