Bits & Bytes
Written By Van Smith
Date: March 15, 2004
Faithful readers of these pages no doubt have viewed Intel's "Prescott" Pentium 4 catastrophe as a slowly unfolding train wreck. For years we have detailed the weaknesses of Intel's NetBurst architecture, so it should come as no surprise that by perversely extending the Northwood P4's pipeline by over 50%, the Santa Clara chipmaker has created perhaps the worst dud in the history of microprocessors.
In previous articles we have contrasted Intel's "speed demon" versus AMD's "Brainiac" approach. With Prescott's much deeper pipeline, the new Pentium 4's performance at any given clock speed is very often lower than its predecessor's (which is really hard to do considering that Prescott has twice the L2 cache!), and its heat production is much higher. This places an even greater need on high gigahertz to reach performance levels comparable with AMD's much more efficient Athlon 64.
Unfortunately, Prescott's heat output is so prodigious that it gates clock speed ramping to no higher than Northwood's without the use of exotic cooling solutions and more costly VRM and socket implementations. Not only is Prescott the hottest consumer class microprocessor by far, but its thermal density, the ratio of total heat output per area of chip space, is scalded dog astronomic.
This leaves Intel with a CPU that is much more complicated than the chip it replaces, produces much more heat, doesn't ramp as well under the same conditions, and is slower to boot. For chip designers, this is a nightmare outcome.
Most observers have been dumbfounded by Intel's perplexing design decisions. Why didn't the chipmaker simply "shrink" Northwood to 90nm, expand its L2 to 1MB, add SSE3, tweak speed paths, bolt on AMD64 extensions (which we, despite Intel's adamant denials, exclusively revealed were in the first Prescott samples nearly two years ago) and call it a day? This would have left Intel with a processor that would undeniably and significantly exceed Northwood's performance at any clock speed (and therefore blow away the current Prescott), use fewer transistors than Prescott, produce less heat than Prescott and probably ramp better than Prescott.
What would lead Intel down Prescott's sorry trail? The answer is clearly that Prescott was a marketing driven product piloted straight over a cliff. This was allowed to occur because Andy Grove's system of pitting opposing camps against each other often leads to the loudest surviving -- not the fittest.
As we've detailed before, NetBurst places a great deal of pressure on Intel's fabrication technology to enable the fastest-switching transistors in the world while simutaneously holding down leakage. But leakage is a direct consequence of higher transistor speeds, so this is not at all an easy goal to reach. Intel effectively spots the competition an IPC lead with NetBurst -- and in the case of Prescott versus Athlon 64 this lead is about 103.2 light years -- placing the onus on the manufacturing side to make up the distance.
Why do this seemingly reckless and pointless exercise? For marketing clout, pure and simple.
From a marketing perspective, NetBurst has, up until Prescott, been a big success. The Pentium 4's high clock speeds have been so useful in hoodwinking consumers that Intel will soon be forced to deploy a model rating scheme in order to "educate" the public of the fact that its Centrino line of CPUs is much, much faster than its P4 line at any given clock speed.
No doubt, visions of 5GHz, roughly the potential of Prescott's 30+ stage pipeline if thermodynamics are ignored, and the propaganda advantages it would give over AMD's hard charging 64-bit assault dazzled the eyes of Santa Clara's top spin-meisters and comforted a few paranoid but remote Intel execs.
Prescott has served a cold, heaping dose of raw humility to a company that many insiders have viewed as arrogant. With recent remarks from Intel's top brass, it appears that some of the more deserving voices inside the company are finally being listened to. Multi-core Banias/Dothan derivatives, a direction we championed in past columns, appear to be the future for Intel desktop processors after Tejas, Prescott's successor (and another item we exclusively reported).
Overclockers face new challenges with Prescott and need to be much more cautious with the little inferno than with previous CPUs. Cooling the CPU alone will often not be sufficient and could even potentially be dangerous. Motherboard voltage regulation circuitry could become hot enough to be a very real fire hazard. It is perhaps this reason why Intel has effectively delayed Prescott until safer conditions exist in the form of new VRM standards and a slightly more power efficient socket.
With Prescott, overclockers don't merely have to worry about frying their processors, but incinerating themselves and their loved ones as well.
Speaking of heat, we took a lot of it when we first disclosed the existence of the Pentium 4's throttling mechanism several years ago. The point we wanted to communicate is that reviewers need to be aware that, with the Pentium 4, the thermal environment can impact performance, so benchmarks should be run under thermal operating conditions that simulate normal usage -- otherwise reviewers were potentially doing their readers a disservice.
X-bit Labs recently produced an interesting article showing that Prescotts might throttle under even favorable conditions. Given that many people are seeing these little chips run close to the throttling temperature while bench testing, those hard to reproduce benchmark results might have a simple explanation, one that we tried to voice years ago.
Another Kind of Prescott
As a sidebar, "Prescott" was also the name of the current U.S. President's grandfather, who, as a member of the elite and extremely powerful "Skull and Bones" Yale secret society, is rumored to have been a grave robber. In May, 1918 at Fort Sill, Oklahoma the patriarch Bush entered the tomb of great Apache leader Geronimo, dismembered the heroic chief's skeleton and stole its skull, elbow and a few associated artifacts as ceremonial relics that are reportedly a centerpiece of Skull and Bones' bizarre occultist rituals.
This act of elitist depravity has made few friends among American Indians.
The remains of Pancho Villa, another American Indian, reportedly were similarly desecrated at the hands of the same affluent cult, an organization said to be the birthplace of the CIA.
For even more disturbing bizarreness, certain Skull and Bones rites are supposedly held using Hitler's silverware. Hitler was financed, in part, by Prescott Bush and as a consequence grandpa Bush and his Union Banking Company ran afoul of the Trading with the Enemy Act in October, 1942.
Although consisting of an extremely tiny conspiratorial gaggle of self anointed American aristocracy, both of this year's U.S. major party presidential candidates are "Bonesmen." John Kerry was inducted into the society in 1966, while George W. Bush, a distant cousin of Kerry, hails from the class of 1968.
Transmeta Efficeon Coming to America
Japan's popular computer technology magazine DOS / V presented a handful of early benchmarks from a laptop containing Transmeta's new 1GHz Efficeon processor. The results were surprisingly poor given the CPU's very large die size and the notebook's relatively robust ATi Mobility Radeon graphics solution.
We added a few data points to their numbers to give a better idea of relative performance. The first graph shown is in same "weird units" presented by DOS / V, while we normalized performance in the second graph against the Efficeon.
The 1GHz Efficeon TM8600 performed roughly as well as a 1GHz VIA C3-Nehemiah running on the new VIA CN400 chipset.
Of course, this is an extremely narrow set of benchmarks -- and we are certainly no fan of PCMark. A somewhat more favorable review can be found here, but the range of benchmarks used there too is frustratingly limited.
The Transmeta Efficeon is scheduled to debut today in America, and we will be sure to test a sample and report our findings at first opportunity.