Contents
From Deschutes to Coppermine
Cache
1GHz... and Beyond?
Enter Tualatin
Identification
The Big Differences
AGTL
Coppermine T
TDP
Chipsets & Motherboards
The 440BX Hack
SMP
Performance
Sources
|
| Pentium III Coppermine die shot. Source: Intel |
Intel's Tualatin CPU is not well understood, despite a resurgence in interest in the last few years. Some say you can run it on a Pentium II motherboard. Some say it's the architecture that became Intel's current Core range. Some say no one pronounces it properly. Some even say it was faster than the first Pentium 4s. Let's find out the truth once and for all. But first, see how good your existing knowledge is with a game of True or False!
1. You can easily tell the difference between a Coppermine and a Tualatin because there's no integrated heat spreader on a Coppermine.
2. Although released 6 months after the Pentium 4, Tualatin was the first Intel CPU to use the 0.13µm manufacturing process.
3. Tualatin CPUs cannot be used on standard Pentium III motherboard because they use AGTL+ signalling instead of AGTL.
4. Intel's controversial Processor Serial Number feature, introduced with the Pentium III, was not implemented in Tualatin CPUs following privacy concerns.
5. Intel won the GHz war, by releasing their 1GHz Pentium III a month before AMD's Athlon.
6. The Coppermine CPU is so-called because it uses copper interconnects instead of aluminium.
7. The 1.4GHz Tualatin-based Celeron was so good at overclocking, you could get it to run at 2.6GHz.
8. Although not in the Xeon family, the Tualatin could be used in configurations of up to 8 CPUs.
(answers at the very end)
From Deschutes to Coppermine
Although we're familiar with the use of CPU codenames today, they were a more recent thing when Tualatin was revealed in 2001. It was the final incarnation of the Pentium III and came at a time when Intel was making great strides in revolutionising its CPU designs, mostly due to strong competition from AMD. The first Pentium III was dubbed Katmai, using a similar package (SECC2) to the Pentium II (SECC), in Feb '99. Although it saw the introduction of SSE instructions and brought noticeable improvements over the performance of the Deschutes Pentium II (both were made using the 0.25µm manufacturing process), it was an incremental release that saw more similarities than differences. Coppermine's technical enhancements were key to Tualatin's success, however, and that's what we're going to be looking at in more detail.
Cache
This is a necessary paragraph but I'll try to keep it brief. Cache is a very small but very fast area of memory that can hold frequently used instructions and data, so that the CPU doesn't have to go running off to the relatively slow RAM each time. It was first introduced on 386-equipped PCs, was about 64KB in size, and resided on the motherboard. One of the improvements that came with the 486 was to put a small amount (8KB) of cache for both data & instructions on the CPU die itself. This was dubbed level 1 cache, while a larger amount remained on the motherboard (level 2), the size of which depending on how much RAM was installed. Both these amounts were increased with the introduction of the Pentium, along with the L1 cache being split into separate areas for data and instructions (8KB each).
|
| The Pentium Pro's CPU die and L2 cache combined on the same package [pcgameshardware.de] |
The most relevant change came with the Pentium Pro, where up to 1MB of L2 cache was located in the CPU package itself. This enabled it to run at the same speed as the CPU, instead of being limited by the speed of the motherboard. A tiny fault in either component, however, would cause a CPU to be discarded resulting in low yields, high prices and relatively scarce availability. The Pentium II was quickly introduced, modifying this arrangement by placing the L2 cache and the CPU on a shared PCB instead, resulting in the Slot 1 form factor and higher yields but half the speed. The same package was used for the Katmai Pentium III. Coppermine (Oct '99) was such a revolution because the move to a 0.18µm fabrication process made full speed, on-die level 2 cache possible. Intel dubbed it Advanced Transfer Cache, which connected to the CPU via a 256-bit bus. According to this source, Tualatins were able to cache the full 64GB of addressable RAM (thanks to Standard Def Steve on Vogons).
1GHz... and Beyond?
Coppermine was a very successful product for Intel but it wasn't all plain sailing. This was a time when CPU frequency (known better as 'speed') was everything, and AMD had stolen a march on Intel by releasing their 1GHz Athlon first. The Pentium III's Processor Serial Number feature had caused more consternation than excitement too, but the really rocky period was still to come. Although clock speeds on desktop CPUs had climbed from 300MHz to 1000MHz in the space of 18 months, the ceiling was hit in a big way. In a brilliant show of tech journalism, HardOCP, AnandTech and Tom's Hardware caused the initial 1.13GHz Coppermine to be recalled by Intel after their standard testing regime showed serious issues with instability in July 2000. It took about a month for Intel to admit the problem, however, with the Pentium 4 being announced during this period. Following a redesign, the recalled CPU was eventually re-released as the cD0 stepping, but it was clear that scalability was going to be an issue at clock speeds over 1GHz, despite Intel previously stating at the 1GHz mark, "there's plenty of headroom left…"
If you want to read more about the enhancements introduced with Coppermine, here's a full write-up from Thomas Pabst.
Enter Tualatin
Intel introduced chips using their new 0.13µm process in Feb 2001. Initially touted as mobile processors, this new Pentium III core was dubbed Tualatin, with some relatively minor differences to Coppermine, at least on the surface: Data Prefetch Logic to make better use of the L2 cache, and the Voltage Regulation Module (VRM) spec was modified (more on this later) brining lower power consumption. Additionally, the extra real-estate on the die made 512KB of cache possible. If you compare the two die shots you can see the differences quite clearly. You could even call Tualatin the Pentium 3.5 (III.V would be funnier). Six months later, server and desktop variations of the chip were announced.
So how do you say it? There are some weird pronounciations out there but, if you want to hear it from the horse's mouth, it's too-ALL-uh-tin. It's quite a well-known fact that the chip was named after one of Intel's manufacturing facilities in the Tualatin Valley, Oregon.
Identification
There are 4 types of Tualatin that need to be discussed because that's one of the most confusing aspects of this CPU: server, desktop, Celeron and mobile. Here's a very brief summary of the distinguishing features:
Server (aka Pentium III-S): 133MHz, 512KB cache.
Desktop (aka Pentium III): 133MHz, 256KB cache.
Celeron (aka Tualeron): 100MHz, 256KB cache.
Mobile: 100/133MHz, 512KB cache, SpeedStep.
Despite Xsome sources saying otherwiseX, Data Prefetch Logic was included with all models as it was an integral feature of the Tualatin design. The principle focus of this article is the server version, because that's the one that produced all these killer benchmarks people talk about. Yes, the Celeron version (aka Tualeron) was cheap and could overclock quite well so was discussed a lot at the time. They're pretty versatile performers but, given that the cost difference isn't so much of an issue these days, you may as well plump for the full 512KB model if you can. Even better, if you get the server version and the right board, you can run multiple CPUs.
|
| Pentium III-S Tualatin Die Shot. Source: Intel |
The Big Differences
On the surface, there aren't many differences so we need to zoom into the die itself. Hilariously Intel doesn't bother with things like brand names in their datasheets as that would obviously be absurd. The product 'name' they use for the full Tualatin is more of a definition:
Intel® Pentium® III Processor with 512KB L2 Cache at 1.13GHz to 1.40GHz
As compared to the plain old Coppermine, which is described thus:
Pentium® III Processor for the PGA370 Socket at 500 MHz to 1.13 GHz
Architecturally, the increased L2 cache really was the biggest new feature and is the thing emphasised most heavily in the datasheet. The Processor Serial Number feature was retired very quietly, with emphasis instead being placed on the newly-introduced Data Prefetch Logic defined as:
add[ing] functionality that anticipates the data needed by the application and pre-loads it into the Advanced Transfer Cache, further increasing processor and application performance.
|
||
| FC-PGA2 with IHS | FC-PGA | |
Both the desktop and server Tualatins were available at frequencies of 1GHz, 1.13GHz, 1.26GHz and 1.4GHz with a 1.2GHz variation exclusive to the desktop version. Compare this to the 20 Coppermine models and you get an idea of how long each family was around. One aspect I find hilarious is the CPUID Intel chose to 'distinguish' the two chips. Coppermine was allocated 068xh. Tualatin's was 06Bxh. Yeah. B vs 8. If the increment was chosen by a pre-defined process, why didn't they make it less ambiguous, and if they chose it intentionally, why didn't they make it less ambiguous? Very odd.
It has been well publicised that, despite its name, Coppermine used aluminium interconnects on the CPU die, with Intel claiming that the design was 'purely transistor limited'. When designing Tualatin, engineers made the shift to copper, which makes sense considering its greater headroom for higher frequencies. Maybe this had a bigger impact than they thought it would.
AGTL
The biggest difference between Tualatin and its predecessors is electrical and is what, ultimately, made them incompatible. This was the transceiver logic they were designed to use, which is most simply defined as 'the voltages used to send signals between chips that are connected to each on the motherboard via the traces'. (If you want a more in-depth technical description, check out the patent describing the technology.) Gunning Transceiver Logic (GTL) had been invented in 1990 and a variation of this (GTL+) was used on the Pentium Pro. The Pentium II introduced Assisted GTL+, which continued to be used on the Pentium III but was unsuitable for Tualatin. The use of AGTL instead made the CPU incompatible with existing motherboards because of changes in the voltages used. Fortunately this didn't constitute a full layout change to the boards themselves, just some updates to the chipset and pin allocations, plus changes to the VRM.
The basic difference is that the Tualatin bus runs at 1.25V instead of Coppermine's 1.5V. This means Tualatin CPUs have an additional pin (for a total of 5) allocated for voltage identification. Depending on the logical state (hi or lo) of these pins, the CPU's core voltage can be set automatically. The additional pin makes twice as many voltage steps available (1.05V to 1.825V in 32x 0.025V steps) compared to Coppermine (1.3V to 2.05V in 16 steps). As mentioned, the VRM design guidelines changed from 8.4 to 8.5 to account for this more detailed level of voltage control.
Coppermine T
Interestingly, Tualatin wasn't the only Pentium III to use AGTL, it was just the only one to use it exclusively. Just to make things really confusing, Intel released a revised Coppermine, dubbed Coppermine T, that could use both AGTL and AGTL+. This meant it would work on motherboards that supported either. These models were clocked at 866MHz, 933MHz, 1GHz, and 1.13GHz. As an aside, 1.13GHz might, at first glance, seem like an odd frequency to jump to, but actually it's 1000MHz plus 133MHz, which makes perfect sense.
TDP
Intel had lost a lot of ground in the server marketplace thanks to competition from more thermally-efficient CPUs, particularly Trasmeta's Crusoe. Thermal performance greatly improved with Tualatin and it was widely reported to be intended for server blade application.
The 1.13GHz Coppermine had a Thermal Design Power of 37.5W, compared to 27.9W for the Tualatin part running at the same frequency. Even the 1.4GHz model ran lower, at 31.2W, so that's quite a significant improvement. There was absolutely no headroom for overclocking the 1GHz+ Coppermines, but it was certainly a possibility on Tualatin and was done frequently.
Chipsets & Motherboards
Official adaptors existed, such as the PowerLeap PL-iP3/T, that could adapt a Tualatin processor to a technically incompatible slot 1 motherboard. There were also interposers that could be used on socket 378 boards. These adaptors made the necessary modifications between the CPU pins and the socket to compensate for the change from AGTL+ to AGTL. They were too expensive at the time for most users ($169+) and are almost impossible to find today so I'm going to assume you don't possess such a unicorn. In most cases, official Tualatin chipsets were existing models but with a 'T' suffix to indicate support for the new CPU.
Much as I would love to also provide a list of compatible chipsets and motherboards for you to peruse, that's beyond the scope of this article. The Vogons Wiki has an excellent list for you to refer to.
The 440BX Hack
Getting Tualatins working on older motherboards was a massive challenge, however. At the time, Intel's 440BX was hailed as Intel's finest ever chipset. It was introduced in 1998 and supported slot 1 boards initially, with socket 370 following later. It achieved legendary status, as board makers were able to provide bus speeds of 133MHz and higher, well beyond its 100MHz specification. This is the chipset that exposed the overclocking abilities of the relatively cheap Celeron 300A CPU, with enthusiasts consistently running it at 450MHz (a 50% increase).
After studying the datasheets and the VRM design guide, an individual called Nightcat from Taiwan was able to work out a way to run a Tualatin CPU on BX-based board, even though they didn't support AGTL. The Tualatin specs required voltage steps of 0.025V, but it was discovered that the 0.05V steps possible on most 440BX-based motherboards were sufficient.
Key to performing this hack on slot 1 boards was the need for a 'slocket' or 'slot-ket', the nickname given to a device that takes a vanilla socket 370 CPU and adapts it to fit into slot 1. Such adaptors are almost impossible to find these days, but were once widely available from a number of manufacturers. There was no guarantee, however, that the slocket you had acquired would work with the mod. The hack was eventually tried on pretty much every Pentium III motherboard people could lay their hands on.
|
| Hacking a Tualatin. Source: James Anderson |
Following one of the popular guides, you can see just how rough and ready this hack is. I mean you have to connect two pins on nearly opposite sides of the CPU together and then insulate three of them, requiring widening of the corresponding holes on the socket so that they fit. Some people opted to remove the pins and I can see the motivation but no. It's a hack in the truest sense of the term, but impressive considering it defeated (enabled?) Intel's engineering. It's also testament to the 440BX chipset that even the introduction of an incompatible CPU couldn't yet render it obsolete.
SMP
The Pentium Pro was arguably Intel's first exclusively 'enterprise' CPU, but the Xeon name was introduced to distinguish server versions of the Pentium II from their desktop counterparts. This continued through the development of the Pentium III.
Tualatin was introduced at a time when Intel's was in multiprocessing trouble: Foster core P4 Xeons were not widely sold and performed badly. Meanwhile, the fastest PIII Xeon was a paltry 1GHz so there was no performance leap available for existing customers. It appears that the server version of Tualatin was intended as a stop-gap solution and it certainly did the job, effortlessly outclassing the 1.7GHz Foster Xeon at most tasks while matching the 1.8GHz Prestonia Xeon as well. Not wanting to completely sabotage the Pentium 4's reputation, they didn't create a full Xeon version of Tualatin because, while Xeons could be used in configurations up to 8 CPUs, only 2 processor Tualatin systems were possible.
Performance
This is the question everyone seems to want an answer to. You may have already read claims that the 1.4GHz Tualatin can outperform 1st gen Pentium 4s. This is something that is routinely reported as fact today, but very few people can point you in the direction of a reliable source. Fortunately reviewers benchmarked the shit out of new CPUs when they came out so there's plenty of evidence to refer to if you can find it. There are also crappy articles that provide no sources or evidence but seem to be cited by people trying to prove a point. I'm not that lazy, so it's time to ask "are the claims true?".
Mostly. Hot Hardware's tests of a 1.2GHz Tualatin saw it beat a 1.8GHz Pentium 4 in Winstone tests, mostly because they're not RAM intensive. It was beaten at encoding tasks, however. Considering the 50% difference in frequency, it still put up a good fight. The big surprise came when they were able to overclock the FSB to 163MHz, resulting in 1.47GHz from the Tualatin (a 22% increase). This enabled it to beat the P4 hands-down in 3DMark 2001 and score at a mere 1.6FPS lower than the Pentium 4 in Quake III. Pretty impressive.
Kinda. Tweakers demonstrated that the Tualatin kicked a 1.7GHz Pentium 4's butt at raw calculations, but lost out big in tests measuring RAM data rate. Additionally, AnandTech's extensive benchmarks demonstrated that the P4 outperforms its predecessor in nearly every test, with AMD's 1.2GHz Athlon being the actual threat. The test only compares with a stock 1GHz PIII so it's actually inconclusive. I needed more data. In IXBT's comparison against the 1.7GHz P4 Xeon, Tualatin came out on top in 7 out of 17 tests
The most decisive results came from Tom's Hardware, who did a group test of 65 CPUs in 2003. I took results from the CPUs of relevance in order to demonstrate how the Tualatin actually performed against the Pentium 4 and made some nice charts. The 1.3GHz Tualatin has a (!) next to it because it's simulated by taking an average of both Tualatins. This was done because I wanted to provide some kind of clock-matched comparison to the lowest-clocked P4. All the P4s are Willamette core (0.18µm) except the 2.0GHz Celeron, which is Northwood (0.13µm, like the Tualatins).
The Tualatins were tested on an Asus TUSL2-C equipped with SDRAM, the lower-clocked P4s used a Asus P4T with RDRAM, and the 1.8GHz P4 and Celerons used an Asus P4PE with DDR RAM. Graphics cards used were the closely-matched ATI Radeon 9700 Pro and GeForce 4 Ti 4600 on all systems.
I'm going to keep score between the 1.5GHz P4 and the 1.4GHz Tualatin to see who the overall winner is at the end.
In DirectX 8 tests, the results are astounding. The Pentium 4's architecture actually seems to decelerate 3D graphics, with the 1.4GHz PIII-S smashing the 1.8GHz P4 in both tests. Clearly the 512KB cache helps, as the others have 256KB or less, but there must be some kind of serious issue between AGP and the Pentium 4 here for such a big effect. My guess would be that it's related to the double-edged sword of the P4's long instruction pipeline. Without getting into it too far, the P4 has the capacity to predict up to 20 instruction in advance which is great until one of them turns out to be a 'bad guess', causing the entire queue to be flushed. Oof. The Pentium III's pipeline is half as long, one of the biggest differences between the two architectures.
Tualatin: 2, Pentium 4: 0
I read this chart wrong at first: lower is better! When it comes to encoding video, the 1.4GHz Tualatin performs admirably. Although raw clocks appears to be the biggest factor in this test (with the 2GHz Celeron winning out), a large amount of cache seems to compensate hugely, with the 1.2GHz PIII beating the 1.8GHz P4. Maybe the software wasn't yet optimised for the P4's SSE2 instructions. Either way, this is exactly the kind of task the Pentium 4 was designed to be better at.
Tualatin: 3, Pentium 4: 0
It's a different story with MP3 encoding, with a linear graph showing that raw speed and a long pipeline takes the crown, but by a narrow margin.
Tualatin: 3, Pentium 4: 1
Another graphics-heavy test here. You can see the almost linear improvement between the group of PIIIs and the P4s respectively but, again, the P4 loses out big to the gang of Tualatins. Raw speed helps out the Celeron, but the big cache of the 1.4GHz Tualatin compensates to put it ahead miles ahead of the 1.5GHz P4. Shocker!
Tualatin: 4, Pentium 4: 1
Here's where the P4 wins every time: memory. The quad-pumped, 400MHz front side bus really demonstrates the speed advantage the next-gen CPUs hold over the Tualatin with its lowly 133MHz bus. Even the Celerons manage to put on a good show.
Tualatin: 5, Pentium 4: 2
The OpenGL test seems to hand the graphics advantage back to the Pentium 4, with a very linear representation of performance. It also demonstrates just how much the lack of cache on the Celerons cripples what is theoretically a 'fast' CPU. The Tualeron truly was the last good Celeron and Pentium 4 generation ones really can be thrown in a fire. They're so pathetic I'm tempted to remove a point!
Tualatin: 5, Pentium 4: 3
An interesting spread of results here, identical in shape to the PCMark test. Again, raw speed (I really struggle to use that phrase) hands a win to the 2GHz Celeron, with the 1.4GHz Tualatin not far behind, beating everything else. Shockingly the 1.2GHz Tualatin beats the 1.5GHz P4. That shouldn't happen really, should it?
Tualatin: 6, Pentium 4: 3
The P4s get a bit of pride back as they whack their FLOPS out for all to see, but there's almost nothing in it between the lower-speed participants. The Tualatins still manage to beat their successors, though. This is getting embarrassing.
Tualatin: 7, Pentium 4: 3
Of course another RAM benchmark leaves the Pentium IIIs eating the P4s' dust. Sad times.
Tualatin: 7, Pentium 4: 4
Sysmark's real-world benchmarks seem to be optimised fully for the P4 architecture, with another linear spread. You'd be nuts to run XP on a Pentium III anyway, so it's not really a fair or relevant comparison.
Tualatin: 7, Pentium 4: 5
Ah, a classic file compression benchmark certainly emphasises the benefit of a large cache. The blistering performance of the 1.4GHz Tualatin suggest it probably didn't access the RAM much at all and, once again, the 1.2GHz Tualatin beats three of the P4s.
Final Score Tualatin: 8, Pentium 4: 5
While these tests show Tualatin out-performing the Pentium 4 with raw calculations and DirectX games, anything slightly memory intensive sees the P4 come out in front with its superior front side bus. Despite this, Intel made any advantage the Tualatin might have had at the time completely irrelevant by pricing it well above the P4. The joke was on Intel in the end, however, as they inadvertently made the P4 irrelevant and impossible to recommend because of its high price and reliance on RDRAM. This eventually changed, with DDR RAM becoming the standard and ever-increasing clock speeds (and temperatures!). What a shit show.
Legacy
Tualatin caused some big waves when it hit the market, probably more than Intel had anticipated. I often wonder if its architects had any idea that it would become one of the most important designs in Intel's history. With hindsight NetBurst, the architecture used in the Pentium 4, was the wrong road from day one. Although it was a completely separate architecture from Tualatin, we kind of need to look at it in detail to understand what happened next. That's an article for another day.
Thanks for reading. See you on Twitter and YouTube.
Sources
Report: Online Shopping Fraud Bites Merchants, Not Buyers (4 Dec 1998)
Serial Number Alone Won't Violate Privacy Concerns at InfoWorld (22 Feb 1999)
Intel's Pentium III Case at Berkley University (Spring 1999)
Intel's New Weapon: The Coppermine at Tom's Hardware (25 Oct 1999)
AMD hits 1 GHz with new Athlon microprocessor! at ITPro Today (5 Mar 2000)
It's official: AMD hits 1,000MHz first at ZDNet (6 Mar 2000)
Intel Pentium III 1GHz at Anandtech (8 Mar 2000)
Intel Pentium III 1GHz Review at Sharky Extreme (8 Mar 2000)
Intel Pentium III 1.13GHz Review at Sharky Extreme (31 Jul 2000)
Intel to detail Pentium 4 at CNN (21 Aug 2000)
Intel Officially Launches Pentium 4 at IDF at Electronics Weekly (23 Aug 2000)
Intel Admits Problems With 1.13GHz Pentium III at Tom's Hardware (28 Aug 2000)
Intel Pentium 4 1.4GHz & 1.5GHz at AnandTech (20 Nov 2000)
Intel Thinks Small in Mobile Spotlight At ComputerWorld (28 Feb 2001)
The Microarchitecture of the Pentium 4 Processor at Intel Technology Journal (Q1 2001)
Intel Samples Tualatin Processors at EE Times (16 May 2001)
SiS635T: A Tualatin Twilight at Aces Hardware (30 May 2001)
Tualatin PIII Outclasses Pentium 4 at The Inquirer (11 Jul 2001)
The Celeron of The Future at AnandTech (30 Jul 2001)
Intel Fights Back With Tualatin Chip at InfoWorld (30 Jul 2001)
Intel's "Tualatin" Pentium III 1.20GHz Processor at Hot Hardware (31 Jul 2001)
Two For The Price of One at InfoWorld (10 Sep 2001)
Last Passing Maneuver at Tom's Hardware (19 Sep 2001)
The New Pentium III - Codename Tualatin at Insane Hardware (1 Nov 2001)
Intel to demo 0.13-micron Pentium 4 at CNN (5 Nov 2001)
The Little Celeron That Could (9 Nov 2001)
Q3 2001 Industry Update at Real World Technologies (7 Oct 2001)
Intel makes strides with Tualatin chips at CNet (2 Jan 2002)
Running Tualatin On CuMine MB w/o Powerleap at Overclockers (17 Feb 2002)
Tualatin on a BX mobo works. No adaptor required at Gamers HQ (25 Feb 2002)
Tualatin In A BX Board at Overclockers (30 Apr 2002)
Tualatin in the Asus P3B-F by Mathias Rufer (6 Oct 2002)
Benchmark Marathon: 65 CPUs from 100 MHz to 3066 MHz at Tom's Hardware (7 Feb 2003)
Intel's Centrino CPU (Pentium-M): Revolutionizing the Mobile World at AnandTech (12 Mar 2003)
Pentium M Dual Core in January 2006 - Summary of the Intel Mobile Roadmap at AnandTech (6 Oct 2005)
The Tualatin Story at OS/2 Museum (6 Apr 2013)
Tualatin Celeron vs Williamette Celeron at Vogons (4 Mar 2017)
Powerleap PL-iP3/T Slot 1 to Socket 370 Slocket Adapter at Ancient Electronics (14 Sep 2018)
Fastest Tualatin Chipset / Best Pentium III Motherboard at Vogons (31 Dec 2018)
Wikipedia article
GTL Patent at Google
AGTL Patent at Google
Pentium Pro Datasheet at dexsilicium
Pentium II Datasheet at dexsilicium
Coppermine Datasheet at Intel
Intel VRM8.5 Design Guidelines
Pentium III Processor Specification Update at chipdb
Die Shots at grafik-feti.de
Intel 440BX at Wikipedia
Intel 440BX Datasheet at Intel
Overclocking Computers with Intel Celeron (Tualatin) at Flylib
Overclocking Computers with Intel Pentium III (Tualatin) at Flylib
Tualatin based Intel Celeron 1.2 GHz for Socket 370 at IXBT Labs
Intel Xeon Processor Review at IXBT Labs
Intel P6 Comparison with charts + additional CPUs at Vogons
PowerLeap PL-iP3/T Slot1 to Tualatin Adaptor at PowerLeap
Server Tualatin Review at IXBT Labs
List of Socket 370 motherboards at Vogons Wiki
Pentium M and Core Interlude at QDPMA
The father of Centrino, Pentium M and Core 2 leaves Intel after 33 years
Intel's 90nm Pentium M 755: Dothan Investigated at AnandTech (21 Jul 2004)
Intel Technology Journal Q2, 1999 at Intel
True or False Answers
1. False. Late model Coppermine PIIIs also used the FCPGA2 package with the IHS. It was officially introduced because of the need for more efficient heat dissipation at frequencies near 1GHz. There is also anecdotal evidence that, without it, the Pentium III die could be chipped during heatsink application on the FCPGA package.
2. True. Tualatin was Intel's first 0.13µm CPU. Initial Pentium 4s and Xeons still used 0.18µm.
3. False. It's a bit counter-intuitive, but Tualatin uses AGTL signalling.
4. True. Although Intel encouraged board manufacturers to turn the feature off by default in the BIOS, the threat of litigation from the EU was enough for them to remove the serial number when Tualatin was being designed.
5. False. AMD released their 1GHz Athlon mere days prior to Intel's release.
6. False. Although it was thought to be necessary to achieve higher clock speeds, Intel found a way to ensure that unreliable aluminium interconnects wouldn't be a problem. Tualatin, however, used copper.
7. True. Some dude called Gradus managed to overclock a Tualatin Celeron 1.4GHz to 2.651GHz using nitrogen cooling and an Asus ST6-RAID motherboard.
8. False. Multiprocessing was a feature that Intel made exclusive to the Xeon line so you could only use up to 2x CPUs in a Tualatin-based system.