Pentium 4: Round 3 - AMD hangs on to lead
Copyright (C) 2001 by Darek Mihocka
President and Founder, Emulators Inc.
Originally posted October 22, 2001
Benchmarks Round 3: October 22 2001
It has been over 6 months now since the last round of benchmarks. During that time both AMD and Intel have been busy to deliver chips for the coming wave of Windows XP driven computer upgrades. If there is one piece of advice that you should have listened to from our Secrets page, it was to hold off buying a computer until now, October, when computer manufacturers are tripping over themselves to sell you a Windows XP computer.
Now is the time to buy. Windows XP is showing up and Christmas is coming soon. With PC prices 60% to 80% below what they were only a year ago, and a poor economy, now is the time to take advantage of fast and inexpensive computer chips.
This price war all started in April when Intel kicked off a massive price cut on the Pentium 4. From April to August, both the AMD Athlon and the Pentium 4 prices were cut again and again. By August, the original $4000 Pentium 4 system from last Christmas was priced at around $1200, as was a comparable Athlon system.
In May and June, Intel and AMD also launched their respective multi-processor capable versions of their chips, Intel with the Xeon and AMD with the Athlon MP. As it turns out, both chips were fairly unremarkable as they carried a high price and only the one additional feature of being dual processor capable. This was of little interest to many computer users, and coupled with the fact that motherboards for these chips are fairly hard to find, they had little impact on anything.
The Xeon, merely a Pentium 4 with a different name, failed to offer any design fixes to the Pentium 4 architecture. For all intents and purposes, it was just a Pentium 4. As of this month, even the latest 2.0 GHz Pentium 4 chips still use the same crippled design as before. And if you paid attention to what I said before, this should not surprise you. Remember, silicon takes time to design and fabricate. Unlike a software fix, you can't just grow a new Pentium 4 chip overnight. It takes months. The first real new change to the Pentium 4 design, the upcoming Northwood processor using .13 micron technology and featuring a 512K on-chip cache, won't see the light of day until next year and will require a different motherboard than the original Pentium 4 systems.
The Athlon MP, based on AMD's new "Palomino" processor core, was similarly more mark than bite. Although it did have the long awaited new core, which is also known as "Athlon 4" because it is the fourth version of the Athlon, suffered from utterly stupid AMD marketing and unavailability of the product.
On June 12 2001 I attended the AMD Road Show when it passed through town here in Bellevue. The AMD Road Show is this ridiculous marketing event for computer builders and resellers (and me!) which as best as I can guess is supposed to excite people about new AMD releases. Unfortunately, the event was a complete dud! The Athlon MP had already been announced and released a week earlier, yet the chip was nowhere to be found. Even AMD's own reps failed to even bring a sample of the Athlon MP to the Road Show, and the representative from the only manufacturer of the dual processor Athlon motherboards, Tyan, failed to show up! So for 5 hours all the poor AMD marketing people could do was show slides from AMD's web site. The few Athlon MP chips that did get out were only clocked at 1.0 and 1.2 GHz, thus losing at single threaded benchmarks to the 1.4 GHz Athlon and 1.8 GHz Pentium 4.
As I told you they would, things have really heated up in these past few weeks leading up to the Windows XP launch. On September 10, the day before the horrific events, Intel bowed to pressure and released an SDRAM based chipset for the Pentium 4. The new i845 chipset uses good old PC133 SDRAM memory, not RDRAM. While this reduced the price of Pentium 4 systems to under $700 (a very good price) it also reduced the performance of the Pentium 4 even further, to levels comparable to plain old Celeron processors. As Tom's Hardware put it so eloquently,
"its lackluster memory performance impacts Pentium 4 so badly, that it makes AMD's Athlon an even more attractive solution than it already is... I personally would consider everyone as close to crazy if he should choose Pentium 4 plus i845 and PC133 SDRAM."
So with only a mild clock speed increase in the Pentium 4 during the past year (from 1.5 GHz to 2.0 GHz) and with these new cheap system that cripple its performance even further, the Pentium 4 is still not an intelligent choice for your new computer. But with manufacturers such as Dell dropping their Pentium 4 system prices to below $700 last month, this has put pressure on everyone to deliver faster cheaper products.
Athlon XP is finally here
On October 9 2001 AMD did so, launching both the 1.53 GHz "Athlon XP" processor and the 1.53 GHz Athlon MP. Unlike the original Athlon MP release, the Athlon XP was immediately available in stores. AMD also launched a rather questionable marketing campaign of showing up early in the morning and giving out free Athlon XP processor and motherboards in the middle of the street. They did this on October 12 here in Seattle, and I, the Emulators booth bunnies, and thousands of Seattle area computer geeks showed up as early as 2am on the Friday morning, in front of a downtown Starbucks of course, to get their free processor. Although we were 5th in line to receive an Athlon XP, and I thank AMD for giving us free stuff, I have to question to intelligence of this marketing. After all, who but a hard core computer geek who already appreciates AMD's products would show up at 2am or 5am to get a free Athlon processor in the first place? Certainly not the mainstream public, the people which Intel has done so well to brainwash with their bogus Pentium 4 marketing.
Speaking of brainwashing, there is a growing effort to stop Intel's false marketing of clock speed as an indicator of computing speed. While yours truly has been screaming, all together now, CLOCK SPEED IS NOT EVERYTHING for a long time, especially on this web site, major players such as Apple and AMD are now joining the anti-clock speed marketing effort. As I explained earlier, both the Motorola G4 processor (used in Apple's computers) and the AMD Athlon deliver more speed for a given clock speed. More efficiency. More instructions per clock cycle. However you think of it, more computer power for each MHz of processor clock.
Apple's education effort is called the Megahertz Myth, which you can read online at http://www.apple.com/g4/myth/. I certainly recommend that you read it, because it clearly shows just how ridiculous Intel's clock speed strategy is.
AMD has a similar education effort called Exposing the Speed Trap which you can read online at http://athlonxp.amd.com/ and I similarly recommend reading it.
AMD's web site goes on to explain why they also are using a rather bizarre numbering scheme for the Athlon XP. The 1.53 GHz processor for example is called the Athlon XP 1800 because it's supposed to remind you that it performs as well as other 1800 MHz (i.e. 1.8 GHz) processors. Personally, I find this marketing scheme to be insulting and very similar to Intel's own exploitation of clock speed. Sigh.
Don't forget the Pentium III - it isn't dead... yet!
Intel not surprisingly, has done absolutely nothing to educate the public about computer performance, and just keep beating the "oh look, we run at 2 GHz" drum. While Intel is doing everything it can to shove Pentium 4 down our throats, they've quietly resurrected the Pentium III! In fact, the Pentium III is now once again Intel's best processor without a doubt.
After the crash and burn of the Pentium III last year, where if you recall the .18 micron process based 1.13 GHz Pentium III chip was recalled, Intel has quietly been working on the Pentium III since then. They announced a new core this year, called Tualatin, which features the .13 micron process and 512K of L2 cache. A good year ahead of the Pentium 4's schedule!
While the news of a .13 micron Pentium III is good news, and will certainly help Intel migrate the rest of their chip manufacturing to the .13 micron processor, it also put Intel in an uncomfortable, yet familiar, situation. Early benchmarks of the chip indicated that it outperformed the Pentium 4! No big surprise, as even the 1.0 GHz chip did in some benchmarks. So imagine the horror that must have resonated through the halls at Intel at the thought of the Celeron / Pentium II price inversion happening again. That is, the "slower" cheaper processor outperforming the "faster" more expensive processor.
So what did Intel do? Two things: they crippled the Tualatin by cutting its on-chip L2 cache from 512K to 256K. And they make a minor modification that renders the chip unusable on existing Socket 370 motherboards. In other words, if you want Tualatin badly enough you will have to put up with having to buy yet another motherboard and settle for a crippled processor.
Well, for this page, I wanted it badly enough! So finally, after months of waiting, months of hype from Intel and AMD, months of price cuts, months leading up to the launch of Windows XP, we have finally arrived at THE MONTH, XP MONTH, October 2001, when we can see how Intel's latest Pentium III compares against the Pentium 4, how the Athlon XP compares to its predecessors, and how well all these chips do on the new Windows XP operating system.
The contenders
I mainly ran tests on 4 different machines: the original 1.5 GHz Pentium 4 machine from Gateway, the 1.2 GHz Athlon Thunderbird machine which won most of the benchmarks in the March 2001 tests, and the two new freshly built Intel Pentium III Tualatin and AMD Athlon XP 1800 machines. Older machines you can just look at the March 2001 results. The specs of the machines are listed here. The machines were all outfitted with 512 megabytes of RAM, 60 gigabyte IDE hard disks, and partitioned similarly with both Windows Millennium (build 3000) and Windows XP Home Edition (build 2600) installed in separate partitions. The Athlon XP and the Pentium 4 shared the same ATI Radeon card. The Pentium III-M used the built-in Intel AGP 4x video.
AMD Athlon (Thunderbird) | AMD Athlon XP 1800 (Palamino) | Intel Pentium III-M (Tualatin) | Pentium 4 | |
Clock speed | 1200 MHz | 1533 MHz | 1200 MHz | 1500 MHz |
Memory size and type | 512 MB PC2100 DDR SDRAM | 512 MB PC2100 DDR SDRAM | 512 MB PC133 SDRAM | 512 MB PC800 RDRAM |
Total L1 cache | 128K | 128K | 64K | 8K |
Total L2 cache | 256K | 256K | 256K | 256K |
Motherboard | ALi | AMI K7T266 Pro-RU | Asus TUSL-2 | Gateway |
Chipset | MAGiK | VIA KT266 | i815E | i850 |
Approximate cost as built | $900 | $1000 | $900 | $4300 (Nov. 2000), $1200 today? |
I did NOT test an i845 based system, as any number of web sites have posted benchmark results that clearly (and obviously) show PC133 based Pentium 4 systems to run even slower than RDRAM systems. Tom's Hardware said it best.
I also did NOT waste my time putting together a 2.0 GHz Pentium 4 system based on RDRAM. Two reasons: the 2.0 GHz processor uses the same flawed design and .18 process of the original 1.5 GHz Pentium 4 and thus at best delivers about a 33% performance boost. It has been the goal of past benchmarks to test different architectures and processor revisions, not to test 2 similar clock speeds of the same processor. Second, Intel has already made obsolete the Socket 423 based Pentium 4 chips by introducing the Socket 478 which will be used for the upcoming 2.2 GHz Pentium 4 and future Pentium 4 releases. These future Pentium 4 releases (due out next year) are rumored to have 512K of L2 cache and are based on a .13 micron process. THAT will be worth taking a look at.
For now, you may simply take the existing 1.5 GHz Pentium 4 numbers and fudge them by a factor of 1.33 to extrapolate 2.0 GHz results, or by 1.47 to extrapolate possible 2.2 GHz results. You will see that applied to the current benchmark results, the Pentium 4 would still lose the benchmarks!
Emulation benchmarks revisited
Running the same Windows Millennium based SoftMac 8.02 emulation tests as before, I've added the new results for the Athlon XP and Pentium III-M to the Benchmarks page. I've copied the relevant numbers here:
AMD Athlon 1.2 GHz Win Me 512M | AMD Athlon XP 1.53 GHz Win Me 512M | Pentium III-M 1.2 GHz Win Me 512M | Pentium 4 1.5 GHz Win Me 512M | |
Word 5.1 Word Count (rounded to nearest second) | 4 s | 3 s | 4 s | 6 s |
Photoshop Load 2.2M JPG file (rounded to nearest second) | 19 s | 17 s | 19 s | 23 s |
Floating Fractals 1000 iteration rendering (displayed time in seconds) | 27.37 s | 23.0 s | 26.4 s | 82.0 s (not a typo!) |
Power Demo DayStars demo, default settings (displayed time in seconds) | 1.48 s | 1.18 s | 1.43 s | 2.37 s |
Apple Personal Diagnostics Tests | ||||
Integer | 232.1 | 317.8 | 252.0 | 236.3 |
Complex Functions | 1050.0 | 1327.0 | 1049.0 | 527.0 |
Floating Point | 550.1 | 697.4 | 579.3 | 388.2 |
Memory | 214.6 | 298.5 | 251.2 | 299.4 |
Processor | 84.8 | 118.0 | 92.9 | 96.0 |
Video | 45.0 | 58.5 | 43.9 | 30.2 |
Speedometer Tests | ||||
CPU | 133.1 | 173.0 | 192.2 | 173.0 |
Graphics | 81.5 | 99.9 | 81.5 | 62.4 |
Disk | 109.3 | 139.8 | 102.1 | 109.3 |
Math | 487.0 | 615.4 | 536.7 | 346.4 |
FPU KWhetstones | 20000 | 30000 | 20000 | 12000 |
FPU Matrix Multiply (seconds) | .300 s | .233 s | .300 s | .467 s |
FPU FFT (seconds) | .133 s | .100 s | .133 s | .233 s |
FPU Average | 24.0 | 33.2 | 24.0 | 14.7 |
Mix KWhetsones | 5454 | 6666 | 6000 | 3333 |
Mix Dhrystones | 88235 | 11111 | 96774 | 73170 |
Mix Average | 221.6 | 278.4 | 236.3 | 174.1 |
Video Average | 36.5 | 46.2 | 36.7 | 27.8 |
While I certainly expected the Athlon XP to win the emulation benchmarks hands down, it is still pretty amazing to see the huge margin by which it trounces the Pentium 4. At the same clock speed, roughly 1.5 GHz, the Athlon just eats the Pentium 4 for lunch. As you can see for yourself, the margin of victory in most of these tests is anywhere from an even tie to a good 50% to 100% faster. May I remind you again that CLOCK SPEED IS NOT EVERYTHING and that the code tested here (SoftMac 8.02 and Windows Millennium) is code which was written last year! Certainly not hand coded for the Athlon XP. Another demonstration of how the Athlon processor does a great job of running existing Windows code without any code rewrites.
Windows XP performance
With the launch of Windows XP this week, you'll be seeing a lot of Intel ads telling you how wonderful the Pentium 4 is for running Windows XP. The problem is I know this to not be the case. Since last December, I've been running kernel level tests on both Windows Millennium and Windows XP and consistently find the Pentium 4 to be slower than both the Pentium III and Athlon at basic kernel operations such as task switching. Both in terms of raw time and in terms of raw clock cycles. And since Microsoft's C++ compiler is known to not produce Pentium 4 optimized code, I can't possibly imagine Windows XP being tuned for the Pentium 4!
So I wrote a simple test app to simulate the kind of activity that goes on in a multi-tasking operating system such as Windows XP. In a single tasking operating system such as MS-DOS, programs spend most of their time in tight loops doing something (whether waiting for keyboard input, calculating something, drawing an image, etc.). When multi-tasking is throw in, multiple programs (or "tasks") are running at any given moment, and within each tasks can be multiple "threads", each thread being a different piece of the programs that is executing concurrently.
Each thread within each task gets assigned a certain amount of private memory (the "stack") and can also share memory with other threads and tasks. Threads and tasks communicate with each other by means of messages which are passed back and forth (messages such as "paint your window" or "here is a mouse click over a windows that belongs to your thread").
The operating system kernel itself co-ordinates all of this, giving short periods of time to each thread, a "time slice", and handling the queuing and dispatching of the messages to each process. Every time a thread in a Windows application makes some operating system call, such as allocating memory, passing or receiving a message, reading or writing a file, a "task switch" occurs from that thread into the operating system kernel. Or from that thread to another thread.
Other operations, such as reading and writing memory within a thread can causes task switching, since the operating system may have swapped that memory out to disk. When this swapped memory is accessed, a "page fault" occurs, and again causes the operating system to juggle some threads in order to read the memory back in and resume the thread which faulted.
The simple act of say, launching Internet Explorer 6.0 and bringing up the cnn.com web site starts no less than 18 threads in the IEXPLORE.EXE task and uses over 15 megabytes of memory. That's a lot of page faults, a lot of messages being passed around, and a lot of task switching going on. You can see the operating system "at work" by bringing up the Task Manager windows in Windows 2000 or Windows XP.
Operating system designers spend a lot of effort to optimize the cost of a task switch, which on Pentium III and Pentium 4 processors can be in the thousands of clock cycles. Multiply that by thousands or even tens of thousands of such task switches per second, and it can start to add up to a significant percentage of the processor time. The Windows XP kernel, really the Windows NT kernel, can it really be optimized for Pentium 4 even though most of the code was written years ago?
To simulate a typical heavy load on the operating system I wrote a simple test application which creates a given number of threads and then proceeds to pass one million 256-byte messages between those threads. In one version it creates 10 threads and has each thread send 100,000 messages, and in another version it creates 2000 threads and passes 500 messages. When all 250 megabytes of messages have been received, the elapsed time is displayed.
This is not a test to see how fast the processor can count to a million. This is a test of the Windows XP kernel code at doing what it is designed to do - create threads, pass messages, task switch. Given that Windows XP when it is simply idling runs about 15 tasks and 250 threads (yes, just sitting there at a blank desktop!) the 10 thread test really won't put much of a load on the system and should execute faster than the 2000 thread case.
So, here are the numbers. Numbers are in seconds, and thus smaller is faster:
AMD Athlon (Thunderbird) | AMD Athlon XP 1800 (Palamino) | Intel Pentium III-M (Tualatin) | Pentium 4 | |
Clock speed | 1200 MHz | 1533 MHz | 1200 MHz | 1500 MHz |
10 thread test (seconds) | 5.1 | 4.0 | 8.8 | 9.6 |
2000 thread test (seconds) | 7.1 | 5.7 | 9.7 | 10.7 |
The ratio of the Palomino and Thunderbird numbers is roughly the 27% difference in their clock speeds. All else being equal (same amount of PC2100 memory, same OS, etc.) this result is to expected. But look at the Pentium 4 numbers! The Pentium 4 requires roughly double the overhead of the Athlon XP, at the same clock speed. How this shows that Windows XP is optimized for the Pentium 4 is beyond me.
To run a more real world test, I did something I do daily, which is to build one of our emulator engines. In this case I built the basic 68000 engine used in Gemulator 2000. This consists of about 20,000 lines of assembly language code which generates about 400K of code and data. This test was run on each machine with no other windows or foreground tasks running. The test was repeated several times on each machine and the lowest time recorded.
The numbers. Again, smaller is faster:
AMD Athlon (Thunderbird) | AMD Athlon XP 1800 (Palamino) | Intel Pentium III-M (Tualatin) | Pentium 4 | |
Clock speed | 1200 MHz | 1533 MHz | 1200 MHz | 1500 MHz |
68000 engine build time | 65 | 58 | 81 | 78 |
MPEG encoding revisited again
I ran the MPEG encoder again, on Windows Me on both the Athlon XP and Pentium III-M and added the numbers to the previous results for the other processors. The numbers offered no surprises, being roughly inversely proportional to clock speeds of slower Athlon and Pentium III processors. As with the emulation tests, the 1.2 GHz Pentium III-M edged slightly past the 1.2 GHz Athlon Thunderbird. The complete list so far:
Chip speed and type | Elapsed time (seconds) | Clock cycles (billions) |
1.53 GHz AMD Athlon XP DDR | 321 | 491 |
1.2 GHz Pentium III-M | 399 | 478 |
1.2 GHz AMD Athlon DDR | 413 | 496 |
1.0 GHz Pentium III Xeon (single) | 473 | 473 |
1.5 GHz Pentium 4 | 484 | 726 |
1.0 GHz Pentium III Xeon (dual) | 520 | 520 |
900 MHz AMD Athlon DDR | 535 | 482 |
900 MHz AMD Athlon | 544 | 490 |
670 MHz Pentium III (single) | 680 | 456 |
670 MHz Pentium III (dual) | 743 | 498 |
650 MHz Pentium III | 757 | 492 |
533 MHz Celeron | 858 | 457 |
600 MHz AMD Athlon | 922 | 553 |
500 MHz Pentium III | 946 | 473 |
600 MHz Crusoe | 1369 | 821 |
Prime95 revisited again - the Pentium 4 finally wins something!
One of my favorite utilities to use as a benchmark, Prime95, was revised a couple of months ago (version 21) to include special Pentium 4 specific code and additional speed optimizations for Pentium III and Athlon processors. The chart below contains the original numbers from March 2001, with new numbers added in with "v21" specified. I've also specified "PIII mode" and "P4 mode" to specify when the Pentium 4 specific code was and was not used.
Chip speed and type | Per iteration time (seconds) |
1.5 GHz Pentium 4 RDRAM (v21 - P4 mode) | .057 |
1.53 GHz AMD Athlon XP DDR (v21) | .075 |
1.2 GHz AMD Athlon DDR (v21) | .087 |
1.2 GHz AMD Athlon DDR | .117 |
900 MHz AMD Athlon DDR | .136 |
1.5 GHz Pentium 4 RDRAM | .136 |
1.2 GHz Pentium III-M SDRAM (v21) | .152 |
1.5 GHz Pentium 4 RDRAM (v21 - PIII mode) | .161 |
1.0 GHz Pentium III Xeon RDRAM | .181 |
670 MHz Pentium III SDRAM | .232 |
When running the generic code, the Pentium 4 again loses to the Athlon and even to the Pentium III-M. Only when the Pentium 4 specific code is activated, which uses Pentium 4-only instructions, does the Pentium 4 finally break out and win for once.
This is perfect example of the dilemma I mentioned last December, that being that the "unique" characteristics of the Pentium 4 will require that software developers not only rewrite code but in some cases write a second piece of code specific to the Pentium 4. Now since the core Fast Fourier Transform routines in Prime95 were written in hand optimized assembly code already, it was (I assume) a relatively easy matter for George to add the Pentium 4 optimized code. Being hand coded, the Prime95 code is not constrained on waiting on C/C++ compilers to be updated to generate Pentium 4 specific code. Although it did take 9 months before this new release of Prime95 was released, just as it took most of the first half of this year for SoftMac to get rewritten and optimized better for the Pentium 4.
It will remain to be seen how well this success story will translate to the thousands of other Windows applications out there and how many software developers will be willing to invest the months of effort in a code rewrite that will benefit little else but the small percentage of Pentium 4 users. Microsoft has still not released the version 7.0 of its C/C++ compiler (with which many Windows applications are built) and it seems to me that even Windows XP is not optimized yet for Pentium 4.
I've said it before, I'll say it again: Intel can save a lot of people a lot of trouble by just fixing the damn chip.
Athlon XP delivers as expected
So how did the new Palomino core compare to the older Thunderbird core? Adjusting for the 27% clock speed difference between the 1.2 GHz and 1.53 GHz chips, most of the Athlon XP numbers are exactly in line with that ratio, roughly 25% to 30% faster. The Palomino core in the Athlon XP does not deliver some great new leap in efficiency. However, it does do an excellent job of scaling the performance of the Athlon.
The other major leap in the Palomino core is significantly lower power consumption and heat generation. AMD I believe claims a 20% reduction in power over the Thunderbird core. I tested this by over-clocking the chip.
Although the Palomino is based on a .18 micron process, I was able to easily over-clock the chip to about 1700 MHz with no problems, using the jumperless BIOS on the AMI motherboard. Running at the 1700 MHz clock speed (295 MHz FSB), and using only the processor fan, the chip did not overheat or exceed 50 degrees Celsius. Both Windows Me and Windows XP booted and ran fine at 1700 MHz and delivered proportionally faster benchmark numbers.
By comparison, the 1.2 GHz Athlon machine, with 2 additional fans, begins to fail above 1.26 GHz.
So AMD has achieved two significant goals here - scaling the performance of the Athlon to above 1.5 GHz, and keeping the power consumption low enough to allow for decent over-clocking without the use of additional cooling.
The chip is also priced to sell at about the $200 price point, well below the $600+ price of a comparable 2.0 GHz Pentium 4 chip.
The other beauty of the Athlon XP is that it maintains pin compatibility with existing Socket A motherboards. If you currently have a Duron or Athlon Thunderbird based system, an upgrade to Athlon XP is simply a matter of replacing the processor. The new Pentium III-M by comparison, and future releases of the Pentium 4, require new motherboards and thus add about $150 to the price of those upgrades.
My two disappointments with the chip have to do with marketing. First, the ridiculous naming, i.e. calling it the "1800". I think this will confuse and possibly offend consumers, as this will appear too much to be an Intel type of clock speed scam. I personally just preferred the way Athlon processors have been sold the past two years - by clock speed along and letting people reach their own conclusions.
My other disappointment has to do with this QuantiSpeed Architecture jargon. What the hell is that? AMD eludes to a 10% to 15% speed gain, on TOP of clock speed increases, over the Thunderbird core. From the various results I got, I don't see this. The Athlon XP delivers the expected 27% or so speed increase over a 1.2 GHz Athlon, but not a 40% increase. It just isn't there in any of the numbers, and I believe that AMD is blowing some hot air here.
Nevertheless, the Athlon XP 1800 is a good solid chip at a reasonable price and it is quite possibly the fastest x86 processor available today.
I also forgot to mention (thanks to Scott Reams for reminding me) that the Athlon XP does add the SSE instructions introduced in the Pentium III. Unlike past Athlon processors (which were compatible with Pentium II and AMD-K6 instruction sets) the Athlon XP is fully compatible with Pentium III optimized code.
SHOCK! Pentium III is far from being dead!
The biggest surprise to me was not the expect speed of the Athlon XP, but rather the great performance of the Pentium III-M (Tualatin). Barely 6 months ago the fastest Pentium III processor available was the very expensive (and very disappointing) 1.0 GHz Pentium III Xeon using PC800 RDRAM. I did not expect the new 1200 MHz Pentium III-M to deliver much faster speed than the Xeon, especially considering that the Xeon used PC800 RDRAM while the Pentium III-M used an Intel 815 PC133 SDRAM based motherboard.
But the numbers speak for themselves. The Pentium III-M, despite the fact that Intel hobbled the larger 512K L2 cache down to 256K, still delivers a solid 20% speed increase over the 1.0 GHz chip, and does so using the slowest and most common PC133 memory.
In fact, a close examination of the emulation results and the MPEG encoding shows the 1.2 GHz Pentium III-M matching or even beating the 1.2 GHz Athlon Thunderbird. This result just amazes me. The Pentium III architecture, as old as it is, scales very nicely and keeps right up with the Athlon.
It is unfortunate for Intel that their older .18 micron based Pentium III Coppermine processors hit a brick wall at 1.0 GHz in August 2000 and that it took Intel a full year to retool the chip on the .13 micron process. This gave AMD the clear lead, and now Intel is about 6 months too late to win the 1.2 GHz battle.
One major feature of the Pentium III-M, as with the Athlon XP, is that its .13 micron core runs cooler than previous Pentium III chips. I was able to over-clock the chip, again using the BIOS on the AMI motherboard, to a full 150 MHz FSB and 1350 MHz clock speed. A 12.5% over-clocking, using no additional fans, and more so than the 5% over-clock I was able to achieve with the Thunderbird and 11% over-clock with the Palomino. I ran the chip at 1.35 GHz for over 24 hours with Windows XP and various looping benchmarks with no problems. After the 24 hours, the chip was still running at a cool 45 degrees Celsius.
My two disappointments with the chip is that I tested the chip with PC133 memory, which did not allow me to see the full potential of the chip. My fault, and I'll try to re-test the chip with a DDR based chipset in the future. The other disappointment is that a dual-processor board is not yet available for the Pentium III-M, since it is incompatible with existing dual Socket 370 boards. Again, this is something that will hopefully be fixed in the future before Intel drops the ax on the chip.
Processor Summary
I'll bring this round of benchmarks to a close by summarizing the basic features and differences of today's gigahertz class x86 processors, as of October 2001. If anyone has any additions or corrections for this chart let me know right away.
Athlon (Thunderbird core) | Athlon XP (Palomino core) | Pentium III-M (Tualatin core) | Pentium 4 (Socket 423) | Pentium 4 (Socket 478) | |
Clock speed | up to 1400 MHz | up to 1533 MHz | up to 1200 MHz | up to 2000 MHz | up to 2000 MHz |
Memory types supported | PC133 SDRAM PC2100 DDR SDRAM | PC2100 DDR SDRAM | PC133 SDRAM | PC800 RDRAM | PC800 RDRAM PC133 SDRAM |
Total L1 cache | 128K | 128K | 64K | 8K | 8K |
Total L2 cache | 256K | 256K | 256K | 256K | 256K |
Street price (USD) | $100 for 1.0 GHz $150 for 1.4 GHz | $200 for 1.33 GHz $300 for 1.53 GHz | $350 for 1.2 GHz | $200 for 1.5 GHz $650 for 2.0 GHz | $200 for 1.5 GHz $650 for 2.0 GHz |
Fabrication process | .18 micron | .18 micron | .13 micron | .18 micron | .18 micron |
Dual processor capable | Not officially (but known to work) | No (??? Anyone tried???) (supported by the virtually identical Athlon MP) | Not yet that I know of (lack of motherboards) | No (supported by the virtually identical Xeon) | Not yet that I know of |
Advantages | Inexpensive and fast. 1.8 GHz Pentium 4 performance with most existing Windows applications, at a fraction of the cost of the Pentium 4. Works with existing Socket A motherboards. | Compatible with existing Duron and Athlon Socket A motherboards. Fastest x86 processor for running Windows XP and most existing Windows applications. Adds Pentium III SSE instructions. Runs cooler, does not require special cooling. Easily over-clocks by 10%.
| Fastest Pentium III ever. Runs very cool, smallest fan and heat sink of any of the processors. Easily over-clocks by more than 12%. | The fastest processor for running the current version of PRIME95. SSE2 instructions can deliver faster speed in some applications. | Cheaper (but slower) systems due to the use of industry standard SDRAM. |
Disadvantages | Limited to compatibility with Pentium II MMX instruction set Runs very hot, needs extra cooling fans | No support for Pentium 4 SSE2 instructions (which at this point is a "who cares" due to severe lack of Pentium 4 optimized software) Really really dumb naming scheme may confuse shoppers. | Not compatible with existing Celeron and Pentium III Coppermine motherboards and this requires a new motherboard Intel chose to cut the L2 cache from 512K to 256K in order to not tread on Pentium 4's turf Intel intending to kill the chip end of this year? | Design flaws limit the performance to about 70% to 80% of the speed of Pentium II and Athlon (at identical clock speeds). Not available for use with SDRAM. SSE2 specific code is not compatible with any other existing processor. The processor is already obsolete due to the switch to Socket 478. All existing Pentium 4 motherboards using RDRAM will need to be replaced to use future Pentium 4 processors. Runs very hot.
| Same crippling design flaws as Socket 423 version. Crippled further by current SDRAM based motherboards. The most ridiculously expensive processor for use with SDRAM. Intel is being slow to develop a DDR SDRAM based chipset. Lack of mainstream software that is optimized for Pentium 4 or SSE2. |
Future outlook | BLEAK! Replaced by cooler-running pin-compatible Athlon XP. Don't even considering buying one at this point. | GOOD .13 micron process some time next year | BLEAK! Pity, it's a damn good processor but its short life span gives most Pentium III users little reason to update to one | BLEAK! Replaced by Socket 478 version. Don't even considering buying one at this point. | GOOD .13 process with rumored 512K L2 cache some time next year. Will be a far more attractive processor coupled with DDR SDRAM and .13 micron based speeds beyond 2.0 GHz |
Conclusion - sorry Intel, I still have to recommend the Athlon
The Pentium 4, once it is switched over to a .13 micron process and beefed up with a large on-chip cache, may finally show promise some time next year. But as it stands, I cannot recommend to anyone to purchase an existing Pentium 4 system, Socket 423 or Socket 478 based. Given that next year's Pentium 4 will likely use a DDR SDRAM chipset, any investment in Pentium 4 today is money wasted. You'll be buying an new processor, new motherboard, and new RAM a year from now.
And almost a year after the Pentium 4's release, there is still a huge shortage of applications optimized to use the Pentium 4, and even of compiler tools to generate Pentium 4 optimized code. Windows XP runs best on everything but Pentium 4. Microsoft and GNU compilers still produce code that runs better on Pentium III and Athlon. Most off-the-shelf applications, the programs that you actually use today, run better on Pentium III and Athlon. Pentium 4 - just stay away from it for the time being.
The new Pentium III-M (with its Tualatin core) really surprised me, because it blows the doors off previous Pentium III processors and even leaps ahead of similar Athlon processors of a few months ago. Being the first .13 micron based x86 processor, it shows the promise of the new fabrication processor be very clearly exhibiting faster clock speed, cooler operation, and lots of headroom for over-clocking.
Yet Intel, in their infinite stupidity, keeps trying to kill off the Pentium III and ramming Pentium 4 at as. And they discourage people from upgrading to the latest Pentium III-M by first developing a 512K on-chip cache and then crippling it to 256K, and also by making a minor modification to make it incompatible with all existing Pentium III and Celeron motherboards. Rumor now is that the Pentium III will cease to be made in a couple of months. WHY????
Given what I've seen so far, I have no doubt that the .13 micron Tualatin core can scale we past 1.5 GHz. Intel could very likely release a 1.5 GHz Pentium III tomorrow if they wanted to and completely catch up to AMD's Athlon XP and outperform the Pentium 4 at the same time. Yet they won't, because some marketing type has decided that Pentium 4 is the future. Bad design or not, they're going to promote Pentium 4. A Pentium III chip that is cheaper and faster than a Pentium 4 is certainly not something Intel wants to promote, even though it would greatly benefit consumers by means of easier upgrades and cheaper system costs.
So despite the Pentium III being the first to be manufactured using a .13 micro process, and despite the popularity of the Pentium III, Intel is likely still going to kill it off anyway. As much as I like this chip, there is little incentive for the average computer user to go spend $500 on a new Pentium III-M processor and motherboard, only to have it be killed off by the end of this year.
AMD Athlon XP, numbering scheme aside, is the best choice for computer buys today. Costing less than Pentium 4 and using existing Athlon motherboards makes upgrading from earlier Athlon system trivial and inexpensive. For new systems, the cost of the Athlon XP processor, motherboard, and RAM comes in at about the same price as a Pentium III-M configuration, but delivering about 25% more speed. And when AMD does release a .13 micron based Athlon next year, upgrading again will be a trivial matter of only replacing the processor.
It should come as no surprise then that I declare the Athlon XP the winner of Round 3!
Note from author, November 1 2007: Fast forward to the summer of 2007. After the end of 2001, AMD's magic touch and technological lead of the PC market ran out. Over the next 5 years AMD's technology faltered as Intel reversed course and did in fact revive the Pentium III as I predicted above they could (and ultimately would) do. The release of the "Centrino" branded notebook computers in 2003, featuring the Pentium III-derived "Pentium M" microprocessor marked a new beginning for Intel, which fully came to fruition in the summer of 2006 with the release of the desktop equivalent of the Centrino - the 64-bit multi-core Intel Core 2. The Core 2 is significantly faster and more efficient than both the Pentium III and Pentium 4 which it replaced, and is even up to 50% faster than AMD's equivalent chips. The PC landscape completely changed in 2006 as well as what is now possible and affordable to do with PCs.
Yet consumers are still not out of the woods, as security, reliability, and malware issues plague computer owners more so than ever. In the rush to outperform each other in benchmarks, both AMD and Intel have failed to address some fundamental design issues that are costing the economy billions of dollars in lost productivity, down time, and security breaches. It makes the whole Pentium 4 disaster moot in comparison. Please read my newer analysis of the PC market in my new blog...
Apple, Mac OS, Macbook, Apple Silicon and Macintosh are registered trademarks of Apple Computer, Inc. Atari is a registered trademark of Atari U.S. Corporation. Athlon, Athlon XP, Opteron, Phenom, and Ryzen are registered trademarks of AMD. Arm is a registered trademark of Arm Limited. Microsoft, Windows, Windows NT, Windows XP, Windows Vista, Windows 7, Windows 8, Windows 10, Windows 11, Visual Studio, .NET, and/or other Microsoft products referenced herein are either trademarks or registered trademarks of Microsoft. Intel, Pentium, Core 2, Core i7, and Atom are registered trademarks of Intel. PowerPC is a trademark of IBM. Additional company and product names may be trademarks or registered trademarks of the individual companies and are respectfully acknowledged.