Original Link: https://www.anandtech.com/show/887
AMD's Athlon XP 2100+ The Last of the Palominos
by Anand Lal Shimpi on March 13, 2002 12:47 AM EST- Posted in
- CPUs
With all of the excitement surrounding AMD's first demonstration of functional ClawHammer silicon, who wants to talk about the Athlon XP anymore? Although the next-generation Athlon processor is already up and running two OSes it is far from complete. There are many more revisions the silicon must go through (it's not 100% functional yet and isn't operating anywhere near its shipping frequency) and then there's the task of validating the part (making sure it runs every piece of software with every piece of hardware out there flawlessly, or close enough). The time for talking about the ClawHammer will come soon enough, but before then AMD has a few more Athlon XP's up their sleeves.
Today being the first day of CeBIT, AMD is announcing the latest model of their Athlon XP line - the 2100+. Just like the launch of the 1900+ and the 2000+, the 2100+ brings nothing more to the table outside of another 67MHz. It does mark the end of the line for the Palomino core however, as the Thoroughbred based Athlon XP 2200+ will replace it using AMD's new 0.13-micron manufacturing process.
On the Road Again
In order to kick off this review of the final 0.18-micron Palomino based Athlon XP we felt it necessary to start off with AMD's current desktop CPU roadmap to put things into perspective.
The roadmap you see above is publicly available from AMD's website and it explains quite a bit about what to expect from AMD in the coming months. With the majority of the most active AnandTech readership currently running AMD processors an idea of upgrade paths would definitely be useful.
If your motherboard currently supports the Athlon XP processor then the highest you'll definitely be able to upgrade to will be the 2100+ that we're looking at today. Whether you'll be able to upgrade to the Thoroughbred core or not is still a bit foggy; the reason being that AMD themselves haven't finalized exactly what the Thoroughbred core will come equipped with.
Since Thoroughbred will be the first 0.13-micron core, AMD could very well equip it with more L2 cache. One potential barrier to this would be AMD's desire to keep manufacturing costs as low as possible. Remember that AMD is still manufacturing on 200mm (8") wafers and thus they want to keep chip sizes as small as possible. Granted that on a 0.13-micron process, boosting L2 cache size to 384KB or 512KB (in addition to the 128KB L1) won't add too much bulk to the die but in the end it's a question of economics.
Rumors of higher speed FSBs for the Thoroughbred cores have also been running rampant, and it's definitely an avenue of performance improvement that AMD is considering. The Thoroughbreds will be put up against Intel's 533MHz FSB Northwoods which already have a 512KB L2 cache, moving to a 166MHz FSB (double pumped) could potentially make those CPUs much more competitive (we'll investigate this issue at the end of the article). It would also give AMD a chance to really begin pushing DDR333 but from what we're hearing, the premium that DDR333 will carry throughout this year will be a bit difficult for most to swallow, especially considering the flurry of complaints regarding recent memory price hikes.
Then there's the issue of what happens after Thoroughbred; at this point, the future of the 0.13-micron SOI Barton core is up in the air. The point of Barton was always a SOI enablement part for AMD, just to prepare things for the next-generation Athlon and server parts which would ship in volume using this process. It seems as if AMD is quite confident in the initial Hammer silicon thus causing them to rethink whether Barton is necessary at all.
Where does this leave the Duron? The CPU that originally entered this market as a Celeron killer has lost much of its meaning courtesy of very aggressive pricing on AMD's Athlon lines. In the time between now and the release of the first desktop Hammer CPUs (ClawHammer) it is unclear to us (and AMD) what role the Duron will play, but it is clear that after the end of this year the Duron as we know it will eventually be replaced by some derivative of a 0.13-micron and eventually a 0.09-micron Athlon XP core. This is definitely necessary as the next-generation Celeron will be Pentium 4 derived with a healthy 400MHz FSB, more than capable of feeding the execution units of that core.
There you have it, that's your Socket-A upgrade path for the rest of this year. The line ends here with Palomino and it'll pick up later in Q2 with Thoroughbred. While it's pretty obvious that as the last 0.18-micron Athlon XP CPU, the 2100+ is not the most desirable, we've put together a comparison of the majority of the XP line to help you pick which CPU is right for you so you don't feel stuck between a Thoroughbred and a hard place.
Modeling a Stylish 2100
As we briefly mentioned at the start of this article, the Athlon XP 2100+ is a generic speed bump in the simplest of terms. The CPU is architecturally identical to all other Palominos and has power characteristics very similar to the Athlon XP 2000+. The reason being that there's only a 67MHz difference between the 2100+ and the 2000+ (1.73GHz vs. 1.67GHz).
With only a 4% difference in clock speed you can't expect to see much of a tangible performance difference between the 2100+ and the 2000+. In fact, the only visible difference between the two chips is that AMD has gone to a different dye color for the packaging substrate.
From now on, all Athlon XPs will feature this green substrate which is reminiscent of the FC-PGA Pentium III CPUs. This change in substrate dye won't change performance at all, but it's always nice to see something different in our CPU drawers.
The Test
|
Windows XP Test System |
|||||||
|
Hardware |
|||||||
|
CPU(s) |
AMD
Athlon XP 1.73GHz (2100+) AMD Athlon XP 1.67GHz (2000+) AMD Athlon XP 1.53GHz (1800+) AMD Athlon XP 1.40GHz (1600+) AMD Athlon-C 1.40GHz |
Intel Pentium 4 2.2GHz Intel Pentium 4 2.0AGHz Intel Pentium 4 2.0GHz Intel Pentium 4 1.8GHz Intel Pentium 4 1.6GHz |
|||||
| Motherboard(s) | EPoX 8KHA+ | ABIT TH7-II RAID (Intel 850) | |||||
| Memory |
256MB
PC800 Mushkin RDRAM |
||||||
| Hard Drive |
Maxtor D740X Ultra ATA/133 80GB HDD |
||||||
| CDROM |
Phillips 48X |
||||||
| Video Card(s) |
NVIDIA GeForce4 Ti 4600 128MB DDR |
||||||
| Ethernet |
Linksys LNE100TX 100Mbit PCI Ethernet Adapter |
||||||
|
Software |
|||||||
|
Operating System |
Windows XP |
||||||
| Video Drivers |
|
||||||
Content Creation Performance
We start out our benchmark analysis with some content creation tests as usual. The first of the bunch is the 2002 edition of Ziff Davis Media's Content Creation Winstone. As with previous versions of Content Creation Winstone, the premise of the test is to run through a script of a handful of applications (in this case 7) and perform various tasks in the applications while switching between one or more of them. The idea is to simulate a present-day multitasking environment using content creation applications, the 7 application tested are listed below:
Adobe Photoshop 6.0.1
Adobe Premiere 6.0
Macromedia Director 8.5
Macromedia Dreamweaver UltraDev 4
Microsoft Windows Media Encoder 7.01.00.3055
Netscape Navigator 6/6.01
Sonic Foundry Sound Forge 5.0c (build 184)
| Content
Creation Performance Content Creation Winstone 2002 |
|||||||||||||
|
33.8The extra 4% boost in clock speed of the 2100+ results in a small and unnoticeable 2.7% increase in performance in CCWS2K2. This gives the Athlon XP the edge over the Pentium 4 2.2GHz but keep in mind that performance differences less than 10% aren't tangible. That means that in this test you'll be hard pressed to notice any difference between the Athlon XP 1600+ and the Athlon XP 2100+.
For most business applications and disk-intensive content creation applications you'd be better suited buying a lower end Athlon XP 1600+ and going for a faster IDE drive and more memory. It's interesting to note that the old Pentium 4 2.0GHz CPU is only as fast as the Athlon XP 1600+ in this test.
Now with enough CPU speeds to look at performance trends as clock speed increases we can include CPU scaling graphs of all of our benchmarks. We graphed the benchmark score against the clock frequency or speed rating of the CPUs. For the speed steppings in-between those we benchmarked we performed linear interpolation to form a straight line.
Here you can see that the Athlon XP has always held a performance advantage over the Pentium 4. It's only the introduction of the Northwood that allows the Pentium 4 to begin to catch up to the Athlon XP in this test. By the looks of the graph it will take a 2.4GHz Pentium 4 for the Northwood's performance curve to intersect the Athlon XP's curve.
SYSMark 2002
With this review we also use SYSMark 2002 for the first time in a CPU review. SYSMark 2002 can be considered to be a much more memory bandwidth intensive version of the Winstone tests. The benchmark is split into two parts, Internet Content Creation which deals with content creation applications (Photoshop, Dreamweaver, etc...) and Office Productivity which is more general usage oriented (Word, Excel, Netscape, Anti-Virus, etc...).
The 2002 update changes things around a bit; first of all the benchmark's total scores are arrived at differently than in the 2001 benchmark. Windows Media Encoder no longer accounts for close to half of the Internet Content Creation test, rather only about 10%. There is also no need for a special Athlon XP SSE patch as the 2002 suite uses a version of the encoding dll that properly detects SSE support on all Palomino cores as well as Pentium 4 cores.
The rest of the benchmark is much more evenly distributed and it is much more memory bandwidth intensive than the old benchmark. The Internet Content Creation tests on average use about 600MB/s of bandwidth vs 300MB in SYSMark 2001. The Office Productivity tests are still stuck at around 580MB/s of memory bandwidth.
For more information on the tests and the applications used consult this whitepaper provided by BAPCo.
| Internet
Content Creation Performance Internet Content Creation SYSMark 2002 |
|||||||||||||
|
Now that the ICC test suite is much more memory bandwidth intensive it's much more sensitive to larger caches. This gives the Northwood a significant advantage over even the Pentium 4 in this test.
The Athlon XP 2100+ does relatively well as it's outperformed by the Pentium 4 2.0GHz by only a 6.6% margin. The Northwoods with their additional cache are much more difficult to beat however as they both score over 16% more than Pentium 4 2.0GHz.
The CPU scaling graph under Internet Content Creation SYSMark 2002 is completely different than what we saw under CCWS2K. As the clock frequency increases the Pentium 4's performance curve begins to separate from the Athlon XP, the use of the Northwood core separates the performance curves even more.
| General
Usage Performance Office Productivity SYSMark 2002 |
|||||||||||||
|
The performance picture changes considerably in the Office Productivity test suite. Only the Northwoods are able to outperform the Athlon XP 2000+ and 2100+ in this test, the additional L2 cache helps offset the long pipeline penalties of the Pentium 4 among other inefficiencies when it comes to executing conventional integer code.
Here we see that the Athlon XP has the ability to scale quite well in this benchmark even as clock speeds increase beyond 1.73GHz (2100+). Additional L2 cache will be necessary however to bridge the gap between the Athlon XP and the Northwood, which could point at a reason for outfitting the Thoroughbred core with more cache.
| Overall
System Performance SYSMark 2002 |
|||||||||||||
|
The overall performance score is derived from the ICC and Office Productivity tests and thus there are no surprises here.
The Pentium 4 and Athlon XP performance curves run pretty much along the same path until the Pentium 4 gets the additional L2 cache courtesy of the Northwood core.
Media Encoding Performance
Our media encoding tests haven't changed from previous CPU reviews, the only slight modification is the use of a newer version of Lame with better MMX, SSE and SSE2 support.
| MPEG-4
Encoding Performance Xmpeg 4.2a - Divx 4.12 Codec - YUV2 720 x 480 - DD5.1 Sound - 29.97 fps Frames Encoded per Second |
|||||||||||||
|
MPEG-4 encoding performance is still led by the Pentium 4 but now the Athlon XP 2100+ falls within 5% of the 2.2GHz Northwood's performance.
After 1.8GHz the Pentium 4 catches up to the Athlon's relatively linear curve and it is then surpassed by the Northwoods but not by a > 10% margin. It would seem that without any further core modifications the Athlon XP can remain very competitive with the Pentium 4 when it comes to MPEG-4 encoding.
| MP3
Encoding Performance Lame MP3 Encoder 3.91 -v -V 0 Time in Minutes to Encode 170MB .wav File |
|||||||||||||
|
The top three performers are about 10% ahead of the next group of three which are then another 10% faster than the next group. A larger cache has no affect on performance here, it's all about raw computational performance. Now let's have a look at how things will scale with clock speed:
With the Athlon XP 2100+ we start to see the Athlon XP's performance curve flatten out here but it'll take a few more speed bumps before we can get a good idea of where the Athlon XP's performance will end up.
3D Rendering Performance
Our 3D rendering tests are also brought over from previous CPU reviews. The only difference here is that we're reporting both scores in terms of images rendered per hour.
| 3D
Rendering Performance 3D Studio MAX 4.2.6 Waterfall.max (Rendered Images per Hour - Higher is better) |
|||||||||||||
|
The Athlon XP has always done very well when it comes to 3D rendering performance, mostly because its performance depends primarily on raw x87 FPU power. The 2100+ manages to pull ahead of the 2.2GHz Northwood by 6%.
Here the Athlon XP's performance curve is placed higher than the Pentium 4, it will take a 2.4GHz Pentium 4 to intersect the Athlon XP's curve but after that the 533MHz FSB and higher clock speed Pentium 4s should make things interesting.
| 3D
Rendering Performance Maya 4.0.1 Rendertest (Rendered Images per Hour - Higher is better) |
|||||||||||||
|
The performance standings don't change much as we move to Maya as our test application of choice...
...and thus the performance scaling doesn't change either.
3D Gaming Performance
To measure 3D gaming performance we've chosen the Unreal Performance Test 2002 and Return to Castle Wolfenstein. For information on the Unreal Performance Test 2002 you can read about it here and here.
| Next-Generation
3D Gaming Performance Unreal Performance Test 2002 - 1024 x 768 x 32 |
|||||||||||||
|
Quite possibly the most important gaming benchmark is the Unreal Performance Test 2002 because it illustrates performance under next-generation games such as Unreal 2, UnrealTournament 2 and other games based on the next-generation Unreal engine.
Here we see that the top performers are the Athlon XP 2000/2100+ and the Pentium 4 2.2GHz. In fact, you can go all the way down to the Pentium 4 2A and still remain within 10% of the top performers.
What's most impressive about this chart is that we're able to attain this type of CPU scaling at 1024 x 768 in a next-generation game courtesy of the power of the GeForce4 Ti 4600.
| 3D
Gaming Performance Return to Castle Wolfenstein - 1024 x 768 - High Quality |
|||||||||||||
|
We originally started using Quake III Arena but with frame rates approaching 300fps at 1024 x 768 it quickly became obvious that we needed a new benchmark. Thus we used Return to Castle Wolfenstein which is based on the Quake III Arena engine but much more CPU and GPU intensive. The end result is pretty self explanatory with the top two performers being the Pentium 4 2.2GHz and Athlon XP 2100+, separated by less than 2%.
The Athlon XP and Pentium 4 (Northwood) performance curves begin to intersect and almost lie upon one another. Going forward we'd expect the Athlon XP and Pentium 4 to be very competitive gaming CPUs with the performance between the two being negligable at higher clock speeds. At lower clock speeds it's clear that the Athlon XP has the Pentium 4 beat.
Introducing PCMark2002
Just yesterday MadOnion introduced their latest Home/Office benchmark suite, PCMark2002. Since we had some extra time on our hands we decided to do a quick runthrough the benchmark. The PCMark2002 suite tests things like file compression, JPEG decompression, MPEG-4 encoding and decoding, among other types of tasks. For more information on the benchmark take a look at MadOnion.com.
| Overall
Performance PCMark 2002 CPU Score |
|||||||||||||
|
We'll leave off our analysis of this test until we become a bit more familiar with the inner workings of the benchmark, but we thought we'd provide the data for you all to compare to.
The Future: 166MHz FSB & DDR333?
At the end of our Pentium 4 2GHz review we looked at the benefits that could be realized simply by bumping the FSB frequency from 100MHz (400MHz quad-pumped) to 133MHz (533MHz quad-pumped). The results were anywhere between a 0 - 15% improvement in real-world performance.
Now armed with memory that will work at 166MHz frequencies (pseudo-DDR333 spec) we are able to perform a similar test on the Athlon. We took an Athlon XP 2000+ and ran it at its native 133.33333MHz x 12.5 frequency, and then at 166.6666MHz x 10 to achieve 1.67GHz using two different approaches. In the second case we kept the FSB at 166MHz and the memory bus at 166MHz as well, thus increasing overall system bandwidth by 25%. This is an additional 25% bandwidth between the CPU and the North Bridge as well as another 25% between the North Bridge and memory.
We were able to do this on our regular KT266A platform making it clear that you don't necessarily need a KT333 chipset in order to use DDR333. The only differences are the ability to run your memory bus at 166MHz while your FSB is clocked at 100 or 133MHz and electrical validation for DDR333 on the KT333 chipset. Do note that neither chipset officially supports the 166MHz FSB.
The biggest performance boosts here are in the memory bandwidth tests taken from SiSoft Sandra 2002 of 21 - 25%. But when we shift to the real world we see that the majority of the performance increases come between 0 and 6%. This is definitely not enough of a performance increase to justify the additional validation of platforms that would have to go into making DDR333 with a 166MHz FSB a standard for future Athlon XP processors.
Granted that this additional bandwidth would be better used once the processor hits higher clock speeds but we'd be willing to bet that AMD would rather focus on ramping up their Hammer based successors rather than worrying about helping the current Athlon XP scale any better with clock speed.
So there you have it, it wouldn't make much sense for Thoroughbred to have a 166MHz FSB or use DDR333 SDRAM not only from a cost standpoint but also when it comes down to performance. In fact, we'd be willing to say that all signs point to Thoroughbred being nothing more than a 0.13-micron version of the Athlon XP to enable higher clock speeds. AMD prides themselves on having very cheap to manufacture small footprint cores and throwing additional cache on the CPU would definitely conflict with that. At this point however it would make more sense for AMD to add another 128KB or 256KB of L2 cache rather than move to a faster FSB for the Athlon XP.
Final Words
Another day, another CPU speed bump. Who would've thought we could've done so much to talk about another 67MHz? The reason we focused on this launch so very much is because of the fact that the Athlon XP 2100+ is indeed the last 0.18-micron Palomino core of the desktop Athlon line. With a $420 price tag the Athlon XP 2100+ doesn't make much sense for most users, even those looking for the highest performance. You're much better off sticking with a 1700+, 1800+ or 1900+ and upgrading later on to a Thoroughbred core or if you can hold off, maybe even a Hammer.
According to AMD's press-release just put out today, the first 0.13-micron Thoroughbred CPUs will be shipping to OEMs by the end of this month. It'll be another few weeks after that before end users can get their hands on the first 2200+ CPUs but those will be a much better buy to look for than this 2100+, the reason being that they'll have a bit more overclocking headroom and will run cooler.
In terms of where things go after Thoroughbred, AMD may need to reevaluate their roadmap once the Pentium 4 exceeds 2.66GHz and Intel introduces the 533MHz FSB. The one route that they could take would be to introduce Barton with SOI support and add some additional L2 cache to the core in order to maintain the utmost competitiveness; another option would be to let things ride out until the release of Hammer before really turning the heat up again.
The one thing that is for sure is that we're no longer dealing with the AMD of yore. AMD no longer plays entirely by Intel's rules and they are in a position where they can control exactly how competitively they want to pursue the performance crown at any given point. While they may not be as competitive in the manufacturing world with 300mm wafers 3+ years away, AMD is definitely able to keep up with (and even exceed) Intel architecturally and on a performance level as well.
Facebook
Twitter
RSS
