Chasing The P4: Athlon Core With A New Design
The smaller rival seems to be slightly weakened. However, the reasons for this are quite obvious: the Athlon design is not meant for such high clock speeds as is the case for its competitor. And the latter is moving ahead by leaps and bounds in MHz.
In our last comparison test (The Die Has Been Cast: Pentium 4/2533 vs. Athlon XP 2100+) Intel’s top model, the P4 with 2.53 GHz, was able to beat the Athlon XP 2100+ in all of the benchmark disciplines. But keep in mind that with modern processors, clock speed alone does not directly correspond to performance. The higher performance of the AMD’s arch enemy, Intel’s P4, was above all the result of a newer chip architecture, a large L2 cache and the optional Rambus platform. Ultimately, it’s no secret that the Pentium 4 runs at top form only in combination with PC1066 memory (533 MHz, currently available).
The AMD Athlon XP 2200+, based on the Thoroughbred core.
Construction of the new AMD Thoroughbred Core.
Chasing The P4: Athlon Core With A New Design, Continued
The test standards for this latest comparison are therefore at least as high as the expectations that many users have set for the new AMD Athlon XP – lately, many rumors have been going around about the Thoroughbred, which is the successor to the Palomino core. Some said that the Front Side Bus clock would be increased to 166 MHz, while others even toyed with the idea that AMD would increase the L2 cache to 512 KB.
The various Athlon CPUs compared, from left to right: Thoroughbred (2002), Palomino (2001) and Thunderbird (2000).
A view of the back side of the three generations of Athlons.
And now it has finally arrived, after a few delays – the long-awaited Athlon XP 2200+ with the T-bred core. This model is called “Athlon XP 2200+”, but the actual processor clock has been increased by 66 MHz to 1800 MHz, compared to the Athlon XP 2100+ (1733 MHz) with the Palomino core. Although officially AMD mostly talks of a decrease in the size of the processor core, there are a lot more modifications behind the design. For one thing, the manufacturer has reduced the structure from 0.18 µm to 0.13 µm, so that the core requires less surface area, because the number of transistors (37.5 million) is identical to that of the previous core. Looking more closely at the Palomino and Thoroughbred cores shows you that while the older core was almost square, the new core takes on more of a rectangular design.
Therefore, the main question in this test is: with the smaller CPU core in the Athlon XP 2200+ (1800 MHz), can AMD overcome Intel’s 733 MHz lead with its P4 (2533 MHz) and offer it some stiff competition?
The boot process of the AMD Athlon XP 2200+.
The Gigabyte board equipped with AMI BIOS, the AMD 2200+ during system startup.
Now Required: Protection Against Thermal Death
On the Soltek board, thermal protection is activated via jumper.
Outdated: temperature monitoring underneath the CPU, using a thermistor.
About nine months ago, THG came out with the article Горячо! Как современные процессоры защищены от перегрева?, which caused quite an uproar in the industry and gave rise to considerable discussion. Accompanying this test was our first downloadable video, which shows the dramatic behavior of an AMD Athlon with the Palomino core (equipped with a thermal diode) when the CPU cooler fails while a system is running.
The alarm bells went off at AMD, and there were frequent meetings between the THG crew and the technical team at AMD and Siemens. A few weeks later, at a meeting with a small number of select attendees, AMD presented a circuit logic, which immediately turns off the power supply when the die temperature overheats to about 85 degrees Celsius. To guarantee failsafe protection, the CPU temperature is monitored frequently, in very short time increments, by the thermal diode. This ensures that the power is switched off in a timely manner. Since a few months ago, AMD has been distributing a new “Thermal Guide” to motherboard manufacturers in order to guarantee thermal protection for all new boards.
Thermal protection with the latest Asus KT333 board.
Thermal protection from Soltek with its new board based on the VIA KT333-chipset.
Now Required: Protection Against Thermal Death, Continued
In practice, even the newest boards with the VIA KT333 chipset cut a sorry picture: in our last KT333 comparison VIA KT333 Put To The Test: 18 Motherboards Compared, out of the 18 different manufacturers, only Asus and Soltek integrate thermal protection on the board. Motherboards from the other manufacturers respond just as we show in our THG video.
Simultaneous to the launch of the new Athlon with the T-bred core, AMD has given the following guideline to the motherboard makers: starting June 10, all motherboards must have integrated thermal protection in order to receive certification from AMD. The costs per board for this thermal protection logic runs at approximately less than $1. Astonishingly enough, the AMD presse kit includes a motherboard (Gigabyte GA-7VRXP, Rev. 1.1), which does not integrate thermal protection circuitry. It should be noted that the Gigabyte board mentioned here was the winner in our last test, but today, we cannot make the same recommendation. The manufacturer needs to quickly modify the board if it is interested in getting support from AMD.
Despite the current situation with thermal protection circuitry, some boards are still equipped with the old thermistors.
In the short term, a general improvement on the market is not in sight, because most of the manufacturers have long since introduced their KT-333 boards, and moreover, they don’t expect to integrate thermal protection circuitry on their boards until a new chipset has been launched (e.g., VIA KT333A or KT400).
No improvement: the Gigabyte board provided by AMD does not switch off when the CPU temperature surpasses 90 degrees Celsius!
Comparison: T-Bred vs. Palomino vs. T-Bird vs. (Barton)
With the launch of the Thoroughbred core, AMD Athlon has reached the third step in its evolution. The first core, the Thunderbird, was already introduced around two years ago. The manufacturer made a further step with the launch of the Palomino core in October 2001, where the new features essentially consisted of an SSE command set extension and optimized cache lines.
Now, in June 2002, there’s the Thoroughbred, which brings us a modified layout of the internal CPU units, as well as a smaller structure (0.13 µm). Detailed information is given in our table. The animated GIF above shows you the difference in size in the CPU die. To make this as complete as possible, we have also included data for the Barton core, which is expected to come out in Q3, marking the fourth and last phase of development for the Athlon program based on Socket 462.
Heat Dissipation: All AMD-CPUs Compared
Exclusive Details On The Barton (Q3 2002)
Looking at the geometric form of the CPU die, as well as the AMD roadmap, you can see that the Thoroughbred is only an intermediate step between the Palomino and the Barton. The dimensions and the rectangular form of the Thoroughbred’s CPU die also make it clear that AMD has not only reduced the size of the die, but rearranged the individual units as well. Regarding the latter, this reduces efforts and does away with the need for a redesign when the L2 cache of the processor is increased in the near future. For this reason, AMD arranged the transistors differently for the L2 cache, with the goal of expanding the L2 cache to 512 KB.
With almost 54 million transistors, the new Barton die (successor to the Thoroughbred) thus catches up (at least in terms of numbers) with the Intel Pentium 4 and its 55 million transistors. Since the Barton is also based on a 0.13-µm process, the surface area of the CPU core increases from 80 mm² (Thoroughbred) to 115 mm2 (Barton).
Still, the surface area of the Barton core is significantly smaller than the 146 mm² required by the Intel Pentium 4 with the Northwood core. The bottom line is that the smaller CPU die allows the manufacturer to produce a larger number of processors from a single wafer, whereas Intel will increasingly be making use of 300 mm wafers in the future. AMD uses only silicon wafers with a diameter of 200 mm in its Fab 30 in Dresden, Germany.
The latest roadmap from AMD. In older versions, an Appaloosa core with a smaller die was also planned as the successor to the Morgan core.
Greater Yield: 322 Athlon-CPUs Per Wafer
Because AMD only uses wafers with a 200 mm diameter, this results in a silicon area totalling 31415 mm². The quotient of wafer surface and the size of the CPU die gives you a theoretical yield without geometric waste. The production process of the 200 mm wafers has shown that there’s a waste of about 18%.
Thus, AMD calculates that the production of the Thoroughbred core in Dresden will result in a yield of 322 processors – based on 0.13-µm technology and a die surface of 80 mm². Note that the yield here assumes an error quota of 0%. Compared to this, only about 201 Athlon processors with the Palomino core (128 mm2 die surface and 0.18-µm technology) can be produced using the same wafer!
In the end, AMD can count on an increase in yield of approximately 60% (322 CPUs vs. 201 CPUs) just by virtue of the smaller die alone! Intel’s Pentium 4 with the Northwood core requires a surface of 146 mm², which means a yield of around 176 processors per wafer. Now, Intel will gradually switch to using wafers of 300 mm diameter in its production process. The reason for this is that larger wafers result in less waste in terms of percentage. AMD has no further plans for change at the moment.
Comparison: More Transistors Per Surface Area With AMD
The table below shows how many transistors AMD and Intel can fit on their respective CPU cores per square millimeter die surface. AMD manages to place about 468750 transistors on a surface measuring 1 x 1 mm.
Manufacturer | AMD | AMD | AMD | Intel | Intel |
Processor | Athlon XP w/Thoroughbred Core | Athlon XP w/Palomino Core | Athlon XP w/Barton Core | Pentium 4 w/Northwood Core | Pentium 4 w/Willamette Core |
Manufacturing Process | 0,13 µm | 0,18 µm | 0,13 µm | 0,13 µm | 0,18 µm |
Die size | 80 mm² | 128 mm² | 115 mm² | 146 mm² | 217 mm² |
Number of Gates | 37,5 Million | 37,5 Million | 53,9 Million | 55 Million | 42 Million |
Gates per size | 468750 gates/ mm² | 292969 gates/ mm² | 468750 gates/ mm² | 376712 gates/ mm² | 193548 gates/ mm² |
Clock Increase and Heat Dissipation
By switching to 0.13-µm technology, the power consumption and core voltage of the Athlon XP with the Thoroughbred core will sink drastically. However, this only applies to the “smaller” models with lower clock frequencies. The table below shows that only the Athlon XP 1700+, 1800+ and 1900+ work with a significantly lower core voltage of 1.5 Volt. The heat dissipation is clearly lower with only three of the processors, if you compare them to the same Athlon XP models with the Palomino core.
The new top-of-the-line Athlon XP 2200+, however, works which a voltage of 1.65 Volt – which is a difference of only 0.1 Volt compared to its predecessor.
Viewed with a critical eye, you can determine that decrease in die size has not made much of a difference in heat dissipation, because at a maximum heat dissipation of 67.0 Watt, the Athlon XP 2200+ with Thoroughbred core emits only 4.1 Watt less than the Athlon XP 2100+ with the Palomino core. In total, the heat dissipation of the top Thoroughbred model has an advantage of 5.7% over the Palomino. Therefore, the Athlon XP 2200+ remains in a hot spot, despite the 0.13-µm technology.
Athlon XP 2200+ in operation during testing.
The Thoroughbred is recognized by the software as “Model Number 8”!
Overclocking: Not A Whole Lot Possible
Prior to the test, the THG engineers had some rather high expectations: overclocking with the smaller CPU core should prove to be easily accomplished. The serial settings run at 13.5 x 133 MHz = 1800 MHz. But our attempt to overclock the processor’s FSB to 148 MHz, similar to the Athlon XP 2100+ (Palomino), failed. Here, it should be mentioned that both systems were operated with a high-quality watercooling setup. The result: the system remained stable only at 140 MHz FSB, meaning that the CPU clock was increased to 1890 MHz. This is an increase of 5%, which is clearly a weaker result than with the older Palomino, which ran with perfect stability at 1918 MHz (11.2% increase).
Not A Topic: 166 MHz FSB
Despite the many rumors of an increased FSB clock frequency, AMD will not be increasing the clock from 133 MHz to 166 MHz for either the Thoroughbred or its successor, the Barton. Also, there are officially no chipsets that have standard support for 166 MHz FSB.
New Guidelines For CPU Coolers: No Cheap Models
A new AMD-approved CPU cooler for the Athlon XP 2200+ with high contact pressure.
When switching from the Athlon XP with the Palomino core to the Athlon XP with the Thoroughbred core, the topic of sufficiently cooling the processor becomes increasingly important. Even though the new Athlon XP has only a marginal performance increase (5.7% less with the top 2200+ model than the previous model with the Palomino), a higher quality CPU cooler is necessary.
To explain this – since the CPU surface has been reduced by 40%, this means that there’s automatically less surface area to dissipate the heat. Here, we wonder why AMD doesn’t use a heat spreader а la Intel Pentium 4 or AMD “Hammer”. For OEM manufacturers and buyers of the new Athlon XP processors, there are new guidelines from AMD, valid starting June 10, 2002:
New Guidelines For CPU Coolers: No Cheap Models, Continued
Only a few select CPU coolers can be used – they must have a copper contact plate for the surface of the CPU die, or use some other kind of contact plating with adequate heat transfer capability. To put it clearly: if you want to equip your PC with a Thoroughbred Athlon XP, you can get rid of your aluminum cooler. As it is, if the new Athlon XP dies a thermal death when used together with a cheap cooler, you will get no guarantee from AMD.
An approved copper contact for CPU coolers.
A CPU cooler from Taisol.
Confusion: Differentiating Between Thoroughbred & Palomino
The confusion is to be expected: there are two different processors covered by a single product name. Take the Athlon XP 1700+, for example, which is available with the Palomino core as well as the newer Thoroughbred core. The performance, however, is the same. The only significant difference that can be discerned is with the heat dissipation – which is, in any case, a difference of 22.8% at maximum performance. It would be helpful to add some sort of marker in the product names for distinguishing new Athlon XP models (0.13-µm) from the older ones. Visually, the two different types of CPU are easy to differentiate:
The Thoroughbred has a CPU core in rectangular form, while the older Palomino has almost a square CPU die. Some further characteristics are that the resistances are placed on the top side of the CPU, and the model name of the processor is found on the OPGA casing.
Price Comparison: AMD Athlon XP vs. Intel Pentium 4
Processor | Price per 1000* |
AMD Athlon XP 2200+ | US$ 242 |
AMD Athlon XP 2100+ | US$ 224 |
AMD Athlon XP 2000+ | US$ 193 |
AMD Athlon XP 1900+ | US$ 172 |
AMD Athlon XP 1800+ | US$ 160 |
Intel Pentium 4/1800 | US$ 163 |
Intel Pentium 4/1900 | US$ 173 |
Intel Pentium 4/2000 | US$ 193 |
Intel Pentium 4/2266 | US$ 241 |
Intel Pentium 4/2400 | US$ 400 |
Intel Pentium 4/2533 | US$ 637 |
Test Setup and Details
Intel Hardware (Socket 478) | |
Processors 1 133 MHz FSB – 533 MHz Memory Clock |
Pentium 4A 2.5 GHz (2533 MHz) Pentium 4A 2.4 GHz (2400 MHz) Pentium 4A 2.3 GHz (2266 MHz) |
Processors 2 100 MHz FSB – 400 MHz Memory Clock |
Pentium 4 2.4 GHz (2400 MHz) Pentium 4 2.2 GHz (2200 MHz) Pentium 4A 2.0 GHz (2000 MHz) Pentium 4A 1.8 GHz (1800 MHz) Pentium 4A 1.6 GHz (1600 MHz) Pentium 4 2.0 GHz (2000 MHz) Pentium 4 1.4 GHz (1400 MHz) |
Motherboard 1 | ABIT TH7II (I850) Revision: 1.0 Bios: TH7H_38 |
Motherboard 2 | Asus P4T533-C (I850E) Revision: Bios: 1001 BETA 007 |
Memory 1 | 2x 256 MB RDRAM, PC800, 533 MHz, 40ns, Infinion |
Memory 2 | 4x 128 MB RDRAM, PC1066, 400 MHz, 32ns, Samsung |
AMD Hardware (Socket 462) | |
Processors 1 133 MHz FSB – 166 MHz Memory Clock |
Athlon XP 2200+ (1800 MHz) Athlon XP 2100+ (1733 MHz) Athlon XP 2000+ (1666 MHz) Athlon XP 1900+ (1600 MHz) Athlon XP 1800+ (1533 MHz) Athlon XP 1700+ (1466 MHz) Athlon XP 1600+ (1400 MHz) Athlon XP 1500+ (1333 MHz) Athlon 1400 (1400 MHz) |
Processors 2 100 MHz FSB – 133 MHz Memory Clock |
Athlon 850 (850/100/133 MHz) |
Motherboard | Gigabyte GA-7VRXP Revision: 1.1 Bios: F6f (19.04.2002) |
Memory | 512 MB DDR-SDRAM, CL2.0, 166 MHz, PC2700, Winbond |
Common Hardware | |
Graphics Card: | GeForce 4 Ti 4600 (MSI MS-8872) Version: 200 Memory: 64 MB DDR-SDRAM Memory Clock: 650 MHz Chip Clock: 300 MHz |
Hard Drive | 40 GB, 5T040H4, Maxtor UDMA100 7200 U/min 2 MB Cache |
Drivers & Software | |
Graphics Card | Detonator 4 Serie V28.32 |
VIA KT333A Driver | 4 in 1 Version: 4.38(2)v(a) |
DirectX Version | 8.1 |
Intel 850E Driver | V 4.00.1009 |
Intel 850 Driver | V 3.20.1008 |
Intel IAA Driver | V 2.0 |
OS | Windows XP, Build 2600 (English) |
Benchmarks and Settings | |
Quake III Arena | Retail Version 1.16 command line = +set cd_nocd 1 +set s_initsound 0 Graphics detail set to ‘Normal’ Benchmark using ‘Q3DEMO1’ |
3DMark2000 Pro | Version 1.1 Build 340 – default Benchmark |
3DMark2001 Pro | Build 200 – default Benchmark |
PCMark2002 Pro | only CPU and Memory Bench (no Video Memory) |
SiSoft Sandra 2001 | Professional Version 2001.3.7.50 |
Newtek Lightwave | Version 7b Rendering Bench SKULL_HEAD_NEWEST.LWS |
mpeg4 encoding | Xmpeg 4.5 DivX 5.01 Pro (YV12) Compression/quality: Slowest Data Rate: 780 Kbit Format: 720×576 Pixel@25 fps 150 MB VOB-Datei, no Audio |
Studio 7 | Version 7.31.6 (MPEG 2) |
Sysmark 2002 | no Patch |
Lame | Lame 3.91 MMX, SSE, SSE 2, 3DNow! |
WinACE | 2.11, 178 MB Wave-Date, Best Compression, Dictonary 4096 KB |
Cinema 4D XL R7 | Version V7.303 Rendering: 1024×768 |
3D Studio Max | Version 4.2 Rendering the Scene “Rabbit” 800×600 10 Build |
MAGIX MP3 Maker | Platinum V3.03 D |
SPEC Viewperf | V7.0 (1280×1024 / 32 Bit / 85 Hz) |
Comanche 4 | V 1.0.0.1.18 |
32 Benchmarks under Windows XP
OpenGL Performance | Quake 3 Arena “Demo 1” and “NV15 Demo” |
3D Rendering | SPEC Viewperf 7 (new) Lightwave 7b Cinema 4D XL 7.303 3D Studio Max 4.2 |
DirectX7 Games | 3D Mark 2000 |
DirectX8 Games | 3D Mark 2001 SE Comanche 4 (new) |
MP3 Audio Encoding | Lame MP3 Encoder mp3 Maker Platinum |
MPEG-2 Video Encoding | Pinnacle Studio 7 |
MPEG-4 Video Encoding | XMpeg 4.5 and Divx 5.02 Pro |
Office Performance | Sysmark 2002 |
Archiving | WinACE 2.11 |
CPU and Multimedia Bench | PC Mark 2002 SiSoft Sandra 2002 Pro |
We performed a total of 32 different benchmark tests in order to obtain the most complete, well-balanced view of how the new AMD XP 2200+ performs. We compared this to all of the AMD Athlon XP processors, the classic Athlon with the Thunderbird core and the slowest of all, the Athlon 850.
We ran four different Quake 3 tests to determine OpenGL performance – the 3D performance from the DirectX package is determined by 3D Mark 2000 (based on DirectX 7) and 3D Mark 2001 SE (based on DirectX 8). A newcomer to the bunch is one of the first games to support DirectX 8, Comanche 4. The different MPEG-encoding benchmarks provide a comprehensive testing environment – the Lame MP3 Encoder and mp3 Maker Platinum were used to encode a 178 MB WAV file into “MPEG-1 Layer 3 format.”
Still a classic, our MPEG-4 test converts a file from a commercial DVD-ROM into MPEG-4 format using Xmpeg 4.5 and the Divx 5.02 Pro codec. We also created an MPEG-2 film using the video-editing software “Pinnacle Studio 7.” A regular in our list of benchmarks is determining rendering performance using Newtek’s Lightwave (version 7b), 3D Studio Max (version 4.2) and Cinema 4D XL 7.303. We also ran the new WinACE 2.11 to test how well the CPU performs when archiving files, a common application in the computing world. The Sysmark 2002 benchmark was used to determine office performance. The new SPECviewperf (version 7) benchmark offers a comprehensive 3D benchmark suite. And last but not least, we used the SiSoft Sandra 2002 Pro benchmarks as well.
OpenGL Performance: Quake 3 Arena
In the four time-demo runs of Quake 3 Arena, the results are consistent: the Athlon XP 2200+ cannot keep apace with the P4/2533. The reason for this is that the Athlon XP lacks memory throughput, causing the AMD platform to bring up the rear. Note: all overclocked processors are indicated by the grey bars and are used for reference purposes.
DirectX 7 Games: 3D Mark 2000
The 3D Mark 2000 determines DirectX 7’s Direct3D performance under Windows XP. The Athlon XP 2200+ outstrips the Intel Pentium 4/2400 with Rambus memory (PC800). Once again, the leader here is the P4/2533. Note: all overclocked processors are indicated by the grey bars and are used for reference.
DirectX-8-Games: 3D Mark 2001 SE
The 3D Mark 2001 determines DirectX 8’s Direct3D performance under Windows XP. This test makes it clear that the Athlon XP 2200+ is a good CPU for games, but it still does not reach the performance level of a P4/2400. Note: all overclocked processors are indicated by the grey bars and are used for reference.
MP3 Audio-Encoding: Lame MP3
The Lame MP3 Encoder under Windows XP is used to convert a 178 MB sound file from a WAV format to a “MPEG-1 Layer 3” format. The diagram shows that the AMD Athlon XP 2200+ reaches the performance level of a Pentium 4 with 2400 MHz. However, it cannot keep up with the top model from Intel, the P4/2533. Note: all overclocked processors are indicated by the grey bars and are used for reference purposes.
MP3 Maker Platinium
MPEG-4 Video Encoding: XMpeg 4.5 and Divx 5.02 Pro
RAM performance is an important factor in MPEG-4 encoding. Here, the AMD Athlon XP 2200+, together with DDR333 (CL2, max. timing), is able to reach the performance of a P4/2400. By comparison, the Athlon 850 looks pretty outdated with approximately 16 fps. Note: all overclocked processors are indicated by the grey bars and are used for reference purposes.
MPEG-2 Video Encoding: Pinnacle Studio 7
The AMD Athlon XP 2200+ creates an MPEG 2 film with Pinnacle Studio 7 in 81.2 seconds, which is almost as fast as the P4/2200. The slowest candidate is once again the AMD Athlon 850, which we included only for comparison purposes. The fastest CPU here is the P4/2533, which encodes the video in 66.6 seconds – that’s a lead of 22%. Note: all overclocked processors are indicated by the grey bars and are used for reference.
DirectX8 Hardcore Game: Comanche 4
Comanche 4 is one of the first games on the market to offer support for DirectX 8. This game puts quite a hefty load on the CPU, and it runs significantly faster on the P4/2533 than on the AMD Athlon XP 2200+. When run with the older processors, such as the Athlon 850, this game can hardly be played. Note: all overclocked processors are indicated by the grey bars and are used for reference.
SiSoft Sandra 2002 Benchmarks: CPU und Multimedia
The SiSoft Sandra Pro Benchmark 2002 shows that the Athlon XP 2200+ nearly reaches the level of the P4/2533 in all but one of the tests, but still, these test results are of theoretical significance only. Note: all overclocked processors are indicated by the grey bars and are used for reference.
Multimedia-Performance: PC Mark 2002
In these two benchmarks, the P4/2533 is clearly ahead of the AMD Athlon XP 2200+. Note: all overclocked processors are indicated by the grey bars and are used for reference.
Office-/Internet-Performance: Sysmark 2002
In all three areas, the Athlon XP 2200+ ranked in the mid-range and performed similarly to a Pentium 4/2000 with Northwood core. A note about all AMD Athlon XP CPUs: compared to Intel’s models, the AMD processors lag slightly behind because they lack enhancements. Note: all overclocked processors are indicated by the grey bars and are used for reference.
3D Rendering Performance: SPECviewperf
In this very sensitive SOECviewperf 7 test, the Athlon XP 2200+ takes the lead in two individual disciplines (DX07 and UGS01). However, the question arose as to whether this new benchmark has compatibility problems with the Intel P4. Note: all overclocked processors are indicated by the grey bars and are used for reference purposes.
Archiving: WinACE 2.11
Archiving is a very practical application. WinACE 2.11 was used under Windows XP to archive a 178 MB WAV file while the clock was running. Here, the P4/2533 was far ahead of the Athlon XP 2200+ with the Thoroughbred core. Again, in this test, the high memory throughput (P4 with Rambus PC1066) shows its positive influence. Note: all overclocked processors are indicated by the grey bars and are used for reference.
3D-Rendering: Newtek Lightwave 7b
The Lightwave benchmark clearly brought the enhancements of the Pentium 4 processors to light – the Athlon XP 2200+ places in lower midfield. Note: all overclocked processors are indicated by the grey bars and are used for reference.
3D-Rendering: Cinema 4D XL 7.303
The AMD Athlon performs aggressively in the Cinema benchmark – here, the AMD Athlon XP 2200+ takes second place behind the Intel P4/2533. Note: all overclocked processors are indicated by the grey bars and are used for reference.
3D-Rendering: 3D Studio Max 4.2
In this benchmark, 10 frames from the “Rabbit” scene were calculated with a resolution of 800 x 600 pixels. Note: all overclocked processors are indicated by the grey bars and are used for reference.
Conclusion: The Competition Bids Adieu
The eternal “AMD vs. Intel” competition has changed in character – what has previously been a close race is now no longer the case. Our comparison of the latest top model, the AMD Athlon XP 2200+, shows that the launch of the new Thoroughbred core, which involves a increased clock frequency, is not enough to attain the level of the fastest Intel Pentium 4/2533. Their respective performance in practice is reflected by the results of the 32 benchmark tests that we ran – the Athlon was only able to beat the P4 in two of the disciplines.
In the transition from Palomino to Thoroughbred, AMD reduced the structure from 0.18 µm to 0.13 µm and rearranged the units of the CPU at the same time. The goal is to pave the way for the new core, codenamed “Barton”, which will soon be released. This will have a 512 KB L2 cache, and this is the only difference between it and the Thoroughbred.
Our extensive tests give the impression that the Athlon design is already a bit outdated and is now reaching its limits. We weren’t able to overclock our sample of the Athlon XP 2200+ by a significant amount. In any case, the lab results were below that of the Athlon XP 2100+ (Palomino core) in our last test, which we were able to run stably at 1918 MHz. Also, it’s no secret that the Athlon design is not meant for extremely high clock speeds as is the case for its arch rival, Intel’s P4. However, the AMD Athlon XP is able to reach the same performance as that of a higher-clocked P4, which goes to show that clock speed alone is by no means an indication for performance. The strength of the Athlon still lies in the parallel execution of internal commands.
A further topic to consider has to do with the thermal protection of the new processors: with the launch of the Athlon XP with the Thoroughbred core, simultaneous to the launch of the new Athlon with the T-bred core, AMD has given new guidelines to the motherboard makers. Starting June 10, all motherboards must integrate thermal protection circuitry, which reacts to the temperature monitoring via thermal diode to switch off the power supply immediately. Otherwise, AMD will not take responsibility for a damaged CPU. The new thermal protection circuitry from AMD is a result of the video from the THG lab, which caused an uproar in the industry nine months ago.
There are also new recommendations for CPU coolers: since the new CPU die is significantly smaller, the heat dissipation of the top model (Athlon XP 2200+) is not any less than with the older Palomino, so a high-performance cooler is necessary. Here, the heat transfer between the CPU die and the cooler is decisive, because the heat has to be transferred over a very small surface of 80 mm. A few cooler manufacturers have responded to this and are already offering models with copper contact plating. A heat spreader, as used with the “Hammer”, would have also helped to increase the contact surface to the cooler. In any case, cheap coolers have not lost out in modern Athlon XP systems! Basically, it’s the same as with integrated thermal protection: if a processor dies a thermal death when used with an unapproved cooler, AMD will not take responsibility for it.
An important argument in favor of buying the processor is the price, but this can be looked at from two different perspectives: purchasing the CPU separately in a retail store, or purchasing a complete (OEM) system with the said CPU. The Athlon XP 2200+ costs $242 (per 1000 units), which is close to a third to the price of Intel’s top model, the P4/2533. But if you look at the cost for a complete system, the difference is less marked. AMD still has a better price/performance ratio if you’re less concerned with getting the absolute top performance. In this case, the true benchmark freaks will want to stick with a P4 system based on PC1066. The T-bred won’t give you the a big performance boost, and we didn’t expect it to, because ultimately, the CPU core has not been changed.
Still, it’s going to be exciting, because the arrival of the Barton, with its larger L2 cache, is imminent. And VIA is working on the KT400 chipset, which is supposed to bring DDR 400 with 200 MHz to the Socket 462 platform. Thus, the race has not yet come to an end – the means are ultimately the goal!