The Thoroughbred core was announced by AMD yet in the second half of 2000. But the promised improvement, which was a new 0.13-micron process, didn't mean a completely new core from a point of view of an average user. However, because of the capabilities brought in by the progress, which are a smaller core surface (which allows adding functional units of a processor, for example, cache), reduced power supply (hence reduced heat liberation), and higher clock speeds, everybody interested in the subject, was looking forward to the new core.

I must say that they were waiting for it quite long: first the core was promised for the beginning of 2002, then it was postponed several times, and, at last, the realists thought it wouldn't be released at all because of the problems with the new fab process, and the optimists considered the Thoroughbred would be replaced right with the Hammer. And it is not far from the truth - some time ago AMD had to stake everything on the Athlon (and it won), but now it can't take the risk anymore and start running different production lines at different factories, that is why the production technology was optimized with the Hammer in mind, and the Thoroughbred was rather a trial of the new fab process, because it became much more difficult to squeeze anything from the old architecture. Well, the processor based on the new core at last appeared on June 10.

Processor

So, the new Athlon XP based on the Thoroughbred (Family 6, Model 8, unlike the Palomino which is Family 6, Model 6) is the same good old Athlon XP based on the Palomino with all its architectural characteristics, including the cache size (L1 128 KBytes + exclusive L2 256 KBytes) and peculiarities of the processor naming according to the QuantiSpeed, expect the 0.13-micron process. The package of the processor with the cheap and having low resistsnace parameters (as compared with ceramic) organic mount (OPGA) and Socket A form-factor (462 pins) are also the same. By the way, AMD is going to use this form-factor at least two years more (up to 2004), thus, making easier the choice for systems integrators and any other owners of processors from this company.

Note that there is no more laser engraving on the die's surface, and all data are written now on the mount. It was done to reduce shortness of the die; AMD informs also about other methods they used to strengthen the core. Physical changes in the processors are shown in the following table with three upper models of each core:

As you can see, the new processors have thrown off almost 10W, while the junior models, which need only 1.50 V, got rid of 15 W. However, there is a new problem: the core's surface has reduced considerably, unlike the heat liberation, that is why the heat flow coming from the processor is of higher density. As a result, AMD says that the Thoroughbred based processors lower than 2200+ have all the same requirements as for Athlon XP cooling, and the requirements are stricter for Athlon XP (Thoroughbred) 2200+ (and the following models), and AMD recommends that you use coolers with a copper base. (Here is a complete list of certified cooling devices.) It should be noted that everyone is able to choose a cooler according to the recommendations and install it correctly, though this innovation will hardly be warmly accepted by users. Note that the permissible core's temperature is 90°C for the models of 2100+ and lower, and 85°C for the 2200+ models (and, evidently, higher).

Performance test

Testbed:

• Processors:
• AMD Athlon XP (Thoroughbred) 2200+ (1800 MHz, 133x13.5), Socket 462 (Socket A)
• AMD Athlon XP (Palomino) 2100+ (1733 MHz, 133x13), Socket 462 (Socket A)
• Intel Pentium 4 2.2 GHz (100x22), Socket 478
• Mainboards:
• Gigabyte 7VRXP (BIOS F6f) on VIA KT333
• Fujitsu-Siemens D1289 (BIOS R1.05) on VIA KT266A
• EPoX 4BDA2+ (BIOS of 05/02/2002) on i845D
• Memory:
• 2x256 MBytes PC2700(DDR333) DDR SDRAM DIMM TwinMOS, CL 2 (used on KT333)
• 256 MBytes PC2700(DDR333) DDR SDRAM DIMM Samsung, CL 2 (used as DDR266 on i845D)
• Video card: ASUS 8200T5 GeForce3 Ti500
• Hard drive: IBM IC35L040AVER07-0, 7200 rpm, 40 GBytes

Software:

• Windows 2000 Professional SP2
• DirectX 8.1
• VIA 4-in-1 4.38
• Intel chipset software installation utility 3.20.1008
• Intel Application Accelerator 2.0
• NVIDIA Detonator v28.32 (VSync=Off) - used on KT333
• NVIDIA Detonator v22.50 (VSync=Off) - used on i845D
• SPEC CPU2000 (tests complied with the Intel C/C++/Fortran 5.0.1)
• CPU RightMark v1.01
• Cachemem 2.4MMX
• RazorLame 1.1.4 + Lame codec 3.91
• VirtualDub 1.4.9 + DivX codec 5.0 Pro
• WinAce 2.11
• BAPCo & MadOnion SYSmark 2002 Office Productivity
• BAPCo & MadOnion SYSmark 2002 Internet Content Creation
• Discreet 3ds max 4.26
• SPECviewperf 6.1.2
• MadOnion 3DMark 2001 SE
• idSoftware Quake III Arena v1.30
• Gray Matter Studios & Nerve Software Return To Castle Wolfenstein v1.1
• DroneZmarK

Remember that we are not going to compare the AMD Athlon XP vs. Intel Pentium 4 because there were a lot of such comparisons, and we have nothing to say about the processor which has the same architecture as its predecessor. Instead, I suggest that we compare performance of the fastest model on the new core with that of the previous champion from AMD on the old core - just to make sure the cores do not differ and to estimate scalability.

Besides, we have added the results of the Pentium 4 of the same level. Although it seems incorrect to compare results such a way, we will show the results, and it will be for you to make the conclusion. The results of the Intel processor were obtained on the system based on the i845D chipset. Note that the results of comparison of the AMD vs. Intel can be found in previous articles of this subject, for example, here.

However, it is very interesting to compare results of the AMD and Intel processors in SPEC CPU2000 and CPU RightMark, where we will try to find equal competitors for the today's heroes.

We also had an opportunity to estimate operation of two systems with the Athlon XP "Thoroughbred" - on mainboards from Fujitsu-Siemens and Gigabyte. The results are quite expectable - the Fujitsu-Siemens D1289 board, which is positioned not as a solution for home enthusiasts, doesn't have enough overclocking settings, and the system's speed is lower on it. We think that it's useless to show results of both systems, and we offer only data on the Gigabyte 7VRXP. But you can look through the characteristics and capabilities of the Fujitsu-Siemens board as well - as usual, the description is available by the link in the list of the test equipment.

Test results

For the SPEC CPU2000 apart from two Athlon XP processors (on different cores) we used Pentium 4 on the Northwood with a 400 MHz bus: one working at 1800 MHz (for comparison of the processors of the same frequencies), and the other - at 2200 MHz (for comparison of processors of the same rating (for the Pentium 4 its rating is its frequency). Both Intel processors were tested on the most efficient system (the i850 based board) with the RDRAM. We give only base test results; the tests were conducted for SSE and SSE2 instructions (provided that the processor supports the latter); iXBT.060202.sse.cfg and iXBT.060202.sse2.cfg were used to run the tests.

It's quite difficult to comment on the SPEC CPU results. If you want ot gain the deeper understanding - see the respective series of articles. In the SPECint the Pentium 4 and the Athlon XP working at the same frequency perform almost equally (SSE2 doesn't help much), in the SPECfp the Intel processors perform much better (by ~16%), although the clock speeds are the same, and SSE2 helps considerably (by ~8%) making the gap equal to ~26%! We also must note that the SPECint and SPECfp have bad scalability (~67% and ~25%, respectively) on the AMD processors (which is similar to the situation on the Intel's processors (~56% and ~39%)).
For the CPU RightMark we have chosen the same competitors, but this time the Pentium 4 was tested on the SiS 645DX based board with DDR333 memory.

In the unit of solving of mathematical equations the Athlon XP leads impressively in floating-point calculations, which, however, turns to zero when the SSE2 are used. The test results are very close for both processors of the same frequency, despite the different architectures.

The SSE instructions used in the rendering unit allow for a smaller gain (as compared with the Pentium 4) for the AMD processors, but anyway, the rendering speed is still high. Remember, however, that the SiS chipset used for the Intel processors allows "levelling" the systems as far as the memory is concerned (DDR333), but it is still not an optimal choice for the Pentium 4.

The overall performance in the CPU RightMark test with the default settings depends mostly on the rendering speed, that is why the Athlon XP falls so much behind the Pentium 4 of the same frequency and outscores the P4 of the same rating.

Estimation of the read and write speeds brings in no surprises: the 2200+ processor has it a bit lower, probably, because of the fractional multiplier.

The other results are worth no comments: the difference between the Athlon XP processors is not great, and there was nothing unexpected.

Conclusion

Well, the Thoroughbred is a logical extension of the Palomino, with the same advantages and disadvantages. On the one hand, it's unpleasant to have stricter requirements for processor cooling, but on the other hand - this is the cost of the progress. Anyway, the Thoroughbred is not the product users are waiting for now from AMD: at present the experts are working on the Hammer.

While the Athlon XP has some frequency reserve, the architecture itself doesn't have any; it's well seen through the tests where performance depends not only on a processor's clock speed. The only thing we shouldn't reproach AMD with is that the speed of the 2200+ model doesn't coincide with the Intel's processor of the same frequency - because nobody promised that.

Write a comment below. No registration needed!