Benchmarks from the Athlon 64 in the c't
It was actually only a matter of time from the start until one of the 500 hammer test samples delivered by AMD to date would fall into the professional hands of an editorial team. The c't surprisingly fluttered such an Athlon 64 including mainboard into the house. A detailed test was the result.
A development sample of the Athlon 64 can be found on page 18/19 in the c't, which will be available in newsagents next Monday a clock of 1.2 GHz against an Athlon XP throttled to 1.2 GHz (comparable to a model number of 1300+) and a 2.2 GHz Pentium 4.
As already mentioned, the hammer tested by c't was operated with a clock rate of 1.2 GHz, which in turn indicates the A1 stepping of the processor core. The first executable hammer prototype in A0 stepping ran at 800 MHz. After the A2 stepping, which was operated with 1.4 GHz, the development has now reached B0 stepping with 1.8 GHz . The c't had a very 'old' hammer in its hands.
The tested Athlon 64 still had a 1 MB L2 cache, which is unlikely to be found on any desktop processor of the Hammer family at the beginning will be. In order to be able to start the home user market with the 64-bit hammer architecture at a reasonable price, there is currently talk of an Athlon 64 with a 256 kB cache .
Furthermore, AMD is silent about the test system used. It is only certain that both the Athlon 64 and the Athlon XP DDR333 modules do their job. With the Athlon XP, too, you can only puzzle over the mainboard used, but it can be assumed that either an nForce2, KT333 or KT400 has been used. The Intel Pentium 4 2.2 GHz, on the other hand, had to be content with a slow DDR266 memory, which, according to the test results, was on a motherboardi845E chipset has started. So you have waived the use of sometimes much faster Rambus. The front-side bus of the P4 ran at 100 MHz (400 MHz QDR), that of the Athlon XP ran at 133 MHz (DDR266), with the Athlon 64 there is no front-side bus in the classic sense, whether it is a memory controller integrated in the processor. An nVidia GeForce 3 or GeForce 3 Ti500 was probably used as the graphics card.
- Athlon 64 1.2 GHz2.306
- Pentium 4 2.2 GHz2,080
- Athlon XP 1.2 GHz2.041
In terms of storage throughput, the Athlon 64 can clearly stand out from the competition , but the impression is deceptive. An old FSB400 Pentium 4 paired with an old PC800 Rambus would achieve around 2800 MB/s, Intel's current FSB533 processors with the PC1066 Rambus even reach more than 3300 MB/s . Even a current Athlon XP with its front-side bus, which has meanwhile increased to 166 MHz, achieves results over 2500 MB/s with good mainboards using comparable memory modules.
In this respect, the results are not exactly exhilarating, especially since the memory controller is located directly in the processor. In return, the latency times of the Athlon 64 are much lower. But more on this in issue 26 of the c't.
- Pentium 4 2.2 GHz9.008
- Athlon 64 1.2 GHz8,753
- Athlon XP 1.2 GHz7.601
In 3DMark2001 SE, which is more of a strain on the graphics card, the Pentium 4 can easily put itself in front of AMD's development pattern. In this context it is very interesting to know that the 1.4 GHz clocked A2 stepping can probably pull the same as a 2.2 GHz Pentium 4 . The real clock increase of 200 MHz should in this case lead to at least 250 points more. That would roughly correspond to the performance that a Pentium 4 increases with such a clock speed . However, the graphics card and its driver play a very important role here.
In the case of a very small L2 cache, incidentally, the performance drops enormously in this test. An Athlon 64 with 256kb L2 cache could deliver completely different values here. However, since the L2 and L1 cache are used significantly more effectively with the Hammer than with the Pentium 4 and processors based on this architecture, the performance drop will remain within tolerable limits.
- Athlon 64 1.2 GHz225,3
- Pentium 4 2.2 GHz218,9
- Athlon XP 1.2 GHz172,4
With the Quake 3 Arena, which mainly uses the memory controller the Athlon 64 cuts a very good figure. The performance can be compared with a P4 with PC800 Rambus . A further optimization of the memory controller, especially with regard to the mediocre results in memory throughput, could ensure significantly higher frame rates in Quake 3 Arena.
- Pentium 4 2.2 GHz51,5
- Athlon 64 1,2GHz49.5
- Athlon XP 1.2 GHz48,3
In the context of our processor tests in connection with Aquamark it quickly became apparent in the past that this benchmark based on Aquanox is very little dependent on the processor clock . Apart from the fundamental differences in architecture, there were only very small performance jumps due to higher clock rates. An evaluation of the c't results is particularly difficult at this point, since a new graphics card driver can provide performance differences, especially in Aquamark/Aquanox.
- Athlon 64 1.2 GHz36.6
- Pentium 4 2.2 GHz36 , 3
- Athlon XP 1.2 GHz28.4
Comanche 4 has been a thing of the Intel Pentium 4 CPUs ( See also ). This was not least due to the SSE2 commands, which the game requires and which the Athlon 64 will also master. In terms of speed, we are now seeing a draw, which can, however, be postponed in favor of Intel when using Rambus. If AMD is able to further improve the hammer's memory interface, another leap in performance can be expected here.
- Pentium 4 2.2 GHz8:17
- Athlon 64 1.2 GHz9:29
- Athlon XP 1.2 GHz11:49
On the other hand, the Athlon 64 has to be outclassed when it comes toEncoding in MPEG4 format. The competition from Intel is much faster here, with faster memory the gap is even greater. If, on the other hand, XVid (results in the c't) is used, the result looks different again. Here the Athlon 64 can catch up with the Pentium 4.
All in all, the Athlon 64 has one cycle from 1.2 GHz to fight hard with the competition from Intel. Further test results in the c't clearly show that the Athlon 64 gets part of its lead over a similarly clocked Athlon XP from the larger L2 cache. The less the application is influenced by the processor's L2 cache, the more the performance of the Athlon 64 approaches that of an Athlon XP. An Athlon 64 with only 256 kB L2 cache could have a much harder time against the Intel power, especially since a technology with Hyper-Threading was presented from this camp, with which the processor can be used much more effectively, which results in some significant performance advantages . Since the Athlon 64 also has to compete against a new Pentium 4 generation with a frontside bus accelerated to 200 MHz (800 MHz QDR), it will not have it easy next year.
However, one should not jump to conclusions about the early stage of development of the Athlon 64 tested by the c't. In addition, one should not forget that all benchmarks were carried out on a 32-bit platform, but the hammer can only reach its top form with 64-bit code. Microsoft is already working on a corresponding implementation of Windows XP .
There are even more benchmarks for the Athlon 64 in the current c't, which will be available from the kiosk on Monday.