Gains Using an Athlon 64-bit CPU Compared to an Athlon XP CPU

We were interested in what performance gains one might expect using an AMD 64-bit CPUs compared to one another and an older Athlon XP 32-bit CPU. After all, the K8 architecture (64-bit) is reported to be an improvement over the K7 (Athlon XP 32-bit) architecture. Unfortunately we cannot test the CPUs alone due to the different pin count on the CPUs. Thus the comparison must include, at a minimum, different motherboards.

The four systems below are compared. The reader should be aware that there are several differences so the comparison only gives you a flavor of the performance differences you might expect. Three variations of the AMD 64-bit CPUs are included, a 754 NewCastle, a 939 San Diego, and an Opteron 148. (NOTE: Opteron processors can only be supported by manufacturers on their server motherboards.) Both the San Diego and the Opteron have an 1MB L2 cache while the NewCastle has a 512KB L2 cauche.

AMD Opteron 148            NOTE: This CPU is not supported by this
DFI LanParty Ultra-D                     DFI motherboard
OCZ EL DDR PC-4200 (2*512Mb), 2.5-3-3-8, dual channel
EVGA 7800GT
RAID0, 2 80G SATA HDDs
Iwaki MD-20RLZT -> Storm -> Chiller -> Reservoir
PC_Power Tubo-Cool 510 SLI

AMD64 3700+ (939 San Diego)
DFI LanParty SLI-D
Mushkin Redline PC4000 (2*512Mb), 2-2-2-6, dual channel
Sapphire X850Pro
RAID0, 2 80G SATA HDDs
Iwaki MD-20RLZT -> Storm -> Chiller -> Reservoir
PC_Power Tubo-Cool 510 SLI

AMD64 3200+ (754 NewCastle)
DFI LanParty UT 250
OCZ PC4200 (2*512Mb), 2.5-3-3-8, single channel
ATI 9800Pro
RAID0, 2 80G SATA HDDs
Iwaki MD-20RLZT -> Storm -> Chiller -> Reservoir
PC_Power Tubo-Cool 510

AMD XP 2500+
Abit NF7
Mushkin PC3500L2 (2*512Mb), 2-2-2-6, dual channel
ATI 9800Pro
RAID0, 2 80G HDDs
Iwaki MD-20RLZT -> WW -> dual rad -> Reservoir
PC_Power Tubo-Cool 510

Results of the comparison are shown in the figures below. The first comparison uses the Prime95 benchmark while the second uses the 3DMark2001SE benchmark.

The first figure is a comparison of the results of the four platforms using the Prime95 benchmark384k FFT test. Lower scores are better. These results are an indication of the calculational ability of the CPU, memory and motherboard combo. Data for these results can be seen here.

It is interesting to note that the results of the A64 San Diego and the Opteron overlie one another as we might expect due to their similar characteristics. Also, a fairly direct comparison of all the platforms is possible at 1.992Ghz. This will represent a comparison of the CPU architectures with their supporting hardware; it does NOT reflect gains due to the different "stock speeds" each CPU might support. This comparison allows us to conclude that the A64 754 NewCastle is approximately 15% faster than the older XP CPU while the A64 939 San Diego (or Opteron) is approximately 35% faster than the XP CPU. Employing the gamesmanship of numbers, we can present the same results from the perspective of the newer platforms so that the older XP CPU is approximately 18% slower than A64 754 NewCastle while the XP CPU is is approximately 55% slower than the A64 939 San Diego (or Opteron).

The next figure is a comparison of the results from the four platforms using the 3DMark2001SE benchmark, higher results are better. Data for these results can be seen here.

These results are an indication of the graphical ability of the graphics card, the CPU, the memory and the motherboard combo. While as shown in the first figure the computational difference of the A64 San Diego and Opteron platforms overlie one another using Prime95, it is interesting to note a difference exists in this graphics test case. This difference between the A64 San Diego and Opteron platforms is attributed to the two different graphics cards used with the respective platform setups.

This graphically-driven comparison allows us to conclude that the older XP CPU-based platform with its graphics card scores approximately 50% lower than the A64 754 NewCastle platform. Because this comparison is made using the same graphics card we can conclude the improved graphical performance is principally due to the computational chain associated with CPU, motherboard and memory. Comparison of the XP CPU platform and its graphics card are approximately 75% lower than the A64 939 San Diego, or Opteron, platforms. These results do further reflect the effect of the newer PCIe graphics cards.

Conclusions to be drawn from these results?  The performance gain that is practically possible with a change from a 32-bit system to a 64-bit system can be reasonably significant. While if you overclock, very significant gains can be expected. Of course the pure % gains are dependent on where you start and where you end.