Athlon 64 3700+ Review

End of the Socket 754 Line?

By Josh Walrath

Errata

            I ran into one major problem with the Athlon 64 3700+ in this review.  I used Mushkin PC3200 Level 2 Black memory, which is based on the Winbond BH-5 chip.  Even with one DIMM installed, I was unable to run with the Command Rate set at 1T.  Everything ran perfectly fine with Command Rate set at 2T, but the performance hit caused the 3700+ to perform at the level of a 3400+.  Now, with the 3400+ CG revision chip I have, it is able to run that particular Mushkin memory perfectly fine at 1T.

            To address this problem, I was able to procure some of the new Corsair XMS 3200 XL Pro modules.  These are based on the latest Samsung DDR chips that perform at PC 3200 speeds at low latencies, and can run upwards of 250 MHz DDR with relaxed timings.  These DIMMs worked fine with the 3700+ at a Command Rate of 1T.

            If a 3^rd DIMM is installed, the motherboard instantly sets the Command Rate to 2T no matter what the user selects.  If a user wants to achieve optimal performance with their Athlon 64 Socket 754 setup, they should never exceed 2 installed DIMMS. 

Testing

            I decided to test the 3700+ against a C0 version of the Athlon 64 3400+ (1 MB L2 cache version).  This should show the average user what an upgrade of this nature can achieve, as well as take a look at the slight differences in performance that the C0 and CG revisions exhibit. 

MSI K8N Neo Platinum (nForce 3 250Gb chipset)

1 GB (2 x 512 MB) Corsair XMS 3200 XL Memory

120 GB Seagate Serial ATA Drive

Toshiba DVD-ROM

Sapphire Radeon X800 Pro (Catalyst 4.9 Drivers)

Audigy 2 ZS Sound Card

Antec 480 Watt Power Supply

Windows XP Professional SP2

DirectX 9.0c

Tests Used

SiSoft Sandra 2004 SP2

Sciencemark 2.0 (latest version, Feb. 2004)

3D Mark 2001 SE

3D Mark 2003

Quake III Arena 1.32

Doom 3

Realstorm Benchmark v. 1.10

WinRAR 

Results

            Sandra is nearly ubiquitous in the benchmarking field, and it has some very good results to help the average user make a better informed decision on their processor buys.

            This result runs exactly opposite of what one would expect.  The new CG revision looks to be slightly slower in overall bandwidth than the older C0 revision.  There are obviously other things at play here that we can’t observe directly which affect the overall performance of this processor.  This result should not scare anyone, as the 40 MB difference between the two should have no noticeable effect on overall performance.

            This result is more in line of what a user would expect.  The 200 MHz difference makes a world of difference when working with Athlon 64 processors.  The 9% increase in clock speed results in nearly 9% increase in performance with these measurements.  As I have said before, the Athlon 64 architecture is very efficient.

Next: More Results

Copyright 1999-2004 PenStar Systems, LLC.