Athlon 64 3700+ Review

End of the Socket 754 Line?

By Josh Walrath

Sciencemark 2.0

            Sciencmark is a solid set of metrics that can be applied to any X86 based processor.  While Dr. Tim Wilkins now works at AMD, he has attempted to keep this benchmark as independent as possible, and in so doing has created a very solid benchmark to measure performance.

            The increase in clock speed has a direct effect on the bandwidth of the L1 and L2 caches, since these of course are integrated onto the die and run at full CPU speed.  Here we also see that the CG revision is again slightly slower than the C0 when it comes to main memory access.  While the Athlon 64 is still based on the Athlon architecture, AMD has really tweaked the cache accesses.  The old Athlon XP can’t match these numbers when it comes to L1 and L2 accesses.

            Here again we see a nearly linear increase in performance with clockspeed.  The overall design and integrated memory controller allows the Athlon 64 to not suffer the bottlenecks that the traditional CPU-Northbridge-Memory configuration often imposes.  The Athlon XP and Pentium 4 cannot come close to the linear performance improvements that the Athlon 64 exhibits.

            The floating point unit of the Athlon 64 is robust to say the least, and these results are still significantly better than a Pentium 4 C clocked at 3.4 GHz (84.92 s for MolDyn and a whooping 401 seconds for Primordia). 

3D Mark 2001 SE

            This (ancient) benchmark has proven itself to be a very solid indicator of processor performance, as it is no longer so video card bound at the standard resolution and color depth.

            The 1088 mark difference only constitutes about a 4.3% increase, but when considering that this benchmark is mainly aimed at video card performance, it is still a pretty impressive increase. 

3D Mark 2003

            This latest version of the 3D Mark is very focused on graphics performance, but it does have two specific CPU tests that really help to show the difference in performance between CPU’s and architectures.

            A 6.2% and 4.5% difference respectively is again nothing to sneeze at when looking at a 200 MHz difference between these processors, running on the same motherboard with the same memory and video card. 

Doom 3

            This latest creation of Id is extremely CPU and video card intensive.  Resolution was set to 640x480 for both tests.

            While the High Quality results show very little difference, the Ultra Quality results show a much more significant difference.  This is because half of the textures used in a scene are stored in main memory and are accessed by the video card when it needs those (Ultra Quality uses over 500 MB of textures, while High Quality is much closer to 256 MB).  This puts a greater strain on the CPU and its memory controller, and this is where we can see the biggest difference between the two CPU’s.

Next: More Results and Overclocking

Copyright 1999-2004 PenStar Systems, LLC.