|
Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home |
. | . |
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ |
LOSTCIRCUITS
|
|
| AMD K6-III+ and K6-2+ Shaptooth's Final Bite | |
|
(Review by Aaron Vienot, July 13, 2001) |
| AMD Athlon64 3200+ At: |
Expendable
Expendable is still one of the most intensive CPU and memory benchmarks. All tests were performed at 640x480x16 and results are given in fps:

The K6-III+ continues to show a modest 9-12% gain over the K6-2. Interesting about Expendable is that, unlike many game timedemos, it is not a recorded demo in playback; rather, it is an executed sequence of commands and consequent events. This means the system starts with pre-programmed user inputs and the system must calculate the enemy reactions, rendering, elimination of the dead, etc. on the fly. Consequently, Expendable timedemo resembles a rapid staccato of alternating read and writes to the system memory with relatively little load on the graphics adapter. The heavy dependence on memory bandwidth and factors like read-write turnaround makes Expendable an excellent benchmark for both memory controller and CPU performance. Thus the bus speed becomes more important than the actual CPU clock speed, which is reflected in the 504 (112) setting outperforming the nominally faster 550 (100) option on either processor.
Quake2
Quake2 results are based on runs of Brett"3Fingers"Jacob's Crusher demo at 640x480, 800x600, and 1024x768. Tests were performed with both the AMD 3DNow! drivers, and with the standard Quake2 OpenGL drivers. The 3dfx mini-GL drivers showed no performance gains over the standard OpenGL renderer, and was neglected. Because of the different resolutions, there are a lot of data that are easier viewed in tables.

640x480
| Speed | K6-III+ Std | K6-2 Std | K6-III+ AMD | K6-2 AMD |
| 500 | 30.6 | 26.2 | 40.5 | 34.2 |
| 504 | 31.9 | 28.3 | 42.5 | 36.2 |
| 550 | 32.5 | 27.2 | 42.7 | 35.5 |
| 560 | 34.4 | 30.2 | 45.2 | 39.4 |
800x600
| Speed | K6-III+ Std | K6-2 Std | K6-III+ AMD | K6-2 AMD |
| 500 | 30.9 | 26.8 | 39.2 | 34.4 |
| 504 | 32.1 | 28.4 | 40.9 | 36.8 |
| 550 | 32.8 | 27.8 | 41.2 | 35.8 |
| 560 | 34.1 | 28.5 | 41.6 | 37.6 |
1024x768
| Speed | K6-III+ Std | K6-2 Std | K6-III+ AMD | K6-2 AMD |
| 500 | 29.6 | 25.9 | 34.7 | 31.9 |
| 504 | 30.5 | 27.9 | 35.2 | 33.2 |
| 560 | 32.2 | 28.9 | 35.8 | 32.8 |
At one glance, here are the results fr the Crusher Demo at 640x480 x 16 bpp:

The K6-III+ with the 3DNow!-drivers is significantly faster, while the K6-2 under the same conditions trails the K6-III+ by several frames; also, with or without the optimizations, the K6-2 still falls behind on 100MHz bus-based clocks while its newer relative gains at every clock although the 112MHz bus-based settings show stronger increases.
Using the standard OpenGL drivers on either CPU, very little change occurs between resolutions at any clock speed; framerates vary by at most 2 fps. For the optimized drivers, however, not only do all resolutions and speeds see gains, but lower resolutions see disproportionately higher gains.
The reason for this strange but well-documented phenomenon is that without optimized drivers, the CPU is unable to take advantage of either the dual, single precision floating point unit for triangulation or the write combine buffers. Thus every calculation utilizes only half of the FPU and is further interrupted by a single bit write after each FPU operation. The two factors effectively reduce the overall power of the CPU to 25% of its total potential; therefore, resolution becomes a negligible factor.
Q3A
Q3A uses the system OpenGL driver, in this case provided by the Voodoo3's driver set. The Voodoo3 only renders 3D at a 16-bit color depth - this is important to remember because the High Quality setting normally specifies 32bit color. Demo001 was tested at Q3A's four standard settings, with no customizations: 512x384 Fastest, 512x384 Fast, 640x480 Normal, and 800x600 High Quality. The average frame rates are shown below.
K6-2 Q3A
| Speed | fastest | Fast | normal | high-q |
| 500 | 49.1 | 42.5 | 41.1 | 40.3 |
| 504 | 52.7 | 45.4 | 44.2 | 42.9 |
| 550 | 51.4 | 44.4 | 43.5 | 42.1 |
| 560 | 55.5 | 47.9 | 46.6 | 44.8 |

K6-III+ Q3A
| Speed | fastest | Fast | normal | high-q |
| 500 | 61.0 | 53.8 | 52.7 | 50.0 |
| 504 | 64.5 | 56.5 | 55.0 | 51.8 |
| 550 | 64.4 | 56.9 | 55.6 | 52.0 |
| 560 | 68.3 | 59.9 | 58.4 | 54 |
A chart hardly seems necessary here. The summary observation is that the K6-III+ enjoys a healthy 17-22% lead over the K6-2 at any clock and any performance setting, and the K6-2 is once again demonstrated to be more bus-speed dependent than the K6-III+.
For the gamer limited to Socket 7 and willing to place raw speed above visual quality, a score of 68.3 FPS, achieved by the K6-III+ running at 560MHz with the Fastest setting, is certainly adequate. However, even at the High Quality setting, a score of 58 fps is closing in on the limitations of the V3 used in the test setup and with a GeForce, these numbers might turn out somewhat higher.
next page: => Conclusion =>
All advice and educational articles on LostCircuits are free, but if you feel you can, please make a small donation to us!