QuickLinks Home  GA-586HX  Hot-Swap  AP5T-3  Q3Test/TNT2  PA-2012  H/W-Hack  JiuZhaiGou  Wallpaper 

Benchmarks

  1. Full Results
  2. Comparision Chart
  3. K6-III vs K6-2
  4. System BIOS Cacheable - Any Difference?
  5. BIOS v1.80 and Write Allocation
  6. The Configuration
Back to Contents

1. Full Results

I did some benchmarking of the K6-III and K6-2, with various clock multipliers, bus frequencies, cache and BIOS settings, to determine if there were any interesting differences in performance. I used Sandra's CPU and Multimedia benchmarks, ZDBOp's Winbench 98 and Winstone 98 to do this. After each test, I defragged the hard disk and rebooted the system, so that each test had an equal starting point.

Here are the full results :-

  Sandra CPU Benchmarks Sandra Multimedia Benchmarks ZDBOp Benchmarks
CPU/Config CPU Dhrystones (MIPS) FPU Whetstones (MFLOPS) Integer MMX (it/s) Floating-Point 3DNow! (it/s) Winbench 98 CPUMark 32 Business Winstone 98
K6-III 411MHz, 75x5.5, WA=on, SBC=off 1279 494 1099 822 1080 25.5
K6-III 400MHz, 66x6.0, WA=on, SBC=off 1242 478 1067 798 - -
K6-III 400MHz, 66x6.0, WA=off, SBC=off 1158 478 1067 798 - -
K6-2 375MHz, 75x5.0, WA=on, SBC=on 1075 446 999 747 623 21.1
K6-2 366MHz, 66x5.5, WA=on, SBC=on 1048 437 978 731 - -
K6-2 366MHz, 66x5.5, WA=on, SBC=off 1048 437 978 731 - -

Note :
WA = Write Allocation
SBC = System BIOS Cacheable
- = Not tested

Back to Top

2. Comparision Chart

Using the Sandra results, I generated this chart :-

Back to Top

3. K6-III vs K6-2

I compared the K6-2 at 375MHz with the K6-III at 411MHz. These are the fastest speeds I tested for both CPUs. Looking at the Dhrystone results, there is a 19% gain by the K6-III. The Whetstone showed a less impressive 10.8% gain by the K6-III. Both the MMX and 3DNow! tests showed a 10% increase by the K6-III.

Winbench 98 CPUMark 32 gave an extreme result, with the K6-III having a 73.4% increase over the K6-2! However, Business Winstone 98 showed a more consistent result of 20.9% increase. The CPUMark 32 test is probably over-affected by the K6-III's on-chip L2 cache.

Back to Top

4. System BIOS Cacheable - Any Difference?

In my benchmarking of the GA-586HX, the required setting of System BIOS Cacheable=DISABLED for the K6-III had a negligible impact on performance. I wanted verify this on the AP5T-3. To do this, I used the K6-2 and ran Sandra with the setting ENABLED and DISABLED. The Sandra results showed no measurable difference.

Back to Top

5. BIOS v1.80 and Write Allocation

The latest available AP5T-3 BIOS does not enable Write Allocation(WA) although AOpen's AP5T-3 BIOS Download Page says that the BIOS can "Recognize AMD-K6(tm)-3 CPUs". Fortunately, WA can still be enabled with utilities like SETK6V2 and SETK6V3.

To determine WA's impact on performance, I benchmarked the K6-III 400MHz with Write Allocation ON and OFF. Sandra's Dhrystone test showed a 7.3% performance increase with Write Allocation ON. Interestingly, WA had no impact on the Whetstone, MMX and 3DNow! results.

Back to Top

6. The Configuration

My test system configuration is as follows :-

Mobo AP5T Rev 3.1
BIOS v1.80
CPU K6-2 CXT 350MHz and K6-III 400MHz
RAM 64MB (4 x 16MB 60ns FPM)
Graphics Card Matrox Millenium 4MB PCI
Hard Disk Western Digital AC22000 2.0GB
Floppy Disk 3.5" 1.44MB
OS Windows 95 OSR2 + AMDK6UPD Patch
Back to Top
Back to Contents

Disclaimer

Adrian
Last revised: April 19, 2000.
QuickLinks Home  GA-586HX  Hot-Swap  AP5T-3  Q3Test/TNT2  PA-2012  H/W-Hack  JiuZhaiGou  Wallpaper