Winbench Benchmarks
Winbench Synthetic Results
| Winbench 99 CPUMark 1.1 |
| Processor |
Score ( CPU Winmarks) |
| Cool K6-3 550 |
55.9 |
| K7 550 |
49.1 |
| P3 Xeon 500 1024KB |
44 |
| P3 500 |
38.3 |
CPUMark Analysis
Very interesting initial results! While the K7 blows away the standard Pentium III, the
Kryotech K6-3 550 we previously tested similarly blows away the K7. CPUMark99 tests not only the CPU itself, but also any of the low level subsystems closely tied to it, including RAM and motherboard.
It's entirely possible that existing settings on the K7 are set conservatively or for debug purposes. It should also be noted that the comparison systems for these tests are not configured identically to the K7 (hence our prior warning) and these numbers should not be considered absolute.
| Winbench 99 FPUMark |
| Processor |
Score (FPU Winmarks) |
| P3 Xeon 500 1024KB |
2580 |
| P3 500 |
2550 |
| Cool K6-3 550 |
1860 |
| K7 550 |
1590 |
FPUMark Analysis
Ugh. Abysmal. A score this low really make no sense, especially when compared to an identically clocked K6-3. Let's take a closer look at the K7's floating point capabilities. The K7 has 3 fully-pipelined FP units as opposed to the Pentium III's 2 pipelined units. What's more, one of the P3's FP units is only partially pipelined, and a multiplication instruction must be immediately followed by an addition instruction in order to reach its peak performance.
The K7 on the other hand, has 3 fully-pipelined FP units, one for additions and subtractions, another for multiplication and complex operations, and another for FSTORES and special instructions, such as MMX and trigonometric calculations. Each of these units can operate indpendently, and in parallel, making for a theoretical maximum throughput of roughly twice a P3-550.
Especially when compared to the following floating-point intensive tests, the most likely explanation is an exception in the FPUMark program that doesn't take particularly to the K7.