SUMMARY: The flexibility of settings FSBs and multipliers gives the Duron some added performance benefits.
As I continue to play with the Duron 650/Asus A7V, I see some value in having the flexibility to vary both the CPU’s multiplier and FSBs. For example, it looks like the stable best I can do with this CPU is 927 MHz (water cooling). Now I can get 927 two ways – 9 x 103 and 8.5 x 109 with the following performance differences:
| Benchmark | 9 x 103 | 8.5 x 109 | % Diff |
| Si Sandra CPU/FPU | 2591/1292 | 2595/1292 | 0.2/NA |
| Si Sandra Mult Med | 3159/4332 | 3159/4330 | NA/NA |
| Si Sandra Memory | 444/548 | 466/581 | 4.9/6.0% |
| 3DMark 2000 | 7399 | 7532 | 1.8% |
| Quake | 127.7/123.1 | 131.4/127.2 | 2.9/3.3% |
| Content Creation | 33.9 | 34.2 | 0.9% |
| Pi Solver | 846 | 818 | 3.3% |
The difference in FSBs of 5.8% translates into 5-6% higher Si Sandra Memory scores – basically a 1-1 benefit. What is interesting is the relatively uneven impact it has on performance; it ranges from about 3% in Quake’s frame rates to less than 1% in Winstone Content Creation – I would have expected more improvement in this app-intensive benchmark.
Kevin Tam sent me a new (to me) benchmark – Pi Solver. What this does is stress the CPU by finding “X” places of Pi – up to 32 million. What’s nice about this one is it not only stresses the CPU but records how many seconds it takes to get “X” places. In this instance, for 4 million places, at 8.5 x 109 it took 818 seconds – 3.3% better than at 9 x 103 (Note: 32 million places @ 927 took 4 hours, 5 minutes and 19 seconds).
The bottom line is that memory intensive apps, like image or sound editing, will show a performance improvement without any penalties (assuming your peripherals like the higher FSB).
Just for giggles, I also benched the Duron 650 with W98SE and W2000 @ 800 MHz with the following results:
| Benchmark | W98 SE | W 2000 | % Diff |
| Si Sandra CPU/FPU | 2256/1125 | 2248/1125 | -0.4/NA |
| Si Sandra Mult Med | 2750/3771 | 2684/3441 | -2.4/-2.8% |
| Si Sandra Memory | 436/537 | 446/522 | 2.2/-2.8% |
| 3DMark 2000 | 6908 | 2971 | -57.0% |
| Quake | 117.7/113.9 | 96.8/90.1 | -17.8/-20.9% |
| Content Creation | 32.2 | 42.7 | 32.6% |
Let’s chalk up the gaming gaps to “immature video drivers” and leave it at that. However, the Content Creation scores are a shocker – no doubt NT is “industrial grade”, but by this much?? We have seen similar results with the T-bird, so we will be doing more testing on this issue.
At this juncture, I am pleased with running a Duron 650 at what appears to be a stable 927 MHz (1.85 v) – a 50% overclock. I am going to set this up as one of my main systems, continue to use it in everyday apps and see how it behaves.
Be the first to comment