SUMMARY: Impressive performance, although the T-Bird 750 at twice the cost does not deliver twice the performance. Duron: Price/Performance champ overall.
One of the questions I was asked when we were doing benches on the Duron 650 @ 935 was “How does it compare to the T-Bird?” Not having one, I could not answer the question. Now with the Blue 750 in hand, I ran some benches comparing the Duron against the T-Bird.
I ran two different sets: One comparing the Duron and T-Bird head-to-head, same speed. The second comparing the Duron’s best speed against the T-Bird’s. All tests were on the ASUS A7V with exactly the same components – only the CPU was changed. Results:
| Bench | Duron 935 | TBird 935 | % chg T vs D | TBird 1100 | % chg T vs D |
| SiSandra CPU | 2609/1304 | 2609/1304 | NA | 3070/1534 | 17.7/17.6 |
| SiSandra MM | 3188/4370 | 3186/4370 | NA | 3749/5142 | 17.6/17.6 |
| SiSandra Memory | 489/599 | 488/603 | NA | 489/608 | NA |
| Quake | 132.2/127.9 | 141.6/138.7 | 7.1/8.4% | 152.5/149.1 | 15.4/16.6% |
| 3DM2000 800x600x16 | 7545 | 8150 | 8.0% | 152.5/149.1 | 15.4/16.6% |
| 3DM2000 800x600x32 | 6397 | 6639 | 3.8% | 6796 | 6.2% |
| Pi Solver 1M | 181 sec | 156 sec | -13.8% | 140 sec | -22.7% |
| Content Creation | 31.4 | 33.6 | 7.0% | 36.8 | 17.2% |
I did encounter some strange anomalies while benching the T-Bird 750; I did manage to get benches at both 935 and 1100, but at 935 the system just was not stable with the T-Bird. I had many freezes in 3DMark 2000 and Content Creation, while I had NONE with the Duron 650 @ 935. However, the T-Bird ran Prime95 for over four hours at 10 x 110 without a problem, suggesting that the issue may be video related.
I suspected that running the FSB at 110 might be a factor, so I ran the benches with the AMD 750 using 11 x 100 instead of 10 x 110. Result: NO PROBLEMS! I infer from this that perhaps the ASUS A7V may have some stability issues with the more powerful T-Bird. Note that these issues look to be video related – running number-crunching routines like Prime95 or Pi Solver did not exhibit any instabilities.
I will do more experimenting with BIOS settings to see if somehow there are AGP issues that can be resolved.
Meanwhile, I did a set of benches to show the performance difference between running 1100 at 10×110 and 11×100:
| Bench | T-Bird 10×110 | T-Bird 11×100 | % chg |
| SiSandra CPU | 3070/1534 | 3098/1494 | 0.9/-2.6% |
| SiSandra MM | 3749/5142 | 3738/5188 | -0.3/0.9% |
| SiSandra Memory | 489/608 | 457/548 | -6.5/-9.9% |
| Quake | 152.5/149.1 | 149.3/144.4 | -2.1/-3.2% |
| 3DM2500 800x600x16 | 8632 | 8510 | -1.4% |
| 3DM2500 800x600x32 | 6796 | 6754 | -0.6% |
| Pi Solver 1M | 140 sec | 142 sec | 1.4% |
| Content Creation | 36.8 | 34.8 | -5.4% |
The biggest difference shows up in SiSandra memory scores, with the impact varying across the benches; worst case, expect about a 5% performance hit. In many instances, I doubt you will perceive a difference on the screen. The hit is not too severe and if stability is an issue in your particular app, running at 100 FSB is an OK compromise.
Bottom Line: AMD makes a very impressive CPU, giving Intel a real run for the $, then decides to irreversibly multiplier lock them – effectively thumbing its nose at the “performance market”.
The issue confronting us is the viability of FSB overclocking Socket A CPUs; right now, Socket A motherboards are running “crippled” – none seem capable of getting beyond 120 FSB. Makes you wonder what’s at work here – something we plan to look into more closely.
Be the first to comment