We’ve said in the past that we thought that many budget-minded current socket Aers would eventually end up with Semprons, once they were x86-64 enabled.
So when we saw this review which indicated that the Sempron falls flat on its face compared to an Athlon 64 and that even a Sempron overclocked to 2.4GHz gets whipped in gaming by a 2.0GHz Athlon 64, this was a matter of great concern to us. If true, we certainly would have to change our tune about this.
So what we did was search for some of the earlier Sempron benchmarking to see if there was any signs of this pattern in earlier reviews.
When we began to compare, the differences and discrepancies stuck out like a sore thumb.
For instance, the Legion Hardware review shows a 10% difference in memory bandwidth running at 200MHz, but other tests run at other sites show no difference at all.
More discrepancies pop up in other synthetic tests. For instance, XBit Labs ran many of the same synthetic tests. The comparisons were a bit different; they managed to overclock their Sempron 5% more, and they didn’t include a 2GHz Athlon64 in the benchmarks, but they did include a 1.8Ghz and 2.2GHz A64.
It only stands to reason that any test which increase the Sempron’s speed by 120MHz, and the A64s speed by 200MHz, would if anything increase the performance gap. On top of that, It wouldn’t narrow it, and most certainly wouldn’t narrow it a lot.
Yet that is exactly what occurs here. The Sempron loses big time to the A64 when Legion Hardware is testing, but usually wins (or at least keeps it close) when XBit Labs tests it. Given that many of the CPU synthetic benchmarks scale directly with frequency within a processor family, the XBit Labs scores make sense (they appropriately scale), while the others (usually) do not.
When it comes to gaming, if you use the XBit Labs benchmarks as a guide, you’ll see that the 2.52GHz Sempron does a bit better or at worst, a little bit worse than 2.2Ghz A64, even in the same tests when a 2.4S gets trounced by a 2GHz A64 over at LH.
You need not just rely on XBit Labs, either. Check any comparison, (here’s another for your convenience); they all show a consistent pattern of the Sempron keeping it close, MHz for MHz. There’s certainly no 15-20% performance gap between the two.
People who can’t see the forest for the trees may point out that configurations aren’t exactly the same, or even terribly similiar, but outside of a GPU-limited game, no difference in configuration would make that much relative difference (and certainly all the other Sempron/A64 comparisons were done with different equipment, too). Nor is cache a reason, the LH benchmarks were done with a Clawhammer with 1Mb cache, but so were the XB benches.
What To Do?
Unless AMD has quietly sabotaged recently-made Sempron performance (which we really doubt, indeed, for some tests, we doubt they even could), there’s something seriously wrong with a lot of the LH Sempron results.
In any event, it’s not like people are going to be running out to buy Semprons today, anyway. Given today’s price differential, if you just have to buy a system today, you ought to buy an A64, anyway.
The Sempron many of you may end up buying will be 90nm with strained silicon, and cheaper than the 3100+ you see today. There will be plenty of time for tests to be run on that, and comparisons with its older brother (we’ll definitely do that when the time comes), and those are the numbers upon which you judge the Sempron.