Comparison

Again, we have a look at the results obtained using Sampsa’s Phenom II X4 965 C3. This time, the amount of comparable data is a bit larger as Sampsa has, just as we, tested the influence of the temperature on the operating frequency in different steps. So, for this article, we can compare the results he obtained with his Deneb system and match it to our results. There was a small issue, however, as Sampsa’s results are obtained with 1.4V Vcore, whereas we only have 1.3V and 1.5V measurements. So, to overcome this issue, we have added an approximate value for the 1.4V overclockability which is based on the average result of the 1.3V and 1.5V.



Judging from the above results, it seems that the Thuban is performing a bit better than the Deneb when comparing at a low voltage level. Given the fact that the Thuban comes with a TDP equal to the Deneb, but 2 extra cores, it looks like the silicon quality of the Thuban is just superior to the one of the C3 Deneb. This may not be that good news for the extreme overclocker as higher quality silicon doesn’t always mean higher ultimate clock frequencies (higher quality = lower leakage), but I do believe that the water-cooling people will see quite good results. Also, given that the temperature scaling at low voltages seems to be a tad better on Thuban, it’s possible that adding a TEC to your system will finally be rewarding again!

We have to note, however, that the results in above graphs are all based on the results of just one Deneb processor and just one Thuban sample. It’s not difficult to see where the problem is located: we have no idea to what extent these figures are representable for the entire group of Deneb and Thuban samples, so we just have to assume that both are a good representation of the average Deneb/Thuban without testing 100+ other samples to confirm this. Fortunately, we do have a very large database which offers us at least a couple of Deneb comparison numbers. Let’s pull some data.

We have run a query to find average and maximum clock frequency using a Deneb core processor in the 3DMark06, 3DMark Vantage or Wprime 1024M benchmark. This information is then compared to the data we have from the two different tests. We have eliminated false data and also tried to remove the results that have not been maxed out (e.g.: lower clock limitation of 3.5GHz on air cooling).



Our expectations and findings are somewhat confirmed when looking at the HWBOT database. AMD’s latest acts as if it were a highly-binned Deneb sample. Our random sample is significantly better than the average Deneb in the HWBOT database until the moment LN2 has been used. The Deneb sample Sampsa used in his test session is slightly better than the average Deneb, it seems.