Let’s say you have a Phenom II 955 BE clocked at its max water-cooling overclock, which is roughly 3.8 GHz. The current voltage setup for the CPU is 1.55v on the CPU with a 2.7 VDD. The CPU-NB is at 1.35v with a Frequency of roughly 2.8 GHz. If you were to disable a core so that the CPU was now a hypothetical “720 BE”, you would think less voltage is required to get the same speed.
In theory this makes sense, but in reality it really didn’t work out this way. What I found out is that when the cores are disabled they are not really turned off 100% – there is still a current going through them, just no data. This means that you are required to use the same amount of voltage even if some of the cores are disabled. This makes sense in the fact that, if you compared a 720BE to a 955BE at the same Frequency on the CPU and CPU-NB, you would need ±.05v on the CPU and ±.02v on CPU-NB.
To help show what I explained above, I have some charts for you. They include tests that are CPU, GPU, and RAM intensive. These tests show the absolute effect of what happens when you downcore a Phenom II.
Phenom II 940 BE
4870 512MB ASUS Matrix Cat 9.7
Windows XP 32Bit SP3 (Fully updated)
How to read the graphs:
Each graph is associated with a different benchmarking tool, and each one has a different scoring guide (see title of graph). On the X-Axis are the scoring numbers associated with that particular benchmarking tool. On the Y-Axis (in order) is the Frequency and then the Number of Cores enabled on the CPU. For example Max 4 means Maximum Frequency acquired with 4 cores enabled.
Max speed = 4.3 GHz (acquired using Dry Ice and Acetone) (Cooling Pot used: K|ngP|n Cooling Venom)
The wPrime and Cinebench Benchmarking tools were the perfect example of CPU intensive benchmarks. These benchmarks require not just speed, but how many cores are on the CPU and how fast each one is.
For Crysis, this was by far the more interesting of the bunch. I used a pre-developed CPU benching configuration. Even though the benchmarking tool was supposed to make the CPU work hard, it seemed that the tool concentrated a lot on the Graphics. So thus, I must conclude with this tool that all data is inconclusive.
For SuperPi 1M, this was a shocker. I thought this tool completely relied on how many cores were present, not the speed. The benchmarking tool completely relies on how fast the CPU is going, nothing else. What is great about this tool is that in some instances I can increase the CPU Frequency much more than with the other tools and still benchmark with SuperPi 1M.
In conclusion with this experiment, I must say that removing cores to gain a higher frequency probably will not affect anything just as long as the programs being used are not multithread intensive. For computer games and all other tools, you could lower the cores and try to increase the frequency, but with my experience there was not too much gain in CPU frequency. Your CPU could be different.
Look for Part Two of this article in the near future. I will be taking the 940BE into extreme cooling to see if my theory still holds true, even at high overclocks. Until then Good Luck, Have Fun, Overclock!
Dolk on the Forums or send me an email at firstname.lastname@example.org