NineBall wrote:
Ever since I heard of these 2 chips recently, I've been wondering how these would stack up against the Phenom II X4 905e. I'm currently using a 4850e in my HTPC and have been considering the 905e as an upgrade. The 2 chips reviewed here have higher frequencies but also higher stock TDP and no L3 cache. Considering I can source a 630 @ ~140 CDN and a 905e @ ~195 CDN (~40% premium), I'm wondering how close they are in efficiency and performance? I'm still leaning to the 905e but am now a little torn.
Anantech has a new tool which will allow you to compare the 4850e with the 910, which as close as you're going to get with their list of cpu's, which is pretty close.
Mike, one thing you didn't mention in the article (it was very good) was how the memory divider circuit works on the AMD processors. Anything other than an even whole multiplier may not be running your memory as fast as you think. Here's the 2 cpus covered in this article.
The 630 at 2.8GHz runs DDR2-800 at it's full speed. Here's the calculation, using 400MHz for the memory, the real speed not DDR2 speed:
2800 (CPU) divided by 400 (memory) = 7.000. There is no remainder, so the divider does not round the number up. So the divider remains at 7 and we have 2800 divided by 7 (the divider) = 400 or full speed.
The 2600 on the other hand is not as good. 2600 divided by 400 = 6.5. The memory divider can't handle fractions and will not overclock the memory, so the divider it chooses here is also 7. So 2600 divided by 7 = 371. So your DDR2-800 memory is actually running like DDR2-742.
The 4850e, a 2.5 GHz model, has the same issue. 2500 / 400 = 6.25. It also gets rounded to 7. 2500 / 7 = 357, or DDR2-714. You can confirm all this with CPU-Z. I run my 4850e at 2.4GHz, so that memory will run at full speed.
The memory divider doesn't take OCing into the calculation so when you up the FSB you up everything, but you still need to know where you start.
Mike, the UV settings seem right in line with the "e" series of CPUs. They run at 2.5GHz with a tiny amount above 1.2 volts, setting a larger safety margin voltage-wise than your findings, but probably a good idea to ensure everything off the assembly line runs without the need for individual testing and settings. You didn't mention if the Q8200S runs at reduced voltages, did you?
I applaud Mike's UV testing. What the people who do full time folding are slowly finding out is that undervolting often leads to far more efficient use of power, which is a big deal when you pc runs under 100% load 24/7.