Thuban Review: Mythbusting if OCing really reduces Efficiency

Very interesting results if you're looking for the ideal performance/watt ratio on Thuban.

First I would like to say this was a blast. Its the first time I ever saw a CPU run lower than 0.85v and see my Kill-a-watt meter hit 300W just from a cpu test (My typical system runs at 300W with the 4850 furmarking and the cpu linpacking).

The results, although are not perfect, show a pretty clear goal of the CPUs. As the system uses less and less power, the TDP becomes more apparent, and we notice how well the lower clocked variation do. While machines with performance parts and power draining parts have the higher TDP chips winning. It makes sense that there are so many desktop parts in the 65W to 125W range, its would be horrible to slap a 25W chip into a motherboard that drains 50W and think it can compete. I would love to try this again with an ITX board and IGP to see what the middle of the road nettop can do.


1

Belgium Massman says:

Here's a bit more. Just check out the original XS thread for all the info.




Here we are simply looking at the TDP power consumption of 1200MHz at different voltages. Each notch on the X axis is voltage step recognized by CPUz, with a label on every 4 notches.

To be honest i started at 1.25VID equaling 1.272v and continued to lower the voltage until i began to freak out about the fact I'm able to complete runs well under 0.8v. Total System Power for this ranged from 101 watts to 134 watts.




This chart shows the minimum voltage needed for a complete run of Cinebench. It was easy to find out the next drop in voltage wouldnt work when I was prompted with a pretty blue screen. The blue dot represents normal Thuban voltage of 1.30v for 3GHz with the axis lined up along him. (Get use to seeing that blue dot in future charts)

There seems to be some trends here, anything from .75v up to 1.10v required a steady bump in volts as the frequency was increased. While everything around 3GHz were all using nearly the same volts. Then in the upper 3.6GHz range drastic changes in volts were required for every tiny bump.

And how did that translate into power consumption?




With the slowest speed and lowest voltage, the system is pulling in a third of the power compared to the highest speed. Compared with the blue dot, a 3Ghz CPU can shave over 40W off the system draw while maintaining the same frequency. It was only until 3.6GHz that the Overclock began to surpass the stock draw.




When comparing the calculated TDP things look even more drastic. The 4.2GHz run is only 3.5x faster than 1.2GHz, but is drawing 7x the power. The charts curve is identical, only scaled slightly and shifted from the conversion between System Draw and CPU TDP, so its no surprise that even on the TDP chart, 3GHz stock is the same as the 3.6GHz OC.

its also interesting how 2.4GHz hexcore cpu is still under 50W, making it quite scary that it can fit into a laptops power envelope.




Thats pretty accurate if you ask me. Since the Bus speed was locked in at 300MHz, and NB, HT, RAM settings were all identical for every increase in MHz, the score should be extremely linear.




The first final results image, and quite shocking. The optimal speeds to run a multithreaded app with thuban seems to be any frequency between 2.4 and 3.3GHz. And you can run as high as 4GHz before its efficiency is lower than stock settings. So basically users who bump up the Volts to 1.35-1.40v and try and max out the speed, will see an increase well over the stock configuration. In some cases it might be possible to see as much as 30% higher performance per watt.




And our last chart shows the TDP instead of System Power Draw with mostly lower clocked highly under-volted results winning. With a CPU running around 2GHz its only ~66% as fast as the 3GHz part, but was done so with less than 0.90v and only about 30-40% the TDP as the 3GHz part

Please log in or register to comment.