Michael Larabel from Phoronix was kind enough to carry out two different testing for the recently released kernel 3.1 on top of our Intel Linux Graphics cards today (thanks again Michael!). He split his evaluations into two different articles – First one is focused on performance and power usage for stock 3.1 kernel; and Second one specifically enabled RC6 on this kernel to see its effects in action.
His testing confirms what I was pointing out with my latest series of the post (however, it always feels great to have independent confirmation of the results). For performance numbers, kernel 3.1 brings around +20% to +40% improvements on top of the previous release for Sandy Bridge architecture. Yeah, this is nice. So if you haven’t updated your kernel to the latest and greatest one (in other words, kernel 3.1), you should really consider doing this.
As for power numbers, we have some good news too. The default kernel power usage on 3.1 is mostly similar to 3.0, with some small variations here and there. However, if one enables RC6 support, the immediate results is that your power usage drops by up to 40% on idle. It is really nice to see less than two-digit watts numbers on current generation of desktop hardware (while it is pretty common to see it on Atom CPUs for some time already, desktop and mobile versions of Intel cards were not that used to such numbers – yet).
One additional bonus Michael has found out is pretty much more interesting one. Apparently, by enabling RC6, the graphics-intensive applications get an additional performance boost of up to 10%. This was a bit unexpected for me, but it is better to have good surprises than bad ones.
From what we have discussed with Jesse Barnes, such improvements could come from several different paths. The first explanation is that, with RC6 enabled, the graphical card can actually consume much less power (down to 0V), so it leaves more room for CPU to use the non-claimed power to do more processing of its own. And the second one is that, thanks to additional thermal bonus which we get from the RC6-provided power economy, the GPU frequency has more room for scaling. So effectively, in both of those case, RC6 allows you to have a better performance, both CPU and GPU-wise – at a cost of somewhat higher power usage coming from such performance.
So, in few words, I would define this situation as battery when you want, performance when you need. When you are on battery, under mostly idle workload (for example, browsing or reading or writing some blog posts), your battery lasts much longer. And when you start some processing-intensive application (say, openarena or angry birds), it gets more horsepower to get the job done, and some extra FPS which could mean life or death – which is specially true in case of said workloads . Of course, those additional horses are hungry, so this incurs in more power being used when such additional performance is there – but it is something to be expected in any case.
While writing this description, I actually came to realize that the feature known as GPU turbo, which is present on Intel graphics cards, could be used to define and control such behavior on a much finer granularity. From what I’ve searched all around, surprisingly, nobody seems to have covered its functionality yet. I guess I’ll try to cover how it works, and how one could use and control it from userspace in one of the next posts, so stay tuned! And meanwhile, enjoy all those nice power and performance-related bonuses which the Intel Linux Graphics team brought to you just in time for Halloween .