Intel SNB RC6 On Linux 3.1 Is Both Good & Bad
As a follow-up to this morning's Linux 3.1 Enhances Sandy Bridge, Preps For Ivy Bridge article that looked at the Intel SNB graphics performance with the new Linux 3.1 kernel, here's a look at the power consumption when enabling RC6.
RC6 is the power-savings feature that was originally enabled during the Linux 3.1 kernel merge window but later reverted as it wasn't stable enough yet across the spectrum of supported hardware. Frame-buffer compression is enabled in the Linux 3.1 kernel for Sandy Bridge (Gen6) hardware, but it won't be until the Linux 3.2 kernel that this power savings feature is on by default. For now you can toggle the RC6 support yourself using the i915.i915_enable_rc6=1 option.
After running some new Intel SNA acceleration benchmarks (will be published on Phoronix next week), I ran some quick tests from the Linux 3.1 final kernel with and without RC6 support. I did some Intel RC6 benchmarks back in August, but since then Intel has made other changes to the RC6 support. The HP EliteBook with its Intel Core i5 2520M CPU had shaved off more than two Watts while idling with RC6 enabled. The power consumption with RC6 also bottomed out four Watts lower than the stock 3.1 kernel.
While RC6 was of benefit while idling, when running the OpenArena game that placed a fair amount of load on the Sandy Bridge graphics, the power consumption was worse with RC6. Enabling RC6 caused an average of seven more Watts to be consumed while running this Linux game.
The frame-rate was just slightly faster with RC6. Having RC6 enabled while running Warsow increased the power consumption by eight Watts.
The frame-rate went up by just three FPS with RC6.
Having RC6 enabled, which will be the Linux 3.2 kernel default for the Intel DRM driver, does lower the power consumption while idling. Under load, however, the power consumption is actually worse.
Eugeni Dodonov, one of Intel's newest Linux employees, wrote in an email to me shortly after sharing the results with him, "Interesting, it is a bit aligned in the sense that when the GPU/CPU is under load, the driver tries to detect it and deliver more power to get the workload to finish earlier. But I haven't related this to rc6 directly so far. Thanks for letting me know, I'll certainly investigate this!"
These new Intel RC6 results are available on OpenBenchmarking.org along with the CPU thermal results.
RC6 is the power-savings feature that was originally enabled during the Linux 3.1 kernel merge window but later reverted as it wasn't stable enough yet across the spectrum of supported hardware. Frame-buffer compression is enabled in the Linux 3.1 kernel for Sandy Bridge (Gen6) hardware, but it won't be until the Linux 3.2 kernel that this power savings feature is on by default. For now you can toggle the RC6 support yourself using the i915.i915_enable_rc6=1 option.
After running some new Intel SNA acceleration benchmarks (will be published on Phoronix next week), I ran some quick tests from the Linux 3.1 final kernel with and without RC6 support. I did some Intel RC6 benchmarks back in August, but since then Intel has made other changes to the RC6 support. The HP EliteBook with its Intel Core i5 2520M CPU had shaved off more than two Watts while idling with RC6 enabled. The power consumption with RC6 also bottomed out four Watts lower than the stock 3.1 kernel.
While RC6 was of benefit while idling, when running the OpenArena game that placed a fair amount of load on the Sandy Bridge graphics, the power consumption was worse with RC6. Enabling RC6 caused an average of seven more Watts to be consumed while running this Linux game.
The frame-rate was just slightly faster with RC6. Having RC6 enabled while running Warsow increased the power consumption by eight Watts.
The frame-rate went up by just three FPS with RC6.
Having RC6 enabled, which will be the Linux 3.2 kernel default for the Intel DRM driver, does lower the power consumption while idling. Under load, however, the power consumption is actually worse.
Eugeni Dodonov, one of Intel's newest Linux employees, wrote in an email to me shortly after sharing the results with him, "Interesting, it is a bit aligned in the sense that when the GPU/CPU is under load, the driver tries to detect it and deliver more power to get the workload to finish earlier. But I haven't related this to rc6 directly so far. Thanks for letting me know, I'll certainly investigate this!"
These new Intel RC6 results are available on OpenBenchmarking.org along with the CPU thermal results.
1 Comment