welcome to bizarro world, enjoy your stay —

Ice Lake GPU underperforming? Put it in powersave mode. Wait, what?

Linux users discovered Ice Lake laptops were actually faster in powersave mode.

If re-envisioning Tux as the most lovable kaiju doesn't improve your day, I don't know what will.
Enlarge / If re-envisioning Tux as the most lovable kaiju doesn't improve your day, I don't know what will.

Back in December, Linux users were starting to notice that Ice Lake-equipped laptops were getting better framerates in powersave mode than in performance mode. This Tuesday, Intel developer Francisco Jerez released a patchset to address the conundrum. Jerez begins by noting the fact that if your system bottleneck is I/O, boosting CPU performance won't help—the CPU can't process more data if the I/O subsystem isn't providing it fast enough.

"In IO-bound scenarios (by definition) the throughput of the system doesn't improve with increasing CPU frequency beyond the threshold value at which the IO device becomes the bottleneck."

Jerez goes on to note that pointlessly boosting the CPU into turbo frequencies when there's no additional data for it to process doesn't just hurt power efficiency. In the case of laptop designs, there's typically no room for desktop- or server-style "overengineering"—you've got limited space as well as limited power. This means, among other things, that there's only so much cooling to go around.

"With the current governors [...] the CPU frequency tends to oscillate with the load, often with an amplitude far into the turbo range, leading to severely reduced energy efficiency."

This is particularly likely to become a problem in games, where the GPU portions of the processor get a much bigger workout than the x86 CPU itself. If you unnecessarily boost CPU frequency, you waste your TDP budget dealing with unnecessary waste heat from the CPU—which may in turn require throttling the GPU, since you've already eaten through your thermal overhead.

The new patchset, which attempts to better target frequency scaling to real-world bottlenecks, is Jerez's second attempt to address the problem, and he's looking for additional testing to confirm his own results.

"Preliminary benchmark results from a Razer Blade Stealth 13 Late 2019/LY320 laptop with an Intel ICL processor and integrated graphics, including throughput results that range up to a ~15% [performance] improvement and performance-per-watt results up to a ~43% improvement (estimated via RAPL)."

If you're a highly technical Linux user or developer interested in this space, we highly recommend looking at Jerez's own announcement in full—it goes into more detail, if with less hand-holding, than we do here. Those interested in building and testing linux-next against his patches can find links there as well.

Channel Ars Technica