Are you implying that when the CPU is idle it drops down to 1GHz?
maflynn, this is from my computer just now:
The first part of the graph is me just reading the forums. Then I fired up a litte probabilistic simulation (middle section, turbo boost kicking to 3.4-3.5Ghz). Then it was done and the frequency drops back to 1.2-1.3Ghz. Yes, this is idling. The CPU spiked soon after since the system decided to start another time machine backup. This is the 2016 i7-6820HQ (base frequency 2.9Ghz).
So again, I might be mistaken, but the picture I see in that tweet is totally consistent with an idling CPU to me. As to why the frequency is so low, no idea, as I don't have an 8-th gen CPU here and I don't know what their normal idle frequency is (maybe someone with a new MBP can check)? Maybe they clock them lower in idle to get even better battery life. And sure, it could be throttling — but there is no way to know just from looking at the graph.
[doublepost=1532098374][/doublepost]
I get what you mean but the point is, even under heavy load, frequency should stay close to the 2.6ghz base mark but if you see the chart, it drops to around the 1.6ghz mark multiple times during load, and at one point dropped to around the 1.2ghz mark. That’s throttling.
Maybe. Or maybe you have burst workloads. Maybe its not one long period of work, but multiple brief periods of work. Maybe its loading/processing parts of the video and then pausing once a buffer has been filled.
I know that I sound like I am trying to talk around the issue, but the point is we
simply don't know what this test actually represents. Its a complex scenario that involves an interplay of CPU, GPU and storage. Moreover, we don't even know what kind of work this even is for the CPU. This is why I think it would make sense to test these factors in separation.
Right now what we have is mostly people doing random stuff, then posting random parts of their results and making random conclusions. IMO, looking at CPU frequencies in complex apps is devoid of meaning since you don't know what the CPU is doing (see above). The only reasonable way to use pro software in benchmarks like these is timing the results and compare them with other machine with the same CPU. So far, the only results I've seen that I trust are those from Noteboockcheck (that illustrate heavy throttling in Cinebench - I will try to replicate it) and the guys from Geekbunch, who ran the benchmark multiple times in succession and didn't see any throttling (
https://www.geekbench.com/blog/2018/07/macbook-pro-mid-2018-throttling/).
When my MBP with i9 arrives, I am going to run it though series of intensive numerical computations (both on a single thread and on multiple threads) — that use only the CPU and do not rely on GPU and SSD — repeating the runs while noting down the frequencies, the power draw, the temps and the time needed to complete the task. This should show how bad the throttling on the CPU is. Then I am going to do the same, but run a little light 3D demo in the background to check whether GPU activation makes the CPU power throttle.
P.S. I am also going to check the "boost agility" time of these CPUs, that is, how often they change their frequency. On Skylake, it seems to depend on CPU utilisation time — if I do short bursts of numeric computations interleaved with pauses, I don't see full boost (and furthermore, the frequency is relative to the amount of time that the CPU is being occupied). The window seems to be around 0.5-1 second. But I'm rambling now
Time to go back to work
