Typically I can run BOINC at 99% CPU Utilization on any machine, effectively turning off that last thread, to keep the system responsive and allow the GPU to have a little something to play with. And with such a setting I usually see a predictable, calculable, CPU Utilization percentage.
My hexcore i7 (12 thread) can be expected to use 8.33% (100% divided by 12) per thread, and 99% setting cuts the utilization down to about 96%, as background processes bump it up some.
I've noticed on my Xeon systems, that disallowing several threads to be used for BOINC still results in my CPU utilization to be shown as 100%. This holds true for the Workstations as well as the newest systems. On BumbleBee and WUSS I have to disable 4 threads (BOINC set to 88%) to get under 100% utilization.
Further, I've noticed, by using Kill-a-Watt meters, that my power usage goes UP (only 1-3 watts more) when I finally disable enough threads in BOINC to obtain <100% CPU Utilization.
And in practice, I've managed to observe that identical systems, running the same projects, do indeed score significantly different depending on if the CPU is maxed out or not. Bee was set to 99% (one thread disabled), WUSS was set to 92% (three threads disabled and still showing 100%, but Turbo'ing higher), and over a four day period WUSS scored about 19,000 higher in Universe than Bee. Worse yet for Bee, he has a 2% overclock.
(Oh, and I just discovered you can keep track of your individual computer's results in FreeDC. How cool is that!!?)
In summary: Having used numerous monitoring tools, the best I can come up with as an explanation, is that the Xeon's are hard locked to the TDP, or thermal design power, and standard (or should I say, overclockable) desktop CPUs are not. With less cores running, the remaining cores turbo to higher clocks, keeping the TDP maxed out. The higher power consumption observed at 98-99% vs 100% can possibly be attributed to the turbo clocks not having to switch up and down to maintain TDP.
So I believe the CPU Utilization on a (Windows) Xeon, is based on TDP, instead of threads in use.
Has anyone else experienced this, or can either confirm or deny my findings and assumptions?
EDIT: Would disabling Turbo result in predictable percentages?
EDIT: Xeon's running WinTen, Hex-core i7 still on Win7.
My hexcore i7 (12 thread) can be expected to use 8.33% (100% divided by 12) per thread, and 99% setting cuts the utilization down to about 96%, as background processes bump it up some.
I've noticed on my Xeon systems, that disallowing several threads to be used for BOINC still results in my CPU utilization to be shown as 100%. This holds true for the Workstations as well as the newest systems. On BumbleBee and WUSS I have to disable 4 threads (BOINC set to 88%) to get under 100% utilization.
Further, I've noticed, by using Kill-a-Watt meters, that my power usage goes UP (only 1-3 watts more) when I finally disable enough threads in BOINC to obtain <100% CPU Utilization.
And in practice, I've managed to observe that identical systems, running the same projects, do indeed score significantly different depending on if the CPU is maxed out or not. Bee was set to 99% (one thread disabled), WUSS was set to 92% (three threads disabled and still showing 100%, but Turbo'ing higher), and over a four day period WUSS scored about 19,000 higher in Universe than Bee. Worse yet for Bee, he has a 2% overclock.
(Oh, and I just discovered you can keep track of your individual computer's results in FreeDC. How cool is that!!?)
In summary: Having used numerous monitoring tools, the best I can come up with as an explanation, is that the Xeon's are hard locked to the TDP, or thermal design power, and standard (or should I say, overclockable) desktop CPUs are not. With less cores running, the remaining cores turbo to higher clocks, keeping the TDP maxed out. The higher power consumption observed at 98-99% vs 100% can possibly be attributed to the turbo clocks not having to switch up and down to maintain TDP.
So I believe the CPU Utilization on a (Windows) Xeon, is based on TDP, instead of threads in use.
Has anyone else experienced this, or can either confirm or deny my findings and assumptions?
EDIT: Would disabling Turbo result in predictable percentages?
EDIT: Xeon's running WinTen, Hex-core i7 still on Win7.
───────────▄▄▄▄▄▄▄▄▄───────────
────────▄█████████████▄────────
█████──█████████████████──█████
▐████▌─▀███▄───────▄███▀─▐████▌
─█████▄──▀███▄───▄███▀──▄█████─
─▐██▀███▄──▀███▄███▀──▄███▀██▌─
──███▄▀███▄──▀███▀──▄███▀▄███──
──▐█▄▀█▄▀███─▄─▀─▄─███▀▄█▀▄█▌──
───███▄▀█▄██─██▄██─██▄█▀▄███───
────▀███▄▀██─█████─██▀▄███▀────
───█▄─▀█████─█████─█████▀─▄█───
───███────────███────────███───
───███▄────▄█─███─█▄────▄███───
───█████─▄███─███─███▄─█████───
───█████─████─███─████─█████───
───█████─████─███─████─█████───
───█████─████─███─████─█████───
───█████─████▄▄▄▄▄████─█████───
────▀███─█████████████─███▀────
──────▀█─███─▄▄▄▄▄─███─█▀──────
─────────▀█▌▐█████▌▐█▀─────────
────────────███████────────────
────────▄█████████████▄────────
█████──█████████████████──█████
▐████▌─▀███▄───────▄███▀─▐████▌
─█████▄──▀███▄───▄███▀──▄█████─
─▐██▀███▄──▀███▄███▀──▄███▀██▌─
──███▄▀███▄──▀███▀──▄███▀▄███──
──▐█▄▀█▄▀███─▄─▀─▄─███▀▄█▀▄█▌──
───███▄▀█▄██─██▄██─██▄█▀▄███───
────▀███▄▀██─█████─██▀▄███▀────
───█▄─▀█████─█████─█████▀─▄█───
───███────────███────────███───
───███▄────▄█─███─█▄────▄███───
───█████─▄███─███─███▄─█████───
───█████─████─███─████─█████───
───█████─████─███─████─█████───
───█████─████─███─████─█████───
───█████─████▄▄▄▄▄████─█████───
────▀███─█████████████─███▀────
──────▀█─███─▄▄▄▄▄─███─█▀──────
─────────▀█▌▐█████▌▐█▀─────────
────────────███████────────────
Last edited: