Turbo explanation on Core i7

[rge]

Many in forums are trying to answer this question, bios bug or working correctly

Issue 1) i7 940 is stock 22x133. With turbo enabled, 23x133 for tasks involving 1-2 cores usually, and 22x133 turbo off for running prime 95 4 cores. so far so good, turbo is +1 multi. Now drop the multi to 15 in bios, so now running at 15x133. With turbo mode enabled goes up to 23x133 still, in other words turbo is now +8 multi. Possibility 1) turbo is +1 over stock or +2 over stock and thus turbo should bring multi from 15 to 23 (and not +1 or +2 over bios setting), in other words that is the way it works. Possiblity 2, bios bug, anyone confirm one way or other.

Issue 2) Bios F3 on mobo, running 22x133 with turbo enabled, loading 1-2 cores with prime 95 and get turbo mode 23x133. Running all 4 cores at max prime load turbo drops off and 22x133...seems correct. Bios F4 however runs in turbo mode even loading all 4 cores with prime, simultaneously linpack, ati, video stress. In other words, it is always in turbo mode 23x133. It is not a software reading multi error which will be the first thought, as it truly is running in turbo all the time confirmed by significantly higher vcore requirements. Is bios F3 correct and bios F4 bugged? Or is it just different hidden settings/parameter changes in bios. Another way to ask is when all 4 cores are max loaded with prime 95/linpack/graphics running simultaneoulsy should it still be in turbo mode?

[TimP]

You're not being very clear about this. I don't think this forum is the place to dive into various motherboard models and BIOS versions, nor to discuss specialized diagnostics which have their own forums. OS also may be involved. I'll guess you're talking about one of the single socket models, as the dual socket models may not have stabilized on a production BIOS.

Running my own benchmarks on a smackover, turbo mode usually gives a 4-5% increase in actual single thread measured performance, and negligible steady state change in 4 thread performance. The performance transition occurs about half a millisecond after the number of active threads changes. Turbo mode is subject to a power consumption limit, and, on some BIOS versions, also a temperature limit. So it is even possible that a CPU which happens to have lower than average power consumption may run a step higher than average in turbo mode.

The cabinet in which my smackover was delivered had unsatisfactory air flow path, and continually produced thermal shutdowns until I figured it out. Much of the air flow was cooling only the RAM sticks, with the CPU drawing hot air from the hot graphics card. My building air flow is shut down much of the time, with ambient temperature probably higher than what these boxes were tested at. My CPU may have above average power consumption, which could prevent it from going into multi-thread turbo state for a significant time interval.

[rge]

Ok, let me ask this more clearly. And this is the only place to ask it, as what I am asking is how is turbo mode supposed to work on core i7. I have read intels white paper explaining it, but it is lacking in a few details.

Question 1) does Turbo mode = 1+ multi over bios setting or does turbo mode = max multi plus 1 regardless of bios setting.

Question 2) Turbo mode is supposed to be limited by TDP, can that be turned off/ignored by bios so that turbo is always active irrespective of load, in effect giving +1 multi or is it likely a bug?

Reason I am asking is I am beta testing utility software for 3 different programmers. If I dont know how it is supposed to work, it is a little difficult to say bug or not, or reading correctly or not. Unfortunately most of programmers dont have core i7 yet.

[TimP]

My BIOS setup screen quotes turbo mode limits, but doesn't allow them to be adjusted. As you say, the multi-threaded limit for mine is 1 step. The only adjustment available to me is enable/disable. I don't see any performance boost when all cores are active, but this could be due to TDP limiting with my CPU already running at the limit. My impression is you take what you get, unless you are capable of getting BIOS code and modifying it, or there is some super priced over-clocker's BIOS with adjustments available on the setup screen and no warranty. As I've already hinted, I think you can't rule out variations in behavior due to variations in baseline CPU power consumption. People dealing with server parts are even more frustrated trying to get repeatable behavior and dealing with kernel/OS related variations.

By the way, I've had people ask questions about stuff like how the Prime95 is used for performance comparisons. Can you supply a reference? Among the many Google hits, I didn't see which are relevant.