32bit apps running 64bit OS, i7/X55x0 CPU underclock in 32bit mode

[20nm]

Using the benchmarking software one of the things we have seen is that when we have a 32bit OS running on a i7 or X55x0 cpu will under clock itself and perform poorly.

Does anyone know why the cpu will typically will clock itself at 1Ghz lower when say running XP32bit on an i7 or a X5550 cpu?

I also need to have the Passmark for Linux to find out what the OS is doing with the cpu/hardware. When will it be available?

[David (PassMark)]

Seems like half your post is missing, as it doens't make complete sense.

New CPU lower their clock speed when idle.

We have no plans for a Linux release of PerformanceTest at the moment.

[20nm]

Seems like half your post is missing, as it doens't make complete sense.

New CPU lower their clock speed when idle.

We have no plans for a Linux release of PerformanceTest at the moment.

When we ran your benchmark software the 32bit OS gave us results of the CPU dropping its cpu speed by a lot in some machines.

Well we don't have a clue or a way to measure what linux is faster, what happens when you run 32bit OS or 64bit OS and how it refelects on the i7/x5500 cpus.

[David (PassMark)]

The CPU clock speed will ramp up under load when the CPU tests are run.

But 64bit will be give better CPU benchmarks for other reasons.

For the CPU test, Linux should score about the same, as the operating system isn't directly used for the test. That is to say we don't make any API calls for the CPU test. The major difference between Windows and Linux, for the CPU test, would probably come down to the quality of the C++ compiler to optimise code during compilation.

[20nm]

The CPU clock speed will ramp up under load when the CPU tests are run.

I am basing this underclocking based on using Passmark 7 benchmark suite on 32 bit XP on a i7 and X5550 cpu. CPU dropped to 1.6Ghz. When we tried XP64 and Vista64 the machines ran at their normal speeds.

Most Linux kernels have big problems with these new cpus and I have reporting that the clock speed is 1.6Ghz instead of the 3.2Ghz of the i7 965 cpu. This is why we need a Linux version of your software. We need to know that hardware is behaving consistently and that software changes to the OS are not underclocking the machine.

[David (PassMark)]

I still think you are just seeing the idle speed. How do you know it is always running at 1.6Ghz? How are you measuing this?

Sorry, but we have no plans to do a Linux release of this software. Have a look at BurnInTest for Linux if you want some Linux software. It isn't a benchmark, but does report some CPU details.

[20nm]

I still think you are just seeing the idle speed. How do you know it is always running at 1.6Ghz? How are you measuing this?

Sorry, but we have no plans to do a Linux release of this software. Have a look at BurnInTest for Linux if you want some Linux software. It isn't a benchmark, but does report some CPU details.

It is your software Performace Test 7.0 (trial and paid version) reports that the machine is running a 1.6Ghz when the OS is XP 32. When we run applications it sure feels slow. When we install a 64bit OS the machine runs at full speed, and also your benchmark reports look in line with what we see on the site.

This behaviour is not just one PC but several of i7 and X5550 (new machines) do the same thing when you put a 32bit OS on them.

Yeah /proc/cpuinfo great, still have no clue what the cpu is actually doing under Linux. It is only thanks your your Windows software that we are able to find proof that there is something wrong. We would like to be able to do the same thing under Linux.

[David (PassMark)]

The CPU speed in PerformanceTest is measured when PerformanceTest starts up. So it is a one off measurement. The CPU speed can vary after this measurement is done. Note that not all CPUs can vary their clock speed dynamically. But certainly the newer ones and chips made for laptops can.

We are actually look at ways to put the CPUs under more load at startup, to try and force them to run at full speed, before we take a measurement. There is a possibility that you are right, and the CPU is just permanently underclocked, but the evidence presented doesn't prove this conclusively.

You should also read this post referenced above, 64bit will be give better CPU benchmarks for reasons other than clock speed.