Enabling and disabling cpu technologies affects performance?

[dguy6789]

Hello. Please take a look at the screenshot below to know exactly what I am talking about.

http://www.elitebastards.com/h...5k-pro/bios/bios03.jpg

I did a benchmark with all of these enabled in the bios, and then a benchmark with all of these disabled except for the execute disable bit. Please note that Speedstep was disabled in Windows so the processor changing clocks is not a factor in the results of the benchmark.

I ran the Nuclear Multicore benchmark and received an overall score of ~18000 with all of these technologies enabled. I ran multiple times to ensure it was accurate. I then ran the test with all of these technologies disabled except for the execute disable bit. The score was very close to ~20000 every time.

Is it common knowledge that enabling the various settings such as C1E Support, CPU TM Function, Vanderpool Technology, Max CPUID Value Limit, and Intel Speedstep decrease performance by a small amount? I'll repeat that my processor's clock was locked at 3.4Ghz. It did not go below that at any time on any of the benchmark test runs. Speedstep wasn't enabled in Windows, only in bios.

Would anyone mind running tests themselves on other systems? I'm curious if this is only with my P5K Pro motherboard, or only Yorkfields.

 

[palladium]

Hmm, that's a very interesting finding. I don't have the nuclear benchmark, but I can do cinebench.

 

[dguy6789]

Any and all benchmarks would be helpful. I'm curious if enabling/disabling these settings has a universal impact on performance or if it is only for very specific platforms.

 

[palladium]

Ok here are my results from Cinebench R10 64 bit ( multi CPU rendering). I ran each test 3 times to reduce errors. NX-bit, HT, Turbo boost were enabled in all tests. ( the ones that were toggled between test are: Speedstep, C1E, C5/C6/C7, Virtualisation)

CPU technology enabled: 19035 CB Marks
CPU technology disabled: 19070 CB Marks

In short, no difference with my Core i7.

 

[dguy6789]

I wonder if it's CPU or Motherboard that is the cause of the differences. I don't have any others to swap and test with at the moment unfortunately.

 

[Rubycon]

Anything with power management particularly things that can change multipliers we prefer to keep OFF.

 

[Flipped Gazelle]

Originally posted by: Rubycon
Anything with power management particularly things that can change multipliers we prefer to keep OFF.

Yup. In fact, AT mentions a huge performance penalty in Photoshop with Cool 'n Quiet enabled, when utilizing Phenom CPU's.

Contrary to AT's findings, however, in the few months I had a Phenom 9850, I found no significant performance decrease with CnQ, and likewise with my PhII. I keep CnQ enabled, though that isn't recommended for overclocking, because I like the thought of power savings. In my own experience, I haven't noticed OCing inhibitions due to either Intel's or AMD's power saving implementations, but then I am not an extreme overclocker.

 

[dguy6789]

I probably could have made it more clear, but the performance changes I noticed had nothing to do with clock speed changes as a result of changing multipliers.

 

[Flipped Gazelle]

Originally posted by: dguy6789
I probably could have made it more clear, but the performance changes I noticed had nothing to do with clock speed changes as a result of changing multipliers.

Have you had a chance to test each setting, one-by-one?

 

[dguy6789]

Originally posted by: Flipped Gazelle

Originally posted by: dguy6789
I probably could have made it more clear, but the performance changes I noticed had nothing to do with clock speed changes as a result of changing multipliers.

Have you had a chance to test each setting, one-by-one?

Not yet, but I'll get around to it soon enough.

 

[Flipped Gazelle]

I tried a few settings, and also used Cinebench r10 for measurement.

CPU - PhII X3 -> X4, 3.5 Ghz, Cinebench r10 64 bit.

CnQ, C1E, VT enabled: single cpu - 4035, multi cpu - 14162
CnQ, C1E enabled, VT disabled: single cpu - 4045, multi cpu - 14698
CnQ, C1E, VT disabled: single cpu - 4040, multi cpu - 14519

Dell laptop, C2D T7250 @ 2.2ghz, Cinebench r10 32 bit. Very few BIOS adjustments available.

VT disabled: single cpu - 2153, multi cpu - 3808
VT enabled: single cpu - 2096, multi cpu - 3805

 

[graysky]

If it is a function of the multipler being stepped down when the app needs it, I wonder if there is a way windows-users can adjust the cpu thresholds for speedstep. Under LINUX, it's called powernowd (don't confuse it with AMD's powernow! technology). In short it's a daemon that controls the multiplier switching and the user can specify low/high cpu % usage for the low to high switch. I have mine set to 5/20 for example. (<= 5 % then 6.0x; >=20 % then 8.5x).

$ cat /etc/default/powernowd
#default file for powernowd, see man 1 powernowd
OPTIONS="-q -u 20 -l 5"

 

[masterbm]

mine using sandra soft was nearly identical 21955/16660 every thing enabled 21899/16669 with it disabled that is intel e6600 runinnig at 3.2 ghz biostar t35-a2 board