Printable View
It's the proof of the compiler bias. Intel Compiler improves performance against ms-compiler only on intel machines. That's why I always stated that real benchs are compiled benchs with information on compiler and flags released. And no, it doesn't mean that specs are the only bench to be trusted, but yes it means that there will be more trustable benchies on linux than on windows.Quote:
Originally Posted by nemrod
But let's avoid a flamme war on OSes here. :)
Not true in general. Intel compilers gives best code on Intel CPUs, thats true, but also gives very very good code on AMD CPUs -probably best as well. Its just AMD CPUs not seeing much benefit from SSE2/3 optimizations, which is not Intel's fault anyway.Quote:
Originally Posted by JohannesRS
Yes and no. Because then why other compilers would give rise to higher performance increasements when told to use SSE2 and SSE3 on AMD machines?Quote:
Originally Posted by Cronos
I mean, if icc gives an increase performance from "do not use SSE extensions" to "use SSE extensions" of, lets say, 15% on an intel cpu and of just 10% on an AMD cpu, then that would mean that you are right. But when the same code, compiled with MS compiler, sun compiler, portland group compiler or even gcc shows that turning on the SSE on both architectures yelds similar increasements (say, 15% and 14%), then you have to begin to consider that, yes, intel compiler is giving problem for AMD machines, consciously. ;)
If you take the previous case I have give on lame compiled with intel or with ms, you see that with Intel compiler, it's faster for Intel but even for AMD this is a little bit faster than MS compiler. The increasements could not be only consider in terms of % if with AMD cpu the best compiler is the Intel one... So I'm agree that Intel compiler is optimized for intel cpu but not for "intel compiler giving problem for AMD machines, consciously".Quote:
Originally Posted by JohannesRS
http://www.tgdaily.com/content/view/34799/118/
Do you believe that when AMD releases new math library to support barcelona, it will increase by the same ratio computing power on intel cpu? Obviously not but I will not say AMD is giving problem for Intel machines, consciously.
"And no, it doesn't mean that specs are the only bench to be trusted, but yes it means that there will be more trustable benchies on linux than on windows."
At least, spec allow to compare best performance for a given set of software at optimum speed, this is probably a better bench to compare 2 architectures. But you will not ask software developper like maxon to sell slower software because it will be most fair for amd. And this is an other side of the problem. :shrug:
Quote:
Originally Posted by tictac
Itīs me again :)
I think that this donīt work.
I try to push my 2344HE to 2GHz but it was not stable, 1.9GHz = not stable, 1.8GHz = not stable :(
I use this methode to give more Vcore.
I push it to 1.8GHz with 1.2V Vcore (original is 1.1 and 1.15V) and it was not stable.
No i use a voltmeter to check the wattage of the complete system and it shows me that there is NO difference between 1.1V and 1.25V Vcore when the system is running at 100% :shakes:
That is not possible, there must be a difference!
The CPU does not change the Vcore :confused:
What can i do now?
OK, let's take a look at this.Quote:
Originally Posted by indiana_74
note. these calculations are by no means accurate. i have assumed the current a CPU draws is constant, etc. But at least it provides an indication
At stock, the procecessor's TDP is 68W. Stock is 1.1V, so 68/1.1=6.18A. Now if you feed it and extra 0.15V, that mean 6.18*0.15=0.93 extra watts, a lot of which isnt actually being used (TDP is a theoretical maximum), and therefore the difference is not large enough to show up when measuring power, especially not of the whole system.
i though that power draw scales ^2 with voltage. Or i was told so a while ago on this forums...
so if its 68W @ 1.1V it should be (1.25/1.1)^2*68= ~88W or 20W more at 1.25V
Current draw is not constant, it depends linearly on voltage.Quote:
Originally Posted by karbonkid
Power draw scales as Clock* Voltage*Voltage.
In this case, increasing voltage from 1.1v to 1.25v gives (1.25/1.1)^2 = 29% increase in power draw. So, CPU with 68W TDP becomes 89W.
when you factor in C/W values... if you are ~30C above ambient at stock, raising the Vcore from 1.1 to 1.25 should add very roughly 8.8C to your load.
again, these are very rough equations, but i found they are pretty close to real world data. even if your cpu uses less wats than its TDP (and it actually does use less) its correct to certain margin of error, because than difference between actual power draw and TDP is on both sides on equation.
I experimented with them on my system to "guess" core temp at certain speed/volts and then i found they are +/- 1C from real values.
if something is unclear in my post, dont shoot me... i tried my best with my english lol
edit: Cronos, you were 3min late lol ;)
Okay, so I understand you right when I say that there must be a difference when I rise the Vcore from 1.1V to 1.2V.Quote:
Originally Posted by Cronos
Note that the CPU are working at there limit when I messure the wattage.
All eight cores are at fully load with QMC workunits and on my other systems i can see a difference when I change it only for 0.05V.
then maybe you're not changing the vcore properly... can you measure the Vcore from VRM mosfets with DMM? that would tell youQuote:
Originally Posted by indiana_74