Anyone have a comparable haswell processor to a 3720qm???

Anyone have a comparable haswell processor to a 3720qm??? If you don't mind running a few tests for me that would be awesome. I wanted to finish my testing of seeing how well intel CPUs scale. This is basically what i would need.

https://www.dropbox.com/s/5375chyb0eplkho/scalablility of IB.xlsx


if you could redo this chart with a a comparable Haswell CPU that would be awesome. I would be more than fine with just the Prime95 and/or TS bench. I also would like to see the test redone as single thread and not 4 threaded. I want to get a good picture of how IVR plays a roll in power management. I have yet to see anyone test this and i am curious in how well the CPUs can scale. You can see from my IB chip 2000 mhz seems to be a sweet spot and 2600 isn't that far off. I also want to really really see how much energy is saved or not saved with IVR in single threaded performance. This is a real interest for me to see if IVR actually delivers in that regard.

Also it is paramount that you reset the base clock for each test. The turbo voltage is not intellectually done. As in if you run at base clock of 1.6GHz and run it at 3.6GHz it will use more power than having 2.6ghz base with turbo to 3.6Ghz. It appears intel set a standard voltage based off a base clock of 2.6ghz and that it applies the exact same voltage boost no matter the base clock so if you turbo from 1.6ghz it thinks you are needing 20 multiplier of voltage at the 2.6ghz level.

Let me phrase that better. It adds voltage like this (making these numbers up to prove a point) at 2.6ghz it adds .1,.2,.3,.4,.5,.....for every added multiplier so normally it only needs 5.5 extra volts for 2.6ghz to get to 3.6ghz but for 1.6ghz to 3.6ghz it "needs" 21. When in reality it only needs 2 volts to reach 2.6ghz from 1.6ghz but since it was created to run turbo off the assumption of 2.6ghz it arbitrarily applies voltage in that way.

I hope that makes sense. It is hard to put in words. It is really easy to just show what it does.

So none the less when you drop below 2.6Ghz you need to set base clock at 2ghz to get an accurate picture. You simply can't just set 1.6ghz base and just use turbo to check 2ghz 2.6ghz 3.8ghz from a bse of 1.6ghz. It will be off by 1-3 w TDP if you do.

 

Err, not really.

I assume by TDP you really mean Joules as TDP is a specification and a set figure.

How are you measuring CPU energy? If using the CPU registers these can be inaccurate as they are based on VID for SNB/IVB and perhaps too for HSW.

I do not understand your reference to base clock, the CPU runs at only one frequency at any point in time.

 

I need TDP number in ThrottleStop. That number is the number that the CPU uses to decided if it needs to throttle or not.

https://www.dropbox.com/s/vlpssixdsw8lfed/intel XTU showing base clock.png

if you set base clock in top right of XTU to 1.6GHz instead of 2.6GHz you will use more voltage resulting in higher TDP at 3.8GHz or at any voltage above 2.6GHz. Now if you set it at 1.6GHz it will let you test 1.6GHz in efficiency. Same at 1.8, 2, 2.2 and so on.

 

I'm not using XTU so not sure why you see variance.

Here's some results of 4700MQ with -100mV offset running TS bench 32M. The multi I set myself but the 4700MQ is limited to 34x on 4 cores so only went up to that. Package power, cores power and vcore shown along with TS score. Hope it is in some way useful.

Code:

		8 Threads				1 Thread
MHz	PkgJ	CoresJ	Vcore	TS32M		PkgJ	CoresJ	Vcore	TS32M
3400	48.1	38.0	1.000	8.9		20.5	10.0	1.000	58
3100	40.5	30.3	0.935	9.7		18.5	7.9	0.935	64
2600	29.1	19.3	0.820	11.5		15.6	5.1	0.825	76
2000	22.0	12.0	0.730	14.9		13.5	3.2	0.730	99
1800	20.7	10.5	0.705	16.6		13.3	2.8	0.710	110
1600	19.2	8.9	0.680	18.6		12.6	2.3	0.685	124
1200	14.9	5.4	0.630	24.8		11.9	1.6	0.630	165	

 

how did you set 1200 and 1600 than? And this includes -100mv? can you undo that? Kinda messes up the test. Also is PkgJ TDP, which is reports in TS?

Also if you use TS bench you have to do 2 thread to actually use a single core...TS bench is very different from prime95 which is why i said not to use TS bench. If you notice TS bench in how it scales from 1 2 4 8 threads it acts very odd compared to other tests.

1 thread TS bench for me was 18-21 TDP stock everything (3.4Ghz though) averaged 19ish TDP. VID is 1.13v+

2 thread is 23-25 w TDP at 3.4ghzc VID is 1.13v+

If we both were doing the same test i sure don't see IVD doing anything. :/

 

I guess I didn't really understand there as well

<scratches head>
TS Bench seems to scale just fine for me. If using HT with 2 threads on the same core of course there is going to be some collision so it will not be twice as fast as one thread, perhaps 1.75x as fast.

If you want stock voltage power then just adjust it in your spreadsheet. For instance core energy at 1V is 38J so for stock voltage of 1.1V it would be (1.1/1.0) ² or 1.21 x 38 = 46J plus 10J for the rest of the package = 56J.

 

i guess..never got that far into the calculations before...just using the w TDP number in TS is always easier...why do it the hard way?

a single thread operation on one core shouldn't be really any faster on one core with 2 threads. 4 core operation shouldn't be 55% slower than an 8 core operation on a 4 core CPU. Most programs the more threads it has doesn't make it any faster. It more so seems a single thread TS bench doesn't actually use a full core not that 2 threads is more efficient per se. That is why prime95 is a bit easier to compare. with single core tests.