Looking to give run some numbers for a new SS to be load tested at.
HT on and OFF if possible, my guess was about 250w?
My system wattage rose from ~150w to 320-340w from stock to 5.0Ghz 1.48v
Yin|Gigabyte GA-Z68X-UD5-B3|Swiftech XT -> GTX240 -> DDC+ w/ Petra's|2600K @ 5.0GHz @1.368V |4 x 4 GB G.Skill Eco DDR3-1600-8-8-8-24|Asus DirectCUII GTX670|120 GB Crucial M4|2 x 2 TB Seagate LP(Raid-0)|Plextor 755-SA|Auzentech Prelude 7.1|Seasonic M12-700|Lian-Li PC-6077B (Heavily Modded)
Squire|Shuttle SD36G5M| R.I.P.
Realtemp shows wattage of just cpu, and seems to be reasonably accurate, as does DES chip on some GB mobos.
Just cpu wattage on my i7950 from DES chip, prime large ffts load, at 25C is:
HT on .......HT OFF
85W...........72W.......stock, 1.12V, 3.03GHZ.
136W........117W.......1.26V, 4.2ghz
180W........160W.......1.34V, 4.4ghz
255W........221W.......1.50v, 4.7ghz
And by comparison, only point i have on 2600K is HT on
HT on 99W, 1.27v, 4.5ghz, which implies about 30% lower power on 2600K vs i7950 for same settings.
Also, when I run my i950 outside at 0C temps, ie 25C temp drop, the power consumption on cpu drops about 20W at higher settings from reduced current leakage in cold, ie colder cpu the less power consumption, that is where it gets difficult to estimate power consumption in cold based on normal ambients.
I think 250-300W would be reasonable guess for max need for cooling. But if Realtemp shows wattage on your cpu, that would help you guess more than anything.
The number tell all
Go big, you have a solid chip there.
300w should be more than enough
Splave, you want a phase tuned closely to the specs of a 2600K this close to X79 launch? No disrespect or anything, but that seems odd to me.
To answer your question, i would imagine no more than 275, but id go for a 300w tune to be a bit more versital.
Current Status - Testing & Research
Last edited by rge; 11-13-2011 at 03:19 PM.
Since there is no ADC on the CPU to measure core voltage then that only leaves VID to calculate the power/energy although the calculation doesn't seem to be a straight forward one.
Might be easier to see with an example. VID can be adjusted with additional turbo voltage and we can use this trick to generate a different VID by setting it high for one test and low 0.004 for the other. Why 0.004, because if we use 0 then BIOS takes that as AUTO and then there is no guarantee it will actually be 0.
By running Linx with fixed vcore, same multi and LLC we can be fairly confident that we will be running at the same power for both tests.
Running with high VID
CPU shows ~150W max peak
Running with low VID
CPU shows ~125W max peak
Now what is going to happen when the OP runs at 1.7V and VID is maybe less than 1.4V?
Note also that the VID above looks like it is calculated using fp. If done by integer you will see defined 5mV steps, well for most of it. For instance there is no 1.3661V in the VID table, probably should be 1.365V. IMO it's not really that important though.
Last edited by some_one; 11-13-2011 at 10:02 PM.
Thanks, that explains the error we saw using that power calc on another forum.
I hate it when RealTemp looks like a piece of crap but thanks some_one for showing us this. Intel's documentation that claims "actual energy use" is obviously a crock since it is based on VID and not actual voltage. About the only time the wattage reported is going to be some what accurate is when VID and actual voltage are the same.
Power usage varies with the square of voltage so would it be possible to apply a correction factor to the RealTemp number and come up with some sort of bastardized approximation when using high core voltages? I think in theory that if the VID was reported at say 1.50 v and actual voltage was 1.65 v then that is a 10% difference so the difference in power consumption should be approximately (1.10 X 1.10 = 1.21). You would need to multiply what RealTemp is showing for watts by 1.21 and might come up with a reasonably accurate approximation.
With the above results it does look like they might use the square law with the voltage. Don't know what to suggest really but IMHO the readings from the CPU are what they are, no need to beat yourself up over it.
Bookmarks