How much Current does Presler draw ?

Decided to do some testing of a different kind this weekend and find out the effects of overclocking and current draw. We all know that Presler draws high power, but exactly how much is what I want to know, during IDLE and LOAD Current is measured at the MOLEX CPU power headers. Power supply is a high quality XPIQ SMC600, 600w 12v @ 50A, and is totally dedicated to powering the cores. The Fluke 87 DVM only has 10A capacity so there is one DVM for each of the 4 PIN molex headers. I think this may not be enough though, being 20A total @ 12V = 240W total and 5GHz should push way past this. In any event I'll plot Current over GHz in Excel and see how if rate increases exponentially. Sorta doing this in real time, on the fly, so it may take a few hours. While I have this stuff here, just let me know if you want me to make some checks under special conditions for you.

One think I have noticed so far that there are heavy current spikes that pin these meters during Windows boot. This is kind of puzzling to me because I'm only measuring core current and not any draws dur to hardware initializing.

I also want to document the effects that this super stable and low ripple (50mv) PSU has on max, OC, stability and temps of the mosfets. First up is the PSU with the DVMs in series with the core supplies:



This is the initial reading for a 955XE, running stock clocks or 3.45GHz and Vcore @ 1.3v. Looks like 3.57A total draw at IDLE or 44Watts with supply @ 12.3v. Note that this is not the true wattage of the cores, but only the wattage on the 12v. The MOSFETS get hot so they can't be 100% efficient, and actual core wattage will be lower.


Looks like 96 watts max load on SP1M run:


3.8GHZ, Vcore=1.3v
IDLE = 50 watts:


Load = 96 watts:


Boot Peak = 135 watts:


4.0GHZ, Vcore=1.38v vcore needed bump
IDLE = 60 watts
LOAD = 122 watts
BOOT = 194 watts...NOTE: 1 DVM pinned intermittently @ 10Amps already
Looks like 200 watts is bare minimum for 4GHz.. And that's only CPU power, no drives, fans, etc...

4.2GHZ, Vcore=1.38
IDLE = 64 watts
LOAD = 130 watts
BOOT = > 240 watts Both meters pinned intermittently @ 20A

4.4GHZ, Vcore=1.38
IDLE = 65 watts
LOAD = 128 watts
BOOT = > 240 watts Both meters pinned intermittently@ 10A
NOTE: Multiplier was changed from 13 to 15, basically lowering the FSB. Lower FSB has the net effect of reducing current draw.

4.6GHZ, Vcore=1.38
IDLE = 74 watts
LOAD = 138 watts
BOOT = > 240 watts Both meters pinned intermittently@ 20A

4.8GHZ, Vcore=1.45
Here's where it gets ugly real fast. Had to go 1.45 for SP1M stable
IDLE = 80 watts
LOAD = 164 watts
BOOT = > 240 watts Both meters pinned continuously @ 20A

5.0GHZ, Vcore=1.52
IDLE = 96 watts
LOAD = >240 watts Both meters pinned continuously @ 20A
BOOT = > 240 watts Both meters pinned continuously @ 20A

5.2GHZ, Vcore=1.58
IDLE = 121-160 watts..large swings
LOAD = >240 watts Both meters pinned continuously @ 20A
BOOT = > 240 watts Both meters pinned continuously @ 20A

5.32GHZ, Vcore=1.60
IDLE = 135-180 watts...large swings
LOAD = >240 watts Not SP1 stable @ 30c+ CPU temps
BOOT = > 240 watts Both meters pinned continuously @ 20A

That maxes out this rig and this post until phase

placeholder.. I'll shrink and cleanup and consolodate pics later so not to look rediculous

Interesting stuff! i wonder if a higher capacity anologue gauge would show on that windows boot.

I bet dual prime instances would generate a lot more amp draw then sp1m. probably be similar to the boot sequence of windows

MikeyLikesItSI said:

Interesting stuff! i wonder if a higher capacity anologue gauge would show on that windows boot.

I bet dual prime instances would generate a lot more amp draw then sp1m. probably be similar to the boot sequence of windows

Click to expand...

I'm gonna get 2 more meters next week and split the load up across 4 wires to have 480w capacity.

I think prime will require the same current, it's the slow heat buildup that creates all the problems

Hey man, update your sig to show that 5GHz. CPUZ verified...

While I have this stuff here, just let me know if you want me to make some checks under special conditions for you

Click to expand...

Can you hook up like 6 meters and check draw at 6GHz @ 1.70Vc for me?

Ross said:

Can you hook up like 6 meters and check draw at 6GHz @ 1.70Vc for me?

Click to expand...

Only if you loan me your DI tube..

dunno if you want some other datapoints for a 920.

I measured all my folding rigs so i could get the right number of rigs on each circuit. So I was interested mainly in full rig draw under full folding load (2 instances on all rigs). since that is how they run 724.



measured through apc rs1000 ups and powerchute software. It seems to read in 6W incrememnts. Minimal no load measurment is 6W.

p5wd2 on a powerstream 420, 1 80 gb seagate sata, 1 120 thermaltake fan, 1 ati pci 8mb rage card, 1GB corsair 5400c4. Same measurment with 5400ul at 295 reading but ups is only relatively accurate +/- 6W.

at 290, vc=1.425, vddr=2.2 boot=174 W, idle = 150 W, load (dual folding) 265 W.
at 295, vc = 1.475 vddr=2.3, load= 294 W stable folding (current reading)
at 300 vc=1.45 vddr=2.3 load = 288 (boot,idle same as above i think since i didn't note them)
at 315 vc=1.55, vddr=2.3, load peaked at 342, then dropped to 282 probably thrashing since it was never stable dual folding for long at 315.

I didn't measure over the whole possible range since that was not my intent.

Also have 830, prescott and northwood measuments, but only at 724 dual folding stable OC's if interested.

pscout, thanks for adding to the thread!

Only if you loan me your DI tube.

Click to expand...

I am actually without it myself right now...undergoing the transformation to Rev. 2

proth said:

I think prime will require the same current, it's the slow heat buildup that creates all the problems

Click to expand...

well you are measuring a dual core CPU, and SP1M is a single threaded program, so i would imagine that dual folding, dual prime, dual stressCPU etc etc would definitely draw more.

MikeyLikesItSI said:

well you are measuring a dual core CPU, and SP1M is a single threaded program, so i would imagine that dual folding, dual prime, dual stressCPU etc etc would definitely draw more.

Click to expand...

Good point...I'll be measuring dual SP later on and then quad SP to see how much current is pulled.

Proth,

Sounds like you have access to lots of equipment - why not use a shunt and not send the current through the Fluke, but measure the voltage at the shunt and convert to current ?

Just a thought, and I could well be missing something.

- Chris

Nice little experiment proth, very usefull. So take a 920 @ 4.3ghz, and we're talkin 200-288watts just CPU when under load.

That is insane, it really is. I'd love to find out what a pair of X1800/X1900 draw when OCed......

BTW proth thanks for all the usefull info you contribute to the forums.......

Look what my 920 Presler@ 4.31ghz does to PC Power & Cooling TurboCool 510w.

I can loop 3Dmark06 all night with the Presler@ 4.2ghz AND the X1800Xt Crossfire cards at 700/800, 12v rail doesn't budge, stays at 12.25v, right where it's set at. AS SOON as I boot up to 4.31ghz 308FSB, clock the cards to 700/800 and run RTHDRBL, along with an instance of STressCPU in the backround, BAM this happens and shortly later it will go down further, then the machine reboots.

This really sucks, that is a $229.00 PSU, so now I either have to: buy the 850SSi unit for $450.00, or be happy where I'm at, 4.2ghz and 700/800 on the X1800XT Crossfire cards..

proth, did you ever do any more testing with this? this looked like it was pretty interesting

so...if you get a 940 and OC to 4.2 you will use less draw/current if you have a 920 OC to 4.2??
But does this hold true for a 660 OC 4.2? Does the 660 or the 940 draw the most current OC 4.2, if you know?

Sorry.. Didn't see the responses to this thread. Have to figure out how to be notified when soimeone responds. Havn't done any more testing because I can only measure 20A with the meters. That's a good Idea to use a shunt, forgot all about it and will check out the cabinet at work.

@Ranger, I can't believe the 12v sag outta that PCP&C 510. Would have thought that unit held up better than that. You might want to check out the zippy 6701p from servercase.com. I picked one up a few weeks ago and it's unbelieveable. Ross and I did a review on it and it's on the front page of the forum.

@ECH. I think a 920 and 940 running @ 4.2GHz would pull about the same current, the 920 may pull a little more because of the higher FSB that it will run at because it has a lower multiplier, compared to the 940, but only a little more. IMO.