Power supply unit and overclocking. [56k Warning!]

PangingJr

let's just start this off... a few weeks ago when i first got a new motherboard, P5Q Deluxe, i stated a setup using CoolerMaster 550W and 2x1GB of RAM and C2D processor,
then i change to use 4x1 GB of RAM and C2Q processor, anyway, i have 3 new power supply units, CoolerMaster 550W, EMACS 500W and Corsair 620W,
all 3 PSUs are working okay for my overclocking, it's just the voltage requirement to support a certain clock frequency are different for each power supply unit,
well, i will not going into any more detail, so, for a couple of examples of what i'm talking about and looking for other people experienced in something similar to this...

all vDimm's is the value that are set in BIOS... testing softwares and method... ran memtest86+ #5 test for 10 passed (that's all to it, i'm not talking about a solid and stable voltage that i will actually use it).

DDR2-1066 ( 333x7.5 5:8 and 444x7.5 5:6 )

CM 550W - vDimm = 2.1 V
EMACS 500W - vDimm = 1.98V
Corsair 620W - vDimm = 2.02V

all hardware setup and rest of BIOS setup include the version of BIOS in use were the exact same, just different in power supply unit in use.

i'm not sure what is the technical reason why voltages requirement in the overclocking are different and depending on power supply unit in use,
there are more things about power supply and my overclocking that i have been noticing but i cannot going into all that now.


here's my setup for overclocking test. (it's a temporary setup since i will probably never finish it)
right now i just start with Corsair 620W...

Judas

interesting.

i typically try and use the coolermaster Extreme power PSU models..

the Real power ones for whatever reason, have shown a poor performance rating

PangingJr

that CM supply are old models i have it (still having, actually 2 more that never been used) a couple of years ago i think.
some of 3 lines start dropping volts quite a bit when system using too many drives and high overclock.

Corsair 620W has +5 line drops abit with a high overclocked and high load, +3.3 and +12 are both seems to be good and stable.

the EMACS has all three main supply lines very solid especially when the system running in a cool room, but 500W is abit too steep now.

Judas

yeah 500 watt is pushing it imo.

500 watt CM extreme power PSU's is what i use in all my general builds... any machine the is going to be pushed a bit more or demanding more power, i usually stick a 600 watt + higher end model such as OCZ GamerXStream or similare.

swimtech

I'm thinking your rams are designed to run on 2.0 - 2.1 volts (the GSkill 8500 PC1066 rams...)

It's unusual to see the amount of variation you're seeing on the voltage from the PSU on the line designed to supply the memory voltage though. The motherboard uses either the 3.3v PSU rail or the 5v PSU rail to supply a regulator on the mainboard that actually delivers power to the memory chips themselves - and we've all seen reviews of decent supplies that show those rails don't change much with regard to load up to their rated output.

That being said then, it is more likely that the issue you are having is related to mainboard memory voltage regulation than the PSU itself.

I know you've used other mainboards - did those boards, other than the P5-Q you are using now - have the same problem? If not, then you know where the problem lies...

PangingJr

nope, my GSkill 8500 HK kits have spec'ed and EPP for DDR2-1066 @ 2.35V,
but myself and everyone i know who uses this GSkill kit, are actually use 2.35V vDimm for ruuning it at speed of DDR2-1200 and above.

lol, anyway, not sure why you think i'm having a problem with my hardwares.
those are just something i found while experimenting my CPU and RAM overcloking with a limited hardwares i already have.

well, if from the beginning i only use 1 power supply, i would not need to try to use a different voltage settings then.
and, if i can recall, yes, this was also more or less the same with my other motherboards, i mean, not just the vDimm, but the core voltages requirement for overclocking as well.
same motherboard and CPU, RAM and etc, just a different PSU, and i mean they all can be a stable system, with a slightly different voltage settings.

right now i'm using the Corsair 620W, and overclock my CPU to 458x7.5, 5:6, DDR2-1100 (4x1GB) @ 2.08V, and pass a quick prime stress test,
when i were having the CM supply i needed 2.16V and 2.02V with the EMACS supply.
same BIOS profile in use, i only change vDimm after loading the OC profile (the overclocked profiles on each version of BIOS has been saved on a pen drive)

with the CM supply unit was in the system, i was not able to get 500FSB and at 4x1GB 5:6 DDR2-1200, and then i was able to after changing the supply to the EMACS unit.

with the Corsair 620W it seems that i'm getting less and less overclocking failed (at BIOS level) than i used to have with the other 2 supply units.

swimtech

Oh, you're saying you set those voltages in the bios on purpose, to experiment. I did think you were experiencing reduced measured voltages - sorry...

Interesting...

PangingJr

you may not notice this in Gigabyte 'cause the vDimm setting in BIOS is just different...
if you have a chance to change or swap your power supply then try lowering the volts and see.
i think it's there more or less.

as for the P5Q-D board, the vDimm can be increase by +.02 at a time.

this will be more clearer on how the different of 2.02V to 2.08 and to 2.16V

2.02V... 2.04... 2.06...2.08...2.10... 2.12...2.14... 2.16V

2.08V... 2.10... 2.12...2.14... 2.16V

2.02V... 2.04... 2.06...2.08

btw, P5Q-D's seems to overvolt Dimm voltage by +.08 V.
this's set at the board, so for BIOS set 2.02V, it's like 2.10V actual.

swimtech

Yeah, I can only increment the Dimm voltage by .05v increments on my board - EX38-DS4.

I've got an Antec Earthwatts 430 in the old machine, but doubt I would ever pull that to try...

I was looking at a schematic for the memory power controller chip used on mainboards, and it states that noise on the PSU memory supply rail gets coupled to the data lines because of the low impedence of the circuit. Evidently, the PSU supplies the memory voltage almost directly with little filtering, and only one or two (now advertised in the latest ASUS boards...) stages of DC regulation. In other words, the controller chip is the most complicated device in the memory power control/data bus system. The regulation of the PSU rail to memory supply voltage is very simple - apparently passing most noise/ripple on the PSU's 5 or 3.3 volt lines into the memory power/data subsystem.

This means the quality of DC on the 5 and 3.3 volt lines is fairly critical to the stability of the memory subsystem - you could be seeing the effects of this with your differing PSUs.

PangingJr

when more higher 12V line amp supply is in use with the P5Q-D, the board acts and overclocks my Q9300 very well,
now, the CPU PLL, FSB Termination and NB voltages can be reduce quite a bit as well.

PangingJr

anyway, speaking of overclocking Q9300 (a bit offtopic, forgive me),
before i got my first Q9300 i used to read this article X-bit labs - The Youngest of Yorkfields: Intel Core 2 Quad Q9300 Processor Review (page 6)
the article itself is good, but for all the voltages they uses in the overclocking,
they needed to use < CPU PLL voltage 2.0, FSB termination voltage: 1.50V, NB voltage 1.65V >, for (only) 467x7.5,
i think those are just crazy voltages.

with P45, P5Q-D, for the same clock, i needed way lower voltages than that,
the P35-DS4 don't have a CPU PLL voltage setting in BIOS, but i sense i must be using more lower CPU PLL and rest of the voltages than that too.

PangingJr

now, at 444x7.5 5:6 and memory runs at DDR2-1066MHz

vDimm = 1.96V
CPU PLL = 1.52
vFSB = 1.24V
vNB = 1.24V

before i couldn't not set CPU PLL less than 1.54 and vFSB less than 1.26V when overclocking this particular Q9300 on this same clock (or even below this), otherwise i didn't seem to able to get in Windows or the overclock would just hang right at BIOS level.

btw, for this particular Asus P5Q-D that i'm now using, i have to manual input almost all the required BIOS overclocking parameters, just couldn't leave them alone at Auto, otherwise the overclock would also fail right there at BIOS.
as far as i know, on this particular model board, same revision boards, some of the boards can use the Auto settings, but some cannot, and probably wouldn’t be able to overclock some certain CPU's FSB to something above 400FSB with CPU PLL, vFSB, vNB and CPU GTL Ref's parameters are left at AUTO.

swimtech

...and this is with the Corsair PSU, correct?

PangingJr

yes.

and the 444x7.5 is done, for now. it's already prime95 stable for a few hours.
i'll be using this setup and the overclock setting for a week and see.



but i'm trying to get Zippy 800W supply from my friend, he's still using it right now to overclock his new EP45T-EXTREME DDR3, i love to play with that board someday lol.

PangingJr

now at 3.6GHz 480x7.5 5:6 and memory runs at DDR2-1152MHz

vDimm = 2.12V
CPU PLL = 1.56
vFSB = 1.32V
vNB = 1.30V

before i had to use CPU PLL 1.62, vFSB 1.36V and vNB 1.34V to got into Windows
and then ran a prime95 stable, but with the EMACS supply and vDimm was at 2.08V

PangingJr

the EMACS (made by Zippy Technology Crop.) has the a very best result, this supply unit has a tested reports that comes with it and showing its OLP test +12V upto 54A, +5V upto 60A and +3.3 upto 45A, which are pretty high, but they are only an overload. however, i think its spec of 12V line at 36A is a bit steep.

the Corsair 620HX comes with +12V line at 50A, it also providing a very solid +12V, btw, i still don't like its 5V rail, which is drop out to 4.88V quite often, and i can't just exchange it at the store anynore since i bought it long time ago so it pass the 30days return policy, i'll just keep using it and see.

anyways, as of now it seems to me that my CPU and RAM overclocking can be easier and better controlled when i use a more higher and stable 12V line power supply unit,
the number of wattage is not what i'm looking for.

[hobo]eclipse

so you talking higher amps? or just better regulation?

PangingJr

don't understand your question, hobo..


i don't know which supply units i have has the best power regulation or anything like that, but to me, as of now, when it comes to a computer power supply for using in overclocking, it's mostly about +12V rail, however as far as i know the 5V can sometime cause a computer system stable problem too if it drops to often and too much. 5V at 4.88V is not that low yet, it's just i don't like to know that this particular supply has a habbit of doing that.

not sure i did answer any of your questions. lol.

i just don't believe that i actually need more wattages on this computer. as of now, the more +12V rail on the supply really do show an improvement on my CPU and RAM overclocking.

cozumel

Really interest thread you started here Panging.

You're like a 'guinea pig' for us all when considering any future PSU upgrades for good overclocks

PangingJr

if you plan for a better system overclock in futurn i would say don't buy modular unit if you don't have to.
i was planning to use the Corsair 620HX with a smaller computer case 'cause it's a modular PSU then i already have another supply in that computer building. i don't want to buy a new PSU yet, and as of now the Corsair 620HX seems to do it for me.
i like to get another 600-650W Zippy or EMACS, the importer don't want to import them in, but 800W which is a bit too expensive. as far as i know, you can even gain higher FSB overclock with a good power supply.

swimtech

I gotta say, I would love to show up at your place with a scope to read ripple and noise on those 3.3, 5v, and 12v lines - we'd have to hunt around a bit to lock on the frequency to see it.

Even so, the ripple figures I was able to find on the 620HX are some of the best around on the 3.3 and 5v lines. You probably won't find any better. The ripple is only 20% of the 50mv specification for allowable ripple/noise. Part of the same reasons I bought the PC Power 610. The two are very similar performers.

The Zippy supply, though a very good supply, does show a tendency to droop the 12v at very high loads a bit, and it's ripple/noise figures are not as low, at max output of the PSU, as the Corsair 620 - from what I've read both here and over the net - but you may have gotten a good sample. It wouldn't surprise me if the Zippy supply has great noise figures at low to mid power output, with noise rising at the higher end of its output range.

If I may venture an educated guess, I would say that the ripple isn't as much of a consideration as the high frequency noise on the DC lines causing some of the tiny differences you notice. Ripple (leftover sine wave "distortion" from AC/DC conversion in the primary circuit of the PSU) is relatively low in frequency and very unlike any data signal the memory or CPU trigger on. High frequency noise is spikey and could mimic random data pulses if leaked to the data lines of your memory or CPU. Slightly higher noise can be overcome somewhat by having more current headroom, which is what the Corsair unit looks to be giving you over the Zippy at the higher current levels you need for your strongest overclocks.