I'm in the market for a 600-700W PS, and I don't want to buy something that will become obsolete in 6 months. Can you point me to any well-designed tests that demonstrate that different PS's have an effect on overclocking stability?
There may be others here (such as proth or crotale) better qualified to help you. I don't know what you want to spend, but you can't go wrong with the Athena I linked to above. OCZ Powersteam 520W seems popular, however; 34A on the 12V railt might be pushing it if you run a lot of peripherials. I just put my PCPower&Cooling 850 SSi back on the Gigabyte G1975X and it's not overclocking as well as it did with the OCZ PowerStream 420W (33A / 12V) I have (i got to review the very first model at madshrimps). FSP Group (Fortron) is releasing the 600W and 700W Epsilon PSU's which are supposed to be power houses. Tagan probably builds the best PSU's on the planet only second to PCPower, except their conservative. Someone wrote that FSP builds Tagan and if this is true I'd be willing to bet their Epsilon PSU is based off that circuitry.
In PSU-roundups (Hexus did a 34-PSU Round-up) that the ATX 2.01 spec is best for the future? It seems to me there are two issues here. 1.) Motherboard power conversion circuitry. Presler's current demand exposes any weaknesses in the motherboard maker's circuit choices. Some might say unfailry so taxing the circiutry beyond "reasonable" levels. However PSU maker's have but one purpose, to design PSU's for TODAYS processors, that includes Presler? What are they (more importantly we) going to do when quad-core processors hit the streets?
The second issue is the PSU itself, whch IMHO has been castrated the 240VA or 18A/20A limits on 12V rails.
Just to be fair here's another hypothesis; nVidia Forums
I'm far from being an expert at PSUs, and it would be wrong to draw any general conclusions from my experiences.
What happend to us was that when the Presler was heavily overclocked our OCZ600W with dual 12v rails tripped the over current protection (just shut down). This happend at "mild" load and "mild" vcore at around 6GHz. The OCZ520W, on the other hand, had no problems even with heavy load from the CPU at 6.3GHz and an X1900XTX in 3DMark.
Our guess, since we don't know for sure, is that OCZ has split the PCI-express 12v from the mobo + CPU 12v rail and divided the accepted load 50/50 to PCI-e/mobo+cpu. The 12v rail to the mobo became too weak in this case.
However, I do not believe this would be a great problem to less extreme systems, but the Presler do draw a serious amount of current and I recommend known good PSUs.
Maybe someday the power supply companies will come up with a sufficient supply for nuts like up. Even PCP&C's 1KW PSU has only 32A dedicated for the CPU via V1+V2 and not sufficient for 6+ GHz. IMO. Good thing I can't afford another x1900xtx because it would require a seperate PSU for video, something easy enough to do. I tried rigging up a seperate "industrial" high quality 50A +12v supply solely for the CPU, but had some problems with startup timing that choked the bios. Thanks for the info..
PS: I've been thinking about a 200AMP Vcore supply, (Not V12). The supply would simply have a voltage follower input, tracking the onboard Vcore exactly, except at a much higher current, bypassint these crappy MOSFETS alltogether The new 1.5 supply can be easily tapped into the modo @ the heavy chokes going thto the power plane. The only thing stopping me is there's really no need for me to do so at this time with my paltry -5Ghz rig
The nvidia forum link has some revealing information on PS pedigrees:
Topower makes some TTGI - Super-Flower, OCZ, EPower/Tagan, RaidMax, Vantec, ACI; Fortron makes Fortron, Sparkle, Cooler Master*, Zalman, Aopen; Channel Well Technology makes Antec, Lead Power, Enermax, Xclio, Turbolink...
But still, the only tests I see are related to power output, not overclocking stability. I know that in theory a clean 1.5v is better than a fluctuating 1.5v for CPU stability, but Intel has wide tolerances... I think +/- 5%.
I would like to see a test rig overclocked to maximum and tested for stability on various PS's. I think we may get a surprise. I believe real tests would show that the PS is almost never the bottleneck in OC.
it was approximately 2-years ago i wrote the article Presscott effects on Motherboard power Circuits, issues we're now discussing aren't very different. While i do not claim to be an expert, I have researched this topic and the issues we're experiencing with Presler result from the same innate, now antediluvian CPU architecture Intel has been building (or rather repairing) since the advent of the Pentium IV.
With each die-shrink and gate reduction Intel's ad-hoc measures to keep the processor "caged" electrically and thermally continue to test the limits on Power Supplies and motherboard power-conversion circuitry. Prescott was an excellent example of this and it's propensity to literally cook mosfetts were just a few of the ramifications. I also discovered while overclocking processors with features such as On Demand and VRD (Voltage Regulator Down) it was better to leave the motherboared BIOS setting for the CPU vcore on "Auto" or "Manual" thereby ensuring no "set" vcore would interfere with normal Vcore fluctuation which is a natural occurance for the feature.
If you read my article linked above you'll find what is happening with Presler may just be history repeating itself, except this time we have conditions which are amplifying (or draining) the issue. Dual Core which practically doubles current (not voltage0, hence thermal demands increase. Throw in the EPS12V 2.91 Design Guide (I've been incorrectly referring to the ATX 2.01 spec) limiting PSU makers to reduce amperes on all 12V Rails at a maximum 22A (750W to 800W PSUs) and you have an overclocking tragedy.
Vdroop isn't simply a matter of inadequate mobo power ciruits, ampere limitations on rails, or processor "features," but a dangerously busy intersecton where all three collide. As fI indicated above Intels ad-hoc fixes disquised as breakthroughs multiply the problems. These include and claim such monikers as, On Demand and the infamous original PROCHOT for P4 (sec2.4) and for the 90nm process PROCHOT aka Thermal Throttling (Sec. 3.4.2). The following quote is taken from Sec. 2.4 of VRD 10.0 paper and is replete with terminology related to the matter at hand;
Thermal compensation allows the processor Vcc voltage regulator to respond to temperature drift in VRD electrical parameters. It is required to ensure that regulators using inductor or FET RDSON current sensing maintain a stable voltage over the full range of load current and system temperatures.
If thermal compensation is not included, the output voltage of the regulator will droop as the resistance of the sense element increases with temperature. With the increased resistance, the regulator falsely detects an increase in load current and regulates to a lower voltage. Thermal compensation prevents this thermally induced voltage droop by adjusting the feedback path based on the temperature of the regulator. This is accomplished by placing a thermistor in the feedback network (tuned with a proper resistor configuration) to negate the effects of the increased resistance of the sense element.
The thermal compensation circuit is to be validated by running the regulator at VR TDC for 30 to 45 minutes. This is to ensure the board is thermally saturated and system temperatures have reached a maximum steady state condition. If the thermal compensation has been properly
implemented, the output voltage will only drift 1-2 mV from its coolest temperature condition. If the thermal compensation has not been properly implemented, voltage error can be measured in the 10’s of millivolts. Care must be taken to avoid over-temperature conditions during this
validation exercise. Sustained current draw above VR TDC can easily result in component failure and/or board damage.
As I've said before these are only my hypothesis and please feel free to correct (or assist me).
I have problems with mine...
My Pentium D can run about 4.6ghz stable 24/7 with about 1,55v core..this draw alot of power..
But a Tagan Tg580u22, delivering 580w 35A single or 20+20 in dual 12v, should be up to the job... but suddenly my pc startet rebooting, not even under load... so i set it to stock and still i got these reboots...
Realy annoying since this acctualy is my second tg580... the first one blew just after i got my pressler...
I presume it is the psu beeing the problem as i have changed it with a "pos" 350w 20A 12v running the system fine, even ocd to 4ghz...
Annyone else using this psu, or seen others with simular problems?
Thinking about getting a Tagan Tg480-U01 had one on my earlyer s478 system and it was rock solid...
I'm working on a upcoming review specifcally on this topic, but my GA-G1975X review should be "on-line" very soon at www.madshrimps.be look for it. Specifically page-7 titled "8-Stage Power Conversion" which gives a pretty good idea of the problems specifcally on the Gigabyte GA-G1975X pertaining to the EPS 12V limitations, multi-VRM power circuits and Presler architecture culminating to cuase problems.
one solution is basically to "freeze" the processor aka phase-change therby giving the power circuits a "break" the other is to find a mobo using a decent 5-stage power conversion circuit like Intel's referce 975X board does, or use a 955X based board and cool the mosfetts, water-cool the Presler and have fun.
I believe the best solution would be the Athena 850W PSu which running a 48A 12v rails is currently the highest ampere 12v rail I know of. of course the PSU is almost $248.
I have problems with mine...
My Pentium D can run about 4.6ghz stable 24/7 with about 1,55v core..this draw alot of power..
But a Tagan Tg580u22, delivering 580w 35A single or 20+20 in dual 12v, should be up to the job... but suddenly my pc startet rebooting, not even under load... so i set it to stock and still i got these reboots...
Realy annoying since this acctualy is my second tg580... the first one blew just after i got my pressler...
I presume it is the psu beeing the problem as i have changed it with a "pos" 350w 20A 12v running the system fine, even ocd to 4ghz...
Annyone else using this psu, or seen others with simular problems?
Thinking about getting a Tagan Tg480-U01 had one on my earlyer s478 system and it was rock solid...
Tagan is an excellent PSU, OCZ PowerStreams ARE Tagans with external adjustable pots. I don't think the tagan's are the problem, the problem is the EPS 2.91 Spec (read section 6.4) limitng 12V rails to 18A maximum continuous. If the PSU is above 750W then 22A is the peak. Peak, not continuous.
the reason your "pos" 20A single rail works is smply the nature of the CPU augmented by low quality multi-VRM power conversion circuitry on the mobo.
For an interesting comparison, look at Intel's refernce 975X chipset board compared to all OEM 975X boards. the reqason people seem to get better results with 955X boards are because they are using power conversion circuitry closer o Intels refernce 975X then to thew 8-Stage circuitry found on new boards. Bottom line Presler is a DC Vampire which sucks the life out of anything in it's path....
I suggest looking for the highest ampere single 12V rail PSU you can find, it's the wooden stake through Presler's heart and garlic necklace for the multi-VRM circuits
The MGE Duro 900 just arrived for review/ I'll have in in my system after taking photos inside and out. I don't know how she'll do because it's 4-12V rails at 16A and were back to the same issue. It depends on how the manufacturer confidured the rails feedig the CPU. Since they all derive from a single source anyway it depends on what type of circuit design is used to "split" the rails.
I'll let ya all know how this unit does in the next few hours. i also asked MGE for their Vortex 700 which WAS in newegg and was stated to have a single 37A 12V rail. this is above the PowerSteam and could have been substantial, then i spoke with MGE and they said the design is actually going to be a dual 12v rail and isn't ready yet anyway.
I would also suggest people look at the PCPower&Cooling TurboCool 510 which is a single-rail design and I'd guess at 34A to 38A-peak slightly better then the Tagan licensed PowerStream, although PowerStream has lighted pots on the rear, of course PCPower has adjustable pots as well with some real "guts."
Also PCPower&Cooling has some excellent "Specials" their refurbished PSU's usually out-last 99% of new units. I just donated a 6-year old Silencer 400W which is still going strong! The new "store" at PCPower is a great place to find certified units for a great price. Just make sure you don;t get the SI version of the TurboCool 510,
...quantity of charge is measured in units called COULOMBS, and the word "ampere" means the same thing as "one coulomb of charge flowing per second." If we were talking about water, then Coulombs would be like gallons, and amperage would be like gallons-per-second...
Were rapidly surpassing the 120A+ peak current demand from CPU's with a Vcore of 1.3V (approx.) Taking into consideration the 240VA safety guideline limiting (split plane) 12V-rails to 18A (22A @ >750W) it's not too difficult to do the math from there.
Just about every review I write includes Overclocking tests, not as an included section with a few CPU-Z screenshots, but as an integral part of the testing process. In five years of stress testing I've discovered the "scalability" of many products not only determines their performance ceiling, but can often predict longevity. Insofar as Presler is concenred "ampere Overkill" is the key.
Many of the "current" (pun intended) issues with Presler are similar to those with the Socket-478 version of Prescott. It's effects on motherboard power conversion circuitry, made worse by the EPS12V Power Supply Guidelines. Features such as On Demand and VID meant to conserve power or current, in an extreme scenario actually have the opposite affect. In other words the physics fee or perhaps it's related to Gauss's Law must be paid somewhere. So what can we as consumers do to circumvent the issue beyond holding Intel engineers hostage forcing an early Conroe release? Well not much, uit's par for the course were going to be FORCED into updating an entire system even with Intel keeping Conroe on the LGA775 package. And i don;t even want to get started there. wait until you've chnaged out processors about seven times. I'm already finding bent leads.
Our other option is to find a "pre-EPS or even a pre Split Plane" PSU hamstrung by the 240VA limit (since most designs wqere executed incorrectly anyway) with the highest possible current flow rate i.e ampere. If your running Crossfire the PCPower 850 SSI has been revised but i'm not sure that will even solve the issue with a Presler at the helm..
The only PSU I found which seems to have the amperes to placate Presler was the Athena AP-P4ATX85FE (48A - 12V) at $289. I've written Athena about five times outlining the issue at hand without any reply. I've written PCPower once and they took the time to write back informing me their working on it. Which I interpreted as new unit coming out.
IMHO PCPower is going to be our best bet, when i wote them stating my 850 SSi was struggling they didn't give me the run around they simply stated they had other options and they were working on the issue. If anyone will do it they will, they build from a purest philosophy which resulted in the worlds first truely proprietary PSU on the market, the Turbo Cool 850 SSI. You can go to their online store now and choose from a number of TurboCool 510ATX (single 12V-rail @ 34 -38A) for $189. There's a TurboCool 510DX-XE refurbished for $169 on the Special page.
When they refurbish it's gauranteed. I obtained a Refurbished PCPower Silencer 400ATX over four years ago, I just donated the unit to a friend who is using it replacing a "650W" unit. That origial refurb is probably going on seven years now. If your running a Presler and want to find the processor's full potential $169 is a small price to pay. Trying to skimp on what Intel's IPC antonymous processor architecture can render your $2000 rig into an overpriced "sand box."
Lets keep this thread alive even manufacturers are reading it maybe we can effect change?
If they have this limit on the 12 volt rail why do they not just revise the specs so that voltage s fed to each key componet via a seprate feed to avoid excessive drain and strain. Have a led directly to the cpu, videocard/ram/chipset ect. Would such a system not allow for smaller and sturdier regulation circutry.
Also From what I got for you blurbs liquid is that the older prescotts for 478 and early 775 were limited not only by poor power supply but insufficient voltage regualtion. I'm suffering from a lack of power on my pent d 805 system with a motherboard built using 478 technoligy could the voltage drops be caused by insufficient regulation circutry as it is by my termaltake 430w with 18v @ 12 or is the powersupply the only problem or would futher increase in power be offset by powerloss in the motherboard
@xTrEmEoVrClOcKr.. good luck getting your Mach in from Chilly1, he's been having alot of shipping problems lately.. If he hasn't shipped it yet you might want to contact him and request a freight service, you might get it in one piece that way.. HTH
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
