-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
undervolting phenom/athlon quads at stock speed
- Thread starter mymorph
- Start date
I've undervolted an AMD® Athlon™ II X3 from 1.4250v to 1.200vCore no problem, very stable etc . . . just testing a AMD® Athlon™ II X4 right now but won't get to undervolting for another few days! 
[edit] as phill0i has stated cool n quiet works great for idle but load will still be higher, best result would be to use both (undervolt + CnQ) . . .
[edit] as phill0i has stated cool n quiet works great for idle but load will still be higher, best result would be to use both (undervolt + CnQ) . . .
Last edited:
Having read this awesome article on Tom's Hardware I've been using a program called K10Stat to both undervolt and modify the speed of each individual p-state (800, 2100, 2500, 3200 seemed a bit too focussed on high end; VLC media player for example tends to jump between the lowest two states, so I've brought the second one down a bit).
In testing my x4 955 BE, running on an Asus M4A79XTD Evo, voltages are utterly solid, never had a board put up such consistent numbers
Defaults:
3.2GHz 1.375v
2.5GHz 1.25v
2.1GHz 1.15v
0.8GHz 1.0v
Now Testing:
3.2GHz 1.225v
2.4GHz 1.000v (may be too low)
1.6GHz 0.925v (may go lower)
0.8GHz 0.825v (may go lower)
Also trying to undervolt the northbridge, defaults to 1.3v, Tom's suggests 1.1v is possible.
Generally very pleased with the result, I'm usually unlucky with CPUs and get hot ones that need massive voltage; this one falls bang into line with the Tom's article, and has lost 5-6 degrees via optimising the voltages
*edit*
Naturally once I've established all the lowest 3 p-state voltages, I'll be overclocking and volting the nuts off the fastest - optimal cool and quiet, tip top speed under high load is the goal
* double edit *
Also of interest, K10Stat can lower the lowest p-state all the way down to 400MHz (maybe lower, haven't tried). Depending on the use of the PC you're building, a 4-600MHz base speed might save a smidge of power/heat/noise when idling, web browsing, etc.
In testing my x4 955 BE, running on an Asus M4A79XTD Evo, voltages are utterly solid, never had a board put up such consistent numbers
Defaults:
3.2GHz 1.375v
2.5GHz 1.25v
2.1GHz 1.15v
0.8GHz 1.0v
Now Testing:
3.2GHz 1.225v
2.4GHz 1.000v (may be too low)
1.6GHz 0.925v (may go lower)
0.8GHz 0.825v (may go lower)
Also trying to undervolt the northbridge, defaults to 1.3v, Tom's suggests 1.1v is possible.
Generally very pleased with the result, I'm usually unlucky with CPUs and get hot ones that need massive voltage; this one falls bang into line with the Tom's article, and has lost 5-6 degrees via optimising the voltages
*edit*
Naturally once I've established all the lowest 3 p-state voltages, I'll be overclocking and volting the nuts off the fastest - optimal cool and quiet, tip top speed under high load is the goal
* double edit *
Also of interest, K10Stat can lower the lowest p-state all the way down to 400MHz (maybe lower, haven't tried). Depending on the use of the PC you're building, a 4-600MHz base speed might save a smidge of power/heat/noise when idling, web browsing, etc.
Last edited:
very interesting! . . . thanks eddiew!
I've not got round to the undervolting stuff yet so most of the above is new to me!
p.s: you say the CPU-NB "defaults to 1.3v" . . . i thought there were 1.175v-1.2000v . . . I could be wrong of course!
I've not got round to the undervolting stuff yet so most of the above is new to me!
p.s: you say the CPU-NB "defaults to 1.3v" . . . i thought there were 1.175v-1.2000v . . . I could be wrong of course!
Hehe, I've only just started myself, this hardware was new in February, only just gotten a decent cooler (zalman 9700) and started to play with it properly
I'm not entirely sure what the NB should be at by spec; my motherboard chose to apply 1.1 on the slowest three p-states, and 1.3 on the fastest. Maybe that's normal for 790X, or maybe that's just Asus deciding to pre-optimise for overclocking.
Either way, since the NB doesn't actually change frequency across the p-states, I'm inclined to believe it could be a flat voltage all across. Am down to 1.05v on the lower three states now, expect I could turn down the top speed to the same.
Er, but I'll do that some time when I haven't got 10 windows open full of unsaved stuff - if there's one thing I have learned so far it's that undervolting too far can drop a BSOD on your head quicker than anything else. I won't forget my unwise setting of 0.6v core in a hurry
I'm not entirely sure what the NB should be at by spec; my motherboard chose to apply 1.1 on the slowest three p-states, and 1.3 on the fastest. Maybe that's normal for 790X, or maybe that's just Asus deciding to pre-optimise for overclocking.
Either way, since the NB doesn't actually change frequency across the p-states, I'm inclined to believe it could be a flat voltage all across. Am down to 1.05v on the lower three states now, expect I could turn down the top speed to the same.
Er, but I'll do that some time when I haven't got 10 windows open full of unsaved stuff - if there's one thing I have learned so far it's that undervolting too far can drop a BSOD on your head quicker than anything else. I won't forget my unwise setting of 0.6v core in a hurry
Following up with results after testing:
Phenom II x4 955 Black Edition undervolted in all p-states
Asus M4A79XTD Evo, typical vdroop of ~0.012v under load (needed the latest CPU-Z to show it, but I still consider this to be pretty good), northbridge at 1.1v.
p0 3.2GHz 1.2375v = 10% under stock (falls to 1.226 under load, so arguably 11%)
p1 2.4GHz 1.0500v = ~16% "
p2 1.6GHz 0.8250v = ~22% "
p3 0.8GHz 0.0750v = 25% "
Also practical (my personal favourite):
p0 3.5GHz 1.3375v = 1% under stock voltage with a 9% overclock, well done AMD
Settings accurate to +/- 0.0125v of minimal tolerable voltage (before vdroop) and each one passing over 2 hours of prime 95. Came out very close to the THG article, and I customised p1 and p2 for what I think is a better distribution that will use less power over the average day.*
P3 and p4 are very pleasing. Idle temps are down 3-4 degrees, and for applications like VLC media player, which hops around the slowest two states, the lowered p2 frequency and big voltage drop there should be a significant power reduction.
I suspect p1 will never see sustained action, but it managed a modest 16% voltage drop compared to it's nearest default state, and p0, well that's "only" 10%, but I think the high frequency means more current drawn (and power = volts * current), so saving 10% voltage at 4x the current is actually a larger saving, and indeed temps come down 6-7 degrees at full load. Lovely to have all fans on silent mode and see 54 degrees max
Overclocking suggestion; When on a locked multiplier, sort out your overclock first with CnQ off, then optimise p-states with K10Stat before turning CnQ on again. No reason you can't have an overclocked top state and a nice efficient CnQ at the same time.
* I would be interested to hear if anyone has an alternative suggestion for p-state frequency distribution. I can't think of an application that puts a consistent 60-70% load on a core... Would p1-3 being in the low half of the range be better than an even spread?
Phenom II x4 955 Black Edition undervolted in all p-states
Asus M4A79XTD Evo, typical vdroop of ~0.012v under load (needed the latest CPU-Z to show it, but I still consider this to be pretty good), northbridge at 1.1v.
p0 3.2GHz 1.2375v = 10% under stock (falls to 1.226 under load, so arguably 11%)
p1 2.4GHz 1.0500v = ~16% "
p2 1.6GHz 0.8250v = ~22% "
p3 0.8GHz 0.0750v = 25% "
Also practical (my personal favourite):
p0 3.5GHz 1.3375v = 1% under stock voltage with a 9% overclock, well done AMD
Settings accurate to +/- 0.0125v of minimal tolerable voltage (before vdroop) and each one passing over 2 hours of prime 95. Came out very close to the THG article, and I customised p1 and p2 for what I think is a better distribution that will use less power over the average day.*
P3 and p4 are very pleasing. Idle temps are down 3-4 degrees, and for applications like VLC media player, which hops around the slowest two states, the lowered p2 frequency and big voltage drop there should be a significant power reduction.
I suspect p1 will never see sustained action, but it managed a modest 16% voltage drop compared to it's nearest default state, and p0, well that's "only" 10%, but I think the high frequency means more current drawn (and power = volts * current), so saving 10% voltage at 4x the current is actually a larger saving, and indeed temps come down 6-7 degrees at full load. Lovely to have all fans on silent mode and see 54 degrees max
Overclocking suggestion; When on a locked multiplier, sort out your overclock first with CnQ off, then optimise p-states with K10Stat before turning CnQ on again. No reason you can't have an overclocked top state and a nice efficient CnQ at the same time.
* I would be interested to hear if anyone has an alternative suggestion for p-state frequency distribution. I can't think of an application that puts a consistent 60-70% load on a core... Would p1-3 being in the low half of the range be better than an even spread?
Number of p-states is dependant on the CPU type (this article has some more info).
You can still look to optimise your idle state though ^^
Last edited:
Tried on my unlocked 555 which is actually 965be and managed to go as low as 1.15 on 2.9ghz, could maybe go lower but didn't try. Was super cool too, max load temp in case with no airflow @ 33-35c.
I wish I could stay that cool... under load I'll touch 58-59 degrees... but that is with all fans on lowest possible speeds. Tis within tolerance, and makes no noise, why should I worry?
Voltage seems to be dependant on motherboard/socket, if you're on AM3 that's awesome, if on AM2 then it's about normal - either way well done for experimenting and finding out your lowest power needs ^^
Caporegime
My 965 can run at 1.3V for stock speeds. Probably lower tbh.
I was wrong. I actually have 4 states but the middle 2 barely get used because it seems to just jump to the top one as soon as it gets any load.
I think playing around with this might be a really good idea. Particularly when I am transcoding and the cpu often sits around 60-80% for a long time.
Is there a good set of instructions somewhere? I am finding the whole FID/DID thing really confusing. If you already have a stable overclock in BIOS and want to use that as P0 what do you do?
EDIT: Also here is another tool. Not tried it yet but it might be worth a look.
Last edited:
Soldato
my AMD Athlon II X4 Quad Core 620 2.6Ghz stock setting rock solid been in use for a month at 1.1Vcore
I see similar; either I'm maxxed out, or on idle speed with occasional hops up.
Consider setting p1 to be 90% of your maximum speed and change the up% setting to 90% of that. It may well choose to hang around in that state instead, and only go up to p0 when p1 absolutely maxes out.
I linked an article in my first post up in this thread that might help
FID is a multiplier - not your typical one though, it seems that 0 = 1600MHz, and each increment after that is 100MHz more.
DID is a divider, you can use it to go below 1.6GHz, they seem to correspond to:
0 = no divider
1 = half
2 = quarter
3 = eighth
4 = sixteenth
Thus to run at...
3.5GHz: FID = 19, DID = 0 [1600 + 1900]
3.2GHz: FID = 16, DID = 0
2.4GHz: FID = 8, DID = 0
1.6GHz: FID = 0, DID = 0
1.2GHz: FID = 8, DID = 1 [ (1600 + 800) / 2 ]
800MHz: FID = 0, DID = 1
I would lay odds that FID is always in 100MHz increments, but whether you'll have the same 1600 base clock as I do I'm not sure. Settings aren't applied until you press the Apply button anyway though, so don't be afraid to play
K10Stat only applies settings when you get into windows, so anything boot-stable in BIOS is fine. It works with multipliers, so your 100MHz increment may vary if you're overclocking your FSB (is it still called that, or is it HT now...?) You should be able to leave your overclock as it is and just modify the lower p-states.
I just left my BIOS on defaults, then overclocked p0 with K10Stat; these processors are expected to change speed while running, so an overclock looks no different to a CnQ change and is just as reliable
Afraid I don't know that one, I'm only just re-learning how to overclock. Last I did this was with an Intel 965 chipset, all the names for stuff have changed
I wish I could stay that cool... under load I'll touch 58-59 degrees... but that is with all fans on lowest possible speeds. Tis within tolerance, and makes no noise, why should I worry?
Voltage seems to be dependant on motherboard/socket, if you're on AM3 that's awesome, if on AM2 then it's about normal - either way well done for experimenting and finding out your lowest power needs ^^
AM3. And cheapo zalman 9700 so not a beast cooler either and case with no airflow : ). Can't complain really.