Undervolting: how low can you go?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Re: undervolting: how low can you go?
Sempron 2500+ "E6" 1.4GHz @ 0.88V
Seemed to be some bug in my TForce 6100 BIOS that prevented selecting any lower voltages. This was actual reported voltage using the 0.900V setting. Prime95 stable. Full-passive Scythe Ninja for "cooling."
I had a 130nm C0 Athlon 64 (Clawhammer) that I got to run at some speed @ 0.850V setting.
Seemed to be some bug in my TForce 6100 BIOS that prevented selecting any lower voltages. This was actual reported voltage using the 0.900V setting. Prime95 stable. Full-passive Scythe Ninja for "cooling."
I had a 130nm C0 Athlon 64 (Clawhammer) that I got to run at some speed @ 0.850V setting.
Intel C2D E2140 - M0 (775) stock - 1.325V
Stock: 1.6 GHz @ 0.925 V, idle=66W Ac, load=79W AC
OC'ed: 2.8 GHz @ 1.325 V
OC'ed: 3.0 GHz @ 1.400 V
All value come from bios setting.
System
MB Gigabyte P35-DS3 Rev=1.0
CPU E2140 - M0 with Ninja
2x1GB Geil PC2-6400
ATI Radeon X300 128MB
Samsung 2.5" 80Gb SATA HM080GI
Pioneer DVR-111D DVD Writer
TvTuner Kworld (SAA7131)
TvTuner Sapphite Theatrix (Theatre 550Pro)
PCI Firewire card
Nexus D12SL-12 exhaust fan @5V
PSU Seasonic S12 II 330W (Fan Swap: Scythe S-Flex 1600rpm running on psu controller)
Case Antec Sonata Designer
Stock: 1.6 GHz @ 0.925 V, idle=66W Ac, load=79W AC
OC'ed: 2.8 GHz @ 1.325 V
OC'ed: 3.0 GHz @ 1.400 V
All value come from bios setting.
System
MB Gigabyte P35-DS3 Rev=1.0
CPU E2140 - M0 with Ninja
2x1GB Geil PC2-6400
ATI Radeon X300 128MB
Samsung 2.5" 80Gb SATA HM080GI
Pioneer DVR-111D DVD Writer
TvTuner Kworld (SAA7131)
TvTuner Sapphite Theatrix (Theatre 550Pro)
PCI Firewire card
Nexus D12SL-12 exhaust fan @5V
PSU Seasonic S12 II 330W (Fan Swap: Scythe S-Flex 1600rpm running on psu controller)
Case Antec Sonata Designer
Intel C2D E2140 M0 - Conroe (775) stock - 1.287V
Minimum: 1.2 GHz @ 0.825V *
Stock: 1.6 GHz @ 0.887V **
OC'ed: 2.66 GHz @ 1.187V ***
* 91/103W idle/occt, boots always, lower (V) only with transition
** 93/108W 0.887V, and 99/129W 1.287V, wattage halved
*** 99/136W
8800GT being idle accounts for 45W included
P31-DS3L
SAGA+ 350P
Minimum: 1.2 GHz @ 0.825V *
Stock: 1.6 GHz @ 0.887V **
OC'ed: 2.66 GHz @ 1.187V ***
* 91/103W idle/occt, boots always, lower (V) only with transition
** 93/108W 0.887V, and 99/129W 1.287V, wattage halved
*** 99/136W
8800GT being idle accounts for 45W included
P31-DS3L
SAGA+ 350P
Last edited by yuu on Thu Jan 03, 2008 11:10 am, edited 1 time in total.
-
- Friend of SPCR
- Posts: 1439
- Joined: Tue Dec 14, 2004 4:06 pm
- Location: New Hampshire, US
- Contact:
I tried a little overclocking in anticipation of my soon to come HD 3870.
I might be able to push the 2.4ghz lower with the vcore, but for now, these settings are F@H and 3D stable.
AMD Athlon X2 4000+ Brisbane G1
2.4ghz 1.125v
2.6ghz 1.2v
Edit: updated the 2.4ghz setting, 1.125v is stable for F@H and gaming. Maybe I'll try even lower after the next WU completes.
I might be able to push the 2.4ghz lower with the vcore, but for now, these settings are F@H and 3D stable.
AMD Athlon X2 4000+ Brisbane G1
2.4ghz 1.125v
2.6ghz 1.2v
Edit: updated the 2.4ghz setting, 1.125v is stable for F@H and gaming. Maybe I'll try even lower after the next WU completes.
Is it possible that I could run Sempron LE-1100 like this?
0.8 GHz @ 0.550 V
1.9 GHz @ 0.800 V (only quick test - cca 0.5 hr Orthos)
How can I check that rmclock is not "fooling" me and the CPU is really going so low? Only thing, which makes me believe, that it is true is that core temperature has dropped from about 12°C to 10°C (but the absolute value is probably incorrect, because there is over 20°C in the room) when I changed 0.8 GHz from 0.800 V to 0.550 V
What do you think?
Updated:
Using CPU-Z I found out, that rmclock was really fooling me. On my motherboard Asus M2A-VM I set vcore to 1.0V thinking that this setting will be just for boot and will be later overloaded by rmclock. It was not. I have to set "auto" in bios to be able to control vcore from rmclock.
After that I got more realistic values:
0.8 GHz @ 0.800 V
1.9 GHz @ 1.000 V (only quick test - cca 0.5 hr Orthos)
0.8 GHz @ 0.550 V
1.9 GHz @ 0.800 V (only quick test - cca 0.5 hr Orthos)
How can I check that rmclock is not "fooling" me and the CPU is really going so low? Only thing, which makes me believe, that it is true is that core temperature has dropped from about 12°C to 10°C (but the absolute value is probably incorrect, because there is over 20°C in the room) when I changed 0.8 GHz from 0.800 V to 0.550 V
What do you think?
Updated:
Using CPU-Z I found out, that rmclock was really fooling me. On my motherboard Asus M2A-VM I set vcore to 1.0V thinking that this setting will be just for boot and will be later overloaded by rmclock. It was not. I have to set "auto" in bios to be able to control vcore from rmclock.
After that I got more realistic values:
0.8 GHz @ 0.800 V
1.9 GHz @ 1.000 V (only quick test - cca 0.5 hr Orthos)
Last edited by milink on Thu Jan 31, 2008 3:37 pm, edited 1 time in total.
Well I'm VERY disappointed with my E6700 compared to my E4300.
Using the same motherboard (Gigabyte P965-DS3), I have not been able to undervolt my E6700 anywhere near my E4300 with the E4300 being fully stable at stock clocking (1.8GHz) with 0.99375V (bios setting) while my E6700 can only get down to 1.075V with its multiplier locked down to 6 (1.6 GHz)
Also, I haven't been able to lower the FSB with the E6700 at all from the stock 266MHz while I was able to drop the E4300 FSB down to 100MHz when I was experimenting.
I really was hoping to be able to run the E6700 at similar voltages and clocking as the E4300 for 'normal' use and take advantage of the higher overclock when I needed it.
Using the same motherboard (Gigabyte P965-DS3), I have not been able to undervolt my E6700 anywhere near my E4300 with the E4300 being fully stable at stock clocking (1.8GHz) with 0.99375V (bios setting) while my E6700 can only get down to 1.075V with its multiplier locked down to 6 (1.6 GHz)
Also, I haven't been able to lower the FSB with the E6700 at all from the stock 266MHz while I was able to drop the E4300 FSB down to 100MHz when I was experimenting.
I really was hoping to be able to run the E6700 at similar voltages and clocking as the E4300 for 'normal' use and take advantage of the higher overclock when I needed it.
Last edited by kaange on Sat Feb 02, 2008 4:16 am, edited 2 times in total.
Well it seems that the E6700 has some weird interaction with the m/b that makes restarting inconsistent when the vCore is dropped below 1.075V (which was what was making me think that this was the lower limit).kaange wrote:Using the same motherboard (Gigabyte P965-DS3), I have not been able to undervolt my E6700 anywhere near my E4300 with the E4300 being fully stable at stock clocking (1.8GHz) with 0.99375V (bios setting) while my E6700 can only get down to 1.075V with its multiplier locked down to 6 (1.6 GHz)
When starting up from cold, the vCore can be dropped down to 0.88375V if I run with the multiplier set to 6 (1.6 GHz). It still seems to run hotter than the E4300 when that was running at 0.99375V but I guess the extra L2 cache will draw more current.
-
- Friend of SPCR
- Posts: 118
- Joined: Wed Jul 04, 2007 3:29 pm
- Location: Lost and Found Bin, Cypress, CA, USA
- Contact:
I have a new processor to add here. I think my previous Brisbane 3600+ (G1 stepping) results can be voided, as I have just purchased a Brisbane 5400+ (G2 stepping) as a last resort to fix months of instability. The voltages that were previously stable for 12h Orthos no longer worked. In fact, even at stock voltage, it was unstable. After swapping out every other component, only the processor was left as the common denominator. I've spent the last few days running Orthos to find the minimum voltages it will run at various speeds. First, a few curiosities:
As with my previous Brisbane, I am still unable to run RMClock's Performance on Demand profile, even if I adjust voltages above what I've found to be stable on Orthos for 12 hours. However, static FID/VIDs run just fine. Anyone else with similar issues?
Unlike my 3600+, if a VID/FID combo didn't fail within 1 minute, it didn't fail in 12 hours. All of these are 12 hours Orthos stable.
CPU: AMD Athlon 64 X2 5400+, 2.8 GHz, G2 stepping, 1.325V
Minimum Voltages
1 GHz, 1.2 GHz @ 0.775V
1.4 GHz @ 0.800V
2.0 GHz @ 0.950V
2.4 GHz @ 1.075V
2.8 GHz @ 1.225V
The 0.775V VID is only available with RMClock, and selecting any lower VID loops it around to 1.550V (confirmed with CPU-Z). CrystalCPUID only allows selection down to 0.800V.
As with my previous Brisbane, I am still unable to run RMClock's Performance on Demand profile, even if I adjust voltages above what I've found to be stable on Orthos for 12 hours. However, static FID/VIDs run just fine. Anyone else with similar issues?
Unlike my 3600+, if a VID/FID combo didn't fail within 1 minute, it didn't fail in 12 hours. All of these are 12 hours Orthos stable.
CPU: AMD Athlon 64 X2 5400+, 2.8 GHz, G2 stepping, 1.325V
Minimum Voltages
1 GHz, 1.2 GHz @ 0.775V
1.4 GHz @ 0.800V
2.0 GHz @ 0.950V
2.4 GHz @ 1.075V
2.8 GHz @ 1.225V
The 0.775V VID is only available with RMClock, and selecting any lower VID loops it around to 1.550V (confirmed with CPU-Z). CrystalCPUID only allows selection down to 0.800V.
#1
E2160 M0
Gigabyte G33 DS2R
Orig:
1.8Ghz (200x9) @ 1.325V
Running at:
1.4Ghz (200x7) @ 0.85V
Cooling: Stock heatsink without the fan. ~65C @ full load in a fanless matx case.
(If i calculated correctly this processor now uses less than 10W of power!)
#2
Q6600 G0
Gigabyte P35 DS3
Orig:
2.4Ghz (266x9) @ 1.325V
Running at:
2.0 Ghz (333x6) @ 1.0V
Cooling: Stock heatsink. ~40C at 1200 rpm @ full load.
All voltages from BIOS setting. Actual voltage about 0.01-0.03V lower depending on load.
E2160 M0
Gigabyte G33 DS2R
Orig:
1.8Ghz (200x9) @ 1.325V
Running at:
1.4Ghz (200x7) @ 0.85V
Cooling: Stock heatsink without the fan. ~65C @ full load in a fanless matx case.
(If i calculated correctly this processor now uses less than 10W of power!)
#2
Q6600 G0
Gigabyte P35 DS3
Orig:
2.4Ghz (266x9) @ 1.325V
Running at:
2.0 Ghz (333x6) @ 1.0V
Cooling: Stock heatsink. ~40C at 1200 rpm @ full load.
All voltages from BIOS setting. Actual voltage about 0.01-0.03V lower depending on load.
-
- Posts: 1
- Joined: Sun Mar 23, 2008 12:17 pm
- Location: Canada
Some numbers from undervolting an HP Laptop:
HP DV9535NR
17" Core 2 Duo T5250 1.5GHz, 2GB Ram, 2 x 120GB (HDD), Nvidia 8600M GS WinVistaHMPremium (Transparency Off)
Power measurements with PowerAngel and external monitor (no LCD power)
Undervolted using RMClock
VCPU: CPU-Z / RMClock__________Power Listed (Watts):
__________________________Idle / CpuBurnX2 / CpuBurnX1 + ATI Tool
Stock_____1.363V / Stock_____23______38_________50
Stable____1.063V / .950V_____20______27_________41
RMClock would let me set Vcore to .950V but CPU-Z showed 1.063V
The fan on the laptop runs slower and the machine is noticeably quieter.
17" Core 2 Duo T5250 1.5GHz, 2GB Ram, 2 x 120GB (HDD), Nvidia 8600M GS WinVistaHMPremium (Transparency Off)
Power measurements with PowerAngel and external monitor (no LCD power)
Undervolted using RMClock
VCPU: CPU-Z / RMClock__________Power Listed (Watts):
__________________________Idle / CpuBurnX2 / CpuBurnX1 + ATI Tool
Stock_____1.363V / Stock_____23______38_________50
Stable____1.063V / .950V_____20______27_________41
RMClock would let me set Vcore to .950V but CPU-Z showed 1.063V
The fan on the laptop runs slower and the machine is noticeably quieter.
a 0.850 setting in crystalcpuid crashes both prime95's tortures.
managed an occt 30min stable with a 5000+ BE G2 brisbane at the 0.875 setting at 2Ghz (10 multi standard 200 fsb) converted by both cpu-z, speedfan and everest to an actual 0.944Vcore
this cpu does kinda hit a wall before the 3ghz with 1.225 stable at 2.8 (I use 1.25) and over 1.4 to pass 3ghz.
I'll add definite results on the max later to this post when I get the time to run it overnight.
managed an occt 30min stable with a 5000+ BE G2 brisbane at the 0.875 setting at 2Ghz (10 multi standard 200 fsb) converted by both cpu-z, speedfan and everest to an actual 0.944Vcore
this cpu does kinda hit a wall before the 3ghz with 1.225 stable at 2.8 (I use 1.25) and over 1.4 to pass 3ghz.
I'll add definite results on the max later to this post when I get the time to run it overnight.
-
- Friend of SPCR
- Posts: 1439
- Joined: Tue Dec 14, 2004 4:06 pm
- Location: New Hampshire, US
- Contact:
That's pretty insane, what kind of power draw do you see from the system with that .875 setting? That's ridiculous. I can only go down to 1.00v at 2ghz, .875 is a big drop.Lostfaith wrote:a 0.850 setting in crystalcpuid crashes both prime95's tortures.
managed an occt 30min stable with a 5000+ BE G2 brisbane at the 0.875 setting at 2Ghz (10 multi standard 200 fsb) converted by both cpu-z, speedfan and everest to an actual 0.944Vcore
It's an actual 0.944 reported though (see screenshot/ previous post) as my FSP Epsilon FX700-GLN is a bit powerful and the motherboard bios isn't quite optimal for tweaking regrettably (latest BETA even locks the HT multi to 5x so no overclocking and thus I'm using a relatively old bios... shame on you asus).ryboto wrote:That's pretty insane, what kind of power draw do you see from the system with that .875 setting? That's ridiculous. I can only go down to 1.00v at 2ghz, .875 is a big drop.Lostfaith wrote:a 0.850 setting in crystalcpuid crashes both prime95's tortures.
managed an occt 30min stable with a 5000+ BE G2 brisbane at the 0.875 setting at 2Ghz (10 multi standard 200 fsb) converted by both cpu-z, speedfan and everest to an actual 0.944Vcore
idles at around 39degrees, however I'd better let it run a dualprime overnight or a longer occt to be completely sure it's >rock<stable at 2ghz/0.944
Also with the 3 hdd's and old 8800gts blazing along without lowering to any 2d speeds I suspect powerdraw still being quite high compared to the G1 rev brisbane in the bedroom (see sig). I do reduce clocks with rivatuner to the minimum for extended periods of plain work though...
edit: before anyone wonders, the weird fanspeed from speedfan is due to the 140mm case-fan idle speed going below 1000rpm (soprano dx frontfan) with asus qfan on active
OK, I forgot that the E4300 would drop back to 1.2GHz when idling. I did manage to get the E6700 to boot with FSB at 200MHz, 1.0V vCore and it ran very nice and cool but then the PC wouldn't post after a couple of reboots so I've had to abandon this quest. It looks like my m/b is at fault here.kaange wrote:When starting up from cold, the vCore can be dropped down to 0.88375V if I run with the multiplier set to 6 (1.6 GHz). It still seems to run hotter than the E4300 when that was running at 0.99375V but I guess the extra L2 cache will draw more current.
I just started testing my X2 4400+ Brisbane rev G2 on an ASUS M3N78-EMH HDMI (GeForce 8200 based board). It seems I struck gold with this 4400+, or how about this:
4400+ Brisbane rev G2 @ 2.66GHz (1.10V)
It has passed OCCT 2h test and Orthos Large FFTs ~1h. Although that's hardly enough to consider it 24/7 stable, it's good enough that a single voltage bump should stabilize it otherwise.
I don't know what's normal for these Brisbanes, but 1.1V at 2.66GHz sure seems good.
EDIT: Another setting that seems stable:
4400+ Brisbane rev G2 @ 2.3GHz (1.01V)
4400+ Brisbane rev G2 @ 2.66GHz (1.10V)
It has passed OCCT 2h test and Orthos Large FFTs ~1h. Although that's hardly enough to consider it 24/7 stable, it's good enough that a single voltage bump should stabilize it otherwise.
I don't know what's normal for these Brisbanes, but 1.1V at 2.66GHz sure seems good.
EDIT: Another setting that seems stable:
4400+ Brisbane rev G2 @ 2.3GHz (1.01V)
Last edited by Mikael on Thu Mar 27, 2008 8:05 am, edited 1 time in total.
I think they're supposed to replace the older G1 revision at some point, but there's probably still a lot of G1 CPUs in stock in most places.ryboto wrote:are the G2 cores replacing all G1s? Is there an indication in the model number to the core revision?
All G2 revision cores have box numbers that end with DOBOX. For example, my box says ADO4400DOBOX. The CPUs themselves should have a part number ending with IAA5DO, in my case ADO4400IAA5DO.
EDIT: Ohh, and the 5000+ Black Edition with box number ADO5000DSWOF should all be G2.
Yes, same here!RedAE102 wrote: As with my previous Brisbane, I am still unable to run RMClock's Performance on Demand profile, even if I adjust voltages above what I've found to be stable on Orthos for 12 hours. However, static FID/VIDs run just fine. Anyone else with similar issues?
Unlike my 3600+, if a VID/FID combo didn't fail within 1 minute, it didn't fail in 12 hours. All of these are 12 hours Orthos stable.
I just picked up a 5000+ Black edition to replace my 3800+ EE, TOTALLY stable with RMClock (I can't undervolt in the Bios). Using the POD profile, it'll appear stable, but then it will crash hard - generally when downclocking from, say, 3 GHz to 1 GHz.
This is in Vista Home Premium, using Media Center.
I tried all kinds of voltages, even close to stock, and couldn't get it stable after a few hours (4 am) of testing. I decided I don't have the free-time to tinker, so out it goes, and out it went. I was a bit bummed, since I'd like the freedom to ramp up to 3 GHz at times.
Dang.. I just realized I could have set it to 3 GHz in the BIOS as a start.. it would cook inside my NSK2400 while it booted, but then C'n'Q would kick in... hmm... must.. resist.. temptatation.. to.. hack..
-Dan
Okay, to summarize:
Athlon64 X2 4400+ Brisbane rev G2
3.00GHz @ 1.23V
2.66GHz @ 1.10V
2.30GHz @ 1.01V (stock clock)
1.00GHz @ 0.83V (lowest possible in Crystal CPUID, CPU may take lower)
EDIT: BTW, this CPU has a VID of 1.300V, which isn't even listen on the amdcompare.com site...
Athlon64 X2 4400+ Brisbane rev G2
3.00GHz @ 1.23V
2.66GHz @ 1.10V
2.30GHz @ 1.01V (stock clock)
1.00GHz @ 0.83V (lowest possible in Crystal CPUID, CPU may take lower)
EDIT: BTW, this CPU has a VID of 1.300V, which isn't even listen on the amdcompare.com site...
Last edited by Mikael on Thu Mar 27, 2008 2:23 pm, edited 1 time in total.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
According to Wikipedia (I have not checked Intel's site) Core 2 Duo E4400 SLA98 (M0) and Core 2 Duo E4500 SLA95 (M0) have the lowest stock min Vcore among all the Allendale/Conroe cpu, a mere 0.962V (the newer E4600 and E4700 have 1.162V, while the former E4300 has 1.225V), and a slightly higher stock max Vcore (1.325V vs. 1.312V for the Allendales).
Have you any evidence that these data may actually impact the respective downvolting and overclocking capabilities?
Any hint is welcome, thanks in advance.
Q.
Have you any evidence that these data may actually impact the respective downvolting and overclocking capabilities?
Any hint is welcome, thanks in advance.
Q.
-
- Posts: 22
- Joined: Thu Jul 19, 2007 4:38 am
- Location: Italy
I had the same problem with a 4800+ G2 Brisbane at a Gigabyte 690G motherboard. Testing with Orthos and fixed multiplier/voltage didn't give problems, but dynamically switching between 0.8v and 1.2v made the machine reboot.RedAE102 wrote:IAs with my previous Brisbane, I am still unable to run RMClock's Performance on Demand profile, even if I adjust voltages above what I've found to be stable on Orthos for 12 hours. However, static FID/VIDs run just fine. Anyone else with similar issues?
Unlike my 3600+, if a VID/FID combo didn't fail within 1 minute, it didn't fail in 12 hours. All of these are 12 hours Orthos stable.
The machine has now become a 24/7 storage server, constantly running 1 GHz @ 0.8v, for a 31W AC power consumption.