Undervolting: how low can you go?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
That is interesting. I have those exact reported voltages. Mine is set to 1.1000 V in the BIOS, though. But I too can't seem to have it stable at any lower than that.Fred wrote:The lowest step i can pick with my E8400 at stock speed(9 x ~333.3) in the BIOS is 1.10625V.
CPU-Z shows 1.072V in idle and 1.056V during load.
I also noticed that cpu-z displays the same values for a range of bios Vcore settings, why is that?
Nice, 1.10000V! When i chose that setting it died a few seconds after windows started. =/
Edit: Hmm. It looks like 1.10625V was to little or something random made my system crash. Anyhow, increased voltage by one step.
Nice, 1.10000V! When i chose that setting it died a few seconds after windows started. =/
Edit: Hmm. It looks like 1.10625V was to little or something random made my system crash. Anyhow, increased voltage by one step.
-
- Posts: 8
- Joined: Mon Jun 16, 2008 12:14 am
- Location: NZ
Undervolter from years ago At various stages I've owned 3 mobile athlons. The best one, a XP-M 2400+ worked well at between 1.225v and 1.25v at 1.8ghz. That was a nice chip. Also, a couple of these went into a Tyan 2466 with 4 gigs of ram. The two ran at stock 2133mhz each at the 1.56v. Individually they could run faster but together that was the stable voltage. That was a nice computer, it had 2x 10k scsi drives and fit all into a homemade plywood and rubber (thick kind the make rubber flooring out of). All ran sweet off 300w zalman PSU.
I also owned a really nice Thoroughbred B 1800+ which ran rock solid at 1.2v and stock speeds which was either 1540 mhz or 1460 mhz depending what the motherboard felt like doing at the time
My latest chip is a MT40 Turion which runs various speeds well. However, the motherboard died (from screwdriver abuse I might add) and I'll have a new one here soon. I'll post when I set it up. From memory, it was doing 0.85v for 1600mhz and 1.0somethingv (less than 1.1v) for 2200mhz on that dungy board
I also owned a really nice Thoroughbred B 1800+ which ran rock solid at 1.2v and stock speeds which was either 1540 mhz or 1460 mhz depending what the motherboard felt like doing at the time
My latest chip is a MT40 Turion which runs various speeds well. However, the motherboard died (from screwdriver abuse I might add) and I'll have a new one here soon. I'll post when I set it up. From memory, it was doing 0.85v for 1600mhz and 1.0somethingv (less than 1.1v) for 2200mhz on that dungy board
-
- Posts: 8
- Joined: Mon Jun 16, 2008 12:14 am
- Location: NZ
Got the new computer setup with the MT-40. It's a foxconn 760GXK8MC. The board has got a few features but not a fully blown setup. Right now compiling gcc-4.1.2 using these speeds 2200mhz at 1.05v. It's running fanless
Incidentally, I want to point potential AMD underclockers to this post by burebista. I haven't seen any mention of it but I came across it on this board as this new board won't post properly with when this formula is out of kilter. That is, if my vcore difference is more than 0.325v. Very interesting. The last board does not have a problem so it's possibly hard coded into the bios.
1/2 the Vdimm is apparently what the memory controller (on die because of AMD) sees. If the difference between the vcore and the memory controller voltage (half vdimm, usually 1.25v for ddr if not altered from default in bios) is more than 0.325v then you are running risk of frying your cpu. So, if your vcore is less than 0.9v you may be in trouble if running standard memory voltages (2.5v/2 - 0.9v = 0.325v).
Incidentally, I want to point potential AMD underclockers to this post by burebista. I haven't seen any mention of it but I came across it on this board as this new board won't post properly with when this formula is out of kilter. That is, if my vcore difference is more than 0.325v. Very interesting. The last board does not have a problem so it's possibly hard coded into the bios.
I'll try and sum up.burebista wrote: @NeilBlanchard An interesting thread.
It seems that there is a protection diode between the memory controller and the CPU that can forward bias if the memory voltage gets too high. This leads to a quick death of the CPU. AMD won't provide numbers -- they just recommend increasing Vcore.
Consensus from the thread is that Vdimm/2 - Vcore < 0.325V (that's 1/2 the memory voltage minus the CPU voltage should be less than 0.325V)...
But I guess on DDRII Vdimm 1.80V should be fine with Vcore 0.900V.
1/2 the Vdimm is apparently what the memory controller (on die because of AMD) sees. If the difference between the vcore and the memory controller voltage (half vdimm, usually 1.25v for ddr if not altered from default in bios) is more than 0.325v then you are running risk of frying your cpu. So, if your vcore is less than 0.9v you may be in trouble if running standard memory voltages (2.5v/2 - 0.9v = 0.325v).
I still think the memory vs CPU voltage is hit and miss. I remember lots of talk about it on socket 939 with the 90nm CPU's and high end memory, but the only people I knew that burnt out CPU's were running memory voltage in the 3.2v+ range.
I've been running 0.8v CPU and 2.8v memory for the last 3-4 years without issue.
I've been running 0.8v CPU and 2.8v memory for the last 3-4 years without issue.
I've recently experimented with undervolting my old Shuttle SN45G
AMD Athlon 2400+ Thoroughbred
Stock voltage: 1.65V @ 2GHz
Min stable voltage (low temps): 1.425V @ 2GHz
Min stable volatge (passive up to 68C): 1.5V @2GHz
No lower voltages were completely stable regardless of speed, even 1.4V at 800MHz was unstable.
Edit:
Although 1.425V was stable running in the low 40's with a fan, after trying the system passively, instablility came in on these low voltages. Raising the voltage to 1.5V made it stable up to 68C which I deem high enough anyway. The mobo itself will shutdown when the CPU hits 75C. This temperature influence may not affect lower frequencies which can't reach that temperature.
AMD Athlon 2400+ Thoroughbred
Stock voltage: 1.65V @ 2GHz
Min stable voltage (low temps): 1.425V @ 2GHz
Min stable volatge (passive up to 68C): 1.5V @2GHz
No lower voltages were completely stable regardless of speed, even 1.4V at 800MHz was unstable.
Edit:
Although 1.425V was stable running in the low 40's with a fan, after trying the system passively, instablility came in on these low voltages. Raising the voltage to 1.5V made it stable up to 68C which I deem high enough anyway. The mobo itself will shutdown when the CPU hits 75C. This temperature influence may not affect lower frequencies which can't reach that temperature.
Last edited by edh on Sat Aug 30, 2008 10:59 am, edited 1 time in total.
E6600 @ 1.06V(after Vdroop) @ stock 2.4ghz
I can probably get it down to 1.01/1.02 stable but feel safer with a bit of leeway as shorting Cmos pin is annoying
Stable using Orthos
It does 3.2ghz on stock voltage with speedstep enabled, pretty typical values for an E6600 from what Ive read.
I can probably get it down to 1.01/1.02 stable but feel safer with a bit of leeway as shorting Cmos pin is annoying
Stable using Orthos
It does 3.2ghz on stock voltage with speedstep enabled, pretty typical values for an E6600 from what Ive read.
Last edited by farns on Wed Oct 01, 2008 7:41 pm, edited 2 times in total.
-
- Posts: 154
- Joined: Fri Dec 23, 2005 3:05 am
- Location: Germany
-
- Friend of SPCR
- Posts: 1439
- Joined: Tue Dec 14, 2004 4:06 pm
- Location: New Hampshire, US
- Contact:
Q9400 in a DFI LP JR P45 T2RS - Set by CrystalCPUID,voltages read by CPUZ/Everest, Load by Prime 95
Stock: 2.66ghz @ 1.212v, 1.168v under load
Undervolt: 2.66ghz @ 1.088v, 1.072v under Load
unfortunately, 1.1v is the lowest CrystalCPUID or Rightmark will let me set the CPU, and the board seems to slightly undervolt. If I attempt to use a lower setting in CrystalCPUID, which have an [x] next to them, the system immediately restarts. Could it just be a board limitation?
edit: I guess since I changed some of the Pico's wiring, the voltage drop has decreased under load, I've changed the numbers accordingly.
Stock: 2.66ghz @ 1.212v, 1.168v under load
Undervolt: 2.66ghz @ 1.088v, 1.072v under Load
unfortunately, 1.1v is the lowest CrystalCPUID or Rightmark will let me set the CPU, and the board seems to slightly undervolt. If I attempt to use a lower setting in CrystalCPUID, which have an [x] next to them, the system immediately restarts. Could it just be a board limitation?
edit: I guess since I changed some of the Pico's wiring, the voltage drop has decreased under load, I've changed the numbers accordingly.
Another E6600 Conroe here. Minimum voltage in BIOS is 1.1V I can get to 2.54GHz before becoming unstable. Quick summary of lowest stable voltage vs CPU speed.
CPU Speed - BIOS voltage - CPUZ voltage (load)
2.54GHz - - - - - - 1.1v - - - - - - -1.016v (minimum I can set)
2.8GHz - - - - - - - 1.225v - - - - - 1.128V
3.0GHz - - - - - - - 1.2625v - - - - 1.136v
3.2GHz - - - - - - - 1.35v - - - - - -1.224v
3.5GHz - - - - - - - 1.475v - - - - - 1.312v
At 3.5GHz+ I need the fans working near full speed (not quiet) to keep load temps 60-70°C, at minimum speed/voltage the fans need to be bearly spinning. Not suprising as my calculation (3.5/2.54)*(1.312/1.016)^2 shows the CPU to be using 2.3x as much watts!
Seb
CPU Speed - BIOS voltage - CPUZ voltage (load)
2.54GHz - - - - - - 1.1v - - - - - - -1.016v (minimum I can set)
2.8GHz - - - - - - - 1.225v - - - - - 1.128V
3.0GHz - - - - - - - 1.2625v - - - - 1.136v
3.2GHz - - - - - - - 1.35v - - - - - -1.224v
3.5GHz - - - - - - - 1.475v - - - - - 1.312v
At 3.5GHz+ I need the fans working near full speed (not quiet) to keep load temps 60-70°C, at minimum speed/voltage the fans need to be bearly spinning. Not suprising as my calculation (3.5/2.54)*(1.312/1.016)^2 shows the CPU to be using 2.3x as much watts!
Seb
-
- Friend of SPCR
- Posts: 118
- Joined: Wed Jul 04, 2007 3:29 pm
- Location: Lost and Found Bin, Cypress, CA, USA
- Contact:
I'd like to update and add some data. First, to update my Brisbane Athlon 64 X2 5400+ G2 with the reported load voltages in CPU-Z/Everest:
Clock @ Reported Voltage (VID) Stock VID
1.2 GHz @ 0.768 V (0.775 V) 1.100 V (stock is 1GHz)
1.8 GHz @ 0.880-0.896 V (0.900 V) 1.125 V
2.0 GHz @ 0.928-0.944 V (0.950 V) 1.175 V
2.4 GHz @ 1.056-1.072 V (1.075 V)
2.8 GHz @ 1.200-1.216 V (1.225 V)
AMD Athlon X2 Mobile TK-57 (Tyler-256) G2 (VIDs only)
Stock
800 MHz @ 0.900 V
1.9 GHz @ 1.075 V
Undervolted
800 MHz @ 0.7375 V
1.4 GHz @ 0.850 V
1.9 GHz @ 0.975 V
AMD Turion 64 ML-30 (Lancaster) E5
Stock VIDs in CrystalCPUID (in RMClock):
800 MHz @ 1.000 V (0.900 V)
1.6 GHz @ 1.450 V (1.350 V)
Undervolted
800 MHz @ 0.900 V (0.800 V)
1.6 GHz @ 1.050 V (0.950 V)
AMD Athlon 64 3400+ (Venice S754) E3
Stock
1.0 GHz @ 1.100 V
2.4 GHz @ 1.400 V
Undervolted
1.0 GHz @ 1.100 V (VID lower limit lock on E3/E6 desktop chips)
1.4 GHz @ 1.100 V
2.4 GHz @ 1.350 V
Clock @ Reported Voltage (VID) Stock VID
1.2 GHz @ 0.768 V (0.775 V) 1.100 V (stock is 1GHz)
1.8 GHz @ 0.880-0.896 V (0.900 V) 1.125 V
2.0 GHz @ 0.928-0.944 V (0.950 V) 1.175 V
2.4 GHz @ 1.056-1.072 V (1.075 V)
2.8 GHz @ 1.200-1.216 V (1.225 V)
AMD Athlon X2 Mobile TK-57 (Tyler-256) G2 (VIDs only)
Stock
800 MHz @ 0.900 V
1.9 GHz @ 1.075 V
Undervolted
800 MHz @ 0.7375 V
1.4 GHz @ 0.850 V
1.9 GHz @ 0.975 V
AMD Turion 64 ML-30 (Lancaster) E5
Stock VIDs in CrystalCPUID (in RMClock):
800 MHz @ 1.000 V (0.900 V)
1.6 GHz @ 1.450 V (1.350 V)
Undervolted
800 MHz @ 0.900 V (0.800 V)
1.6 GHz @ 1.050 V (0.950 V)
AMD Athlon 64 3400+ (Venice S754) E3
Stock
1.0 GHz @ 1.100 V
2.4 GHz @ 1.400 V
Undervolted
1.0 GHz @ 1.100 V (VID lower limit lock on E3/E6 desktop chips)
1.4 GHz @ 1.100 V
2.4 GHz @ 1.350 V
Last edited by RedAE102 on Fri Oct 24, 2008 11:54 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 just swapped my DFI LP JR P45 T2RS for an Asus P5Q-EM. At stock, the same undervolt works,
2.66ghz Bios set to 1.1v(1.07-1.08 actual by Everest)
3.0ghz Bios set to 1.13v(1.12 actual by Everest)
The system is rock solid at 3.0ghz, I was hoping to achieve 1.1v, but power consumption only increased 5W so, I'm not sure what the 0.02v would gain.
2.66ghz Bios set to 1.1v(1.07-1.08 actual by Everest)
3.0ghz Bios set to 1.13v(1.12 actual by Everest)
The system is rock solid at 3.0ghz, I was hoping to achieve 1.1v, but power consumption only increased 5W so, I'm not sure what the 0.02v would gain.
Re: e8400
How low can I go with power consumption of underclocked / undervolted Intel E8400 / E8500 without risking the stability? I'd really like a rough estimate before I commit to buying it...Conroy wrote:Intel E8400
Stock: 3.0 Ghz @ 1.2250V (cpuz reports 1.216v), 73/113W
3.0 GHz @ 1.0375V (cpuz reports 1.024v), 70/98W
3.6 GHz @ 1.2250V (cpuz reports 1.216v), 77/125W
Remember your overall system power consumption depends on the rest of your system as much as your CPU, so you need to ensure all components are as low-power as possible.
My E8400 runs stock speeds at 1V in the bios (slightly lower in cpuZ). The overall system runs fine on an 80W brick, but without much headroom.
I haven't got measurements of its current configuration, but I'd guess it idles at ~65W, load ~75-80W (from the wall). Burning a DVD while both cores are at 100% sometimes (but not always) causes a reboot - I'm guessing that's because I'm pushing the brick to its limits.
My E8400 runs stock speeds at 1V in the bios (slightly lower in cpuZ). The overall system runs fine on an 80W brick, but without much headroom.
I haven't got measurements of its current configuration, but I'd guess it idles at ~65W, load ~75-80W (from the wall). Burning a DVD while both cores are at 100% sometimes (but not always) causes a reboot - I'm guessing that's because I'm pushing the brick to its limits.
Hi all..im new on this forum and building my silent pc now. This results are with the stock cooler of Intel still installed, since my Noctua cooler was missing parts (great service btw).
e8400 stepping E0 on Gigabyte P45-DS3r
3.0 ghz not sure which voltage in bios
CPU-Z -> 1.008v and 0.992v under load
14+ hours stable with Orthos blend test
So im pretty satisfied, undervolt made the temps drop from ~59 to ~51 degrees.
When i receive my last parts i will post in the gallery with more info
e8400 stepping E0 on Gigabyte P45-DS3r
3.0 ghz not sure which voltage in bios
CPU-Z -> 1.008v and 0.992v under load
14+ hours stable with Orthos blend test
So im pretty satisfied, undervolt made the temps drop from ~59 to ~51 degrees.
When i receive my last parts i will post in the gallery with more info
Pentium 2160 stock speed stable at 0,91v.
Idle 30C and load 44C with stock fan running at 900-1000rpm.
Small microatx case with ventilation only from a FSP PSU and there's also an 80mm fan running at 1000rpm mounted
inside the case blowing down on the expansion cards to keep a passively cooled graphics card from overheating.
http://img517.imageshack.us/img517/9007/cpusz1.jpg
Idle 30C and load 44C with stock fan running at 900-1000rpm.
Small microatx case with ventilation only from a FSP PSU and there's also an 80mm fan running at 1000rpm mounted
inside the case blowing down on the expansion cards to keep a passively cooled graphics card from overheating.
http://img517.imageshack.us/img517/9007/cpusz1.jpg
-
- Posts: 1
- Joined: Thu Oct 23, 2008 8:55 pm
- Location: Australia
Has anyone undervolted a Q9550 quad? I have an E0 step, and am interested in seeing if it's possible to undervolt it while remaining at 2.83GHz (or even when bumped to 3GHz).
I'm waiting to buy a watt meter first (Jaycar here in Australia has a device, similar to the US "Kill-A-Watt"), so I can measure stock system consumption "at the wall", and consumption after tweaks. I'm interested in tweaking my system for efficiency at this stage rather than for performance gains, but a faux-Q9650 at 3GHz would be nice
I'm waiting to buy a watt meter first (Jaycar here in Australia has a device, similar to the US "Kill-A-Watt"), so I can measure stock system consumption "at the wall", and consumption after tweaks. I'm interested in tweaking my system for efficiency at this stage rather than for performance gains, but a faux-Q9650 at 3GHz would be nice
Well, it has taken a lot of weekends without being able to report anything.Cistron wrote:Hmm, my E2160 currently runs with 1.285V core voltage at 3GHz, but that value was set by default. So I'm wondering how far I can go down. I'll experiment on the weekend.
Is it possible that RMClock doesn't support my motherboard (Abit IP35)? Whatever I set the voltages to, nothing changes. There is still a voltage drop by speedstep in idle, but underload the voltage still climbs to 1.285V. Besides, I can only drop to 1.185V in the programme.
Any ideas?
edit: I'm running Vista Home Premium.
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
Athlon X2 5400+ BE
Stock: 2.8ghz (200*14) 1.35v
My CrystalCPUID Settings:
3ghz (222*13.5) 1.325v
2ghz (222*9) 1.00v
1ghz (222*4.5) .8v (lowest I can set)
I find it really amusing that I could increase the frequency AND drop the voltage from stock.
Spends probably about 80% of time in the lowest setting since I can do pretty much anything short of gaming on it.
WoW with all the settings gets it up to the middle setting.
Higher end games: GRID, Crysis, etc. are about the only things that get it up to the highest setting.
I do have pretty strict settings on CPUID, it has to be using an average of 80% to go up and an average of 30% to go down. I know this probably gives me some performance loss, but I have yet to notice it, when I do I'll start tinkering again.
Stock: 2.8ghz (200*14) 1.35v
My CrystalCPUID Settings:
3ghz (222*13.5) 1.325v
2ghz (222*9) 1.00v
1ghz (222*4.5) .8v (lowest I can set)
I find it really amusing that I could increase the frequency AND drop the voltage from stock.
Spends probably about 80% of time in the lowest setting since I can do pretty much anything short of gaming on it.
WoW with all the settings gets it up to the middle setting.
Higher end games: GRID, Crysis, etc. are about the only things that get it up to the highest setting.
I do have pretty strict settings on CPUID, it has to be using an average of 80% to go up and an average of 30% to go down. I know this probably gives me some performance loss, but I have yet to notice it, when I do I'll start tinkering again.