Throttlestop overclocking Desktop PCs

[Retrorockit]

From what I saw in the VRD11 datasheet VID6 which has a value of -.400V. (from 1.6125V.) is always high on LGA1366 and then excluded from what the CPU can select. It looks like the MSI mother board BIOS uses pull down resistors on the Voltage regulator to create an offset from whatever CPU Voltage is already there. So the Voltage IS fixed on the unlocked LGA1366 CPUs. At least as far as OEM systems are concerned. It looks like any changes will need to be done in hardware. I wonder if the W3690 is just binned higher than the W3680, or if it could have a higher VID setting also?
Can Throttlestop still reduce the Voltage on these CPUs if it was modded a little too high?

 

[unclewebb]

Can ThrottleStop still reduce the Voltage on these CPUs

ThrottleStop has zero control of voltage on the the 1st Gen Core i CPUs.

The W3680 and W3690 are basically identical coming down the assembly line. It is likely that the CPUs that needed the least amount of voltage to run reliably became W3690 processors and the ones that needed a little more voltage became W3680 processors. A W3680 would likely have a voltage table that is slightly higher than a W3690.

Intel has been doing this since the beginning of time. They make slight changes to MHz and voltage so they can create a wide variety or retail products. This allows them to sell as many CPUs off the wafer as possible. They do the same thing with the 10850K and 10900K. The CPUs that can run each core reliably at 5.3 GHz at reasonable voltage become 10900K. The ones that cannot run this speed reliably are given extra voltage, they are slowed down 100 MHz to 5.2 GHz and these become 10850K.

I have one of these 10850K. Below 5000 MHz, the default voltage curve is very similar to a 10900K. Trying to run at 5.3 GHz is possible but it takes high voltage which creates an insane amount of heat. Not worth it so it is best to run this chip at a maximum of 5.2 GHz.

 

[Retrorockit]

OK thanks for the reply. That means if i want to try a VID tapemod I will have to go directly to the desired Voltage, and not overshoot and then adjust back down.

 

[Retrorockit]

I looked into the VID tapemod option and found that they have multiplexed some signals onto the VID pins. But they're used before Powergood, and not to run the computer after boot. One of them activates VRD11.1 which is more restrictive than VRD11. A VRM hardmod might be the easy way out.

 

[frankr2994]

So can I make what I think is a reasonable assumption? Doing a vrm hard mod on something with no documentation such as my 5810. I have a 6 phase vrm, 6 MOSFETs, 6 chokes for the cpu. There are 4 caps placed in front of them assume they feed the vrm. Infront of the caps are 4 resistors. Would this be the same situation as I see with GPU modding where I figure out the resistance of those resistors and just solder on the same resistor on top to trick it into thinking half current? I don't need to trick voltage as I believe I don't have a lock there. Or going up past 1.3v is going to give me more heat than I can deal with anyways. If these are just shunt resistors this should work shouldn't it? I can only find Intel documentation for vrm layout on first gen xeons. The schematics I found appear to be much more simple than what I have.

Other than that here is a photo of the vrm and I have the data sheet on the MOSFETs.

The MOSFETs are rated for 45amps. Right now as far as I can figure with efficiency loss the voltage and wattage I can see that I'm running I'd be pulling about 22 amps through each MOSFET.

 

[Retrorockit]

Most Dell computers are not operated by the people who own them. Some of them ( maybe all?) are mission critical to someone. So anti tampering is a legitimate consideration for those applications.. If you don't own it don't do this. Overclocking without Voltage increase is fairly benign. This is not!

It's done between that area and the 12V power connector. There will be some MOSFETS and capacitors there too. The VRM chip is usually square with 36 or 40 pins. Unfortunately I only know the specific pinout for the Dimension E520. But the VSENSE pin is a feedback to the VRM chip to confirm proper regulation. If you pull that down with a resistor to GRD (trim pot.) you will have a Voltage adjustment. But it may need a direct measurement of Voltage installed also. The mod is done at the step down regulator from 12V. to CPU Voltage, not at the CPU itself. Since most enthusiast motherboards already have Voltage control it's more common among GPU modders. Understand that if you install a 20k trim pot. your initial Voltage will be that of a 20k pulldown resistor. Your 5810 may be quite different than the LGA1366 systems. You might have other options. Since the CPU reads the VSENSE pin it should be spoofed bt this mod also. If you install a fixed resistor it will create an offset.
The reason for the mod on the E520 is the 65nm CPUs it ran can actually use more than Intels 1.6V. cutoff limit. Especially for short CPUZ validation runs. The VRD 11.1 table the 32nm X58 chips use is 1.2V max. due to a .4V pulldown pin (VID 6) that is active in all instances on VRD 11.1. Any more than that needs to come from the MB itself. Dell for legitimate warranty and service contract considerations does not provide this.
You might be the first person in the history of the world to do this mod on that system. That's not unusual in this thread. The data sheet for your socket, your CPU, and your VRM chip will probably give you what you need.

I'm a retired truck mechanic with no electronics background. Someone else explained this mod to me. You may get better advice form someone else.

 

[frankr2994]

I guess what I'm confused about is the voltage vs current. In GPU hard modding they are only going after shunt resistors. Now the mod is still changing the voltage out of the resistors. It's programmed in that if you pass 12v through 20k ohm the voltage out of the resistors will vary based on load. I'm not in tune enough to make quick determinations on figuring out how many amps it takes to drag it down to 11.8 or anything. But all I'm getting at is if you change the voltage sense coming from a shunt resistor then all your doing is altering the sensed load. Voltage seen going through the vrm should be untouched and read accurately. Just playing around today I managed to get a stable 4.3ghz all core. Really dragged the voltage down to do it. As long as it stays under 140watts sustained it's good. And I know I have a good chip since all core 4.3 is happening at 1.16v. from what I read on the i7-5630 which is really the same cpu people have to push way more voltage to get clocks like that. Also my fans never ramped up and temps stayed at 70c for an hour of cinibench.
If it matters I'd like to stay away from using a pot in anything. If I used a fixed resistor of the same resistance then I know my shown current draw just needs doubled. I honestly bet that getting close to 280 watts wouldn't be all that hard with this chip as you approach 5ghz. If possible I'd like to stay at or below the 1.3v threshold. If I can get there that should lead to a stable 4.8ghz based off the 2 thread test I've done with that voltage.

And for anyone else and I'll verify this next week on the 5810 I just bought my son. I believe that the 1650v3 can be clocked to 4.2/4.3 completely stock in this machine. With as low as I have the voltage I don't really see the need for the cooling mods that I did. 6 cores hyper threaded at that speed is still pretty respectable in 2022 no matter what you need the PC to do.

You might be the first person in the history of the world to do this mod on that system. That's not unusual in this thread. The data sheet for your socket, your CPU, and your VRM chip will probably give you what you need.

Fyi everything I search for custom anything for my system goes back to all the sites I've posted on. Since this has been going on for over a year now sometimes I think I've found something only to come across my username in the thread lol.

This is the controller I found on my board. Anyone care to help me make sense of it?

And this. its a 3566A. The only difference I can see is that the B version was certified for something intel wanted. otherwise I believe the are identical. except the data sheet for the B (the one I have) does not show anything other than that pinout.

 

[lexluthermiester]

The MOSFETs are rated for 45amps. Right now as far as I can figure with efficiency loss the voltage and wattage I can see that I'm running I'd be pulling about 22 amps through each MOSFET.

I would glue heatsinks to the top of them and call it good. Replacing them would render minimal gains given their ratings and you expected usage level.

 

[frankr2994]

I would glue heatsinks to the top of them and call it good. Replacing them would render minimal gains given their ratings and you expected usage level.

Photo is old. Had a heatsink on them for some time now. Wasn't questioning replacing them but more on how can I trick them into reporting lower current. Was only making reference to the headroom I have with them to pull some more power.

 

[lexluthermiester]

Wasn't questioning replacing them but more on how can I trick them into reporting lower current. Was only making reference to the headroom I have with them to pull some more power.

Ah. Missed that part of the conversation. Sorry.

 

[Retrorockit]

The mod I'm discussing is Voltage regulation. I haven't looked at your system so I can only discuss the general modification. The VSENSE pin reports the regulated Voltage back to the VRM chip. A parallel resistor running to GRD will lower the returned value. The regulator will increase Voltage output to "correct" for this shortage. It's like putting a calibrated leak in a turbo wastegate actuator line ( I know you're a car guy).
Volts x Amps=Watts. So the current A can stay the same, but the power W will increase with V. So "Current" is a distraction for this mod.The Volt Mod calculator I posted earlier will show the resistance needed to produce a desired Voltage offset if you know the existing Voltage, and the resistance to GRD of the VSENSE pin. Those tables are in mV. Just add your own decimal point.
Be aware that Dell PSUs are multi rail and have 216W per rail (12V x 18A=216W). If you don't have 2- 12V rails ( or an aftermarket PSU) feeding the CPU connector you will only get so far. The MB fan header is running off of this also. If you increase the Voltage the fan may speed up due to heat and it can crash due to the extra draw on that circuit.. I've run the power leads from the fan to a different 12V. rail, and run the power leads from the MB fan connector to a small chipset fan to avoid a fan fail message. I also make sure all my drives are on a different rail than the CPU. Usually this means SSDs in or around the front drive bays. The 12V rails will be Yellow, White, or Blue. Then striped to add more if needed. With your newer CPU some of this may not be needed. An 800W Dell PSU may only be providing 216W to the CPU and fans. In the case of my original TS QX6800 project getting the fan off of the MB header was the biggest improvement there was. That was even with an EVGA single rail PSU installed.

 

[frankr2994]

The mod I'm discussing is Voltage regulation. I haven't looked at your system so I can only discuss the general modification. The VSENSE pin reports the regulated Voltage back to the VRM chip. A parallel resistor running to GRD will lower the returned value. The regulator will increase Voltage output to "correct" for this shortage. It's like putting a calibrated leak in a turbo wastegate actuator line ( I know you're a car guy).
Volts x Amps=Watts. So the current A can stay the same, but the power W will increase with V. So "Current" is a distraction for this mod.The Volt Mod calculator I posted earlier will show the resistance needed to produce a desired Voltage offset if you know the existing Voltage, and the resistance to GRD of the VSENSE pin. Those tables are in mV. Just add your own decimal point.
Be aware that Dell PSUs are multi rail and have 216W per rail (12V x 18A=216W). If you don't have 2- 12V rails ( or an aftermarket PSU) feeding the CPU connector you will only get so far. The MB fan header is running off of this also. If you increase the Voltage the fan may speed up due to heat and it can crash due to the extra draw on that circuit.. I've run the power leads from the fan to a different 12V. rail, and run the power leads from the MB fan connector to a small chipset fan to avoid a fan fail message. I also make sure all my drives are on a different rail than the CPU. Usually this means SSDs in or around the front drive bays. The 12V rails will be Yellow, White, or Blue. Then striped to add more if needed. With your newer CPU some of this may not be needed. An 800W Dell PSU may only be providing 216W to the CPU and fans. In the case of my original TS QX6800 project getting the fan off of the MB header was the biggest improvement there was. That was even with an EVGA single rail PSU installed.

I'll try and draw a diagram up tonight if how I think it should work. But I don't believe I have separate rails. My PC takes a server psu. 12v 1 rail. My cpu power is a 10 pin eps. That's huge and should be able to handle anything. But my idea doesn't change any reported voltage only reported current....which is technically done by voltage but not in the sense your talking about.

Ok I think I'm yanking the motherboard tonight. Stumbled across this by accident but if you look at my physical photo of the vrm you will see a "large" resistor next to each one. Those almost have to be the shunt resistors I'm talking about. One side should have no resistance back to the eps plug and the other end should go straight to vccin of the MOSFET. I can check all this with my meter. If that's the case I should be able to measure the resistance across and just buy 6 more shunt resistors and solder them on top . The vsense off of these should just be millivolts. That millivolt reading is what is relaying current drawn. If I cut the millivolt reading in half then half current should be reported without changing any actual voltage regulation.

Ok mb pulled. 2 things. Sorry you are right I have multiple 12v rails on that psu. I'll figure out how many the cpu uses. Next I believe the 4 components Infront of the 4 caps for the vrm are the shunts. Very low resistance. Too low for my fluke to measure accurately. I'll have to see if the fluke I have at work goes lower. But I have a direct path to vccin to my MOSFET and a direct path to my eps socket. Designation on the board is "LP"

 

[Retrorockit]

Your at the point where my ability to help is probably at it's end. You might try a TV repair shop, or cell phone repair shop and see if you can find someone local with experience doing this sort of mod. Maybe Badcaps.net could help. I'm afraid my vocabulary is limited to simple electrical work. We seem to be using 2 different vocabularies. With a newer CPU 1- 216W rail may be all you need if you can move other components off of it. Some of my LGA775 stuff goes beyond that power level. The PSU on that is very proprietary so your options are probably limited there.

 

[frankr2994]

I did do a pinout on my psu. 5 wires 2 on 1 rail 3 on another. So yep have 2 rails on my 10 pin. And ya i think I may find a radio repair shop. I've talked with some of those guys in the past local to my area on older car circuit boards. I found out today that my other fluke meter nor any meter in the shop can accurately read milli ohms so testing is pretty much impossible. I could really use a full schematic on the motherboard.

 

[Retrorockit]

Dell publishes very few pinouts. Mostly the standard connectors. Schematic is probably not going to happen.
My CB radio guy is afraid of 6 layer boards. But does reflash/repair automotive ECMs. It's hit or miss with those guys. But usually someone will know someone who does this. He referred me to a TV tech for MB work. He used to repair TVs there himself. So it varies. The younger guys tend to be in the cell phone shops.

 

[frankr2994]

On another note I'm still waiting for the e5-1650v3 to show up for my boys PC. I got sick of waiting and since mine is tore down I just took my cpu and stuck it in his. I have found in a short amount of time I became spoiled. After installation I got TS up and did a 4.0 overclock. Easy on this chip real easy. First thing is the stock cooler with arctic 5 on it runs at 77c at 4ghz. My cooler has no fan ramp up and keeps it at 70c at 4.3 full load. Next I installed cinibench for a little test. Not really needed since I know where this chip clocks. I have a 2.5 SSD in this system. Just extracting the cinibench files drove me mad. I've gotten so used to my nvme drive ripping through things like extracting a bunch of files. So in all I said a bit ago I bet I could run 4.3 on a bone stock system. That doesn't work. I didn't try it but since the fans were already ramping up I think it's safe to say it needs a bigger cooler which unfortunately is a PITA to do. So 4.0ghz for anyone browsing is what you can do with TS using a bone stock T5810 with an E5-1650v3 cpu.qnd I almost forgot. Just having voltage control makes this system 10 times better. Without changing anything this Dell is at pretty much every throttle at idle with the stock power curve. Even if you don't overclock a touch more voltage will make this chip stay up high.

Also I just posted a thread over at badcaps. Thanks for the recommendation. I've never been on that forum before and it looks promising by some of the info floating around over there. I'll keep you guys posted on any updates.

 

[frankr2994]

So I found this. https://www.sqltechbloq.com/motherboard.html#contact/

Requires some survey Bs to get the download. Seemed endless to me and I don't know if I'm being redirected or what. I tried posting a comment saying that I'd be willing to straight up pay the author for the download but I was not able to do so. Would someone else mind checking this out and telling me if you think it's legit?

So I found this. https://www.sqltechbloq.com/motherboard.html#contact/

Requires some survey Bs to get the download. Seemed endless to me and I don't know if I'm being redirected or what. I tried posting a comment saying that I'd be willing to straight up pay the author for the download but I was not able to do so. Would someone else mind checking this out and telling me if you think it's legit?

NVM it's total Bs found the thread where the user posted the link....for a totally different PC with the exact same post and user replies. It's just an auto generated scam replacing the PC name with whatever your searching for.

 

[frankr2994]

Ok found this today. Back side of the motherboard there are 6 resistors measuring approximately 2.4 ohms. They are right underneath the 6 MOSFETs. Only difference is I was looking for something inline with the positive circuit. These are on the ground side and by design anyways they look much more like the shunt resistors I've seen.

Also found another pwm controller ncp6133 . I don't know how this fits into the circuit yet. But this is off the data sheet for it. Looks like I'm on the right track. My office is being occupied by my 3 year old niece tonight so tomorrow is another day lol.
"Total Current Sense Amplifier
The NCP6133 uses a patented approach to sum the phase
currents into a single temperature compensated total current
signal. This signal is then used to generate the output voltage
droop, total current limit, and the output current monitoring
functions. The total current signal is floating with respect to
CSREF. The current signal is the difference between
CSCOMP and CSREF. The Ref(n) resistors sum the signals
from the output side of the inductors to create a low
impedance virtual ground. The amplifier actively filters and
gains up the voltage applied across the inductors to recover
the voltage drop across the inductor series resistance (DCR).
Rth is placed near an inductor to sense the temperature of the
inductor. This allows the filter time constant and gain to be
a function of the Rth NTC resistor and compensate for the
change in the DCR with temperature.
Figure 8.
-
+
CSN1
CSN2
CSN3
SWN1
SWN2
SWN3
Rref1
Rref2
Rref3
10
10
10
Rph1
Rph2
Rph3
Cref
1n
CSREF CSCOMP CSSUM
Ccs1
Ccs2
Rcs2 Rcs1
82.5 k 35.7 k
Rth
100 k
The DC gain equation for the current sensing:
VCSCOMP−CSREF -
−
Rcs2 Rcs1*Rth
Rcs1Rth
Rph * IoutTotal * DCR
(eq. 2)
Set the gain by adjusting the value of the Rph resistors.
The DC gain should set to the output voltage droop. If the
voltage from CSCOMP to CSREF is less than 100 mV at
ICCMAX then it is recommended to increase the gain of the
CSCOMP amp and add a resister divider to the Droop pin
filter. This is required to provide a good current signal to
offset voltage ratio for the ILIMIT pin. When no droop is
needed, the gain of the amplifier should be set to provide
~100 mV across the current limit programming resistor at
full load. The values of Rcs1 and Rcs2 are set based on the
100k NTC and the temperature effect of the inductor and
should not need to be changed. The NTC should be placed
near the closest inductor. The output voltage droop should
be set with the droop filter divider.
The pole frequency in the CSCOMP filter should be set
equal to the zero from the output inductor. This allows the
circuit to recover the inductor DCR voltage drop current
signal. Ccs1 and Ccs2 are in parallel to allow for fine tuning
of the time constant using commonly available values. It is
best to fine tune this filter during transient testing.
FZ -
DCR @ 25° C
2 * PI * LPhase
(eq. 3)
FP -
1
2 * PI * Rcs2 Rcs1*Rth@25° C
Rcs1Rth@25° C
* (Ccs1 Ccs2)
(eq. 4)"

 

[frankr2994]

I need to bring my needle point leads home for my meter to continue testing. I really do think if it happens it's going to be all me lol. I have all this info on 2 separate electronics forums and no response. The one I've been on for a while and I know there are very knowledgeable people. But getting into 50+ pins on a pwm controller is probably out of many people's scopes of knowledge. I have some feelers out there on getting a motherboard schematic and I'm going to call a company tomorrow that specializes in motherboard repair. It's a long shot but maybe I can pay a diagnostic fee for them to share something if they have it. Right now it isn't much help as all I have for components in question is typical circuit examples. And obviously on our multilayer circuit boards I can't visually follow traces. There is a chance if I can figure this out it can be applied to the 5820 as well. I've already read on the Dell forum the new w series xeons are " too powerful" to be used without constant throttle.running stock.

 

[Retrorockit]

I looked at some high resolution photos of your board and I can't see where it goes from 12V to CPU Voltage. So I can't help. The PWM controller, and the VRM controller are probably 2 different things. Dell does a very good job of locking down the PWM. That's the whole reason for Throttlestop multiplier overclocking. The Nehalem based CPUs brought some things with them from LGA775 to X58. Your's is a lot newer, so much has probably changed.

 

[frankr2994]

I have 12v all the up to the MOSFETs I had a heatsink over. The main pwm controller for those is a 6+1 phase controller. I'm not sure if they are even using the +1 phase for the memory due to there being more than 1 set of MOSFETs for the memory. The other pwm controller I found is listed as a 1/2/3 phase cpu controller. It's the only thing I've found that had pins to gather and report current usage.

What I really don't know is if 1 controller is in charge of the other or if they are totally separate. Hopefully with my needle leads tonight I'll be able to sort it

 

[Retrorockit]

I think the trend at Dell has been to eliminate the lower Volatge rails, and drop everything straight from 12V. Buck transformers don't seem to care what the input Voltage is. You can probably get the datasheet for your CPU socket, and specific CPU and learn something about how things are being done in that generation of CPU. Dell has been moving away from the 24 pin ATX connector for some time now. Unfortunately they haven't standardized on anything else.

 

[frankr2994]

Ya cpu pinouts on x99 are well not great. Some basic things are identified from what I can find but I haven't found a full schematic.

 

[Retrorockit]

Here are 900+ pages of it. Theoretically it will contain some useful information.

X99 dtatsheet at DuckDuckGo

DuckDuckGo. Privacy, Simplified.

duckduckgo.com

Of course VID now stands for Vendor ID register.

 

[frankr2994]

Here are 900+ pages of it. Theoretically it will contain some useful information.

Quote of the day there. Everything I look at work or at home theoretically helps.... usually never does lol. I will start browsing through it regardless. Thanks for the homework in tracking that down.

Ok think the 2nd pwm controller I found is for the memory. I've found 2 of them. So back into the 6 phase vrm controller. Pins 5 and 6 are vsen and vrtn. I have 100 ohms of resistance to my 2.2 ohm resistors in question. This was measured with everything off the board so 100 ohms being alot still means it's in the circuit. I really just don't know how to make this schematic out on what I should be seeing at those pins.