Overclock.net banner
1 - 20 of 52 Posts

·
Registered
Joined
·
26 Posts
Discussion Starter · #1 ·
Hello all,

I have a GT1030 (GDDR5 version) which I am looking to perform a modification to increase the power limit. Here is my predicament:

This card can hit ~2000mhz core at only ~70C with old thermal paste & stock air cooler, however that is only when the memory controller is being virtually unused as any vram usage draws power away from the core. The only limiter to higher clocks appears to be the card's 30W power limit as the card sits at ~30W in HWInfo64 under load

Despite the solid core max, the card only comes close to 2000mhz in non-vram intensive loads as any memory usage siphons power away from the core and slows the core down to ~1200-1700mhz depending on the level of memory load.

Although this is the stronger GDDR5 version, the 2gb of vram is quite limiting in most games. Due to this, the card's memory controller is under near constant >70% load in any sort of vram intensive application (games) as the card is constantly swapping out textures from the vram. This constant memory controller load siphons power away the core which significantly hinders performance given the minimal power budget of this card.

HWBot records show the card hitting ~2700mhz (on LN2, not aiming for this), which I assume is impossible on the card's 30W stock power budget, so somebody in the world must have figured out how to bypass the 30W power limit.

vBIOS modding is currently impossible on desktop pascal cards due to encryption, so I'm looking to do some sort of shunt mod, however none of the videos or forum posts online about shunt mods are geared towards cards that only draw power from PCI-E.

I've looked through BuildZoid's videos, but he also has not done any PCB breakdowns on cards which only draw power from PCI-E, nevermind a breakdown on the GT1030. Despite this, his videos have been massively helpful in explaining the basics of a shunt mod.

What I've gathered is that in every Graphics Card PCB, there is a resistor somewhere across the board (usually 10 ohm) which an IC measures the voltage drop across in order to calculate the current and thus the power draw. The shunt mod involves removing that resistor to trick the Power IC into thinking that less power is being drawn than actually is, thus allowing the card to draw more power than the threshold in the BIOS, 30W in this case.

Here are the front and back of my card's PCB:
Front:
Passive circuit component Circuit component Hardware programmer Electronic engineering Electronic component

Back:
Circuit component Electronic component Font Electronic engineering Hardware programmer

I've looked closely, and my best guess is that the blue resistor marked "10" (Pointed at by green arrow) is the 10 ohm resistor in question, due to the fact that it has a "10" printed on it, but also due to its proximity to the power pins of the PCI-E slot. Does anyone have any ideas as to whether this may be the correct resistor? I have a multimeter and if needed can check resistances across the board, its just that I have no idea what values I am looking for since there is no documentation for this card so I have nothing to report yet.

If this is the correct resistor, what would I do to trick the IC in this case? Based on the videos of higher-end cards online, it seems that every card is different so I don't know if I can just remove/short the 10 ohm resistor, or if I need to replace it with some lower/higher resistance resistor. Any help/information is greatly appreciated!
 

·
Registered
Joined
·
1,101 Posts
Hello all,

I have a GT1030 (GDDR5 version)
How much GPU usage %, you get when gaming.
Use MSI afterburner for the collection of statistics.
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #3 · (Edited)
I get 99% on the core and 95-100% video memory usage. The memory controller is also almost always at 70% or greater load due to having to constantly swap textures to and from the low amount of vram. Temperatures never exceed ~70C.

BTW I would recommend HWInfo64 for in game stats, it has much more info than MSI afterburner and it can also hook into RTSS. (Also lower CPU/memory overhead in my experience)
 

·
9 Cans of Ravioli
Joined
·
22,334 Posts
  • Rep+
Reactions: Falkentyne

·
Avid Something-or-Other
Joined
·
210 Posts
Shunt resistors are actually in the range of milliohms because they need to pass a lot of current with minimal voltage drop. Judging from this thread and others, I think the BIOS power limit is enforced through a different method than shunt resistors and the white SMD is actually a 10A fast blow fuse to minimize damage to the PCIe slot/mobo in case of a short. You can check by removing that SMD and cleaning the pads with solder wick to determine if there are multiple pads.
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #8 ·
wouldn't shunt mod a card without 6/8-pin power. people avoid the bottom shunt by the PCIe slot on cards with 6/8-pin power for a reason.

not that long ago AMD had to release a fix for RX 400 series drawing >75W from PCIe and causing problems.
I understand the inherent risk in a shunt mod for the PCI-E slot, however the cards stock power limit is 30W, so even if I doubled the power limit with a shunt mod I would still be well within the 75W PCI-E safety limit.
I know that GPUs still can spike above their power limit for a millisecond or two, so that would be my only concern about exceeding 75W, however I don’t plan to push the power usage much farther than 45-55W based on the cards current thermal performance, so I don’t anticipate that even a power usage spike would be dangerous here, as I would likely run into thermal issues far before any sort of Power overdraw event could occur.
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #9 ·
Shunt resistors are actually in the range of milliohms because they need to pass a lot of current with minimal voltage drop. Judging from this thread and others, I think the BIOS power limit is enforced through a different method than shunt resistors and the white SMD is actually a 10A fast blow fuse to minimize damage to the PCIe slot/mobo in case of a short. You can check by removing that SMD and cleaning the pads with solder wick to determine if there are multiple pads.
I have read that thread, however the PCB shown there has quite a few differences from my PCB. It looks to be a DDR4 variant, which is a few years newer than my GDDR5 (2017) variant.
Judging from that forum post, it doesn’t seem like anyone is quite sure what controls the power limit for these cards, except perhaps the people on HWbot who are running the gpu at 2700mhz as thats just not possible on 30W without maybe removing/disabling the vram.
Yeah I figured that 10 marked component wasn’t it as none of BuildZoids videos show any shunt resistors as white. I may reach out to him to ask
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #11 ·
I've gone through the PCB and marked up all the resistors on the board:
Front: One marked in Red is 1.8 ohms (Code = 1R8)
Passive circuit component Circuit component Hardware programmer Electronic engineering Microcontroller

Back: The one in green is 15ohm (Code = 150) and ones in blue are 2.2 ohm (Code = 2R2)
Circuit component Electronic component Font Hardware programmer Electronic engineering

Is there any way to test via a multimeter if these resistors are used to calculate power draw?
 

·
Registered
Joined
·
1,101 Posts
I get 99% on the core and 95-100% video memory usage. The memory controller is also almost always at 70% or greater load due to having to constantly swap textures to and from the low amount of vram. Temperatures never exceed ~70C.
You will not like the news, but shunt-mod this is not for you, +3.5% at FPS cannot help anyone, when GPU is all ready at 100%.
The solution to your problem, this is my GPU on your system, and I have no intention to share it.
I am aware that you are determined to push your luck, but its pointless, there is no benefit.
You might burn your motherboard ....
What game are you playing? What are the minimum VGA specifications ?
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #13 ·
You will not like the news, but shunt-mod this is not for you, +3.5% at FPS cannot help anyone, when GPU is all ready at 100%.
The solution to your problem, this is my GPU on your system, and I have no intention to share it.
I am aware that you are determined to push your luck, but its pointless, there is no benefit.
You might burn your motherboard ....
What game are you playing? What are the minimum VGA specifications ?
Let me give you some background about this project. This GT1030 system is a learning project rather than my main system. I’m not concerned with blowing my motherboard, not just because I do not plan on pushing the GPU that hard, but mainly because this is just a learning project, if I destroy a component then that is fine with me, this is an LGA775 system that isn’t worth as much as the GT1030 itself.

Where are you pulling the 3.5% figure from? If this power mod was successful then this card could easily see a 25-40% increase in core clock under load which would likely translate to a higher lift in performance. Even if it was just 3.5%, I would be fine with that, like I said this is just a learning project to understand how much value can be extracted out of these low power cards.

I have the same CPU as you in this system, a Q9650, and your GTX 1060 is way overkill for that processor as the Q9650 bottlenecks even a GT1030 in certain situations.

Not to be rude, but if I was you I would take advantage of this hyper-inflated GPU market we are in and sell your GTX 1060 and replace it with a GT1030. You would likely see near identical performance in many workloads outside of any sort of GPU-compute task.
 

·
Avid Something-or-Other
Joined
·
210 Posts
Even if you didn't physically desolder the component and check to see if it is connected to multiple pads like my first comment said, I believe a multimeter can identify if the component is a shunt resistor by probing out what it is connected to. You can then use your multimeter + the pinout and/or functional diagram of your GPU's switching controller to determine which passive components are hooked up to the current monitoring sense pins and compare your results to see if the suspected shunt resistor is connected to the current monitoring circuitry. Having a full datasheet of the controller and the boardview of your PCB makes this arduous task vastly easier; but reverse engineering of the PCB layout is technically possible if you have lots of time and patience.

Be warned that without the pinout/functional diagram telling you what pin you are probing, you will need to start altering the characteristics of the passive components (attaching parallel resistances, replacing resistors with potentiometers for small value adjustments, etc) and looking for changes in voltage/current draw/card behaviour, etc to tell you how the component influences the switching controller. It goes without saying that physically modifying your card is always risky; but this haphazard reverse engineering of the switching controller is at extreme risk of killing something before you even figure out what stuff has to be modified for the current sense circuit to register lower than actual current draw!

In any case, what is the switching controller for your GPU? Posting a clearer picture of this component might help us identify what company makes the controller and potentially find a datasheet/pinout for it.
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #15 · (Edited)
Even if you didn't physically desolder the component and check to see if it is connected to multiple pads like my first comment said, I believe a multimeter can identify if the component is a shunt resistor by probing out what it is connected to. You can then use your multimeter + the pinout and/or functional diagram of your GPU's switching controller to determine which passive components are hooked up to the current monitoring sense pins and compare your results to see if the suspected shunt resistor is connected to the current monitoring circuitry. Having a full datasheet of the controller and the boardview of your PCB makes this arduous task vastly easier; but reverse engineering of the PCB layout is technically possible if you have lots of time and patience.

Be warned that without the pinout/functional diagram telling you what pin you are probing, you will need to start altering the characteristics of the passive components (attaching parallel resistances, replacing resistors with potentiometers for small value adjustments, etc) and looking for changes in voltage/current draw/card behaviour, etc to tell you how the component influences the switching controller. It goes without saying that physically modifying your card is always risky; but this haphazard reverse engineering of the switching controller is at extreme risk of killing something before you even figure out what stuff has to be modified for the current sense circuit to register lower than actual current draw!

In any case, what is the switching controller for your GPU? Posting a clearer picture of this component might help us identify what company makes the controller and potentially find a datasheet/pinout for it.
Thank you for the well thought out answer. Unfortunately the component is too small for any of my cameras to focus on, however this is what is written on the chip:
Top Line: vP16660 (Possibly uP16660)
Bottom Line: PGA69T

Judging from google search, PGA69T is the name of the socket that the chip is attached to, but I could be wrong.

A quick google search shows that uP16660 is part of a family of GPU power controllers beginning with uP1666. Here is the data sheet for a similarly named model:

A visual comparison shows that the two models at least have the same pin count: 20 with 5 on each side. That’s all I’ve gathered from quickly looking at the data sheet.

Edit: Upon closer inspection, it looks like the top line actually says uP1666Q, which matches up perfectly with the chip family in the datasheet.

Edit 2: Judging by this post on R/Overclocking, this chip runs the memory controller:
database/geforce20/2060 - overclocking (reddit.com)
If you CTRL-F the uP1666Q, every instance of this chip on a card is being used to control memory voltage, rather than regulate total power draw which I assume is what we are looking for here.
 

·
Registered
Joined
·
1,101 Posts
I have the same CPU as you in this system, a Q9650, and your GTX 1060 is way overkill for that processor as the Q9650 bottlenecks even a GT1030 in certain situations.
I have demolish this myth with my own bare hands.

GTX 1060 stays at my system as future-proof upgrade.
Good luck with your learning project.
Personally, I would giveaway a good working computer to an orphanage instead.
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #17 ·
I have demolish this myth with my own bare hands.

GTX 1060 stays at my system as future-proof upgrade.
Good luck with your learning project.
Personally, I would giveaway a good working computer to an orphanage instead.
Have a nice day!
 

·
Registered
Joined
·
26 Posts
Discussion Starter · #19 ·
But for real the only way they are hitting 2700 on that card is ln2 a zombie vrm card or epower.
I agree it’s impossible to get near that frequency without extreme cooling and an extreme power mod.
That being said I would still like to learn as much as possible here and attempt to discover a less extreme modification to increase the power limit of these cards.
 

·
Registered
Joined
·
2,830 Posts
I agree it’s impossible to get near that frequency without extreme cooling and an extreme power mod.
That being said I would still like to learn as much as possible here and attempt to discover a less extreme modification to increase the power limit of these cards.
There really isnt. You change one shunt your gonna have to change them all.
 
1 - 20 of 52 Posts
Top