Shunt modding

Looking to start my shunt modding journey, I have been browsing online (mostly youtube) I can't find a good tutorial that mentions what I'll need and how to get started.

can someone please advice me on this. I want to mess around with a broken board before messing with my mxm gpu

 

@seanwee is your guy

 

What do you want to shunt mod?

 

rtx 2060 mxm.

only picture I've got of it for now, taken quite a while ago.

on a side note I knocked 2 components off accidently around 2-3months ago while taking out a damaged screw, should I try to fix it or just ignore it? it has been working perfectly file without any issues

 

The shunt resistors are on the back on the corner, you use a small amount of liquid metal (the less the better as it does corrode the solder) to reduce the resistance.

Those seem to be filter caps, you should replace them if you can.

 

do I use LM to remove the current resistors or to add them?

also what do filter caps do the filter caps? i have no idea how to go about finding a replacement or replacing it. OEM said they don't have the PCB map/layout

 

DO NOT USE LM FOR SHUNT MODDING IN LAPTOPS

The Lm can shift and short out components potentially killing the whole laptop.

For your gpu I wouldn't recommend shunt modding. The vrms in mxm gpus are built cheaply and the stock power limits are already very close to the limit of what they can handle.

 

ah that sucks. I was really hoping to run them at, at least 115w. flashed them from 80w to 90w with alienwares bios but it really irks me that they're so underpowered.
btw what are you running your rtx 2080 max q at?

 

The power limit is completely removed but I reckon it hits 140w under heavy games with RT in metro Exodus. I'm held back by the laptop total power limit enforced by msi which is 230w even though I'm using a 330w AC adapter.

 

wdym completely removed? won't it voltage underclock if there's no limit?

 

What do you mean voltage underclock?

Nvidia gpus operate on a set voltage/frequency curve. Overclocking/undervolting the gpu raises the curve so the gpu clock will be higher at a given voltage.

 

what happens if I bypass the limits of the VRMs? I've been thinking about this since I realised that I got cheated by clevo with only 80w rtx 2060 in my machine that has a 330w adapter.
its I really want to shunt mod or rather I find it hard to give up on it at this point

 

I was trying to refer to 2D clock speeds

 

They will either
1. Shutdown due to thermal/current limits being hit or exceeded
2. Blow up

Realistically its around the 300w mark which is as good as unlimited for my gpu.

2D clocks only happens if you do a direct connection bypassing the shunt resistors.

 

If you are using liquid metal enough that it can move, you are using too much, especially for a shunt mod.

The way I put it on leaves a skin of liquid metal on the card, it does not move even if you shake the card about

 

are cheap vrms limited to only 2060's for mxm modules? I have seen some posts of people shunting rtx 2070s and some flashing rtx 2080s from 150w to 200w. (I think this was by bios flash)

I can't really have this laptop blowing up on me really disappointed

 

They all use the same chips, it's the number of phases and draw draw of the actual chip itself. The 30xx series is a little better provisioned than the 20xx series for draw of core vs number of phases.

 

I don't understand your explanation, from what I understand it reads every board can run 150w since that is the baseline for rtx 2080 but that doesn't sound right.
I'm assuming there is a baseline for the actual chip but offsetting that relies on the quality of the chip. if they're all using the same chip then each chip should support 150w?
what do you mean by the number of phases?

 

A single chip can't supply the power needs of a GPU, so several of these are run in parallel to supply the chip. The basic arrangement is to have each chip paired with an inductor and capacitors to smooth power delivery. These form a phase and the exact provisioning of these depends on the data sheet of the chip and the design constraints (cost, space, efficiency etc).

The 2060, 2070 and 2080 could all have the same basic design of phase but have different numbers of them, say 3 for the 2060, 5 for the 2070 and 7 for the 2080. Due to the margin left in the design they could all end up having the same amount of headroom.

But the exact choice of chip and headroom to design in is down to the manufacturer and you would use that and the knowledge of the power target to make any adjustments.

Each chip will have an optimal point of power delivery and a max point and it's finding the range between these that's safe that's the goal of power modding.

 

So for instance I have 8 phases on the board, a couple will be used for vmem and then the rest are used for different parts of the core, so 6 phases. I want to say have a total power draw of 175W, 25W of that will go to the memory and the two phases so put that aside.

150W/6 = 25A per phase at full pelt assuming a 1v core voltage (this is an assumption, look at your clock/frequency curve to get a better figure). I can expect looking at the data sheet I can calculate the heat output of that to be about 3.5W per chip. With a large heatsink with the aid of watercooling I can cope with that level of heat and should be able to keep it in spec.



That's the chip that every 20xx and 30xx series uses, notice the exponential increase in heat as the output current goes up (the input voltage is 19v).

This is also an output of 1.8V, our chips are more like 1V so these numbers are on the optimistic side a bit.

 

So for instance if I have 5 phases with 1 phase supplying 25w but the manufacturers tdp for the card is 100w, does that mean I have 25w headroom to power mod?
How would I find out this information on my board out of curiousity

 

I have high clocked GDDR6 and more chips, your mem is probably closer to 15w on a single phase.

You need to look at your TDP, a memory power calculator, the number of phases and the graph above and estimate your cooling. Of all of them the 2060 does have the least headroom from what I remember and these cards do aggressively turbo so just take care.

 

are these my vrms? refer to attached picture, tried to identify them online but people only refer to the phases on desktop modules so I'm coming up blank. guessed those simply because of the thermal pad placement. does that mean I have 6 phases? there's also 6 of that greyish coloured modules. I think the grey ones are the VRMs

I don't really understand the memory power calculation stuff, nothing came up online other than calculating power consumption for PC parts
don't think cooling will be in issue in this DTR laptop p775tm-g

I am still a bit confused about this phases stuff

wouldn't 300w be far above the limit of his phases added together, or in this case does the GPU draw only what it needs? wouldn't a voltage surge fry the whole system?

 

You are 4 + 1, the 2060 and 2070 have the mem phases top right, 2060 uses one and the 2070 uses two.

As I said the 20xx and 30xx use the same VRMs.

So the 2080 uses 11 and I think two of those are vram (might be 3 due to the different memory type). Say it is 3 and it's hungry stuff so it eats 40w.

We take that out and we have 8 30 amp VRMs (a phase is a VRM, an inductor and some capacitors at a minimum) the VRMs are the chip that gets hot. So we know as a max that 30*8 = 240w so a peak of say 280w but that's assuming sufficient cooling for the 5w per VRM which is a LOT. A safe max would be more like 180w + 40w ram so a total of 220w say or roughly 22 amps per phase for air. Which is a nice uplift over stock.

 

if that's the case since the rtx 2080 is stock 150w aren't a lot of these phases going to be made redundant or are they there for stability/heat spreading?

 

so with referce to the graph before, the more you push a phase to its max tdp the higher the temps, only way to negate this is adding more phases?

 

See how the efficiency curves? If you have more VRMs you can push them to be lower down their load curve where they are more efficient and at load you will get lower overall power draw.

And yes the only way is to either change chip or change the number of phases. Or one of the other conditions like input voltage.

Clevo are somewhat limited by feeding 19v into the card and stepping it down from there, where as most VRMs expect 12V (as that's the desktop standard) but it saves having to step the 19v down to 12v in between.

 

@Meaker@Sager explained it eloquently. Basically more phases = higher efficiency and higher current handling capability.

The vrm current rating represents the peak current draw it can handle, sustained current draw is lower by around 5-10A depending on the vrm’s cooling

I shunt modded mine as such because I know I can carefully monitor the power draw and I set a limit in msi afterburner rather than letting it run free.

the only way you could potentially increase your card’s current handling capability would be to solder more vrms onto the empty vrm pads on the pcb but that would be near impossible to do without an smd soldering station. And you’ll need to find a source for the vrm mosfets and inductors.

 

You also need to tell the controller to activate more phases so there is a citcuit there you modify.

Not as simple as the 980M that had 4 phases that could then run with 2 VRMs per phase instead of 1. That you just got them soldered on like I did and it would take a beating at that point and just carry on.

 

Thanks for taking your time explaining it to me @Meaker@Sager @seanwee I understand what seanwee meant when he didn't recommend shunting the 2060.
soldering on new vrms are definitely too extreme for me guess my shunting adventure ends here, I learned a lot though.

btw random question - is it better to leave CPU on adaptive or fixed power when undervolting

 

I don't know how modern CPUs are like, but I get better stability + temperatures on adaptive on my desktop Intel 6900 series CPU. Fixed gave me higher temperatures and I couldn't do a stable overclock as high as on adaptive. By stable, I mean, I ran my desktop for 24 hours with the overclock and it passed all tests successfully.

 

Fixed is for high end overclocking, if you want a quiet idle you want adaptive.

 

just leave it at adaptive

 

thanks will do, need those temps down