2019 Mbp VRM mod power draw results and comparison

[gardenmash]

If it's true that the VRM mod allows you to sustain higher wattage (and thus performance), it would be interesting to collect data points. Namely, the maximum combined power draw that one can sustain during a heavy, mixed load that asks for 45W+ from both each of the CPU and GPU.

Unmodded as well as post-VRM-mod results would be useful for comparison, please post either. Results from my machine are below.

I have performed several iterations of the VRM thermal pad mod. The first started with only partial contact to the black thermal insulation tape of the bottom case, and the most recent one has up to four layers of pads for better contact with the bare (tape-stripped) bottom case.

Test: Test duration is 3.8 minutes
Test thoughts: Machine is prewarmed. This test run and resulting data, though short, is a good representation of the usual throttling I see when gaming.

Machine: 2020 Macbook Pro 16, i9 2.3GHz, 5600M, 32GB RAM
Hardware setup: Clamshell, resting on a laptop cooler with a fan (pushes air up to bottom of case)
Software setup: Bootcamp, CS GO [I used CS GO because it starts out drawing 45W GPU and 40W+ CPU from cold start before throttling]
Ambient temps: Lowest at 31C on battery sensor 1

Mod info: Fujipoly 17W/mK 1.5mm and .5mm. Only padded VRM parts, and not CPU or GPU. Bought a second bottom case on eBay for $80 to preserve warranty after removing tape.

Conditions: Machine warmed up until power draw throttling seesaw occurs, which is after 10 minutes watching a CS GO replay at 5K res + highish settings. Then, machine is idled for a few minutes to configure and set up logging, and tabbed back to the game for full active load for the start of the log data.

Software: Not running any modifications like QuickCPU etc.
Logging software: XTU, GPU-Z, (Afterburner- data not used)

Result
----
Averages taken from ~229 one-second-interval data points (229 seconds = 3.8 minutes)
Average CPU draw (W): 18.45946378
Average GPU draw (W): 45.71491228
Combined average CPU+GPU draw (W): 64.17437606

Analysis
----
- During the throttled period, combined power draw is around 57-58W. The up periods seem to average around 70W.
- Up periods last 4-6 seconds, down periods are 8-16 seconds.

Thoughts
----
Even with my best attempts to perform the mod, this machine still experiences throttling such that it is unable to sustain 70W. In CS GO (at 5K), this means dipping to an unacceptable 50 fps at 58W versus the 60+ fps at 70W. While I believe there is an improvement in performance (more on this later), this mod is not sufficient to sustain 70W+. As 70W seems to be the stable point where the GPU maxes its load at 45W and the CPU boosts to an acceptable 4GHz+ at 25-35W, this mod by itself is not a definitive solution to fixing performance throttling.

If you believe me when I say I can run games at higher settings stably where I was unable to before, and considering that the combined average draw is now at ~64W versus the 57W (IIRC) max that I was limited to before, then there is still a decent improvement to be had with this mod.

How to test
----
There are two ways, the easiest is to install MSI Afterburner [1] and enable on-screen display to show CPU and GPU temps and watts [2]. The other is to use Intel XTU and GPU-Z file logging but this is involved and requires data massaging and Excel.

[1] https://www.msi.com/page/afterburner
[2] Seems like a decent tutorial

Please post your sustained combined power draw below so you can determine the max performance achievable by your Macbook!

 

[993]

Hello! I enjoyed reading your experience with thermal throttling on this laptop. I had a similar experience, also with the Pro 5600M. It was intended as an all-in-one desktop replacement for both Mac OS for content creation/general productivity, and Windows for VR and gaming. But I don't think it was made for the latter usecase; there's very little thermal headroom for that, and more often than not sustained loads causes the whole upper case (including the VRMs) to get heatsoaked. My unit in its vanilla state has shut itself off a few times during long gaming sessions at sustained temperatures, at slightly over 80 °C (I believe the GPU and CPU was heavily throttled at the time as well). I think some users with the 5600M report no throttling, some users also report that they experience no throttling after padding the VRMs. But I think that it's more probable that everyone just uses their unit differently, and will find that pushing their machine in more intensive workloads will cause it to throttle, like with any other Pro 5600M model.

When I decided to look into modifying this computer, my goal was to reduce temperatures in order to enable those long sustained sessions. So while I had been inspired by the VRM mod, I took it a step further and drew up plans to add thermal pads to the CPU and GPU as well. This would dissipate even more heat to the bottom case, so I wanted an aluminum heatsink and a fan in the designs. A few weeks later, I had DIY'ed a solution that mounted a large aluminum heatsink and a USB-powered Noctua Fan underneath those areas. This gave me better temperatures; I don't have graphs like the OP, but under sustained gaming loads for many hours (~45W - 50W GPU, ~15W - 20W CPU, so maybe ~60W - 70W combined), my temperatures seem to settle at around 70°C - 76 °C, and the area above the touch bar while hot, isn't frighteningly so anymore.

In terms of software, I use Throttlestop + Task Scheduler to disable Turbo boost + BDPROCHOT and MorePowerTool along with Blue edition drivers (either the default ones on AMD's website or from bootcampdrivers, doesn't matter). I made these choices so that the units would clock more consistently and predictably - Turbo boost has an erratic CPU power draw which is problematic and BDPROCHOT throttles both CPU and GPU even in the low 60s for temperatures. If I enable Turbo Boost, the temperatures on the GPU would increase as well, as both processing units are on the same heat pipe, or maybe just very close together especially post-mod. Blue edition drivers while compatible with MorePowerTool for tuning, poses an inconvenience since many games don't work with those drivers, and you don't get access to as many settings as the Red edition drivers - I have to uninstall and reinstall drivers frequently in order to play the games I want.

In response to the OP's call, I've tried to draw even more power on an earlier iteration of my cooling project, and while I could draw upwards of maybe ~85W (with a heavy game, Prime95, and Furmark running simultaneously), my unit would get gradually hotter and hotter, without reaching any sort of equilibrium which was worrying. If I recall correctly, at around the half-hour mark, I was climbing from the high 80s to the low 90s in temperature. I don't want to do that again and risk hardware damage (or even failure) or Applecare. I think this laptop was $4000+ USD, and all the materials going of my cooling project weren't cheap either. In total, I think I spent nearly $250 USD on all the materials I needed just to get my temperatures under control for my intended usecase (the iFixit kit, thermal pads and other thermal interfaces, aluminum heatsinks, Noctua fans and USB adapters, etc.) The OP also decided to buy an extra bottom plate from ebay (in my case I didn't - I still have the original insulative layer). I don't think they're cheap or easy to get a hold of, so I think that goes to show how much time, effort, and money we're spending to go against the grain here.

In hindsight, reading all this must really be super ridiculous. But I think people who already have this laptop are just really keen on trying to make do with what they have, because they had some sort of usecase in their mind for this machine to fulfill. Perhaps they even knew it wouldn't be the perfect solution, but they were willing to accept that, because it's a Macbook, and can do both Mac OS and Windows Bootcamp. But like, Windows gaming is just not the intended usecase and I think we're just really really stubborn if we have to go to such lengths to get better performance. But in my case, in my stubbornness, I just had to do this silly project so that I can have this mediocre PCVR setup that I can fold up and throw in my bag. I guess it's alright.

For anyone who's entertaining the idea of trying to get more performance out of your Macbook, browsing this thread and other related threads for ideas, perhaps by way of the VRM mod or even engineering their own solution... I would strongly recommend against it. Thermal pads are not inexpensive; opening your computer and doing all those things is not without risk of losing warranty or much worse. Nobody really has owned this computer for longer than a year - certainly not with these modifications. How long this will last and whether or not we'll have to get the unit serviced in the future is all up in the air.

I think my lesson here seeing how the OP and I have spent this much time and effort on this is just, not just on the initial cost of the Pro 5600M laptop, but all these modifications and iterations as well is very telling... Instead I would say probably go for an eGPU or something else entirely. Though I haven't tried it myself, I think others have figured things out and have working setups and acceptable temperatures too, since it's only one of the two main processing units inside the chassis working. If you have to have a Macbook, the 16" is really nice and you can have all that with the 5300M/5500M. If you want additional performance, an eGPU which will net much greater performance with lower temperatures, likely with a fraction of the effort and at a lower cost. You would still retain all the benefits of the laptop itself. If an eGPU is too much trouble and you still haven't bought this laptop yet, I think a machine dedicated to gaming would be a wiser decision.

While gaming can be done, the screen while large and pretty isn't optimized for gaming, so if your framerate isn't locked at 60 or 30 FPS, it will appear stuttery (you can set it to some other value, but then you'd have to switch back and forth). The CPU and GPU processor while quite decent and unassuming consider that it's in a Macbook, isn't much when compared to a desktop or laptop, especially when you get to the higher-end like this machine is. Windows bootcamp has a whole host of problems - trackpad, keyboard, touch bar, speakers (aren't as punchy and sound more tinny), everything is a bit "off" on this OS. All in all, I think I did what I did because I don't know any better having used nothing but a Macbook since forever, and I don't know what it's like on the other side to own and use a desktop or a Windows laptop.

I think this reply turned out to be more of a review of the machine and my experience in modifying it, rather than about how much power draw it can handle. But I guess I found OP's post relatable, considering that modifying the laptop is something that many owners of the machine become interested in doing from browsing reddit and other communities where its owners congregate.

 

[gardenmash]

Good to hear your thoughts and experience. You are right that this is pretty ridiculous and way too involved for us when we just want something that works without throttling past 10 minutes.

The macbook in many other ways aside from this is extremely nice, from OS to parts, use experience and quality. And the allure of the 'dream' behind it as I see it, to be a mobile all-in-one powerhouse capable of Windows gaming, is pretty strong. Apple seems to have made an unfavorable tradeoff for sustained loads, and instead favored sleekness and quietness. However, they are also capped by some things like (1) thermodynamics (2) not able to use bottom case as a heatsink to avoid the risk of burning customers. If this flaw can be remedied with a lightweight fix and careful handling, it would take us much much closer to what I and maybe others considered as the idea behind this machine.

 

[993]

Yes, I absolutely agree with you. Having to be in the laptop market again in 2020, there weren't any options that really checked all the boxes. Especially so for the basics - it seems that there's always compromises to be had in the keyboard or the trackpad. Once you've heard the speakers (and also for me the convenience of a decent built-in microphone), it just becomes really hard to go back to anything else? But I guess this laptop doesn't check all the boxes either.

I know this is like a forum for tech enthusiasts where this may all seem really obvious. But coming into this purchase as my first machine that's like, actually capable of running modern programs... I didn't know about stuff like thermals and why you would want to consider a desktop or a dedicated laptop for gaming, because I didn't have any firsthand experience. The Youtube videos of early benchmarks and people using the 5500M to play VR, definitely pulled the wool over my eyes.

Anyways, that aside, I wanted to properly contribute to this thread, so I torture tested my modified unit with a combined torture test of Furmark + Prime95 + OBS recording at lossless (AMD encoder). I noticed that this kind of sustained load is much heavier than gaming, because those torture tests always request the maximum power available from the GPU and CPU. GPU seems to be able to clock at 50 watts fairly consistently without major thermal issues. The CPU on the other hand, seems to be the cause of thermals getting out of control very quickly.

***Note: These charts are not representative of what you would get out of the machine vanilla or with the VRM mod. My unit is modified with a custom cooling solution that I detailed in my earlier reply.

I started with 85W of sustained load for around 5 minutes. As you see around the 3 minute mark, power drops drastically. This is BDProchot kicking in, in an attempt to get temperatures under control. You'll observe that BDProchot is basically this cycle of alternating between drawing more power and kicking in when temperatures get too hot and/or for too long. You can lower the length of time the computer is throttling, as well as the frequency it does so by its direct correlation to temperature. Software can control things a tiny bit, but it doesn't really change anything drastic; only to reduce your performance in order to delay the onset of throttling. Modifying the hardware is the other option so many others have gone to but... opening the case is just something that doesn't sit too well with me, especially when people have spent so much on this machine. Whatever...

The slope down at the 5 minute mark is when I stopped the torture tests. This was after the 2nd throttle (when BDProchot kicks in). Anyways, if I disable BDProchot here, the temperatures would likely steadily rise past 90 and eventually shut itself off, probably like a few minutes later or so... let's not do that lol. BDProchot is saving my device here.

I'm running these tests back to back in order to find how much sustained wattage is safe for my cooling solution. The internals components do have a moment to cool down, but the case is still fairly warm between each test (the built-in temperatures don't indicate this well btw). Turbo boost was disabled for the remainder of the torture tests because it Prime95 is really crazy and there wasn't really anymore power or thermal headroom for it?

With the 70W test and BDProchot ON, my temperatures are 68°C - 76°C. This is actually not too bad temperature-wise, but the machine will throttle by BDProchot anyways. These are the green humps at the bottom of the graph, and it does cause GPU clocks and power to dip as well.

So after that test, I did 65W with BDProchot ON. Actually, I know this is probably safe, because I use this for VR and gaming, with BDProchot OFF so that throttling doesn't kick in. You can see here that BDProchot does start to throttle even though temperatures are not too bad I think...

As one last test, I ran 70W with BDProchot OFF, for twenty minutes straight, just to know how far I can push. I don't think this is safe for my system as temperature were around 80°C - 85°C. Those 5 watts make such a huge difference... I guess heat capacity/transfer is my bottleneck? Anyways, I wouldn't run this for any longer. But it does go to show how turning BDProchot OFF when your temperatures are under control was kind of like the final piece of the puzzle in my case.

I think my temperatures would probably be better if my room was more air conditioned to be cooler. A bottom case without the black insulative layer on the inside may help but, you know, I don't think that would improve it by much? Because when I look at other people who have peeled that layer off, it wasn't like a significant improvement. And that piece is like anodized CNC piece of aluminum I think, so it isn't at all cheap so... maybe not.

I guess this thread so far has like, two modified 5600Ms with pretty extensive hard data in graphs. It's kinda weird since we lack data on an original vanilla model. And this entire thing of people feeling so compelled to mod their Macbooks is so silly lol.

 

[GumaRodak]

Instead of these mods, why you just dont buy the desktop? With better cooling, made for ongoing load? I think the notebook is meant for ppls which needs work on the go or be mobile. But i see ppls attaching 2-3 monitors to it etc...for this purpose ofc the desktop or egpu help works best...

 

[gardenmash]

@993 Interesting results. By '70W test', do you mean that you limited to watts to 70W or is it a way of naming the test runs? By your third back-to-back test run (the one named 65W test), it seems that the average sustained combined power draw you experienced is similar to what I saw, ~65W.

I know this is like a forum for tech enthusiasts where this may all seem really obvious. But coming into this purchase as my first machine that's like, actually capable of running modern programs... I didn't know about stuff like thermals and why you would want to consider a desktop or a dedicated laptop for gaming, because I didn't have any firsthand experience. The Youtube videos of early benchmarks and people using the 5500M to play VR, definitely pulled the wool over my eyes.

Funny that you say this, because none of this stuff is obvious and I got pwned by this hard. My 2-year-old 2018 mbp was throttling hard while gaming, and since I wasn't aware of thermal throttling and didn't have the time to figure out what was going on, I thought the computer was just still not good enough and I had to upgrade. Now, only after having spent $4K+ on a machine that is blindingly obviously powerful enough to handle the 5+ year-old games that I play, now it's obvious that the problem isn't that the computer isn't strong enough - there is something else going on. I could probably have used the machine a few more years by doing this mod. And this reason, VRM throttling, is hidden from us as VRM sensors are disabled.

@GumaRodak I like: being able to use my mac anywhere, having less bulky 'stuff' mass a la minimalism, and gaming, and a laptop can meet all of these needs. But perhaps with some experimentation, risk and learning required.

 

[993]

@gardenmash Ah, so I should've clarified but the name of these tests assume that the GPU is running at a constant 50W. The remaining watts is the CPU, managed via Throttlestop. But on these tests, you can see that while I tried to control that throughout with software, there's actually a bit of variance and if you look at the actual power draw, the Radeon Pro 5600M GPU was pushed very hard by both Furmark and OBS, taking 55W - 58W instead of the 50W TDP. In the last 70W Test, the whole machine is actually drawing closer to 75W - 80W in the last 12 minutes. It would be more accurate to call that test like a 70W - 80W Test.

I would say that my machine can definitely sustain 70W under 80 °C, but I would be worried doing so and instead go for a sustain of no more than 65W instead. My temperatures would not exceed 75 °C, which I feel more comfortable with leaving on for extended periods of time.

The long term issue here in my opinion is not so much the power draw or even throttling, but the temperatures. If temperatures are too high, you can't really sensibly turn off BDProchot to stop the up and downcycles of throttling. If you do, it'll just keep increasing and shut itself down or forcibly throttle anyways.

Instead of these mods, why you just dont buy the desktop? With better cooling, made for ongoing load? I think the notebook is meant for ppls which needs work on the go or be mobile. But i see ppls attaching 2-3 monitors to it etc...for this purpose ofc the desktop or egpu help works best...

I think we both mentioned it, but for me especially it was kind of a mix between wanting that do-everything machine and ignorance and stubbornness as well. And also like, the Macbook 16" is actually like, at a price point that kind of makes sense if you want a high end Mac computer. The Mac Pro is just too much when it comes to price-performance ratio. And I think I'm beginning to see from @gardenmash as well that I'm not the only one who only wants a Macbook for everything and will do anything to not purchase and make room for a desktop tower. Also I think like people who use laptops a lot like to be able to work anywhere, whether that's the couch or up and about or over with family or in bed tucked in with a comforter.

@993 Funny that you say this, because none of this stuff is obvious and I got pwned by this hard. My 2-year-old 2018 mbp was throttling hard while gaming, and since I wasn't aware of thermal throttling and didn't have the time to figure out what was going on, I thought the computer was just still not good enough and I had to upgrade. Now, only after having spent $4K+ on a machine that is blindingly obviously powerful enough to handle the 5+ year-old games that I play, now it's obvious that the problem isn't that the computer isn't strong enough - there is something else going on. I could probably have used the machine a few more years by doing this mod. And this reason, VRM throttling, is hidden from us as VRM sensors are disabled.

Yea... I really relate to that. I came from a 2014 Retina" 13 Pro myself and I only really used it as like, a really nice Chromebook for typing I guess. I would play some games now and then, but almost nothing worked so I naturally thought it was because of the computer's poor performing hardware. Even within the first month of owning this new version of the 16", it would throttle on me but I didn't really know what was going on. So I'd just restart Windows and it'd be (sort of) fine again.

I think a lot of people might think as well because as you mentioned, the temperatures aren't actually all that bad on the surface judging only by the temperatures of the sensors, so you wouldn't think it was thermal throttling at first. But it definitely is and the whole upper case of the chassis definitely gets too hot, even if the sensors (hidden VRM sensors or not) don't seem to corroborate that. I'll mention this again but I had my computer shut down on me (before outfitting it with a cooling solution) at around 80 °C because things were simply getting too hot overall.

I think what I'm basically saying is (and don't get mad at me but), the VRM mod by itself is not really going to cool things down. To me, it's kind of like all those other solutions that people propose on these forums saying to run this driver or this program and that... it's like trying to solve a hardware problem with software all over again, not exactly but quote, "The VRMs are throttling, so attach them to the bottom plate with thermal pads to cool them down a little bit". In this case, I guess it does help the hardware a little bit but... I think it just kinda delays the inevitable by adding a bit more heat transfer to the bottom plate, which has a little more heat capacity, but not much. I think a real solution is to dissipate that heat more effectively, and that's why I chose to go with an aircooled heatsink design.

 

[GumaRodak]

Ok, got your points

 

[A MacBook lover]

Saving to add graphs later

 

[tkd4444]

Hey Guys,

So I bought the MBP 16" 8Core i9 2.3Ghz, 32GB, 5600M on December 2020 (I'm not sure which revision is it). I did the VRM (without removing the black tape from the bottom case) mod but haven't noticed any performance gains. I Also tried removing the bottom case while the MBP was resting on coolermaster active notebook cooler stand (big fan in the middle) so the VRM parts where actively cooled. In both cases I still see the same behavior. When the MBP is cool the GPU boosts and stays ~1000Mhz until the GPU reaches 90+°C this is usually after 10-20 minutes. When 90+°C on the GPU is reached it automatically throttles down to ~500Mhz and the GPU temp descend to ~70°C and will never boost again to ~1000Mhz. This was tested under BOOTCAMP + Windows 10 (up-to-date) + Turboboost disabled. I have no idea what else I should try to no throttle the GPU.

 

[gardenmash]

@tkd4444 The gains are probably there, just too small compared to your workload to be noticed.

With a high workload, your CPU is asking for 50+W constantly, and GPU asks for 50W. Your VRM can give out 75W+ initially but quickly gets exhausted/overheated and like a really bad treadmill runner that is trying to maintain a 4-min mile for hours, alternates between walking slow and sprinting at max wattage until it gets too exhausted. The Mbp is overworked and gets overheated, and needs to throttle to reduce heat and catch up on its heat debt.

So your workload (= power load = heat load) is ~100W, your unmodded VRM can maintain a constant max of ~50W, and your modded VRM can do ~60W. This mod may improve your sustainable power draw by 20%+, but you need to hit 167% in this workload to stop throttling. Since the modded VRM is still completely unable to provide the required power while dissipating its equivalent heat, you will see the same end-result result of throttling.

How can we fix this? Like fixing our relatively-weak and bad-strategy treadmill runner, we need to do two things together to be able to run continuously without throttling: (1) make requirements realistic (1) pace ourself.

(1) Instead of a 4-min mile, 8-min mile.
Instead of 5k-ultra settings at 100W, we do 4k-med or 2.5k-med at 60W. If you used 640x480, you would never throttle because your CPU+GPU workload is like 20W max. So find roughly the highest settings you are ok with that is also realistic for your system to handle. E.g. for one game I have, it's 5k-mostly-high at 120fps with CPU draw at 20W and GPU at 40W. For CS GO, it's like 5k-med-high with CPU 16W and GPU 45W or something.

(2) Instead of always trying to sprint at max speed that makes us overheat and stop, we need to make our pace something we can maintain forever.
Instead of CPU asking for 100W all the time, we limit it to exactly the max it needs to give us the fps+quality we want. E.g. In my experience, limiting the CPU to ~22W lets me sustain around 3.5ghz or so. This limiting of the CPU causes the GPU to use the rest of the power, and the GPU uses up the rest and usually ranges around 35-45W. I use QuickCPU for this.

The result of doing these two things together is that now, your workload is 60W, your system is using CPU=20W, GPU=40W stably. The system is not overworked and thus doesn't need to throttle anymore.

 

[tkd4444]

@gardenmash I got your point but the problem is that when it's throttling it's far from 100W even when it peaks. The GPU when at ~1000Mhz consumes ~45W and CPU is around ~15-20W and it does throttle anyways. (these numbers I'm getting from MSI afterburner) I was thinking to limit the max clock speed on the GPU somewhere on 800Mhz to never reach 90+°C and thus no throttle on the GPU. But until now I couldn't figure out a software to have control over this, I tried the Morepowertool but the 5600M is not supported.

 

[gardenmash]

Are you limiting your CPU watts?

What you mean by 'far from 100W', do you mean your current total power draw?

 

[tkd4444]

Are you limiting your CPU watts?

What you mean by 'far from 100W', do you mean your current total power draw?

Yes I'm. Turboboost is off.
Not sure about the total power draw, I guess I/O, Display and other components add up into this as well. But when added before throttling GPU+CPU power draws it's far from of 100W of power draw in total. I'd say max 70W.

 

[gardenmash]

By total power draw I mean CPU+GPU only. I think the CPU+GPU only ever reaches 100W+ temporarily when the CPU turbos which it could never sustain without liquid nitrogen.

Interesting that you can play with TurboBoost off, maybe your game has low CPU requirements? For me no games could handle higher res+fps on 2.3ghz. Maybe your GPU is being held back by the CPU?

If it's not that, I think no matter what if you're throttling, your workload is too high compared to the sustained watts your VRM supports. GPU 45W + CPU 17W = 62W, and from more usage I think 57-59W is what my modded system can actually sustain, with unmodded numbers being less.

 

[tkd4444]

@gardenmash I’m playing Red Dead Redemption 2. Doesnt look like the CPU would be the limiting factor. Nice 60fps with 2,3ghz right until GPU hits 90+°C then FPS drops.

 

[ZiPpYMaT]

Does anybody know what's below the area highlighted in red here? I found this video on YouTube with a good tutorial on the mod, but it seems he's the only one suggesting to add the pads also there.
Thanks

 

[Mr. Incandenza]

I've got the older model with the 5500M and my main issue is throttling while exporting video or large photo batches in lightroom... I just find the fans constantly winding up to be pretty annoying. I think if I wasn't hooked on Final Cut I'd have moved my workflow over to PC by now but... Anyway, my solution was to do the VRM mod, but also add a heat pad to the processor. I place the computer on a little hand towel upside down so the bottom is facing up and then I used another thermal pad on top of the internal heat pad above the processor, where I've put a spare Hyper 212 tower cooler.

You still end up with the power limitation at 90 watts of sustained load but it seems to hold that, plus the high CPU boost clocks when there isn't much GPU load. It also basically completely stops the fans from winding up past being audible. I just run it passively with great results but I imagine it'd be even cooler with some airflow to the heatsink.

This is all absurd and in retrospect I should have just bought an iMac, but if you're looking to eek more performance out of this thing and are desk-bound most of the time, it's a possibility. The main benefit is the increase in all-core CPU boost for me. In stock form mine would throttle down to 2.2ghz and sometimes below that on a Handbrake render. With the VRM pads, CPU pad, and tower cooler on the back it can fully sustain 3.2ghz all core in a Handbrake render. The difference for a long one can be drastic.