I love projects like this. I’ve always swapped out the thermal compound in my laptops, because it does make a difference, and have been curious about tinkering with the thermal solutions.
I don’t have the money, or equipment, to try to weld or mill anything, so, really, the extent of any tinkerings I could do would be limited to try to strap one of my spare heatsinks (Thermalright Silver Arrow) to the die and let-er-rip. It’d be fun, but, impractical. Perhaps one day, when I retire a laptop and don’t need to carry it around anymore.
Going to use the two shims from the original sink. Got the GPU one down to ~0.25mm. Height difference between the two is 0.47mm now. Close enough to 0.5mm for me.
Tried the shims and plate with the 1mm thermal pads that turned up today, and it all fits nicely. Thermal pads compress properly over taller components. Just got to put it all together now.
I was doing some measurements on my own system and I noticed something interesting. Though you may already know this, I figured it was worthwhile to share. All of Zavar’s measurements lined up on my 9570 except for the top right spring-loaded screw, and the gem-shaped cutout next to it. On my motherboard, there is no top right screw hole, but instead a MOSFET (I think that’s what it is anyway) in its place. There’s also one less of the black rectangular component with a plus-sign in parentheses on it (sorry, I don’t know what it is) near the top-rightmost motherboard mount.
After looking into this further, I found that there are actually different motherboards in the 9570 based on whether it was made earlier or later in the year last year. The difference is clear between these two images.
Soldered it all together today. Managed to get the plate 3mm too low… Everything else is on well, though, so that’s nice. Can’t be bothered now, so I’ll fix that tomorrow, and try to line up the pipes properly with the fins.
Been having trouble with the plumbing blowtorch, because it just kept heating everything too quickly, and making the pipes swell up. Ended up soldering it with a kitchen torch meant for crème brûlée haha.
I know it looks terrible. Main thing is it’s all together.
Got it installed with thermal pads on everything. It’s looking good. Hit 95 with no fans on at 3.9GHz @ 65W, then the fans kicked on, and it dropped to high 80s 55W @ 3.7GHz on all cores. The fans didn’t go above the lowest setting. Going to try gaming for 15 minutes. I’ll reset the values in Throttlestop to give a better idea.
I’ve used 2mm spacers around the mounts to keep the whole thing level. Could take them off to increase pressure…
I’m going to message the solder manufacturer and ask what flux they use. There’s still some left in the gaps that I want out, and need to know what solvent to use.
War Thunder on the high preset at 1080p, CPU highest core temp was 78°C, with all other cores between that and 75°C. Had Vsync on, so GPU also wasn’t taxed fully. Max GPU temp was 68°C. Fans stayed on low again.
Overwatch on medium at 1080p capped to 60. CPU maxes at 86°C, same 5°C core difference. GPU maxed at 74°C (Throttles at that temperature). Fans stayed on medium.
The palm rest is staying cooler than it normally is, but the heatpipes are contacting the bottom panel, which is getting quite warm in the centre. I’ll try removing the copper sheet on the bottom panel, and replacing parts with the kapton tape where it’s touching tomorrow.
Running into our first potential firmware issue with rendering. Fans aren’t maxing out because it’s not thermal throttling constantly, so it’s sitting at 3.3-3.4GHz on all cores (8750H is spec’d for 3.9GHz on all cores). When the heatsink does get saturated, I expect it would then kick up the fans, which would cool it more and stop the throttling, but then it would drop the fans down again, and end up in a cycle. Unless that’s all bullshit, and increasing mounting pressure would fix it…
Holy smokes. I applaud this entire thread.
Awesome work dude. Most people would have just cleaned dust and maybe try to lower voltages and clocks and called it a day. @Zavar
If I can fund the research, I want to do basically this with my Asus G50VT-X5 once I have a Core 2 Q, more ram, and a 260m in it. I already beefed the thing 2 times and I think I’ll just need to do what you did here TBH. But here I’d use your unit as a puppet
But, part of my design makes my unit a bit taller and puts some heatsink into the bottom panel for the 9800GM in there. It gets as hot as the desktop chip easily.
Tried messing around with the pipe and fin placement, and got to the point where I’ve realised that it isn’t going to fit without changing up some parts of it. I need to make the plate and pipes as low profile as possible, so they don’t contact the bottom panel.
My first thought is to scrap the shims, and get the bottom of the plate milled out where the GPU die sits. Go back to the original idea of having the pipe’s path on the plate slightly milled out, too. This would drop the height of the whole thing by about 1.3mm.
The only problem then is the fins. I’ve emailed a company that sells a full copper version of the XPS 15’s heatsink (shoutout to htaige on Notebookreview forums for linking that). Hopefully I can buy just the fin stacks from them. If I can’t do that, I’ll take some tools to the ones I have, and make them fit.
Until then, I’ve put the stock heatsink back in, and bridged the VRMs to the bottom panel with some stacked thermal pads.
@SgtAwesomesauce, -138mV is the best undervolt I can do without it throwing errors up. I could up the ICC Max, but that’ll make temps shoot up, and I don’t think it’ll change the max boost duration.
@FaunCB, What I’ve done is really not difficult to do. You’ve just got to balance adding thermal mass with adding extra dissipation. Otherwise you’ll just end up with a heatsink that takes longer to heat up, but doesn’t perform any better over long periods.
Did a write up a couple of weeks ago. If anyone wants them, the files for the newest revision (made for CNCing) can be found at the bottom of the page.
Don’t judge the completely non-defined sketches in the SolidWorks model. You’re good as long as you don’t try modifying the sketches in any way.
Never got around to actually making the final version (due to a lack of funds, mainly). Been running with the VRMs padded to the bottom, which isn’t ideal.
Probably selling the laptop soon, so I don’t think I’ll actually make the rev 12, sorry. Saying that, I can guarantee that the final version 100% fits using the stock heatpipes and fins.
The latest BIOS (1.11.2) has changed the fan curve quite significantly. The fans only ramp up properly until the CPU’s in the low-mid 80s, which means that even changing the heatsink isn’t going to do that much to help CPU/GPU temps.
Dell seem to be trying to compete with Apple to see whose devices can thermal throttle first (at least Apple are undervolting their new MBPs themselves).
Thanks for the update, I think I’m gonna do it. I’ll be able to force the fans to ~5k rpm with HWinfo, between that and the ‘cool’ setting in Dell command power manager it should hopefully provide a more suitable fan curve.
It might be worth using a sacrificial toaster oven for soldering, rather than a blow torch. Less swollen pipes, and it should evaporate the flux more evenly.
Just in case people want to override this on their own Dell XPS or other laptop, you can use this software on Windows to do it, but it may require a bit of research and troubleshooting to get working.
