I removed the cooler (i know the entire like, liquid metal thing) so I opened up the heatsink while it’s running. It hit 103, but dropped down to 90, drawing 0.7W. I was able to shut windows down, so it “was working”
Did the whole cleanup, alcohol, etc. And now my framework won’t post, S4 Deassert Fail.
I want to believe it’s a small alcohol patch, but I did blast it with 100 psi compressed nitrogen, so it can also be a chunk of cooling pad.
I don’t think I cracked the die, I hope not.
Now we wait 10 billion years for it to check hardware (took me like 8 minute of waiting to get the blink code out idk what’s going on)
Well thats pretty bad and totally on your Side. Leave the Thermal Solution off and check the Liquid Metal around your Die. Possibly there came something through ther Conformal Coating. Check if any Sprinkles got out and on the Board. There could be a short. Maybe you gained an expensive Paperweight.
It couldnt be the Alcohol Pad i drenched my Board in IsoP and it didnt budge. Dange removing the Heatsink while its on without an IHS like on Desktop CPUs can totally destroy your Chip.
Okay today it is 1 Month since i modified my Heatsink with removing its Shim and installing a coppershim with a PTM7950 Sandwich (Thermalgrizzly Phase sheet)
With every Backgroundtask running after Boot Screen Recording and just Best Performance Power Profile here a Video of the CBR23MC Round + HWinfo
My Modification holds strong.
-Cleaning alcohol
-Razor Blade
-PTM7950
-Variation of Sandpaper (400-2000grit)
-Even Surface (white brick thing for me)
-Coppershim (20x20x0,8mm)
I pried off the Original Shim with the Razorblade.
Sanded and lapped the Vaporchamber to get rid of the residual Solder.
Deburred and lapped the new little Copper Shim.
(Clean everything in between with the ISO)
I made a paper stencil of the Die to easy cut the PTM for the Die Side. Put the PTM on the Die.
Layered the Coppershim on the PTM and put a precut PTM Sheet on the shim (the whole size of the Shim)
And screwed the Heatsink on Top.
Done, heatcycle is like the first 5-10 full Heat and Roomtemp Cooling Cycles.
== Performance
I didnt even remove the Board or anything it took like 20min for me.
After a CMOS reset it’s back in business.
Apparently, you don’t just un-power the board. You are to jab (carefully) the chassis intrusion switch 10 times. “You must press it slowly, so press 2 seconds. release, wait for the red blink on the Mainboard LEDs. repeat.”
And it just worked.
To get to here, you must have the cooling pad be in a liquid state. I accomplished this by removing the fan and playingarras.io. Then loosen and remove the heatsink. It come off without too much trouble. The CPU will spike to 100 degrees, but if the thermal protection is anything to go by, it’s fine.
Then, you should probably lift that plastic shield off first, then clean up the mess. I didn’t do that, so I contaminated the plastic shield. I have it sitting on some waxed paper (sticker backing stuff)
It’s actually not bad. It “never dries out”, and I can stand to that. I have found this stuff on 22-year-old Xeons (Supermicro X5DA8 to be exact) in derelict workstations. As fresh as day 1.
The main reason is because this is from our electronics lab, so it’s on the house.
I just smeared this onto the thing, I didn’t bother removing the liquid metal interface thing (the thing with all the dots).
Interesting spread pattern. I mean the mounting pressure never have to be perfect. But this is actually … not amazing. I guess it works, I can pull 45W consistent. But if any of this thread goes by (almost 500! by this point), hm.
I think Framework got scammed. The cooling sheet claims “72W/mK”, which is an order of magnitude higher than any paste. Results show … not that.
If you are having terrible thermals, maybe try re-pasting. But maybe talk to customer support first.
My Lord you Lucky B* I am happy its not a glorified Paperweight.
I noticed the Same Problem with the Spread Pattern, while doing my PTM Experiments. I think thats one of the Major Factors contribuiting to the Run off. But hey all in all better than before thats a Win. Now get get you some Phasesheet or do the mod i posted above You are too far in to turn around.
I have hit 100C on all 3 boards I have owned since day 1. I have also hit 16k or higher in Cinebench R23 on all 3 boards on the first day of ownership. I would see 50W or more on each new board for the first few days. My current board is at about 13500 while seeing a 37W package power.
I agree that FW got scammed here. The heatsink vendor (I believe this is Cooler Master?) has a problem with their design and needs to issue a recall. But FW is not innocent either. They should have been running longer tests to see how the machines held up long term. From what I have gathered in this thread, it would have been quite easy for them to identify this issue and make the vendor correct it.
I use this machine for a several things.
I do play games on it. I played through Horizon Zero Dawn and saw core 4 hit 100C while gaming. I am currently CPU limited in Cyberpunk 2077 due to core 4 hitting 100C and limiting the package power to 38-39W. To be clear, I did not get a dGPU, I am using an RTX 3080 in a Razer Core X Chroma to drive a 3440x1440@120Hz display.
I also use this machine for development. I don’t really see many issues here since that is more of a burst load when I compile any code.
I also use it for video encoding, particularly HEVC compression. I am aware that will use the AVX512 instructions which can be very hard on the cores. However, the cooling solution needs to take that into account and deal with it. Don’t give me a CPU with AVX512 and not provide adequate cooling for that workload.
My 4th board will arrive this week and I intend to torture test it in ways FW should have already done.
they are cool. but it’s expensive as damn, Why do that when I can have industrial-grade “heat sink compound” for free. Plus, I know for a fact it never dries out.
It look hilarious. Some 20-year old stuff in like a beat-up toothpaste tube, with a literal missing lid. But it goes.
Probably not even that expensive, this is measured in fluid ounces.
Well 1 Phase Sheet costs about 18 Bucks. It is expensive, but worth it as it is just about 2-5% off of Liquid Metal and stays Maintenance free.
The Trade off for any Paste is, that it is silicone based and it will bleed off its liquid components and it most definitly will degrade.
DIRECTIONS: For use on Radio, FM, TV, Automobiles, Aircraft and Tractors. Increases the efficiency of electronic devices, especially semi-conductors, by lowering overall temperature by means of heat to heat sink or chassis. High heat conductive metallic oxide sus-pended in silicone dissipate ‘hot spots’. Temperature range -65F to +400F (-64C to +205C). Will not dry or harden. CAUTION: Keep out of reach of children. Do not take internally. May be harmful if swallowed. CONTAINS: Silicone
I wonder when is my tube made, it doesn’t have a concept of a computer. It’s at least 10 years old, found a 2015 website selling it.
Says .0018 CAL/SEC cm K. Do math, turns out about 7.5W/mk. Checks out.
But not having to worry about anything at all is pretty cool. And PTM 7950 is non-conductive, so we don’t even need this crazy, like, baffling and stuff.
For testing anything is great
The PTM is 8,5w/mK, That sounds not great, but its working best under pressure and heat. It starts at 250ym and with every heatcycle its squezing out. Leaving at best a bondlinr of 15ym meaning 0,015mm. After a lot of cycles it is not getting liguified anymore and stays at its lowest bondline thickness. Thats alot lower than with normal Thermalpastes.
PTM is never the most bestest. But it’s comparable to regular paste, and it actually require absolutely no maintenance, and, in fact, is not conductive. These alone make it a solid choice for pretty much everything.
Still waiting for this to become available. I am hoping it is redone with better quality materials. However, I am not seeing anything in the description saying “new version or improved design etc…”
Part of the problem is not many are even aware they have this is issue. Only awareness will generate enough noise for Framework to do something about it. It’s going to be a costly fix if there’s even to be one.
I don’t want them to go bankrupt, I don’t want to “press for refund/free replacement part” but they should do something about it.
A well written guide on how to re-paste the CPU, for instance. Goes a long way.
As I have demonstrated even with really cheap (and not super high-end) paste, you can get okay results. Certainly better than 35W, and probably a lot more stable. It doesn’t exactly flow easy.
I wish Framework would actually say somewhere what minimum Cinebench R23 score should be expected (under a long list of expectations, eg the 7940HS, a particular Windows ISO with no additional software whatsoever installed which might run background tasks, certain ambient temperature, etc.).
I paid an extra $200 for what I knew was marginal performance increase, but what’s $1800 vs $2000 anyway so I yolo’d it, and I support Framework’s mission.
But then I look at NotebookCheck and see they got 16519, and a minimum of 16352, and meanwhile on my 2nd board I’m getting max 15737pts (albeit with some software installed, no longer clean Windows). And I’m just so tired of this, since they took my computer to a service center for 12 freaking weeks to resolve my macropad not working.
I am happy its not a glorified Paperweight.
You are too far in to turn around. 