Delidding Zen 4 Ryzen, 7950x etc

[User5566]

Other than by famous YouTubers I haven't seen much info about delidding these cpus. I've seen the huge improvement in heat transfer and I'm tempted to do so.

However, I'm really not a fan of the "delidding by ripping the ihs by force" method used by "delid-die-mate" tool. It is really convenient and the tool is both available and not that expensive, but I've seen some pics online of dies ripped off the pcb when attempting a delid. It makes sense. These dies are soldered to the pcb. So by using sideways force we're essentially hoping the solder in top of the die fails before one on the bottom.

However, I've seen another delid method used by Gamers Nexus (shown very briefly in a youtube vid about self destructing am5 cpus). That method involves cutting the glue between the ihs and the pcb with dental floss, then suspending the cpu in a vice by the pcb with ihs pointing down and carefully heating the ihs with a heat gun while monitoring the temperature with a thermocouple. At some point the indium solder melts and the ihs falls off. This, to me seems like a much more gentle method. If the cpu dies are soldered to the pcb with normal lead free solder there is good 60C between the melting points so this should be safe.

Has anyone here delid, or have broken their zen4 cpu trying to delid? What is the delid-die-mate success rate? Are my concerns unfounded? Is the indium solder so soft there is little danger of ripping the dies off and people who put such pictures sound are full of sh***(because they put in the cpu wrong etc) ? What do you think?

 

[EarthDog]

I've delidded a few Intel CPUs (successfully) back in the day with the delid tool, never tried AMD AMx though. It's a jarring feeling the first time you hear the IHS 'pop' off, lol, let me tell ya. But I only did that to get better scores in benchmarking... not a daily driver-type PC.

Just remember, even when successful, this isn't a miracle worker. The yields (what 100-200 MHz?) may not be worth the risk.

 

[Woomack]

I can tell you it's not really worth it, unless you see the CPU run at lower-than-expected frequencies or at the edge of throttling.
I delidded Ryzen 7600, and compared to a regular one, I see lower temps and nothing else. I mean, there is no difference in OC or performance. Of course, it's a lower chip and has other limits.
Remember that to have the best results, you need to run it without IHS. Then you need an additional bracket for delidded CPU (Thermal Grizzly is widely available, 1-2 days shipping from Amazon or Caseking). Delidding tools cost a bit too (also Amazon or Caseking). Then you need a cooler that will fit = almost only selected AIOs or custom wc, or something like Noctua with their kit for delidded AM5 CPUs. Most cooler mounting kits will leave a 0.5-1mm gap between the cooler and CPU dies.
The solder is soft but still hard enough, so you can damage something if you make it wrong or put too much pressure. I recommend something plastic to remove it, so you won't scratch anything in case you make it wrong.

 

[User5566]

Interesting take that it's not worth it from someone who's done it. You're probably right.

My main motivation for considering it is that it seems such a nice cpu handicapped somewhat by 3mm of that copper and thick layer of solder on top. I think if I saw 100MHz improvement in top frequency during all core oc I'd be rather disappointed. 200mhz actually sounds pretty good, but that's the top boundary.

I already have the bare die mounting frame (a local guy was selling it for less than half price, I guess he decided against delidding). Thermal grizzly direct die waterblocks are available around here for 30% more than an equivalent "normal block", but I'd probably make my own shorter spring loaded standoffs instead.

My main reservation is the possibility of breaking the cpu. I sure would feel stupid if I broke it.

I have some hobby board repair experience, bga rework etc, so I'm not totally new to working with delicate chips. That's why if I decide to do it it will not be using the mechanical method. I saw videos where der8auer was using the thermal method with a much older ryzen and he said he broke 2 first ones. I wonder how did he brake them? In theory one should be able to unsolder/resolder that ihs at will... I'm guessing he knocked out some small parts under the ihs when cutting the glue?

 

[Woomack]

I saw that Thermal Grizzly direct-die blocks were all sold out, and delivery dates were not confirmed for the last 3 weeks. At least this is what the Caseking store shows, and they are a direct distributor of them. They also have mixed reviews from users, but I can't confirm that.
I was using Alphacool Core 1 block and Noctua NH-U12A with direct-die + offset mount kits. I couldn't find anything else that would fit, and I have multiple other coolers. Most new coolers (air coolers, custom liquid, AIOs) have "perfect fit" mounting kits with standoffs that often don't even use springs. Because of how the AM5 backplate is designed, it's not so easy to find an out-of-the-box cooler that will fit, and the easiest way is to use a custom liquid kit and or buy different screws from older blocks or make a trip to a metal store.
If you decide on a custom liquid loop then I recommend Alphacool Core 1 blocks as they're quite cheap (at least compared to some others) and they're about the best regarding performance. I was using one with screws from Eisblock XPX Pro ( I guess the same as for older Alphacool blocks). Screws are double thickness (check the photo), and you can adjust the pressure/height with springs and additional nuts, so it works well and doesn't look like a "ghetto mod". Core 1 block like this (there are various options with RGB or not) and I guess this mounting kit as there is no other one included with the block. I thought it was for AM5, but I just took it from the box and there was no description. Now I see in the XPX Pro manual there is only one AMD kit marked as AM4, but the compatibility says that the block works with AM5.

If I was still taking part in competitive benchmarking, then I would care about it, but for daily usage, it's not worth doing it ... especially if you want to sell the CPU in a while, as the next Ryzen generation will probably work on current AM5 motherboards. I'm still not saying no, but I wanted you to think some before delidding. I even had an idea to delid 7800X3D that I'm using in my gaming PC right now, but it runs below the max temp even on smaller coolers like DeepCool AN600 (it's surprisingly good cooler for the size).

I gave my 7950X to my brother a while ago as I was using it for tests only, and no one wanted to send any Ryzen-specific samples like motherboards or RAM kits with EXPO for a long time. Anyway, it was boosting fine up to ~5.8GHz, and with +200MHz PBO/curve it was going up to 5.95GHz on specific cores. Switching from 360 AIO to a custom loop with 2x360mm rads changed pretty much nothing. These CPUs don't overclock past 6GHz, so all I could achieve from delidding would be a longer boost frequency under a higher load. It would be good almost only for benchmarks, as games balance cores, and most of them don't fully use more than 6-8 threads. The best cores would still be boosted for longer to the maximum frequency.

Almost everything around the dies is protected on new Ryzens. Of course, there are thermal limits and how much mechanical pressure they can stand. The only problem is cleaning dies, as the solder is maybe not so hard, but you still have to put some pressure. I guess you could melt it, but if it was so easy/safe, then others would show videos of how to do it. Everything I saw was suggesting scratching off the solder with various tools.

 

[Zerileous]

I just would note, going from remote memory, that GN video was about diagnosing the failure on already dead units. I can't say for sure if they were de-lidding a working CPU or a broken one just to look at it (or a CPU they were planning to kill to see just how it blew). I would just recommend verifying the status and purpose of that CPU before you adopt that technique. Also as everyone said, you likely won't notice any meaningful difference.

 

[User5566]

After more searching online I decided my idea to melt the solder is actually more risky than using delid-die-mate. It seems the people that destroy their cpu and write about it online were:
A) not using delid-die-mate, but alternate methods
B) post and dissappear so no one can ask

Personally I'm not convinced either way. I've seen a thermal grizzly direct die water block as well as a heat spreader in Netherlands so they are available in Europe. I did order one, if I decide not to go ahead with it I'm pretty sure I can sell it locally pretty easily for the original price as I can see local sellers are putting lots of markup on this item and they are all "not in stock". That's the store: https://www.highflow.nl/watercoolin...-am5-mycro-direct-die-cpu-waterblock-rgb.html

I guess we'll find out if they really do have it, or is it one of these shops that has "everything" until you place an order.

At this stage I'm sort of convinced using delid-die-mate is a fairly safe method if one is not going to crank on the screw with maximum force on the first try. If I go for it and I do destroy the cpu you'll definitely read about it here...

@Woomack ,in these not so great reviews for the direct die thermal grizzly block, what was the main complaint? Lack of expected gain in temperature?

If so I think it may be caused by the fact 10C of gain is caused by removing the ihs metal, the other 10C is highly variable because it depends on the thickness of solder under the ihs. This comes out from simple calculation (copper conductivity over 200mm.sq over 220W equals about 10C not 20). So I suspect some cpus can gain more than others.

Or was the complaint something else?

 

[Woomack]

They have it from here https://www.caseking.de/en/thermal-...cpu-water-cooler-rgb-socket-am5-fsd8-060.html
"This item in currently not in stock and there are no information available about an expected delivery date by the manufacturer. This can delay the delivery to our customers by several months. We cannot guarantee any delivery date. If delivery dates are foreseeable, we will update the stock level display."
There is a chance that other stores have it from a partner/distribution channel as Caseking is a distributor with some unique brands on the list. Not everything can be listed in a retail online store.
Some some reviews are there and under the block without RGB. Some I saw around the web. Mainly about the temperature difference, but some can be because of the "user problem" as this is not a regular product for beginners.
It's like ~10-12°C because of delidding, ~2-3°C with offset mounting (not counting if the block covers dies properly), ~2-3°C when you use liquid metal over top regular TIM - the top TIM is already like 2°C better than anything more standard (I dislike liquid metal, but the temp difference isn't so big problem for me). Then is also a specific CPU factor. When you were unlucky with the core difference over 10°C then you may see 20°C+ difference. If you had nearly perfect CPU then you may expect closer to 10°C improvement.

Anyway, good luck with the block. I was thinking of getting one too, but eventually, I decided to move away from water cooling of my daily PC.

 

[jarablue]

I have a 7900x running quite cool with no delidding. I have 0 temp problems. I just play games and nothing is an issue. I don't get that whole scene.

 

[User5566]

I have a 7900x running quite cool with no delidding. I have 0 temp problems. I just play games and nothing is an issue. I don't get that whole scene.

And that's great too

At least in my case it's 10% utility and 90% just wanting to see how this nice cpu will run unencumbered without its extra heat resistance

The delid-die-mate has arrived...

I hope this goes smoothly and I'll not be buying another 7950x soon

Other parts haven't arrived yet so I can delay the final decision some more...

 

[User5566]

I couldn't wait too long... As they say in the English translation of delid-die-mate manual, the cpu has been beheaded But it works (so far).

It was a difficult decision, but eventually I found the shear strength of indium solder online and I saw it is 5 times weaker than normal solder so I thought its a safe method.

As mentioned, I haven't received the direct die waterblock yet, but I had both a direct die mounting frame and a normal mounting frame. So the plan was to delid, clean up, put some liquid metal on, put the lud back on, and use the normal mounting frame with my AIO until the direct die waterblock arrives.

But plans have changed as the situation developed and currently I'm running direct die (in the thermal grizzly mounting frame) and AIO on top of it with no heat spreader. Also I'm using a two point AM4 mount so definitely not an ideal situation, but it's only temporary. Just as it was said here before the AIO mount was under a mm too short, so I simply bent the metal "ears" slightly upwards and it was fine then.

The deliding itself was very easy. No great force was necessary. I used the short end of the hex key provided to turn it and after under 10 repetition the cpu was free.


However, it turns out removing the remainder of the indium solder is not soo easy. I used various plastic picks to get rid of as much as I could and eventually I used liquid metal to dissolve the rest.

But then I've managed to get tiny balls of liquid metal everywhere(even under the tape) and it didn't really dissolve all of the solder. So I cleaned the cpu very thoroughly. If I was doing this again I would coat the components on the pcb first, before removing the solder!
But I didn't think of it. Thankfully I have a good microscope and various short micro bristle brushes so I cleaned it up very well.

I then coated it with a conformal coat (silicone 422c, it can be removed with acetone and it can be soldered through).

The results so far? Well, I got 38600 Cinebench r23 points with stock pbo, no undervolt and ram at jedec 5200MT/s with the Aio. However, the contact is not ideal. One ccd runs at 80C, the other maxes out at 95C. Still I'm going to leave it like this until my direct die waterblock arrives.


Also, the first time I switched it on I must have used too little liquid metal (or too little mounting pressure), because my cpu was idling at 43C in the bios and the IO die hotspot was 60C just sitting in windows. I was so worried about cracking the die I had the aio way too loose. After adding some liquid metal and cranking down on the screws ever so slightly it seems fine.

So let me say first, I think it is a bit insane to use an AIO on top of bare die with a two hooks am4 mount, as it has no springs there is real risk of crushing the die. So I do not recommend this to anyone. As for deliding itself. So far ok.

 

[EarthDog]

You're a trip man... you worry about the risk in a proven method, but then have the chutzpah to run bare die without a die kit??? Damn, son......

 

[User5566]

You're a trip man... you worry about the risk in a proven method, but then have the chutzpah to run bare die without a die kit??? Damn, son......

Thanks

BTW, other than cracking it while putting the cooler on, are there any ongoing dangers with running bare die under a stock aio? (I've flattened it somewhat, so it is much flatter than from the factory).

Hopefully I'll be receiving my direct die waterblock soon so it's only going to be few days.

Other than gpu blocks I have all the other parts of my watercooling loop. I'll probably put it together just for the cpu initially while I'm waiting for the gpu blocks to arrive.

Also, as I was undecided if I'm going to delid I got a standard waterblock too (that I'll return or sell). It's an Alphacool xpx rgb. Looking at that block I can tell it is not super flat. I read somewhere normal blocks are not really flat but concave/convex slightly. Is that true that it is on purpose? And if so is it just because it's easier to make it this way or is it actually helping anything (in case of a non-delided cpu).

 

[Woomack]

If you install that frame for delidded CPUs then you won't crack it, as dies are barely above the frame.

New blocks are barely ever perfectly flat. It's because the cooling is slightly better, and TIM spreads better. The XPX is worse than Core 1 series. Core 1 is flat, I was using it with my delidded CPU. XPX has a good mounting kit for delidded CPUs, Core 1 not ...

 

[User5566]

Thanks. I've now received the Mycro direct die. I have to say the packaging and build quality sure do look premium. But first let me warn people about the dangers of indium metal in your liquid metal .
I had the tiniest bit of indium solder left on my cpu, you can see it as grey color on my die photo. Normally these dies are gold coated and yellow. This stuff has dissolved in LM and formed a sort of stalagmite? (like those things that grow from the ceiling in caves). Have a look at that. This is the surface of my aio after one day only.

You see these things sticking up? These were not there when I was putting it together yesterday! And they were right next to one of the ccds. Had I not put conformal coating on. It would've probably shorted something...

And here is the direct die Mycro rgb. Somehow it is half the size of xpx, but weights a lot more...


On the bottom where the die touches it is diamond milled. One can see the milling path, but there is no ridges etc that could be felt so it appears very flat. I might measure the flatness properly, but unfortunately due to it being recessed I couldn't make it flatter anyway. So I have count on it being sufficiently good.

 

[Woomack]

I dislike liquid metal in general. It reacts with various things, and sometimes the reaction takes a long time, but almost always, surfaces are uneven. I'm using more regular but higher series TIM, even with delidded CPUs. I know I lose a few degrees, but I don't care. Another thing is that I switch motherboards or cooling more often.

I don't know what XPX block you have, but I have pure brass+copper one, so Mycro can't be heavier

 

[User5566]

I dislike liquid metal in general. It reacts with various things, and sometimes the reaction takes a long time, but almost always, surfaces are uneven. I'm using more regular but higher series TIM, even with delidded CPUs. I know I lose a few degrees, but I don't care. Another thing is that I switch motherboards or cooling more often.

Me too. The worst of its properties is that it creates these tiny balls that are very difficult to pick up. Touch them with a piece of dry tissue - they smear. Touch it with tissue moist with alcohol they just move and are not absorbed

But I was curious if that 20C gain is true for an already good cpu and it is. For a "normal person" I think I would explore these "graphene" sheets and normal paste of course.

I don't know what XPX block you have, but I have pure brass+copper one, so Mycro can't be heavier

Perhaps it's the quality of packaging that influenced the way it felt in my hand

I have to say. So far this delidding proces has met all of my expectations. I have no gpu blocks yet so I'm currently running just the cpu on a pair of 480 and a 360 radiators (just the 480 has fans on as I need more fan splitters). Cinebench r23 runs at 71C (stock pbo, but undervolts on, jedec 5300MT ram) and scored 39390 points on my first run with hwinfo running

Of course once I add 700W of heat from two gpus it's not going to run this way but still it shows what 7950x is capable of on stock locked pbo, with undervolt.

Few more bits of info for anyone interested in details. After I pulled out the cpu I wanted to check flatness as I had one ccd significantly hotter than the other. I've done so by gently rubbing the die on a precision ground lumina bench stones and look at that. Dies are not flat. You can clearly see contact points on two dies(shiny patches) .


So I sanded a bit more and eventually I decided the below is good enough.

It is not ideal, but Its the best I can do with just precision ground bench stones. I have diamond nd laps etc I could use, but I thought this is good enough and it appears it is.

So, final pic. Is my current setup. It's not very pretty, sorry for mismatched rgb, I just pulled out both gpus in case my new custom loop leaks. Yes, my 480 is missing the top fan and I forgot to order a drain valve... So I'll have fun in few days once time to drain it comes.

Cables will get sleeves and it will look much better once my gpu blocks arrive. This is just temporary so I can use the direct die waterblock.



Edit: I forgot to mention, temperate difference between cores during Cinebench doest exceed 5C. Thats pretty nice too as all cores run at the same freq.

 

[Woomack]

One tip, as maybe you missed it in the store. You can order a short AM5 backplate for this block - https://www.caseking.de/en/thermal-grizzly-amd-am5-short-backplate-fsd8-041.html?flxSmartSuggest=1
Nothing really needed, but you won't have to keep the plastic AM5 cooler holders. My luck is that Thermal Grizzly backplates (regular and short) don't fit my ASRock motherboard, and I'm using air cooling on ASUS and Gigabyte.

 

[User5566]

One tip, as maybe you missed it in the store. You can order a short AM5 backplate for this block - https://www.caseking.de/en/thermal-grizzly-amd-am5-short-backplate-fsd8-041.html?flxSmartSuggest=1
Nothing really needed, but you won't have to keep the plastic AM5 cooler holders.

Thanks

My luck is that Thermal Grizzly backplates (regular and short) don't fit my ASRock motherboard, and I'm using air cooling on ASUS and Gigabyte.

I might go to ASRock with my next platform in few years time too.

As much as I'm happy with my current cpu results with custom loop watercooling I would've stayed with air too if not for the two GPUs that have to run their fans at full blast when running my stuff (and I still have to drop the max power down to 90% to avoid overcooking them). So the GPUs were the main motivation for this, the cpu is just an extra. Although I'd be very reluctant to go bare die with a monster air cooler. A heat spreader seems necessary negating some of the delidding improvement in thermal transfer.

My gpu blocks have been posted 2 days ago by the very slow "DHL EuroPacket" it'll probably be a week before they arrive... I'm not a fan of DHL in my area. They pay their couriers very little in comparison with other carriers and seem to have double the number of packages per courier. As a result there is constant employee rotation there and results are predictable (almost never deliver on time around here etc).

What do people think about running a custom loop on just distilled water with no biocide for up to about 10 days? Am I asking for trouble?

I was going to add the biocide once I filled my loop, but then I found out gpu blocks were posted on the same day so I'll be draining it in about a week anyway.

What do you think?

 

[Woomack]

I might go to ASRock with my next platform in few years time too.

Looking at the current offer ... you don't want to. Hard to say what they will release for the next-gen. I somehow hate my ASRock mobo, but it was a review sample, so at least I didn't pay for it. Right now, I'm not using it as I have no backplate, and the 3rd party ones don't fit. I keep it as a spare one ... or I will use it for tests with delidded CPU and modified mounting.

What do people think about running a custom loop on just distilled water with no biocide for up to about 10 days? Am I asking for trouble?

I was going to add the biocide once I filled my loop, but then I found out gpu blocks were posted on the same day so I'll be draining it in about a week anyway.

What do you think?

I used to run my loops with nothing but demineralized water from a local store. I usually had no problems for a year+. If there is no dust or other things in the loop and the water temperature isn't high, then it will be fine. This is why loop flushing kits exist, but you can use some homemade mix with a little vinegar or other things to clean the loop before you connect everything with graphics cards. There are guides around the web.