I do have a planned piece I need to finish up on the effects of GPU scaling on frame rates in games (average and 99th percentile). Overclocking, delidding, CPU frequency scaling, Memory scaling and iGPU scaling are the plan :)
This would be amazing! Would love to see the absolute peak performance (with gpu clock and ram speed and latency as priority over cpu speed). 3600 cl 14 on the ram would be ideal, or using the stilts 3466 llc setting for bdie.
A recent article on delidding Intel CPUs on [H]ardOCP suggested using clear nail polish to paint over SMD components to protect them against stray TIM.
Yeah that does work, it's the exact same method as der8auer uses, but I wouldn't want to test the theory out of purposely trying to short SMDs with liquid metal; that's the main issue other than damaging the die.
Given the cost of the CPU it's probably better to buy actual potting compound from somewhere like Element 14. This will also provide moisture protection.
What I would like to see is a kit to delid, get some indium solder, some spring loaded clamps, and a very precise baking temp... Solder that sucker on.
Interesting read for sure from the link you provided. I am not sure if it is good or bad but all I know is my Sandy Bridge [email protected] (soldered) has been running at this speed and has had the same temps since day one of me getting it way back then. I am not sure if there are micro cracks in the solder or not but it performs the same today as it did in 2012. Then again my temps 98% of the time are always at or around 45-63c under full load and idle temps 21-23c
I wanted to add that I also do not spend very time doing stress tests that bake the CPU to the point of cooking itself. I spend most of my time just messing about in Windows and Gaming which is far easier on the whole system than running stress tests 24/7 like some people tend to do.
That was liquid N2 testing that he was cracking them so fast under, air/water cooled will have much smaller thermal gradients and thus much slower rates of crack formation.
Properly-formulated and applied solder is irrelevant when it comes to "crack risk" unless you're using extreme cooling like nitrogen. If it's not properly-formulated, though, there can be a problem. That's what happened when Nvidia released a lot of mobile GPUs that failed. Apple was nice enough to replace the defective boards with defective boards. The industry doesn't do real customer service, like recalls.
In this chart, you show the stock/retail chip consuming 132W at 4.2GHz, but I think that is an error because the other charts don't show it capable of reaching 4.2GHz. Is this correct? https://images.anandtech.com/doci/12640/Delid%20Po...
Each test bed is different, the chip used in this piece was a second Ryzen 5 2400G sample that could achieve this. The original sample from the review Ian published couldn't attain; I have both the launch samples and a pair of newer retail samples from AMD which we used all 4 in the overclocking piece: https://www.anandtech.com/show/12542/overclocking-...
The CPU cooling plate wouldn't sit flush on top of the die. This is why companies such as EKWB make 'naked die' mounts for their water blocks. When removing the IHS completely, what you're essentially doing is adding clearance between the CPU cooler and the die.
To make it sit flush, you would need the cooler to screw closer in as to actually make contact. Too much pressure on the die however will crack it which = dead APU/CPU
There are a lot of Pentium 3/AMD Athlon/AMD Duron/AMD Athlon XP's that would disagree. In the 2k I messed around with a lot of Athlon XP's and where not carefull at all. Yes, I scratched and chipped a lot of chips while soldering pins and spots to unlock special features. But they tended to keep on working as normal.
Interesting article, thanks! The thing I don't get is why AMD, trying to gain market share from chipzilla, wouldn't spend the extra cents per chip and use the solder instead of the paste to begin with. Heck, they could even make it part of their marketing: they use gooey paste, we use quality solder. On another note, as they provide a stock cooler: are there any systems out there that do away with the lid entirely, and go straight to something like a vapor chamber approach in a sealed setup? Is this what they do for the Xbox X?
Hi Ian, I should have been more precise: it's exactly because AMD "does the right thing" and uses solder for its non-APU Ryzen 2000 chips that this use of paste sticks out. I wish they would do so also for their Ryzen APU lines. Would make anyone considering buying a 2200G or 2400G feel less like a second - class customer, and emphasize the perception that this is a premium product
"Would make anyone considering buying a 2200G or 2400G feel less like a second - class customer, and emphasize the perception that this is a premium product"
Intel hasn't had much trouble convincing people to spend a lot more on its second-class customer CPUs. It has been selling quite a lot of them since Sandy.
Well the article was exaggerating the problem. There nowhere was a package difference of 12c in their numbers, as far as I remember in the upper area of 4.1 GHz it was 9c (from 67 down to 58). Sure they got another 100MHz out of it, but that is playing the silicon lottery area. So in reality depending on your workload the difference between the liquid cooling method and the a stock cooling via the usual ways is 4-9 degrees. This is not worth it. They might have shaved off 2-7 degrees by going full solder, but we are talking about cheap apus here. This is not an Intel like situation where the difference often is in the 15-25c degrees range due to cheap stuff being used.
Why techsites and techtoubers don't understand, temps is not what is holding Ryzen or Ryzen APU's, is the way they designed the 1st iteration for a mobile oriented node.
OCing the cpu part of the APU is WORTHLESS when the base clock is already 3.6Ghz, without delidding, any 2400G will do 1500Mhz+ on the gpu (good ones 1700 @1.3v for the soc voltage, the max for safety)
The benefit for delidding the APU is basically lower temps, specially when using the stock HS or a low profile one.
Agree. I am eagerly awaiting a GPU-only overclock article for these, both stock and delidded. Also, second the interest in seeing that comparison with the stock cooler. My main interest in these (especially the 2400G) is the price/value aspect, and I don't think that many will want to shell out well over 100 Euros or dollars for a fancy cooler.
Great article. Was an interesting read but FYI: If you add the cost of all the parts it totals $247 dollars. For that price you could easily step up to a much faster Ryzen 7 1700 for the same price.
Yes and no; a key reason why I am even considering a 2400G is that, once you have a 1700 or 2700, you still need a dedicated GPU, and that means an extra $120 - $150 for an underwhelming dGPU (geforce 1030), all better ones being way overprized thanks to the crypto-craze. Also, the delidding tool doesn't wear out (buy once, delidd many), so my plan would be to borrow or rent it from someone for a fraction of the cost, or buy one and resell it for $10 less after I am done. That leaves the costs for the conductonaut, which won't break the bank. Overall, still a good value proposition to get through the cryptomining dGPU price craze.
The 2400 is suitable for many situations, but for my desktop 4 cores 8 threads was not enough... Usually you have the graphics card already. I am considering going for a 2400g however later this year for a homegrown nas. This littly processor provides enough power to run vms decently a thing which most commercial home nases are not able to pull off decently due to their underpowered processors.
Gavin mentioned that he didn't know of large vendor offering pre-delidded processors in the US. I believe that Silicon Lottery does this, though they may not offer this service for AMD processors. They also do binning of CPUs. From the little bit I've heard, people speak pretty highly of them.
"a direct soldering has the best performance but might not withstand mulitple years of cycling."
Bogus FUD. That concern only applies to liquid nitrogen or a similar extreme.
People have been recycling a slipshod PDF "article" with that concern-trolling nonsense for years now, pretending that it applies to normal cooling methods like water and air. Intel marketing at its finest. In an alternate marketing universe, AMD's use of solder is a mistake because it's more convenient to delid polymer TIM parts. And, in this fantasy zone, indium solder doesn't yield enough performance.
"standard TIM 'goop/paste' has sufficient longevity but lower performance"
Lower performance than what? There is a wide enough range among polymer TIM alone, without comparing the best polymer TIMs to the best indium solders. This is one of the reasons people weren't pleased when Intel marketed Devil's Canyon on the basis of a new-and-improved polymer TIM but an enthusiast website's testing found that not only did it have the identical low performance of its predecessor, it was outperformed by a rather inexpensive 3rd-party polymer TIM. The US government exposed Arctic Silver 5 for having a paltry thermal transfer rate when compared with its company's claimed rate. The 3M polymer TIM, by contrast, performed exactly as advertised (much better than Arctic Silver 5).
"I could see delidding a Ryzen 2000 series APU viably applicable would be in a situation" where rich people have nothing better to do with their time and money.
Yeah, it does seem like a lot of effort and cost for a minimal reward. In fact, taking the results at face value -- the CPU was already operating at a reasonable temperature under load and the performance increase was insignificant given the risks. However, there's more to this effort that we haven't seen yet like the upcoming iGPU results. I'm curious about what's to come.
I would like to see fewer silly screen names on products though. It's hard to take a name like Der8auer with a glaringly obtuse number in it very seriously. When I was a teenager, I remember seeing children calling themselves "sk8ters" and I thought it was stupid back then. It looks even more ridiculous now that I'm getting older. Maybe someday there'll be another segment of the PC hardware market that forks off the LED and screen name festooned man-child branch that is understated yet still premium.
About relidding: You already reduce the clearance by removing the glue. Ok, so you need a naked-die-mount if you leave the heat spreader away, but why not go all the way when you already went this far? And it would eliminate the worry that the heat spreader is inadvertently moved in a bad way.
Maybe stuff for a future article? How are temperatures affected by naked-die-mounting relative to relidding?
Yeah I was wondering about that as well. It used to be the norm from factory. I’m sure it’s common knowledge among enthusiasts who do this sort of thing regularly.
The included Ryzen mounting system uses springs, which have a spring rate. If you remove the spreader, you change the CPU stack height, and lower the clamping force. So you have to somehow modify the mount, with heavier springs or shims behind the springs, or install a shim to increase the CPU stack height again.
Other mounting systems have their own issues, which may be easier to modify.
I’m not convinced. I’d be more willing to believe that maybe it causes the heat sink to not clear a motherboard or socket component. There’s a decent amount of pressure on that heatsink clip and I doubt it’s going to notice such a small variance.
Also, given the complexity of this mod it seems strange that a spring loading issue would stop you from getting the lid out of the way. If that were a problem, after all the trouble of delidding and using special paste surely they’d be able to figure out how to slide a toothpick under the spring clip.
I remember all the old Athlons (e.g. the venerable “Barton” core) basically came delidded in the retail box and your heat sink went straight on top. I can understand why AMD and Intel started putting lids on, but it seems weird to me that one would go to the trouble to delid and then put the lid back on. Certainly heatsinks scan still be screwed down to make up the difference in thickness. Is it because these sockets don’t balance pressure across the die as well?
Heatsink mounts are height sensitive. If you reduce the net CPU height then you can reduce, or even totally eliminate, the downforce on the CPU, which is not good! Direct contact between the CPU and heatsink is always going to be preferable, because it reduces the thermal resistance. If it's not possible to achieve the proper downforce, then use a small metal shim rather than using the original heat spreader. Probably most any metal shim would do, but it's going to need some kind of heat-transfer compound. It should be the same thickness as the original heat spreader and cut to the size of the CPU and perfectly flat. Removing the original heat spreader brings the added benefit of convection cooling for the other little chips surrounding the CPU.
I’m skeptical. I’ve dealt with a lot of cpu mounts and I doubt the spring loading will notice 1mm difference. Heck a lot of them load the springs with screws and there’s going to be a wide variance there based on how hard people torque the screws.
Tests like the POV-Ray results are nice because they show not all "stable" overclocks produce the same performance results, even at the same clockspeed.
I don't know if it actually helps or not but in the case of switching my laptop CPU to liquid metal I followed the LTT recommended method and applied some conformal coating to the area around the die, two layers especially covering the visible passive components on the chip. Supposedly this helps protect it in case any LM does come out and contact it. Haven't had any of the laptops I've done this to fry yet, thankfully.
Please no one actually delid a 2400g, you would be wasting your money.
$78 dollars worth of tools and paste. And a $180 cooler.....all to make a $170 chip lose a few degrees at load and not overclock any fruther...
Please spend $260 somewhere else. 260+170 for the 2400g is enough to buy yourself a 2600 and still have $230 for a gpu(which before the mining craze would get you a 580 8gb/1060 6gb). Which would run absolute circles arround a delided 2400g.
About relidding vs. naked die. Looking at some other delidding articles, the reasons for relidding are related to the LGA sockets: the chip needs to be pressed on the springs for electrical contact (not just for good cooler contact; the mounting bracket of the LGA sockets is designed to press onto the heat spreader, and one has to find a replacement for that if the heat spreader is missing. Also, the mounting bracket is in the way of the usual coolers if you go for direct-die mounting, so you need to remove it, and replace it with something else.
Antway, none of these problems is there for PGA sockets. I have recently looked at an AM4 CPU in its socket, and at least the cooler we use (Thermalright AXP-200R ROG) would not collide with the socket if mounted a little lower, and I guess this will be true for most other coolers. The only issue is that the mounting system would have to be adapted to mount the cooler that bit lower. The Thermalright AXP-200R uses a bracket that presses the cooler on the CPU; I would put something of the appropriate thickness between the bracket and the cooler.
Concerning the fear of breaking the die, I know of only one case of a broken die from the Athlon/Pentium III days: A friend of mine built his own water cooler, and when he was finished, he mounted it with a screwing mechanism. In the end, he wanted to give a screw another turn for good measure, but that was too much, and the Athlon broke. A case where the die survived was a cooler that was mounted in the wrong way, and sat one-sided on a Celeron. This was noticed because the Celeron was very slow, thanks to thermal throttling.
Concerning whether it pays off: It seems obvious to me that one would do that with a Ryzen G not as a particularly cost-effective way to get more performance, but because one wants to play around with delidding and other exotic cooling techniques. And in that context, a Ryzen G is a not-too-expensive playground.
Here a a bunch of corrections to your spellllling grammar. "and the Delid Die Mate 2 AMD Kit. This to our knowledge is" An introductory clause needs comas. "and the Delid Die Mate 2 AMD Kit. This, to our knowledge, is"
"Making sure the chip is placed into the tool the correct way allows the tool can do its job properly; putting a case fan the wrong way" Wrong word. Substitute "to" in place of "can". "Making sure the chip is placed into the tool the correct way allows the tool to do its job properly; putting a case fan the wrong way"
"between the die and IHS. One wrong jerk could or misplacement could get" Double "could". "between the die and IHS. One wrong jerk or misplacement could get"
Please go Extreme. What is the point of adding liquid metal on the cpu core when deliding you can remove a hole layer og TIM? Over the past years I have delided many K8 and K10 AMDs and adddind a copper thermaltake volcano just on the cpu core the results were impressive. Great overclocking ability and at lowest fan speeds. You have a AIO watercooler and removed the amd retaining base. The AIO waterblock has a copper base. I would like to see it attached directly on the cpu core. And I would like to see something I never had the chance to test.. Water flowing directly around the cpu core! I had never the chance to make such a container but the results would be interesting for sure.. Please go EXTREME!
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
73 Comments
Back to Article
MajGenRelativity - Thursday, May 10, 2018 - link
Thank you for the very detailed process description and results.gavbon - Thursday, May 10, 2018 - link
Not a problem :)Cellar Door - Thursday, May 10, 2018 - link
Thanks Gavin, top notch guide!eva02langley - Thursday, May 10, 2018 - link
How about the iGPU results? I was actually only looking for those to be honest.As for CPU scores alone, it doesn't worth it. However, for gaming... does it worth it?
gavbon - Thursday, May 10, 2018 - link
I do have a planned piece I need to finish up on the effects of GPU scaling on frame rates in games (average and 99th percentile). Overclocking, delidding, CPU frequency scaling, Memory scaling and iGPU scaling are the plan :)polyzp - Thursday, May 10, 2018 - link
This would be amazing! Would love to see the absolute peak performance (with gpu clock and ram speed and latency as priority over cpu speed). 3600 cl 14 on the ram would be ideal, or using the stilts 3466 llc setting for bdie.Lolimaster - Thursday, May 10, 2018 - link
Aside from temps delidding makes nothing else for the APU, you can get the same max igpu clock on any random 2400G.What we need to see is the limitis of the MC, 1700Mhz igpu OC + 3466 CL14 or 3600 CL15 memory if higher.
Lolimaster - Thursday, May 10, 2018 - link
OC the worthless cpu part hinders the max igpu clocks. It's 3.6Ghz base anyway...Ket_MANIAC - Thursday, May 10, 2018 - link
This is the real stuff. OCing the CPU to modest clocks like 3.8 is more than enough. Its the GPU OC which matters the most. Waiting for that part.wumpus - Thursday, May 10, 2018 - link
I'd certainly like to see OCs done with the stock heatsink. There's little point in watercooling a GPU with all the performance of a nvidia 1030...JoeyJoJo123 - Thursday, May 10, 2018 - link
>OUT>OF
>BOX
>THERMALS
DanNeely - Thursday, May 10, 2018 - link
A recent article on delidding Intel CPUs on [H]ardOCP suggested using clear nail polish to paint over SMD components to protect them against stray TIM.gavbon - Thursday, May 10, 2018 - link
Yeah that does work, it's the exact same method as der8auer uses, but I wouldn't want to test the theory out of purposely trying to short SMDs with liquid metal; that's the main issue other than damaging the die.kupfernigk - Saturday, May 19, 2018 - link
Given the cost of the CPU it's probably better to buy actual potting compound from somewhere like Element 14. This will also provide moisture protection.bill.rookard - Thursday, May 10, 2018 - link
What I would like to see is a kit to delid, get some indium solder, some spring loaded clamps, and a very precise baking temp... Solder that sucker on.gavbon - Thursday, May 10, 2018 - link
Would be interesting to say the least!edzieba - Thursday, May 10, 2018 - link
Probably a very bad idea: http://overclocking.guide/the-truth-about-cpu-sold...rocky12345 - Thursday, May 10, 2018 - link
Interesting read for sure from the link you provided. I am not sure if it is good or bad but all I know is my Sandy Bridge [email protected] (soldered) has been running at this speed and has had the same temps since day one of me getting it way back then. I am not sure if there are micro cracks in the solder or not but it performs the same today as it did in 2012. Then again my temps 98% of the time are always at or around 45-63c under full load and idle temps 21-23crocky12345 - Thursday, May 10, 2018 - link
I wanted to add that I also do not spend very time doing stress tests that bake the CPU to the point of cooking itself. I spend most of my time just messing about in Windows and Gaming which is far easier on the whole system than running stress tests 24/7 like some people tend to do.DanNeely - Thursday, May 10, 2018 - link
That was liquid N2 testing that he was cracking them so fast under, air/water cooled will have much smaller thermal gradients and thus much slower rates of crack formation.Oxford Guy - Thursday, May 10, 2018 - link
Properly-formulated and applied solder is irrelevant when it comes to "crack risk" unless you're using extreme cooling like nitrogen. If it's not properly-formulated, though, there can be a problem. That's what happened when Nvidia released a lot of mobile GPUs that failed. Apple was nice enough to replace the defective boards with defective boards. The industry doesn't do real customer service, like recalls.nathanddrews - Thursday, May 10, 2018 - link
In this chart, you show the stock/retail chip consuming 132W at 4.2GHz, but I think that is an error because the other charts don't show it capable of reaching 4.2GHz. Is this correct?https://images.anandtech.com/doci/12640/Delid%20Po...
gavbon - Thursday, May 10, 2018 - link
Each test bed is different, the chip used in this piece was a second Ryzen 5 2400G sample that could achieve this. The original sample from the review Ian published couldn't attain; I have both the launch samples and a pair of newer retail samples from AMD which we used all 4 in the overclocking piece: https://www.anandtech.com/show/12542/overclocking-...Ian Cutress - Thursday, May 10, 2018 - link
We can take a power reading at say 30 seconds, but it might not be stable for a full 5 minutes load.medys - Thursday, May 10, 2018 - link
Is there a reason for installing the lid back on the cpu?Why not install heat sink straight after deliding?
gavbon - Thursday, May 10, 2018 - link
The CPU cooling plate wouldn't sit flush on top of the die. This is why companies such as EKWB make 'naked die' mounts for their water blocks. When removing the IHS completely, what you're essentially doing is adding clearance between the CPU cooler and the die.To make it sit flush, you would need the cooler to screw closer in as to actually make contact. Too much pressure on the die however will crack it which = dead APU/CPU
Lord of the Bored - Friday, May 11, 2018 - link
BRING BACK THE THUNDERBIRD SHIMS!...
No, seriously. That's exactly why we had shims. Bare die for better cooling, but a greatly reduced crush risk.
nathanddrews - Friday, May 11, 2018 - link
They still exist, but they're called "direct die frames".Foeketijn - Wednesday, May 23, 2018 - link
There are a lot of Pentium 3/AMD Athlon/AMD Duron/AMD Athlon XP's that would disagree. In the 2k I messed around with a lot of Athlon XP's and where not carefull at all. Yes, I scratched and chipped a lot of chips while soldering pins and spots to unlock special features. But they tended to keep on working as normal.eastcoast_pete - Thursday, May 10, 2018 - link
Interesting article, thanks! The thing I don't get is why AMD, trying to gain market share from chipzilla, wouldn't spend the extra cents per chip and use the solder instead of the paste to begin with. Heck, they could even make it part of their marketing: they use gooey paste, we use quality solder.On another note, as they provide a stock cooler: are there any systems out there that do away with the lid entirely, and go straight to something like a vapor chamber approach in a sealed setup? Is this what they do for the Xbox X?
Ian Cutress - Thursday, May 10, 2018 - link
AMD does use solder on the Ryzen CPUs. It's the APUs that have paste. They do market the Ryzen-2000 CPUs as having higher-grade Indium-Tin solder.eastcoast_pete - Thursday, May 10, 2018 - link
Hi Ian, I should have been more precise: it's exactly because AMD "does the right thing" and uses solder for its non-APU Ryzen 2000 chips that this use of paste sticks out. I wish they would do so also for their Ryzen APU lines. Would make anyone considering buying a 2200G or 2400G feel less like a second - class customer, and emphasize the perception that this is a premium productOxford Guy - Thursday, May 10, 2018 - link
"Would make anyone considering buying a 2200G or 2400G feel less like a second - class customer, and emphasize the perception that this is a premium product"Intel hasn't had much trouble convincing people to spend a lot more on its second-class customer CPUs. It has been selling quite a lot of them since Sandy.
werpu - Wednesday, May 16, 2018 - link
Well the article was exaggerating the problem. There nowhere was a package difference of 12c in their numbers, as far as I remember in the upper area of 4.1 GHz it was 9c (from 67 down to 58). Sure they got another 100MHz out of it, but that is playing the silicon lottery area.So in reality depending on your workload the difference between the liquid cooling method and the a stock cooling via the usual ways is 4-9 degrees. This is not worth it. They might have shaved off 2-7 degrees by going full solder, but we are talking about cheap apus here. This is not an Intel like situation where the difference often is in the 15-25c degrees range due to cheap stuff being used.
Lolimaster - Thursday, May 10, 2018 - link
Why techsites and techtoubers don't understand, temps is not what is holding Ryzen or Ryzen APU's, is the way they designed the 1st iteration for a mobile oriented node.OCing the cpu part of the APU is WORTHLESS when the base clock is already 3.6Ghz, without delidding, any 2400G will do 1500Mhz+ on the gpu (good ones 1700 @1.3v for the soc voltage, the max for safety)
The benefit for delidding the APU is basically lower temps, specially when using the stock HS or a low profile one.
eastcoast_pete - Thursday, May 10, 2018 - link
Agree. I am eagerly awaiting a GPU-only overclock article for these, both stock and delidded. Also, second the interest in seeing that comparison with the stock cooler. My main interest in these (especially the 2400G) is the price/value aspect, and I don't think that many will want to shell out well over 100 Euros or dollars for a fancy cooler.gavbon - Thursday, May 10, 2018 - link
There will be a iGPU scaling piece, it is in the pipeline, along with CPU frequency scaling and memory scaling on APU, iGPU and dGPUDiji1 - Friday, May 11, 2018 - link
>the way they designed the 1st iteration for a mobile oriented node.Hopefully that means we'll see some tablets with 3D gaming possibilities then!
Matt G - Thursday, May 10, 2018 - link
Great article. Was an interesting read but FYI: If you add the cost of all the parts it totals $247 dollars. For that price you could easily step up to a much faster Ryzen 7 1700 for the same price.Ian Cutress - Thursday, May 10, 2018 - link
Therein is the rub, unless you know someone with a delidding tool. But hey, people buy $2000 cars and spend $10k in mods.sonny73n - Thursday, May 10, 2018 - link
Isnt the sole purpose of delidding is to reduce temp and improve performance? It’s quite different from modding a car.How about a blade which costs only 20 cents.
Lord of the Bored - Friday, May 11, 2018 - link
A lot of car mods are dedicated to improving performance. Just like building a PC, there's more to a car than adding RGB LEDs.sonny73n - Saturday, May 12, 2018 - link
You’ve never modded a car, have you? I wouldn’t spend more than 10% of that $10k on a $2k car’s engine, unless you mod a lemon.eastcoast_pete - Thursday, May 10, 2018 - link
Yes and no; a key reason why I am even considering a 2400G is that, once you have a 1700 or 2700, you still need a dedicated GPU, and that means an extra $120 - $150 for an underwhelming dGPU (geforce 1030), all better ones being way overprized thanks to the crypto-craze. Also, the delidding tool doesn't wear out (buy once, delidd many), so my plan would be to borrow or rent it from someone for a fraction of the cost, or buy one and resell it for $10 less after I am done. That leaves the costs for the conductonaut, which won't break the bank. Overall, still a good value proposition to get through the cryptomining dGPU price craze.werpu - Wednesday, May 16, 2018 - link
The 2400 is suitable for many situations, but for my desktop 4 cores 8 threads was not enough... Usually you have the graphics card already. I am considering going for a 2400g however later this year for a homegrown nas. This littly processor provides enough power to run vms decently a thing which most commercial home nases are not able to pull off decently due to their underpowered processors.boozed - Thursday, May 10, 2018 - link
A very valid point, which is addressed in the article's final paragraph.Diji1 - Friday, May 11, 2018 - link
I've never done it but there are delidding services available apparently which presumably cost less than buying the tool to do the delidding.VoiceOfReason116 - Thursday, May 10, 2018 - link
Gavin mentioned that he didn't know of large vendor offering pre-delidded processors in the US. I believe that Silicon Lottery does this, though they may not offer this service for AMD processors. They also do binning of CPUs. From the little bit I've heard, people speak pretty highly of them.Oxford Guy - Thursday, May 10, 2018 - link
Probably a stealth subsidiary of Intel.Oxford Guy - Thursday, May 10, 2018 - link
"a direct soldering has the best performance but might not withstand mulitple years of cycling."Bogus FUD. That concern only applies to liquid nitrogen or a similar extreme.
People have been recycling a slipshod PDF "article" with that concern-trolling nonsense for years now, pretending that it applies to normal cooling methods like water and air. Intel marketing at its finest. In an alternate marketing universe, AMD's use of solder is a mistake because it's more convenient to delid polymer TIM parts. And, in this fantasy zone, indium solder doesn't yield enough performance.
"standard TIM 'goop/paste' has sufficient longevity but lower performance"
Lower performance than what? There is a wide enough range among polymer TIM alone, without comparing the best polymer TIMs to the best indium solders. This is one of the reasons people weren't pleased when Intel marketed Devil's Canyon on the basis of a new-and-improved polymer TIM but an enthusiast website's testing found that not only did it have the identical low performance of its predecessor, it was outperformed by a rather inexpensive 3rd-party polymer TIM. The US government exposed Arctic Silver 5 for having a paltry thermal transfer rate when compared with its company's claimed rate. The 3M polymer TIM, by contrast, performed exactly as advertised (much better than Arctic Silver 5).
"I could see delidding a Ryzen 2000 series APU viably applicable would be in a situation" where rich people have nothing better to do with their time and money.
MDD1963 - Thursday, May 10, 2018 - link
100 more MHz! Great Scott...!!!!!!!!!!!!!!!!! :/PeachNCream - Friday, May 11, 2018 - link
Yeah, it does seem like a lot of effort and cost for a minimal reward. In fact, taking the results at face value -- the CPU was already operating at a reasonable temperature under load and the performance increase was insignificant given the risks. However, there's more to this effort that we haven't seen yet like the upcoming iGPU results. I'm curious about what's to come.I would like to see fewer silly screen names on products though. It's hard to take a name like Der8auer with a glaringly obtuse number in it very seriously. When I was a teenager, I remember seeing children calling themselves "sk8ters" and I thought it was stupid back then. It looks even more ridiculous now that I'm getting older. Maybe someday there'll be another segment of the PC hardware market that forks off the LED and screen name festooned man-child branch that is understated yet still premium.
AntonErtl - Thursday, May 10, 2018 - link
Nice Article, thanks.About relidding: You already reduce the clearance by removing the glue. Ok, so you need a naked-die-mount if you leave the heat spreader away, but why not go all the way when you already went this far? And it would eliminate the worry that the heat spreader is inadvertently moved in a bad way.
Maybe stuff for a future article? How are temperatures affected by naked-die-mounting relative to relidding?
sor - Thursday, May 10, 2018 - link
Yeah I was wondering about that as well. It used to be the norm from factory. I’m sure it’s common knowledge among enthusiasts who do this sort of thing regularly.vext - Friday, May 11, 2018 - link
The included Ryzen mounting system uses springs, which have a spring rate. If you remove the spreader, you change the CPU stack height, and lower the clamping force. So you have to somehow modify the mount, with heavier springs or shims behind the springs, or install a shim to increase the CPU stack height again.Other mounting systems have their own issues, which may be easier to modify.
sor - Friday, May 11, 2018 - link
I’m not convinced. I’d be more willing to believe that maybe it causes the heat sink to not clear a motherboard or socket component. There’s a decent amount of pressure on that heatsink clip and I doubt it’s going to notice such a small variance.sor - Friday, May 11, 2018 - link
Also, given the complexity of this mod it seems strange that a spring loading issue would stop you from getting the lid out of the way. If that were a problem, after all the trouble of delidding and using special paste surely they’d be able to figure out how to slide a toothpick under the spring clip.boozed - Thursday, May 10, 2018 - link
A hammer and a bad habit you say?sonny73n - Thursday, May 10, 2018 - link
Why did you have the lid back on after going thru troubles of delidding?sor - Thursday, May 10, 2018 - link
I remember all the old Athlons (e.g. the venerable “Barton” core) basically came delidded in the retail box and your heat sink went straight on top. I can understand why AMD and Intel started putting lids on, but it seems weird to me that one would go to the trouble to delid and then put the lid back on. Certainly heatsinks scan still be screwed down to make up the difference in thickness. Is it because these sockets don’t balance pressure across the die as well?vext - Friday, May 11, 2018 - link
Heatsink mounts are height sensitive. If you reduce the net CPU height then you can reduce, or even totally eliminate, the downforce on the CPU, which is not good! Direct contact between the CPU and heatsink is always going to be preferable, because it reduces the thermal resistance. If it's not possible to achieve the proper downforce, then use a small metal shim rather than using the original heat spreader. Probably most any metal shim would do, but it's going to need some kind of heat-transfer compound. It should be the same thickness as the original heat spreader and cut to the size of the CPU and perfectly flat. Removing the original heat spreader brings the added benefit of convection cooling for the other little chips surrounding the CPU.sor - Friday, May 11, 2018 - link
I’m skeptical. I’ve dealt with a lot of cpu mounts and I doubt the spring loading will notice 1mm difference. Heck a lot of them load the springs with screws and there’s going to be a wide variance there based on how hard people torque the screws.sonny73n - Saturday, May 12, 2018 - link
Heatsink mounts are height sensitive, NOT.Alexvrb - Friday, May 11, 2018 - link
Tests like the POV-Ray results are nice because they show not all "stable" overclocks produce the same performance results, even at the same clockspeed.Wall Street - Friday, May 11, 2018 - link
Gavin, I like your writing style and enjoyed the article. I look forward to reading more from you in the future.wonderbread2 - Friday, May 11, 2018 - link
I don't know if it actually helps or not but in the case of switching my laptop CPU to liquid metal I followed the LTT recommended method and applied some conformal coating to the area around the die, two layers especially covering the visible passive components on the chip. Supposedly this helps protect it in case any LM does come out and contact it. Haven't had any of the laptops I've done this to fry yet, thankfully.wonderbread2 - Friday, May 11, 2018 - link
also this stuff looks incredible, photos from my first application on a 5th gen i5 u series processor:https://imgur.com/a/ydSD9yI
a little drop goes a long way, you can see I probably had too much here.
SilthDraeth - Sunday, May 13, 2018 - link
Do you actually have to reinstall the IHS? Or could you simply install the heatsink directly to the processor like days of old?none12345 - Tuesday, May 15, 2018 - link
Please no one actually delid a 2400g, you would be wasting your money.$78 dollars worth of tools and paste. And a $180 cooler.....all to make a $170 chip lose a few degrees at load and not overclock any fruther...
Please spend $260 somewhere else. 260+170 for the 2400g is enough to buy yourself a 2600 and still have $230 for a gpu(which before the mining craze would get you a 580 8gb/1060 6gb). Which would run absolute circles arround a delided 2400g.
Outlander_04 - Thursday, May 17, 2018 - link
YesAntonErtl - Sunday, May 20, 2018 - link
About relidding vs. naked die. Looking at some other delidding articles, the reasons for relidding are related to the LGA sockets: the chip needs to be pressed on the springs for electrical contact (not just for good cooler contact; the mounting bracket of the LGA sockets is designed to press onto the heat spreader, and one has to find a replacement for that if the heat spreader is missing. Also, the mounting bracket is in the way of the usual coolers if you go for direct-die mounting, so you need to remove it, and replace it with something else.Antway, none of these problems is there for PGA sockets. I have recently looked at an AM4 CPU in its socket, and at least the cooler we use (Thermalright AXP-200R ROG) would not collide with the socket if mounted a little lower, and I guess this will be true for most other coolers. The only issue is that the mounting system would have to be adapted to mount the cooler that bit lower. The Thermalright AXP-200R uses a bracket that presses the cooler on the CPU; I would put something of the appropriate thickness between the bracket and the cooler.
Concerning the fear of breaking the die, I know of only one case of a broken die from the Athlon/Pentium III days: A friend of mine built his own water cooler, and when he was finished, he mounted it with a screwing mechanism. In the end, he wanted to give a screw another turn for good measure, but that was too much, and the Athlon broke. A case where the die survived was a cooler that was mounted in the wrong way, and sat one-sided on a Celeron. This was noticed because the Celeron was very slow, thanks to thermal throttling.
Concerning whether it pays off: It seems obvious to me that one would do that with a Ryzen G not as a particularly cost-effective way to get more performance, but because one wants to play around with delidding and other exotic cooling techniques. And in that context, a Ryzen G is a not-too-expensive playground.
ballsystemlord - Wednesday, May 23, 2018 - link
Here a a bunch of corrections to your spellllling grammar."and the Delid Die Mate 2 AMD Kit. This to our knowledge is"
An introductory clause needs comas.
"and the Delid Die Mate 2 AMD Kit. This, to our knowledge, is"
"Making sure the chip is placed into the tool the correct way allows
the tool can do its job properly; putting a case fan the wrong way"
Wrong word. Substitute "to" in place of "can".
"Making sure the chip is placed into the tool the correct way allows
the tool to do its job properly; putting a case fan the wrong way"
"between the die and IHS. One wrong jerk could or misplacement could get"
Double "could".
"between the die and IHS. One wrong jerk or misplacement could get"
Thanks!
Bill.Amd - Saturday, September 22, 2018 - link
Please go Extreme.What is the point of adding liquid metal on the cpu core when deliding you can remove a hole layer og TIM?
Over the past years I have delided many K8 and K10 AMDs and adddind a copper thermaltake volcano just on the cpu core the results were impressive. Great overclocking ability and at lowest fan speeds.
You have a AIO watercooler and removed the amd retaining base. The AIO waterblock has a copper base. I would like to see it attached directly on the cpu core.
And I would like to see something I never had the chance to test.. Water flowing directly around the cpu core! I had never the chance to make such a container but the results would be interesting for sure..
Please go EXTREME!