Fixing the 13900H (ES) Overheating - 10729 ITX Motherboard
Vložit
- čas přidán 30. 04. 2024
- Grab yourself a Pint Glass or Xeon Coaster at craftcomputing.store
My pint-sized gaming rigis one of the fastest gaming PC's I've ever built. Featuring a 10729 ITX Motherboard + 13900H (ES) 14-Core CPU, along with an Asus TUF RTX 4070 Ti Super, there's no game that's off limits.
But, not everything was sunshine and roses... the 105W 13900H proved to be a but too much for the be quiet! BK034 low-profile CPU cooler, with the CPU regularly thermal throttling, and the side of the case getting too hot to touch.
Today, we're going to remedy that with a cooler upgrade and a shim mod to the motherboard IHS. But what kind of performance gains will we see on the other side?
But first... What am I drinking???
Lapsang Souchong (Harney & Sons) amzn.to/4cYocfT
Links to items below may be affiliate links for which I may be compensated
ID-COOLING IS-55: amzn.to/3U1yTWu
ID-COOLING IS-40X: amzn.to/3JsgxZy
10729 Motherboard/CPU: Unavailalbe :-(
Fractal Design Terra: amzn.to/43VAdi8
ASUS TUF Gaming RTX 4070 Ti Super: amzn.to/43Ve8QO
Patriot Viper VP4300 1TB: amzn.to/43YSATt
Western Digital SN770 2TB: amzn.to/43YSQBz
Cooler Master V850 SFX: amzn.to/3JiNhV8
be quiet! BK034 Cooler: amzn.to/3TZlIFg
OLOy 2x16GB DDR4 3600 MT/s: amzn.to/3Jd8uQx
Follow me on Mastodon @Craftcomputing@hostux.social
Support me on Patreon and get access to my exclusive Discord server. Chat with myself and the other hosts on Talking Heads all week long.
/ craftcomputing
Music:
"Loopster" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 4.0 License
creativecommons.org/licenses/b... - Věda a technologie
Alright, I heard you all loud and clear. I removed the heat spreaders from the 32GB 3600 kit, and here are the results.
be quiet! Pure Rock LP:
Overall: 10,285
Graphics: 12,047
CPU: 5,6,25
ID-COOLING IS-55 (w/2666)
Overall: 10,226
Graphics: 12,029
CPU: 5,531
ID-COOLING IS-55 (w/3600)
Overall: 10,331 (+1.0%)
Graphics: 12,020 (-0.1%)
CPU: 5,753 (+4.0%)
Still not the butt-kicking I expected with the CPU staying at 95W vs 79W, but now at least it beats the be quiet score.
Any chance you can remove the heat spreader on the cpu?
Margin of error stuff. I highly doubt anyone would notice any difference in performance between these two sticks in real life gaming.
Are the be quiet results also updated to exclude heat spreaders on the ram?
Kidding! 😎
Great videos 👍
Wow, thanks so much for the update!
Really appreciate you following through here, since its probably not worth making another video.
Keep it up!
You should look into pure copper options and noctua swapping the fan in the future maybe?
15:30 the memory speed did impacted the score i think
If you pair this beast with does 3600cl14 b-die kits
Or it's run to run variance.
I was thinking the same thing, it is impressive that, even with the slower memory, it was within a margin of error of the smaller cooler with the faster RAM.
Intel isn't sensitive, they learn their mistake from C2, they moved away from gluing.
Meanwhile, the competition, introduce a lot of unnecessary bottlenecking, IF-MC-FCLK cannot run at 1:1:1 (since Zen++ and Zen+++++, FCLK for Zen+++++).
Firestrike always yielded better results with lower frequency AND timings RAM settings. So, I highly doubt it was memory.
And also firestrike score scales a lot more with graphics overclock than CPU. Couple hundreds more CPU Mhz is indeed a mere couple percent above baseline score (that is if overclock is even stable, you may very well go down in score otherwise).
But Cinebench is a surprise though.
Run be quiet now with this memory.
I'd say the IS-55 is working really well, so well that u got basically same results with 2666 ram vs 3600, now find lowprofile 3600 ram and retest for the real diff :)
love the troubleshooting homelab content! Can't wait for the Hyper-V video to drop!
yes!!!
Its the slow memory yo
Im impressed it even got same results
Lots of youtubers sell merch, but yours is far more inventive.
I think most CZcamsrs just slap their face on a thing and call it a day. It's rare to find one that has merch that shows that they get their audience.
Thanks!!! I appreciate you noticing the effort I put into it.
Jeff's smart, because he also *makes* the merch. No middle man with their fingers in the pie.
Maximize profits be seizing the means of production.
I'm all here for this revision type video, getting more tweaks for it to get better!! Btw great tech tech potato shirt.
The conclusion is very true, one has to test everything
"Standard height memory"
What I've done in the past is to shuck the heat spreaders off the memory, that usually reveals something close to, if not exactly, standard height sticks.
Probably voids your warranty, but, I've never had any memory issues doing this. Also never tried it with DDR5 which, tends to be thirsty for cooling.
Those thermalpads are usually glued to the heatsinks, so it is a one-way ticket.
Older Corsair Dominator RAM sticks were glued down with thermal epoxy and in an attempt to remove them I removed two of the chips on the board. I would say that was the exception, rather than the rule though.
Yup. I've done the same to my two 32gb ddr4 dimms to make more space. Been running fine at 3600mt/s for years.
Just be careful when doing this. If you're not careful, the memory chips might stick to the heatspreader instead of the pcb.
There's plenty of fast memory that comes with a heatspreader with almost no additional height. A basic metal plate is enough, these tall fins are just for show.
I like Corsair Vengeance LPX, those are low to begin with.
lol 2666mts? thats slowing thins down
Not sure the bios can handle anything outside of jedec specs
@@oscarcharliezulu The 3600 kit was working at that speed in the last video
RAM speeds are largely a marketing thing these days. The difference in real world performance is in the single digits on a percentage scale. In effect, the performance increase between these kits would get lost within the margin of error (at least in most cases).
@@jameslewis2635That's somewhat true, if you don't run any CPU heavy application (only play games and edit documents). Also memory hungry applications like fluid simulation and VFXs.
@@jameslewis2635 it does affect @2666mts ;D his board only supports dualchannel, a "normal" Ramspeed for the used CPU tanks min fps but mostly obersvable in synthetik benchmarks, which he used to measure performance.
I am glad I watched this video purely for the fact that I now know I am not crazy with the ID-40X issue. I had the same problem. Also thank you I love your videos!
TimeSpy is very memory sensitive. As others have pointed out, that's the difference. You can easily get 10% CPU score in TimeSpy by overclocking RAM and tuning timings.
For reference, I have a system with an i3 10100F and 2666Mhz RAM without tuned timings that scores ~4500 in TimeSpy CPU. An FX8350 system with everything overclocked and RAM properly tuned down to the tertiaries scored around 3600. That's on a Wraith Prism, not an amazing cooler or anything.
Jeff: "You have to test everything!"
Also Jeff: *doesn't re-test the Pure Rock LP with the heat spreader shimmed*
Maybe the RAM is reason why the 3dmark benchmark score is higher when using bequiet
I can confirm from my in person viewing of the machine at PDX-LAN that she was a toasty boi of a machine...
The CPU side is cold to the touch now! Crazy difference.
great video! for pdxlan i did a build using a similar board and saw the best results with that same is55 though swapping in a slim noctua did help temps vs the stock fan though that being said I did use the next one down from the 13900h, either way I also happened to pick the terra though in green!
Tea review! An unexpected but welcome surprise. Gonna order a tin!
Harney's "Holiday" black tea is my current go-to for black tea.
I'll have to give that a shot as well :-)
Great video bud. I honestly think your ram is your draw back. Might be worth finding a good set without heat spreaders. Maybe even a higher speed like 4K minimum
You should check out the Thermalright AXP90-x53 Full Copper. No clearance issues and performs better than the IS-55.
You could also just remove the heat spreaders off the 3600 ram. That cooler has direct airflow over the dimms
Big fan of the IS40 have one on my 13600K, running 5.3GHz undervolted to 1.2v (P cores only/E disabled). Keeps my cpu under well under 85c while gaming and pumps out frames.
Good but it would have been interesting to just test with your added shim and the bequiet cooler also a feeler gauge (for checking the gapping of spark plugs) can be super handy for checking contact
I had a similar problem with the height of the cold plate on the 12900k ERYING with a Thermalright low profile cooler. I was getting better cooling with a Hyper 212 EVO, so I'm still using that.
Have you consider using a Graphene sheet like Thermal grizzly Carbonaut or Kryosheet? You would have to cut one to fit the actual CPU but, it may yield more stable results than thermal paste in some cases. I have a hunch it may help with the BeQuite Heatsink and you wont have to sacrifice on RAM.
I'm a big fan of the IS-55. I have one in a SFF OEM system cooling an i7 9700. Removing the power limits, and running R23 for 30 minutes straight it has it sitting in the mid 80's pulling down about 115 watts. I'm thinking about swapping it out for the IS-67 XT, which might even be a little better. I put one with my nephews new system I'm helped him put together in the ridge and it was cooling the i5 13600K @ 150 watts in the low to mid 80's as well. ID Cooling is really killing it. Too bad that won't fit in this setup, but it seems the IS-55 is holding up.
love the techtechpatato shirt.
Dude! Pull those heat spreaders off of the fast RAM and get that thing working the way it should! You have direct fan flow over the RAM now so the heat spreaders are useless anyway.
8:12 erying has a video on this topic with thier Vapor chamber video. they litterally had to because they made the ihs too low and alot "lga 1700 ready" coolers are actually not usable.
noctua l9i (lga 115x/1200) with a 1700 bracket on it will have the mounting holes line up, but the cold plate is thinner and actually comes into contact with some parts with the board.
the l9i lga 1700 fits without touching any board components
bend the brackets a bit which will create more downward preasure as you are creating space to screw down before bottoming out
The RAM change is why the performance is lower. If you put back in 32gb of 3600 you will have a new higher score with the new cooler.
These gen boards has trouble working over 3000
Great video
I know its not worth another video, but i was curious how much of a temp/clock difference those washers would have had on the be quiet.
Oh, and every time i try the BC-250 i move one step forward and 2 steps back.
Sometimes it will show up in windows if i use 6650S drivers, but it doesnt actually do anything. in linux i cant get SSH to work when it works fine on other machines, what i'm probably going to try next is installing linux on a zen2 machine, install SSH, and then move it over to one of the BC-250 sleds
Nice desktop background
I wonder if you might be able to use LM between die and heat spreader for better conduction of heat?
Your reduced memory speed could be having a small impact on your overall performance scores. Other than that, I am impressed with how much cooling compacity you're getting with the IS-55's super slim fan and nearly zero clearance from the side panel. Have you tried testing the thermals again but without the side panels in place?
Would love to see low profile 3600mhz ram :D
Maybe 3800-4000, just to see what the motherboard can handle
@CraftComputing
While the goal you were going for is easy to see, I would have gone with a dual slot GPU for this particular build. A 4070(non ti/super) dual slot, triple fan model. That's just me though. I game at 1440p and my 4070 does more than enough. Also would have gone with a lower profile kit of 3200 RAM. Regardless, great build Jeff! EDIT: Video's like this are excellent. Experimentation is always fun, even when you have a failure.
if you know how big that gap was between the mid cooler and the heat plate, might be worth using a 1 or 2mm alu/cpr plate as a shim. But I would be sort of iffy about putting another shim in to a system.
I'm not familiar with these tests, but could the slower RAM have influenced the results? (As I understand it, the second set is slower than the first.)
Just put that tea to my Amazon cart for Tomorrow. As a scotch drinker, i am excited to try it. :) Thanks
(Used your link in the description, to be sure i got the right one. ;) )
I’m wondering if you had tested the be!quiet with the shim mod on the heat spreader and maybe get better temps on that one too? Might change the conclusion
The IS-55 is an absolute champ. Probably to optimize you’ll want some Corsair Vengeance 3600-16 or something, not a lot of options given the height restrictions
On the first new cooler tried it looked like you could have flipped the mounting tabs on the cooler to make them higher than the cpu and then shim it back that way?
Can you compare this to a standard 14600 or 14600, + itx mobo both in performance as well as idle power and peak power consumption? Specifically when you control all other variables. I want to know if the H processors will use less power at Idle but still perform competitively with the 14600 on a performance per watt basis. If you set power caps on the 14600 would you achieve the same performance as this setup?
I had an IS-55, did you try flipping it around the other way over the IO so the taller ram was free?
Would be interesting to see it with a water block on the cpu and the good memory and a big fan blowing at the motherboard, set an ideal score, to understand if cooling really has any headroom left to give.
Why didnt you do a multicore cinebench compare? if the scores are within error it means the bequiet was enough to fully tap the chip. A question I have is what idle wattage is like from the wall?
is it possible to mount any of these coolers direclty on the cpu die?
You need faster ram. Those 2666 sticks slow down things and cpu cooler does the rest so its almost the same scores.
The secret with most of these boards is to get better paste between the die and the vapor chamber, I use ptm7950 and it works great, no more thermal throttling, I even overclocked mine.
I wonder if the fact the be quiet is smaller, the case has better air flow. Maybe with the bigger fan the air flow is a bit blocked and maybe the guy is overheating. Sat the gpu is causing a throttling. Who knows? What was the high temp on the cpu given all the cores were fine?
SDFX SHOW used Thermal Grizzly Kryosheet on die and another one under the cpu cooler. The temps went really low after.
Even without improved performance I suspect the better handling of heat will give the components a longer life.
I'd love to build a little machine like this. Looks like a lot of fun :)
Dude, you should. This thing is a BEAST!
Did you try Windows 11? It may help distribute the load between E cores and P cores. At least I noticed this on my 13600KF
On the IS-40X can the shiny metal brackets be flipped so the threaded studs are facing away from the motherboard? Because that looks like how BeQuiet did it on theirs.
I argue that the objective should be sufficient cooling for the most demanding game workload, without throttling or being noisy. Cinebench wouldn't be an ideal test, maybe something that uses 50-60%. If it works there, people tend to be okay with the solution.
I suspect the CPU is fluctuating very quickly between 95 watts and lower wattages, and is doing it quicker than the granularity of the monitor program. Did you try an R15 single core run to see how those compared?
Could you not do direct die cooling with the BeQuiet? Would need some modifications to fit, but would take up less space and possible better performance?
I was wondering if direct die is possible too
Not enough air flow and heat not getting expelled fast enough. Have the same issue with my threadripper. heatsink is very capable sustaining decent overclocks under all core loads but the fans just cant move the air fast enough, the my case fans were garbage and was not removing all the heat fast enough.
Hey i dont know if you check their page they have a new chamber available for $17 that according to reviews it solves the heating issues
Er ying is selling their vapor chamber heat spreader made for these itx motherboard nowadays. You should give it a try.
please do more videos on ES/hacked cpu's?
maybe even some insight into QTJ0/QTJ1 chips? they are pretty much i9 9900k's but for 110 bucks.
not worth it anymore unless you happen to have the parts to do so. do you want to buy a usb programmer as well that increases cost?
i own both qtj1 and the qqls and xeon 2286m, they are "good" but you cannot justify them anymore as a most 5000 series ryzens are so cheap now to do so its in your best interest to go with the newer platforms / parts.
@@mikeymaiku i already run a 5950x as my main rig but i just build a testing rig for those cpu's.
getting a few in the mail. and got a extremely high end board for it.
worth it? maybe, it depends.
if you got a good z170/z270/z370 board. you can spend 77 bucks for a 6c12t or 110 bucks for a 8c16t unlocked chip
then resell your current chip (which will still fetch a decent penny. the used market is just fucked)
which in turn leads you to a 50 bucks upgrade in the end. maybe even less if you have a i7 8700k or 7700k to sell. you probably get that QTJ1 for free then in exchange of the higher end chips.
you just gotta mind that you wont be doing 3200 mhz ram on these chips.
also. currently testing differences between interposer designs. i feel like there's a huge difference that people are not aware of.
are you sure its a problem with the cooler and not a problem with the die contact to the IHS that comes with the motherboard?
Should have repasted it / checked if it needed more liquid metal. whatever the case
With the fans now being very close to the sides, is there an increase of fan noise?
ID-Cooling low profile coolers’ mounting mechanisms are far from perfect, had issues with couple thin-mitx boards where mounting brackets or heatpipes were resting on MB elements around the CPU socket which lead to improper contact or no contact with CPU at all.
maybe test "amd equivalent" BD790i from minisforum with mobile ryzen 7/9 on desktop ITX board ?
Can't you invert the orientation of the IS-55 , I/O seems to be low profile enough to allow that
If you can undervolt that ES sample give it a shot. I have seen better overall performance with newer intel parts when undervolted.
lol if your lazy you can run -.200v in intels xtu or whatever, both my samples can run 5.4 all core at 1.24v and i use a substantially smaller cooler than the is and still remain under 90 degrees
I had the same thoughts. Getting the bigger cooler to run at the same wattage may result in a surprising increase in test scores.
All of the modern chips tend to like a little undervolt, as a treat
Wow. I got here within an hour of the video going up. Unheard of for Australia, yay for insomnia.
2 free variables? 😎👍
Cooling
Ram speed
I wonder if the issue with performance is memory speed. I know that these mobile chips really like having as fast of memory as they can get. Going from 3600 MT/s to 2666 MT/s is a massive difference
The RAM speed dude, run the test again with the SAME speed RAM.
So un noticeable performance gains while using more watts… upgrade or nah?
I was more baffled to the questioning such a reduction in memory bandwidth. At least it’s addressed in the topic comment. That kit put in was far too slow for any modern platform.
Erying sell a vapor chamber heat spreader, i think this one only a copper slab. Idk if it's compatible with this one tho
Did you get IOMMU working or PCIe Passthrough?
Hi just an idea for another video. Why don't you try to use water cooling both your CPU and GPU in that tiny case. It would be interesting if is even possible😄
PSA, another way to do this is a PIO motherboard from taobao and a JF13K cooler using a standard 12700k CPU for about the same price. This puts the GPU and the CPU intake on the same side and reduces the wasted space (and no need for auxiliary fans or risers). Could do a "dual chamber" and put the 30 dollar FSP-500 "NUC Extreme" PSU under the motherboard for the GPU, and a Dell/ Alienware GaN brick for the system. Well, that's the dream anyway.
Do you have a model number for the motherboard by chance? I tried looking it up on taobaos site and it didn't show up but I also don't speak Chinese so idk if it was because I used English characters lol
you could sand that metal heat spredder a little bit maybe it will reduce a degree or 5 who knows.
I don't know about you, but I would consider that plate which goes over the CPU an IHS and have liquid metal there. As much good as Arctic MX-6 is, I don't think it's doing much in this instance.
Are these mobile CPUs decent for plex? I need multiple 4k encodes capable CPU. I know mobile CPUs have 96 execution units vs. desktop versions with 256 eu's. Just want to know how much hit I will take by going the mobile route.
Im guessing the extra watts went into the e-cores to do their stuff that's not in the test.
It's probably down to the speed of the memory. Using the slower 2666 MT ddr4 is probably holding the cpu back a bit compared to the 3600MT memory. That's about the only thing I can think of that might be causing the less than expected scores.
Maybe I didn't watch closely enough, but why do you reattach the stock heatspreader b/s? Go direct die cooling like in the old Athlon days. Careful of course, but I wonder how much better that would actually be.
you can only really direct die these with tower coolers. there is some elecrital components right by the cpu. and thats why the is40 did not fit
all you need to do is remove the heat spreaders on the oloy ram and you'll get the best of both worlds. Ram speed matters ton in alot of scenarios when you are comparing 1000mhz of a difference.
On paper it’s like the be quiet is swinging critical hits and the 55 is just plowing through best thing to to now is swap to the be quiet with the less effective memory and see how it works as that will be the truest apples to apples.
NR200P with 280mm AIO and temp problem solved for good?
yup its beeen thatr way awhile. Sure if you can get a 1ghz higher clock that's worth it, but if you have to add 20% more power for 2% performance it just isn't. The extreme guys have even seen where backing off the clocks actually gives better performance in certain cases.
My thought was to flip the brackets on the IS-40 and try to figure something for that solution.
You have piqued my interest in these CPU/motherboard combo's. I was thinking about trying one of these in a Lian Li air mini. A little bigger case would allow for better air flow, plus you can use a 240 or 360 liquid CPU cooler. Thermalrite has AIO's running from $50(low end)-$80(top end). Which would still keep it cheap.
Nice
As soon as you said shim i was thinking wtf i have the same case with a i9 13900k at default settings my max cpu temp is 65-70c using a nactua cooler
Please try to remove the heat spreader from the 3600 MTs ram. Most ram doesn't need it und the cpu cooler blows direct air over the ram.
Upgrade the RAM and benchmark again please!
16:36 some artefacts 🌝 why that may happen?
Perhaps memory is why underperforming results with the IS-55. Buy a new kit which is matching DDR4 3600
i blame the ram speed and want a follow up video!!
Its that case you have it in
Next video.... Machine a new block, N2 cool it... Just to see what the Max prefirmance is.... 😂😂
Super small form factor is such a measure twice, cut once situation.
Still... a 7700 gets around 6200 points in Timespy Extreme with a lower TDP...
It's the memory speed. Clearly