Etern4l

A few delidding questions guys: 1. Has anyone directly compared LM/Conductonaut with PTM7950/7958 between the die and the IHS? 2. Rockit recommends "seasoning" the copper IHS with their Quicksilver gallium paste (I suppose LM would work as well, albeit more expensive), ideally for a week. This is supposed to prevent the die and the IHS getting welded together, as well as improve the thermal performance. Have you done this, and if so, how long for? 3. What's your recommended re-lidding glue/technique? I've come across quite a few: a) No glue - sounds like a bad idea as it would be difficult to keep the IHS still on the die b) 4 small dots of super glue on the PCB, some people say super glue has low flash point and can damage the PCB c) 4 small dots of silicone RTV glue on the PCB d) 4 small dots of silicone RTV glue applied externally on the corners of the IHS (wouldn't that interfere with the contact frame?) Thanks for any thoughts and tips.

Very interesting, so Nvidia launched an attack on the CPU front... 144 cores. Respect, although no clocks data given. Would be interesting to see what the pricing and performance of workstation-grade ARM CPUs looks like. I always liked RISC architecture. It was miles ahead of Intel back in the day. I think it almost died for a while mostly because they didn't manage or care to break through to the mainstream (Apple is changing that now). The CPUs were mostly used in things like Silicon Graphics workstations. You can see those machines in office scenes in the original "Jurrasic Park", along with Jeff Goldblum and the other dinosaurs :) As Intel fans we don't talk about enterprise CPU prices @Mr. Fox lol

Companies do change. Remember it was AMD who gave us the 64 bit x86 architecture, while things at Intel went sideways for a while with Itanium. What is the underlying architectural problem with Zen 3 and 4? I mean something is not quite optimal if they are merely trading blows with Intel despite the massive process advantage. If you are just looking at it from the perspective of overclockability/tweakability, then still Intel every time I guess. BTW guys, what's the story with ring voltage and ratio? What's the default ring ratio for the 13900K (for some reason MSI board doesn't show this, or I missed it), do you guys touch the voltage or leave it on auto, assuming fixed CPU ratio mode+voltages scenario.

The This would be pointless, just like all those sham revolving door AI ethics committees. The time to pull the plug on the whole thing, starting with the companies responsible, is now. Will humanity do it? Of course not, due to globally insufficient natural intelligence.

Well, let's hope Intel is not gone. Certainly the stock market doesn't think much of their prospects (as opposed to AMDs). Seems to me their loyal customer base, often going back decades, is their greatest asset in the current testing conditons. I wonder if "progressive business ideas" such as disabled features, activation fees, and seriously compromised quality to shave a few bucks off a $700 CPU etc. constitute the best strategy for them to retain those customers.

Well, not really if AMD would just be catching up, plus Intel would increase 13th gen unit sales, and they would have probably been able to charge more for the overwhelming performance, just as NVidia does with the 4090... They would seriously put AMD on back foot, and still actually have one more shot around Zen 5 release (provided there would be anything to actually offer lol). Think about it bro :) Anyway, I look forward to seeing if DVLR is not just a gimmick which does UV for me (which I am reasonably good at by now, by necessity lol, so probably wouldn't be too interested in that feature alone). That said, my fine 13900K will probably soon die a horrible death during a delid procedure, and if not just fail later in the year, as I was running it at over 253W for quite a while :)

Well, they could have just enabled it from the start. The overwhelming performance delta would mean nobody would buy AMD and as a consequence, their main competitor would suffer losses, ergo Intel would make more money off folks who otherwise went with AMD. Intel management team does comes across like a bunch of shifty tools in this story TBH. BTW if I can run CB23 at 200W and get 38K, what benefit will I see from the DLVR feature? Is it more than just an undervolting trick?

Which is already in Raptor Lake but disabled. Great. This might be along the lines of additional AI features being physically present on some Sapphire Rapids chips but disabled by Intel, and requiring a subscription or activation <facepalm> Was that another one of Raja's brilliant business ideas perchance? Keep it up Intel, and one of these days even @Mr. Fox might throw in the towel, and emerge from Fort Fox, proudly carrying an "AMD" banner lol To conclude: sadly, tech is richly saturated in (not full of, fortunately) scum whichever way you turn.

One would hope they would offer some material design upgrades rather than bin/OC more out of the box and call it a day.

Tell me about it bro. My kingdom for a proper socket lol! Yes, the contact frame helps, but it's not without expected serious pitfalls (requires careful calibration by repasting, checking fit, etc. until optimum). Interesting point. I don't know why, but my undervolted 13900K periodically downclocks to... 2.2GHz under *particularly heavy* load, then bounces back to 51/41 I'm currently running.... Ah, now I remember - the way this works on MSI boards is that on entry to BIOS after CMOS reset you are presented with a pictorial question about your cooler type. If you select AIO/water (the picture shows a 280 AIO...), it sets power limits to 4096W on the 13900K, which is optimistic to dangerous. MSI should have made it clear which settings exceed Intel's specs. TBH I wasn't aware of that nominal power limit until today... This also raises the question: when reviewers bench, do they do so with power locked to 253W? Probably not, and at those higher powers all bets are off w.r.t. reliability as far as Intel is concerned. I don't recall seeing "Massive 253W of power inside!" on the box either lol. Nice, a self-propelling false advertising machine. 100%. Reliable stock performance should not require any of that. What's infuriating is that it boils down to probably the simplest aspects of CPU manufacturing. It's like spending 2h writing a beautiful homework essay, and the spilling cola on it right before handing in or something - seems like a silly schoolboy mistake, although possibly has something to do with old production lines they couldn't upgrade because of other priorities. Just guessing. As always, take great care of your core business first. I'm glad you have found Indigo useful. Less of a killer on the Intel side (and it's kind of short), but cross-platform and supports multi-GPU setups. BTW if CB23 runs on AMD without breaking a sweat, that just means that AMD is very efficient at handling CB23 load (and yes, probably CB23 is not super-optimised for AMD, if it were it might run even faster). Your AMD also does great with Indigo (although with greater strain). I am not aware of a single great tool that would reliably test stability for me. I believe that neither CB23 not Indigo are particularly taxing on the memory and its controller. However, when it comes to pure RAM testing Google GSAT comes close. They use it to test their servers: https://www.overclock.net/threads/gsat-google-stressful-application-test-on-a-tiny-bootable-linux-iso.1718166/

Ha, mine did low 40k on just stock ratios, I just didn't realize the undervolt had such an effect. 41k was with 45 on the e-Cores (and long term unstable UV, actually it might have been closer to -80 or -100mV). What do you even mean by "stock power limits"? The CPU has no power limits. Edit: notebookcheck corroborates what you said: https://www.notebookcheck.net/Intel-Core-i9-13900K-Processor-Benchmarks-and-Specs.666495.0.html Median 38200. I guess maybe my chip wasn't bad to begin with, and clearly UV helps. I don't know what the score on default voltages was I think I UVed straight away. Expecting great things from the delid then, the kit already shipped.

Leaving Nvidia aside, this is surely a biased statement. AMD is a viable alternative. In general they offer three clear advantages over Intel: 1. Reasonably uniform cores (no baby cores) 2. AVX512 3. Larger cache To be clear, Intel CPUs have advantages of their own. From your point of view that might be the case, but we live in a microbubble here. 99.99% of PC builds don't have gigantic external water cooling setups that might just mask off a lot of the flaws. Additionally, most users probably wouldn't want to mess about with contact frames, never mind things like delidding in order to achieve adequate stock performance. Pretty much every review and forum is full of statements confirming the chip overheats. I believe Intel are guilty of false advertising here too. Check this out: https://ark.intel.com/content/www/us/en/ark/products/230496/intel-core-i913900k-processor-36m-cache-up-to-5-80-ghz.html Processor Base Power125 W Haha, running minesweeper perhaps lol Maximum Turbo Power253 W Another joke. 13900K would probably top 350W in CB23 on default volatages. My specimen was breaching 335W on stock ratios with -55mV dynamic UV, back when it still functioned at that level. Now, you hastened to add "compared to the alternative option", but really let's not resort to a strawman AMD argument. I would expect them to deliver a well engineered product, that works to spec (looking at the ark entry, clearly it doesn't!) for years without any unreasonable cooling requirements ("delidding, custom loop and chiller required" lol) on stock settings. Sure, there should be enough OC potential for people who have extreme cooling setups to get more performance, but the chip should not require any of that on stock settings. Not to generalise, but my now (not initially) problematic 13900K is indeed still performing better than 12900KS, which a bit of a lame excuse really. All I am really providing here is constructive feedback for them (for their refresh and MTL, where applicable): 1. Make the socket/IHS larger and more resistant to bending to improve cooling 2. Solve the IHS/die contact such that no delidding is required unless someone wants to go direct die 3. Implement a robust ILM/mounting mechanism, to eliminate the need for any aftermarket upgrades in this area 4. Make the die/package larger to improve cooling (sounds like that's going to happen in MTL) 5. Improve fab process to reduce the currently idiotically high variance in chip quality I hope we can all agree those are sensible requests. No more nonsense marketing gimmicks to try and show you are better than AMD, no more funding for Linus to try and mansplain to the entire YT how difficult it is for Intel to manufacture a properly functioning IHS (jk, the funding for that came from contact frame and 3rd party IHS manufacturers), just get it done. BTW on the point 5. above, hilarious latest video from Linus where he is showing a 13900K score 37K in CB23 with EK direct die cooling and something like 36K with a custom loop and stock ratios. ROTFL, that must have been funded by Intel to set the expectations low. The guy is really losing credibility in my view. As posted earlier, I got 38K (on Linux!) with 51/42 ratios and heavy undervolt at 200W. Stock performance of a properly functioning 13900K on Windows 11 is around 40-41k.

Crazy. At 1.17V my 13900K runs at 5.2/4.3 lol Have you filmed the deliding by any chance? Would be a cool video. BTW Why don't you set the voltage for the L2 cache? What's the default ring ratio for the 13900K? I know you guys love intel and want to defend it, and so do I, but let's be realistic. Just by the very fact that he was Intel's chief architect, he is a pretty smart guy, not a looser. Looking very briefly, he had a vision and tried to push the envelope by expanding/diversifying, which is technically impressive but unfortunately mistimed from the business perspective. Intel's CPU business is at risk and the CEO should have (probably) said 'no'. They need a reality check on where they are at the moment. In addition, we ought to avoid underestimating the power of the finance people in a corporation, especially in a crunch, and boy are we in a crunch (doesn't apply to bro @Papusan, Norway is raking it in on oil and gas), and tech crunch specifically. In this kind of situation, non-core R&D the guy bet on is the first to suffer cuts, and TBH fair enough. IMHO they just need to focus pretty much all resources on the CPU business, as this has been declining due to competition from AMD and ARM/RISC. On a personal level, I'm quite disappointed with the engineering quality of their flagship consumer product: failed ILM, failed IHS, failed process/too much heat. All this YT influencer talk about Intel not being able to afford to engineer a robust ILM and IHS, because they do it at scale. LOL, more like they have been cutting corners to fund peripheral research. They just really need to put in some hardcore work over there, if it's even about hardcore work - could be a TSMC patent thing and they just need to write off their legacy fabs (ouch!) and join the queue at the master chip maker, which would be very bad as obviously TSMC needs competition. Good luck dear Intel, you will need it.

This is nuts. I was just reading the same thread! Basically I am dealing with a variant of this with 3 rogue cores. Pretty good guess it's the cracked solder problem I've found some references to on other threads. If you read through, ultimately he fixed his rogue core issue with the Rockitcool IHS (he lapped it a little on the die side after a prior test LM application). People are also saying the latest iteration is shrunk (a little bit smaller than Intel IHS), but fits the Thermalright contact frame, which would be convenient.

Understood. Hearing/reading some mixed opinions about direct die cooling, seems tricky to get to work. Anyway, not an option for me for now. Ordered the Rockit kit, goodness knows when it will arrive from the States, but I will be ready, heat gun in hand lol

Custom cooling might help, but given the huge core deltas and the fact that the heatsink fit is confirmed to be pretty good, with the coolant being pretty.. cool, my working hypothesis is I need to delid this and possibly lap/replace the IHS first. Rmemeber I don't care about OC. If I can just get stock ratios to run at 100% under 90C I will be very happy. Prior to the degradation I was able to run CB23 at 320-330W.

Tried to quickly google but no joy, e. g. no IHSes there. http://www.supercoolcomputer.com/en/category/8/cpu-water-block Where did you get it from, and why that vs the rockit IHS?

Rockit tool and original iIHS?

I am partial to Noctua but it has to be said that Phanteks T30 beat even IPPC hands down on performance while being significantly less noisy (if the extra 5mm of width is workable). I hope Noctua makes a 30mm fan at some point (in black!).

Same here, but I do prefer just plain black, everything else being equal. I realise Noctua brown is a thing for some people, but thankfully both their the Chroma.X and iPPC fans are black or predominantly black. That said, I did enable ARGB, just the default breathing rainbow, on my new AIO pump, and it does look pleasantly mesmerising through the side panel mesh lol

Ebuy7 is killing me. Basket items have poorly translated with names such as Suitable for Honeywell PTM7958SP phase change silicone grease cpu thermal paste notebook radiator graphics card 7958 or Honeywell PTM7950SP phase change silicone grease cpu thermal paste notebook radiator graphics card 7958SP silicone grease Bro @Talon, would you have a link for the item you so successfully ordered? @Mr. Fox doesn't look like derBauer has any LGA1700 delid tools available yet.. https://der8auer.com/oc-tools/

BTW an excellent in-depth 13900K tuning guide, focusing more on the dynamic ratios side: https://www.overclock.net/threads/asus-maximus-z790-extreme-and-intel-i9-13900k-a-tuning-guide-for-beginners.1801569/#lg=attachment_xfUid-12-1679227139&slide=0

What's funny is that Noctua has the same fan in Black (chroma.x). Exact same specs as the brown A12...

That's interesting. When upscaling videos, they can dedicate the entire GPU/all RR cores to the task, and usually there will be fewer FPS to deal with. Anyway, here is a FAQ: https://nvidia.custhelp.com/app/answers/detail/a_id/5448/~/rtx-video-super-resolution-faq No Linux support mentioned, so a bit of Linus love for them for now. No wonder though, as Nvidia is basically at Microsoft's mercy now and kind of vice versa which is ironic. An evil dual monopoly.