Etern4l

Ah, OK. Still, the point stands - DLSS-quality upscaling of generic video content would be an impressive achievement, as DLSS 2+ in turn requires very rich and precise additional metadata which is not directly available when upscaling general video content, and thus upscaling video games is an easier task.

Well, all depends on how well it works in practice. Did they take the unorthodox measure and optimise it for the likely target content? :) On a more serious note, it would be quite impressive if this worked well on a variety of source material. "Super resolution" feature on my LG TV does zip, and AFAIK DLSS has to be trained for each specific game. Also wondering if it was technically necessary to restrict this to Ampere+, given that 2060 Super performs about the same as 3060 (and clearly 2070 and 2080/Ti outperform it), or just another d... move from Nvidia.

I don't remember what the extra Netflix charge is, but it's fairly minimal, and clearly any Netflix user with 4K display should just get that, instead of falling back to smart upscaling. As for the main use case for this, I suspect there will be an overlap with the main uses of 3D and VR lol

Thanks for that bro @tps3443. Obviously, always best to stream at native 4K, but wait, what content is typically provided at a variety of resolutions below 4K? :D

For reference, a full single run of memtest86 consists of something like 13 phases, would run probably north of an hour on 32GB, and would still sometimes miss issues (which is why they default to 4 runs). Might be worth a check overnight for the peace of mind. It's not that the RAM might necessarily faulty, but perhaps the settings are too aggressive.

Very sorry to hear man, well benchmarks can make things come out of the woodwork. FWIW, it looks to me like we cannot exclude a hardware issue, possibly RAM (remember that out of bounds error?). In your shoes, I would ensure cooling is 100% in order and run at least one full cycle of memtest86 to begin with. I'd also try running Linux from a live USB. If issues continue there, it's likely hardware (would be an unwelcome conclusion, but at least you'd know where you stand)... Good luck bro, this puts my 13900K issues in perspective. In this case we can assume the IHSes we are comparing are made of the same material. Are you saying that an IHS 10% larger will provide zero additional benefit, or that the benefit will be small? The latter I would agree with. Surely cooling efficiency is not linear in the area, but still in principle more IHS area is better, with diminishing returns. By the second law of thermodynamics, heat flows from hotter to colder objects (whoa!), but what that means is that initially heat will prefer to flow to that excess copper, and from there - through the extra contact area - transfer to that coolant at a higher rate than it would have otherwise. This is clearly preferable to a situation where this extra route of escape for the heat does not exist. We haven't been bashing BGA manufacturers on account of their small heatsinks for no reason. That's the exact same idea. Now, clearly there are diminishing marginal returns, but that depends on the initial size of the IHS. Imagine the IHS is infinitely large, you are not going to gain anything by adding an extra 10% lip around it, as heat will never get there in the first place. In contrast, take a look at that pitifully small LGA1700 IHS being asked to pump 300W+ lol Going back to the original question: what improvement would an additional 10% IHS size on LGA1700 yield. I am afraid, we are not going to be able to derive the exact formulae here, nor will we be able to test this ourselves unless we find an aftermarket original 12th gen Rockit IHS, but yeah - if I were to guess probably somewhere around 5% (but not zero!). I mean apparently even Linus managed to use it correctly (then proceeded to do his usual thing: drop stuff etc. lol) But then of course that's an ad. What does that even mean? Do you mean a mod expressed a concern? It's not like has been publicly tainted somehow, there is nothing on his profile. On the other hand, a VAC ban is for life :) Edit: nm, Reciever clarified. For not observing a prior warning about fairly strong 1930s style content he posted (from what I gather).

Apart from the fundamental physics, if not common sense, which dictate that a larger IHS is strictly preferable (when matched with an appropriate coldplate, sorry Arctic), the opposite has also been shown: https://www.gamersnexus.net/guides/3238-custom-copper-ihs-tested-on-intel-i7-8700k-cpu-rockit-cool https://www.gamersnexus.net/hwreviews/3295-ryzen-custom-copper-ihs-tested-on-raven-ridge-apus BTW This is an interesting thread: https://linustechtips.com/topic/1448543-psa-the-rockit-cool-copper-ihs-for-the-i9-12900ks-will-make-your-cpu-incompatible-with-most-custom-socket-upgrades/ People are saying that the Rockit copper IHS is not compatible with contact frames (how could it be if its 10% larger? The frame fit over the original IHS is very tight). The the Rockit owner presumably shows up (around March 2022) and says they developed a new IHS which is compatible with frames. People then ask for data on whether the new smaller IHS + frame performs any better than the old larger IHS and get crickets. Perhaps what @Mr. Fox used was the new shrunk IHS? Edit: yeah, looks like Rockit deprecated the larger IHSes. What people have been getting lately is the standard size IHS (which apparently still doesn't fit well!), defo the IHS size discussion is academic with regards to LGA1700.

Jay wasn't impressed with the EK delidder, might not be the same as advertised by derBauer: I'm not sure anything is available directly from derBauer (as a rule of thumb, very little is ever actually in stock). Will check later. Care to elaborate? In my case the coolant is cold-ish, meaning there is not enough heat exchange happening. Additional IHS and coldplate surface areas should help with this? If I look at the Xeon and EPYC, and even Threadripper IHSes, they are substantially larger than LGA 1700, while the CPUs don't necessarily push that much more power by default. Similarly, CPU water blocks typically have much larger coldplates than most AIOs. Nice, good luck and please keep us posted. Yeah I'm not sure warranty would help me here. "Oi Intel, used to be able to pull 330W in CB23, now 2 cores are hitting 100C and unable to pull more than 240W". I doubt this would be an acceptable claim. Even if the warranty is honoured in this case, how long will it take to process, and what will be the quality of the replacement CPU? Warranty would probably help in case of the CPU dying, which is something there has been no recorded case of in the last 40 years or so ;)

Probably not the ideal thread to ask this, but does anyone have a good solution to monitoring fans in Linux? Hardinfo is lacking the info at the moment.

One thing I discovered quickly is that the contact was poor on the top 25% of the IHS. I was able to correct that to near perfection by varying the tightness of top vs bottom contact frame screws (over a couple of time consuming iterations), and this resulted in significant temp improvements (10C?), however, not enough to restore performance to the initial state where I was able to pull 320-330W in CB. My current hypothesis is that because of essentially incorrect mount, the solder under the IHS sustained some sort of damage over time, truly necessitating a delid, which does actually seem like a reasonably straightforward process. I am assuming that Rockit is still the best tool to use at this time, and this seems like the only place to get it from: https://rockitcool.myshopify.com/products/copper-upgrade-kit-intel-12th-gen BTW am I correct in assuming that the Rockit copper IHS will not fit the contact frame? They advertise it as having 9% larger contact area (something Intel should have considered, in addition to using a less ****** ILM and general IHS design). If that's the case, would filing the interior of the contact frame to make it fit the larger IHS work, or would it just destroy the frame? Failing that, I would have to go back to using the ILM (if I wanted to benefit from the larger IHS), potentially risking further bending issues. What do you mean by that? Reapplying shortly after the first application (in case of any copper surfaces)? BTW @Papusan I'm assuming you delided your 13900K. Which tool and IHS (if any) did you use?

Looks promising. Their pre-built boxes are definitely on the more premium side, probably worth considering for particularly heavy NAS use.

Zen has nothing to do with getting stoned, bro - just saying ;) Didn't know this stuff was legal in Canada. It's really bad for adolescents and brain development, hope there are some age restrictions at least.

Of course, but the less obvious question is whether running CB23 in Wine slows it further down or not. Not really too in-depth on the tuning details, but interesting nonetheless:

That's a very good Bedroom score on Windows. My 5500 is specific to Clear Linux (normally it would be 6k, but my CPU has issues and running in "eco mode"), on Windows 11 it was around 4000-4200 for me IIRC, same on Ubuntu and other slower distros. Indigo is taxing (too bad you couldn't run the Supercar scene, that's notably more intense), but this must be platform-specific, since on Intel I find CB R23 to be more demanding still.

Not VM, technically it's Wine, which is a light Windows emulation layer, although I actually used a handy tool called Lutris which sits on top of Wine as well as other emulators and software repos: https://lutris.net/games/cinebench-r23/ I'm tempted to try running 3DMark, but would need to dig up my Steam account from somewhere. Re Indigo, should be able to run this on AMD no problem: https://openbenchmarking.org/test/pts/indigobench-1.0.2&search Are you running version 4.4.15 from here: https://www.indigorenderer.com/indigobench ? Ah, perhaps it's some sort of issue with Win7 :(

Sorry this was on Linux, so there might have been a penalty due to the emulation layer (I'm not sure, my Windoze 11 died trying to upgrade itself against my will). Also bear in mind max power draw was around 200W, not 236W. Do you want to have a go at Indigo Benchmark? Would be interesting to see how well 7950X does there.

A few entries from me, CPU runs in eco mode at around 200W (voltage limited):

Those new to Windows on Linux will probably appreciate Lutris. CB32 was click and play. Gaming-oriented there are ready to use presets for apps as well, e.g. https://lutris.net/games?q=photoshop Thanks to @Mr. Fox for the tip.

I wrestled with my 13900K for a while today, and managed to get it to a usable state by taking a few hints from derBauer's video on tuning this CPU for efficiency, as well as your advice guys, @Mr. Foxin particular. Fixed ratios(51/42), fixed voltage ((ring 0.975V, Vcore: 1.125V). On top of that a new AIO with 25% larger coldplate, 50% faster pump, and T30 fans, but 11mm thinner radiator (cold during tests anyway). Used some throwaway paste there, don't remember MX-5 or TF7. The result: 38K in CB23 at around 200W... Very usable. Temps still reach 90C so something is wrong and deliding will be necessary. Indigo is also not far off 300W results: GPU is doing OK vs @Mr. Fox's OCed 6900XT on Linux, roughly in line with the spec advantage. conservative OC +70/+2000

"Today Gelid proudly introduces a phase-change TIM you don't have to buy from ebuy7". Oh wait.... Fair enough, I didn't realise laptop report VRAM temps now...

I reckon the only temps the poster could have been referring to there were GPU. That only makes sense if the K5 Pro application failed as expected (low pressure/density), since less or no heat from VRAM would transfer to the heatsink which in turn would result in deceptively lower GPU temps. Then one day space invaders will arrive and the poster will wonder which planet they came from ;)

Does bro @Talonrun it 24/7 close to full load? I was quite happy for the first 2 months or so. The 12900KS lasted 3 times as long in much more severe summer conditions with worse cooling. I'm not disputing Intel being a nice fun hobby platform though. Dealing with all the overheating, bending, plus risking ruining the CPU during a delid can be very exciting ;) You can't run an EPYC in a workstation? The rule of thumb on Intel pricing seems to be take equivalent AMD price and multiply by 3: https://www.amd.com/en/processors/epyc-cpu-selector I would still prefer an Intel CPU due to Thunberbolt and better numerical software support, but boy do they need to up their game by a lot and fast.

I appreciate Intel have to fight with the tools they have at their disposal, but Sapphire Rapids is broadly a fail vs AMD EPYC Genoa. A dual Xeon system gets beaten by a single AMD CPU. Just like they failed with the 13900K - a CPU which requires delidding out of the box for serious use.

Thanks for this. Any idea about the difference in specs between 7950 and 7958? Thermal conductivity is the same. @cylix

As per @Mr. Fox is more a slap in the face than a meaningful reduction. My guess is that it's a combination of foreign exchange adjustment (the dollar has lost a bit of value vs the euro recently), and an exercise in operant conditioning of sorts. The idea could have been to train consumers to associate a change in price of 2.5% in this case, with the words "price reduction", whereas the correct terms are "rounding error", "spare change" and "sad joke".