Etern4l

Yeah, 1 day would capture any curing behaviour (very pronounced with the Honeywell stuff), and1 week+ would be useful to assess pump-out resistance, although this is hugely dependent on heatsink fit, contact area size, heat flux, workload pattern etc.

The new ACE Z690, made in 2022 rather than 2021 (kind of along the lines of what you guys were saying about the Apex) is too good to be true. 1. The socket is different, the pins specifically look somewhat more slivery - hope that will prove more resilient and stable long term 2. Heatsink fit using the ILM ranges from very good on the original IHS to perfect on the copper IHS - no contact frame needed at all, and with just hand-tightened heatsink screws. Day and night vs the old setup with the contact frame and basically foolproof. 3. Memory stability is hugely improved: running 5000MHz with 28-36-36-36. Can't believe it, was never able to even train at CL28. With this it seems like it either doesn't train or if it does it works with no errors, in starrk contrast to the old board 4. CPU runs at lower voltages - currently testing 53/43 which achieves 40500 in CB23 (in Linux, which does about 500 points better than Windows), at 1.16V/235W, with temps in the eighties. Pretty good for an AIO. Dlidded CPU is absolutely fine on PTM7950+Apex, and definitely part of the early success story. The experiment continues..

Agreed if you are building a workstation with a $20,000+ budget. Which is what you would need (OK, maybe a bit less) to get in the vicinity of the raw performance provided by 13th gen + gaming GPUs rig if you are limited to using overpriced 2 slot compute/workstation GPUs. It' s a bit of a hack, I know, but a pretty cool one, if you learn to navigate things like the idiotic LGA1700 socket, and to manage power to avoid a meltdown. Host I/O bandwidth is often of little concern in compute tasks, so there is no appreciable penalty due to the TB connection. Notice how miners (used to lol) use 1x raisers to connect their GPUs.

Data analysis/ML stuff, quite memory intensive. TB ports provide the option to connect additional GPUs, 10GbE (if the board lacks this), or fast external storage - very handy and a major gap in the AMD offering from my perspective. Xeon boards also lack TB support from what I have seen so far which is a shame.

All makes sense. I incorrectly assumed Apex was the equivalent of Godlike. It's a XOC 2 DIMM board, a competitor of EVGA Dark, which is why I never took a closer look. I need 4DIMMs, E-ATX, at least 2 PCIe slots, two TB ports and best reliability/build quality/QC. ASrock Taichi aside, the main options are MSI ACE as well as Asus Hero and ProArt. Asus Hero failed the smell test due to the burning cap issue, while ProArt is weakly built, and had lesser memory OC potential. The ACE also did consistently well in benchmarks against those two Asus boards. MSI board specs are a bit more detailed/technical as well, which I like, for example they state things like TB controller model etc. For some reason though, Asus probably is valued by a lot of people, since availability tends to be worse. I suspect BIOS / Windows software or perhaps more aggressive launch pricing could be the reason (at the cost to QC possibly). Whatever, I didn't even review Asus options this time around since I saw @Mr. Fox's complaints about Linux compatibility of an Asus board and want none of that.

Clear. All this crazy depreciation (how much was the board at launch? $1k+?) is just stealth tax on enthusiasts and early adopters. Instead of releasing a properly working product, they push half-baked stuff out out as soon as possible, knowing full well people will lap up anything new and shiny (c) @Papusan, only to fix things in a later revision or refresh. Not really an Asus-specific point either, I'm afraid. I guess this puts EVGA in good light with their delayed releases, some things in life are worth waiting for a bit.

Cool, hopefully just a whale offloading. $800 on Amazon.

Looks like a crazy discount on the mobo? The question is: what's the catch?

This is just after application, right? Not at least a day, week, or ideally a month on?

Define "nice difference".. For a bencher engulfed in LN2 fumes, fighting to the death for the top spot on HWBOT, that extra 1% of realized performance would be very nice indeed. For most people though the difference would be very difficult to tell apart from noise, particularly in gaming and most other heavily optimised numerical benchmarks used in typical enthusiast scenarios. Whatever you do, do you require lots of RAM? If not, then marginally faster RAM is mostly a vanity purchase, as is a modern Alienware laptop itself :)

Just reordered the tried and true Z690 board at 40% discount from Amazon. Pretty much the same performance and features as the Z790 model when it comes to handling 4 2R SODiMM config, and zero risk of compatibility issues with the newer chipset. Heavily discounted means £400 lol Thanks for all the inflation Vlad BTW. Will then get a chance to get in touch with MSI support about the socket replacement. Will get a chance to see what @Mr. Fox has been raving about lol

Tell me about it bro. "Phew, I wanted to upgrade to Z790 anyway" Yep, my mobo is dead - broken pin. Needed 30x magnification to confirm this, nuts. Actually, not clear if it's worth getting a Z790 replacement. Interestingly MSI now supports 24GB and 48GB modules on both Z690 and Z790. Z690 is now almost half the price of Z790. Not sure the latter is worth the premium, def don't need more PCIe 5.0, or 7800 RAM support.

Def don't lose hope, was visible in large parts of the UK recently.

You'll know something is wrong if you are in like Texas or Southern Europe and see an aurora ;)

Good thing about worrying about such events is that you can actually meaningfully preprare, as opposed to say AI and/or nuclear apocalypses (probably going to experience both simultaneously, if I were to hazard a guess).

This seems significantly harder than LGA1200. It's very difficult to even make out the pins and a sewing needle is almost too thick for the job. I had to use a magnifying glass + strong glasses to make sense of what's going on. The pins are incredibly fragile, much easier to bend than on LGA1200 from what I could see on YT videos. Another one for the list of Intel's engineering fails with this platform, unfortunately, and again due to the socket being inadequately small. The AI advancements came too late to help Raja with that one. The biggest issue with the contact frame, as opposed to the ILM, is probably the fact that not being 4-handed robotsx and not typically having 4 wrenches/screwdrivers of the appropriate type available, we are unable to lower the frame evenly on our own, so there is always some lateral force which could induce some pins to bend. If the board survives, I will be taking a minimum 30 min to lower the frame lol Not tightening too much is also good advice, albeit it's less of problem with the Thermaltake frame, since it's thicker than der8auer's, hits the mobo and probably impossible to catastrophically overtighten.

@Rage Set @Mr. Fox I'm not sure if that's the root cause for sure, since I don't have the LGA1700 pin map, but upon repasting and reinstalling the CPU as a sanity check, I started getting code 55 RAM not installed, which seems to be a telltale sign of a socket contact issue. Cleanup of the CPU with acetone didn't help, and then I looked closely at the socket itself and spotted a bent pin, then proceeded to develop a new valuable Intel enthusiast skill involving a magnifying glass and a needle lol. It's nuts when you look at how tiny those pins are, but it seems I may have figured out how to fix this. I'm not sure when and why this problem developed, you kind of have to look closely at the right area to spot it, but if the board survives I will sure be more gentle with the contact frame, and take extra care to centre the CPU (of course, Intel couldn't have possibly designed the socket so there is less play!). Kind of regret installing the contact frame in the first place, and if I need to treat myself to a replacement board, I won't do it again.

We should draw the line somewhere, since some of this high-end hardware is outright scam. Take the 13900K as an example: advertised as running 54/43 clocks, but max 253W of power draw is supported. This CPU would maintain nowhere near those clocks at 253W, even with the heaviest undervolting. Most boards remove the power limit by default automatically voiding the warranty, and for a good reason, since the CPU (the solder/IHS) likely won't survive extended loads at higher power. An Intel laptop that doesn't support UV will perform terribly, so I would draw the line there, no matter the possible existence of any warranty-voiding hacks that work around that.

Pretty cool, thanks bro. The thing though is that to fully leverage this, would need some additional/external rads anyway. The D5 pump is great though. 4500rpm. At the moment battling new I/O-related instability. Standars benchmarks are fine, temps are fine, however, heavy I/O results in the SSD disappearing from the system. Could be the paste, could be the contact frame (it's a tight fit this time), or could be that I damaged the CPU. Will know soon.

Sounds like there is a price to pay for the toughness. I played around with K5 Pro recently. Wanted to make a liquid metal dam out of it. Almost completely useless. I was hoping it woild be reasonably dense, kind of like playdough. Nope it's more like toothpaste, horribly messy. I'm not clear how this would help unless used like a normal paste, in which case there are better pastes out there obviously.

Graphics cards. The setup above would require about 18cm of clearance above the fan. No way. Could probably fit a second pump there though if needed. How about something like this: https://www.watercoolinguk.co.uk/p/Alphacool-Eiswand-360-Solo-Black_65661.html https://www.overclockers.com/alphacool-eiswand-360-watercooling-kit-review/ Would need to be able to disconnect without draining the system to enable maintenance, and this does have quick release connectors. 2 pumps. The slow fans can be replaced presumably. Maybe this is even better an idea than getting a new case to do an internal setup, from the cooling performance point of view, especially if the graphics cards can stay on air (and they do).

Appreciate the idea bro, wouldn't work unfortunately, there is just about 10-11cm vertical clearance over there. Would need to reclaim all that space lost to the 3.5 drive cage area to be able to install a pump/reservoir internally. A bit of a lost cause. The only thing that would work with this is an external cooling setup, for instance something like this: https://www.performance-pcs.com/water-cooling/external-cooling-system/aquacomputer-aquaduct-240-eco-mark-ii-ext-watercooling-system-microprocessor-controlled-12v-pump-aq-11234.html Lame on its own (240 rad), but in combination with a hefty internal 360 in lieu of the AIO it would likely provide good improvement, maybe in conjunction with a second internal pump if I can find space for one. Something like that might actually be a good upgrade, will take a look.

People should just reject products which require heavy modifications out of the box to work properly to the advertised spec and reasonable expectations, in order to disincentivise manufacturers and vendors from doing a poor job. Otherwise it is a moral hazard situation of sorts, where certain consumers are saying: whatever crap design you guys put out, we, the very best of enthusiasts, are so enthused that we stand ready to buy it and fix it as best we can lol

Another, likely more realistic, way of looking at this would be: the cooling is inadequate, it's unlikely to be fixable, especially in view of the wretched inverted mobo, therefore the turdbook lacks further potential. High end 13th gen is just difficult to cool due to crazy heat flux. That's the idea I think. They don't want people to be able to service their laptops, not least because they make a lot of money on their service plans. The good news is that those CPUs are quite efficient at lower power levels, so people won't be sacrificing that much by running them at 125 or 150W.

I'm not sure using k5 pro instead of pads is that logical by default, depends on the gap I guess. Played around with it on the desktop and the mess is terrible. If you are doing something on a turdbook that leads it you into thinking k5 pro might be the solution, you probably shouldn't be doing it on that turdbook.