G2 LBC for AM5, not the standard. Shouldn’t be much behind stock Phantom Spirit with fan swaps on the Phantom Spirit costing way less and running a little cooler noise-normalized. G2 LBC also gets performance demolished at lower cost by plenty of midrange 360 AIOs that’ll probably be about as reliable. If you’re really worried about EPDM leakage Light Loop 2’s roughly G2 pricing and has a supported fill port.
If going 140 air, which is 150 mm fin width with the G2, make sure you’ve checked the dGPU, case, DIMM, and M2_1 clearances. M2_1 in particular’s often overlooked and I wouldn’t set up to have to change KryoSheet every time the socket’s accessed. Haven’t measured much ΔT on the DIMMs with top mid-front intake but it can worth giving some consideration to anyways.
Not really, for the reasons discussed. How much risk attaches to ordering the dGPU later depends on return windows and how well you’ve checked all the dGPU clearances and done thermal pre-design.
Have you done the fan affinity maths or comparative measurements? Unless alternate parts selection’s poor my experience is the stock 2x180s’ll noise-normalize a little warmer than 3x120, consistent with what you’d expect from fan affinity. I’ve also measured with and without an exhaust fan and at different exhaust speeds. While there’s a range of temperature changes, the main result I got was somewhat lower M2_1 temperatures and somewhat higher VRM temperatures. Pretty decent tradeoff.
The positive-negative pressure thing doesn’t seem to be evidence based, just people repeating the idea absent supporting measurements. Never seen anyone do the math or measurements to get the actual airflow balance at the operating point, either. It’s not a well posed idea anyways as cases aren’t constant pressure volumes because of the presence of interior fans and obstacles. Ignoring that single and dual tower fans and GPU fans are active devices which pull their own air is weird because all you have to do to verify they’re influential is turn up those fans up until you can feel the flow.
FWIW, about half of what I build would be considered negative pressure and half positive. There isn’t a clear difference in dust accumulation between the two, even in builds that have been running pretty much 24x7 for 2+ years, but if anything it’s the negative ones which have less dust. The amount of dust available for intake aside, the best predictor of dust accumulation seems just to be flow volume.
Can you describe your workloads and ΔTs? Having built and measured 5950X, 7950X, 9950X, and some other recent Ryzens I’ve found there’s no difficulty pushing any of them to their thermal limits in a range of AVX workloads, including the one mentioned above, under dual tower in typical ambients with a range of airflow cases. Including Torrent Compact.
As pointed out upthread, full load engages all cores and thus maximally spreads power density. So it’s less thermally demanding than lower thread counts that don’t reach PPT. All cores active’s also more likely to hit EDC or TDC rather than PPT in SIMD workloads, making it less thermally stressful. However, my experience is Granite Ridge’s boost profile pulls consistently more power at the current limit than Raphael or Vermeer.
Essentially what I see in all that is 65 to ~75 °C ΔT’s common with multithreaded AVX at ~23 dB(A). At ~142 W cores might stay a couple degrees under Vermeer’s 90 °C at the lower end of typical ambients. ~162 W Granite Ridge can clear 95 °C if you’re careful about the mount and paste and thread distribution. So it’s unsurprising ~200 W easily throttles without going rad. I don’t usually see 9950X under Phantom Spirit and GX-14 throttle below ~185 W at ~23 dB(A) though.
Yes and no. Raphael topside vcache makes cores hotter than Granite Ridge bottom side but that’s also why 7950X3D’s 162 W default PPT instead of 200 W PPT like the 9950X. That +38 W is a tipping point for a quiet dual tower build. Granite Ridge’s layout changes help but aren’t a resolution on their own.
I know people who do 250+ W air cooled in Torrents without exhaust fans and fairly low core temperatures. But they’re not running AVX, they’ve got the intakes at 1200 RPM, and have 3000 RPM fans on the cooler. That’s not quiet.
Is this a measurement based assertion? Because every time I’ve tried blocking off mesh in the airflow cases I’ve tested temperatures have increased. Since by default the fans spin up to try to compensate noise increases as well. Unless you recurve to stay noise-normalized and take additional temperature increase.
GN’s fan normalized testing finds much the same thing when compared across open area, so it’s not just me. Manufacturers’ and reviewers’ inability to communicate how airflow actually works is IMO a rant worthy topic. The measurements are easy and circumstantial evidence suggests quite a few (but definitely not all) case design teams are doing them. GN and a couple other reviewers are pretty close.
Mmm, Torrent [Compact]'s performance competitive with the alternatives mentioned but costs more and runs comparatively high on disadvantages and quirks. That’s enough tradeoffs I’d hesitate to call them great.
Lancool II mesh performance released about a year before Torrent (July 2020 versus August 2021, I think) and ~1.5 years before Torrent Compact. Lancool III was a few months later (July 2022) and Lancool 207 and Flux were around October last year. There’s a few other competitive releases in the past year or two, so the Torrents’ market position isn’t as commanding as it was the first part of 2022.
Torrent Compact’s one 3.5’s thermally good if set up right (see upthread) and Nano might be fairly similar but, yeah, the full Torrent’s two unventilated 3.5s are useless if the drives are doing much of anything.
My experience with 2.5 SATA is unventilated mounts are ok due to the ~550 MB/s limit. If it’s U.2/3, then no, but most cases have problems there. Lancool 207 might be the best two mount option I know of, followed by III, Flux, and II, but I don’t have enterprise drives to test with. For more than two mounts Meshify seems to be the only option without substantial tray, cage, or case obstruction.
