|
VelociBacon posted:I know that coffee lake comprises more than just the 9900k but I'd be pretty surprised if many people can run at 1.4v with the 9900k even with doing some serious custom loop cooling and even then I feel like you'd need to remove the solder TIM and replace it with liquid metal etc etc. Even deliding only nets you another 5C or so based on what I’ve seen. Like you said nearly everyone will run into temp issues before voltage becomes the problem, custom loops included.
|
# ? Oct 8, 2019 15:23 |
|
|
# ? Apr 24, 2024 13:18 |
|
VelociBacon posted:I know that coffee lake comprises more than just the 9900k but I'd be pretty surprised if many people can run at 1.4v with the 9900k even with doing some serious custom loop cooling and even then I feel like you'd need to remove the solder TIM and replace it with liquid metal etc etc. That depends on what you mean by "best", I guess? What he's saying is that at least on that Asus board he tested, LLC 3 (which has quite a lot of droop) has the nicest transients - you get neither overshoot over the idle voltage nor much undershoot below the load voltage (although he didn't really test with a high load, as far as I can tell). With the flatter LLC's you'll get a higher load voltage (closer to your configured voltage), but right when the load starts the voltage will undershoot the target voltage by a large margin, which means there's a risk of crashing. There will also be an overshoot above the idle voltage at load release. LLC3 on that particular board does have a lot of droop, but it doesn't have unpredictable and impossible-to-measure-without-an-oscilloscope voltage spikes and drops. Higher LLC's might enable you to get a higher load voltage, but that doesn't necessarily help if the undershoot before it stabilizes makes you crash. In that particular case all of the three highest LLC settings drop to 1.12v at load onset, even though they later stabilize at progressively higher voltages. If you're unlucky that might mean that LLC6, 7 and 8 all crash at load onset. LLC3 on the other hand drops to 1.08v and pretty much stays there (or stabilizes only slightly higher), making it more predictable. What I take away from this is that the three highest levels are probably not that useful, at least not for daily use. TheFluff fucked around with this message at 16:00 on Oct 8, 2019 |
# ? Oct 8, 2019 15:52 |
|
VelociBacon posted:I know that coffee lake comprises more than just the 9900k but I'd be pretty surprised if many people can run at 1.4v with the 9900k even with doing some serious custom loop cooling and even then I feel like you'd need to remove the solder TIM and replace it with liquid metal etc etc. Yup, it's probably hard to damage a 9900K with too high voltages if you keep an eye on the temperatures. TheFluff posted:What I take away from this is that the three highest levels are probably not that useful, at least not for daily use. Your posting sums it up perfectly. This is also what I've observed with my Maximus X Hero. 1.380V LLC7 (=supposedly almost flat under load but 1.430V indicated due to the way Asus Z370 measures) is barely stable but 1.370 LLC4 (indicated 1.312V) is absolutely stable, runs cooler and is likely better for overall longevity. Sadly many people — including der8auer — still suggest flat LLC settings.
|
# ? Oct 8, 2019 16:24 |
|
You talking about the Maximus X Hero (which I also have) reminds me that I still never bothered to get my 8700K stable at 5GHz. Today is the day it's finally happening though. Off to a promising start with OCCT small dataset just instantly hanging, but then I realized the culprit seems to be uncore. It hates 46x but 45x seems fine. Time to run some long rear end realbench sessions.
|
# ? Oct 8, 2019 21:33 |
|
TheFluff posted:Time to run some long rear end realbench sessions. We're you successful?
|
# ? Oct 9, 2019 04:02 |
|
TheFluff posted:You talking about the Maximus X Hero (which I also have) reminds me that I still never bothered to get my 8700K stable at 5GHz. Today is the day it's finally happening though. Off to a promising start with OCCT small dataset just instantly hanging, but then I realized the culprit seems to be uncore. It hates 46x but 45x seems fine. Time to run some long rear end realbench sessions. It helps to remember the graphs by multiple people that show that actual Vcores are ~132mV lower than what the software readout indicates and can’t be higher under load than idle because the VRM isn’t capable of it. That also explains to me why Asus’ auto OC settings are running seemingly high voltages of 1.45V and more, because that’s closer to 1.3V after measurement errors and vdroop and thus might be fairly safe.
|
# ? Oct 9, 2019 07:09 |
|
GeneticWeapon posted:We're you successful? Not looking all that hopeful, unfortunately. Set 1.4v adaptive mode in BIOS (which means VID under AVX load will be ~1.44v), LLC 4. Realbench pulls 165W or so (plus 150W to the GPU) and ends up at 1.38v vcore, software measured, and although it's not crashing or BSOD'ing it's still throwing out the occasional WHEA error. OCCT small is still a complete no-go, instant errors. I could go even higher on voltage - cooling isn't a problem at all, I'm at like 70C with the NH-D15 at 165W to the socket - but it's starting to feel a bit uncomfortable. Already at these settings I'm seeing vcore max out at 1.43v or so under low loads. Will probably gently caress around with it some more, but at this point it looks like I need an AVX offset for this (which I don't want because it fucks with the transients even more and it's annoying to stability test), or I'll have to go back to the 4.9GHz I've been stable with for almost two years. TheFluff fucked around with this message at 09:34 on Oct 9, 2019 |
# ? Oct 9, 2019 09:19 |
|
The AVX offset on the MXH also seems to throttle all gaming workloads down to the offset frequency, making he whole overclocking exercise beyond the offset bit pointless depending on what workloads you run. Other Z370 boards aren’t as sensitive. My 8700K seems to be about the same as yours, though I run it at 5GHz/1.375+0.01 LLC4 and 165W PL1 so that it clocks down only for sustained AVX workloads (which I never run outside stability testing).
|
# ? Oct 9, 2019 09:41 |
|
eames posted:The AVX offset on the MXH also seems to throttle all gaming workloads down to the offset frequency, making he whole overclocking exercise beyond the offset bit pointless depending on what workloads you run. Other Z370 boards aren’t as sensitive. Hm, lowering PL1 is an interesting idea. I currently have it at 180W but as you say, sustained AVX workloads are really unrealistic. Worth a shot. e: hm, nope. Doesn't really seem to have improved things TheFluff fucked around with this message at 11:45 on Oct 9, 2019 |
# ? Oct 9, 2019 09:48 |
|
Palladium posted:Whats the consensus for max safe 24/7 VCCSA/VCCIO voltages for CFL chips? My Micron E-dies are coming and I would wanna drive them at 3600C16. I have an 8700k on a z390 Aorus Master. Running a 4 dimm set of 3200/cl16 e-die at 3800/cl16 with pretty decent timings (16/20/20/40) but I haven't spent much time trying to tighten it up further.
|
# ? Oct 10, 2019 03:53 |
|
Been a while since I messed with overclocking -- I've got a 9700k with a MAG Z390 Tomahawk -- is the OP still accurate so far as principles are concerned? Any air cooling recommendations with a ATX mid-tower case (Fractal R5)?
|
# ? Dec 4, 2019 20:53 |
|
tehinternet posted:Been a while since I messed with overclocking -- I've got a 9700k with a MAG Z390 Tomahawk -- is the OP still accurate so far as principles are concerned? Any air cooling recommendations with a ATX mid-tower case (Fractal R5)? Most of the principles I'd say are accurate still, but the practical side of things has generally been streamlined quite a bit these days. For a daily overclock you don't really want to disable any of the power management features (C-states, etc) or set a fixed vcore voltage. Basically, you want to use adaptive vcore mode, and the knobs you can tweak are basically the frequency multiplier, the target vcore voltage at the max boost frequency, the power limit and the load line calibration (don't go too high with this, leave it at one of the middle settings). Don't gently caress with the bus clock (BCLK), leave it at 100MHz or you're likely to get all sorts of weird issues with PCIe devices and other peripherals. And make sure you are clear on the difference between VID (basically target Vcore) and actual Vcore and that you're reading the right values in HWinfo64. Noctua NH-D15 is the gold standard of air cooling. Main downside is that it's big, heavy and sort of expensive, but it's a really solid buy and can almost certainly be carried over to your next build - the warranty for the fans is like 6 years IIRC and if your next build's CPU socket has an incompatible mounting system they've historically had a standing offer to ship you mounting brackets for new sockets for free as long as you have proof of original purchase.
|
# ? Dec 4, 2019 22:15 |
|
The NH-D15 is also reasonably quiet. As long as you can fit it, it's one of those things that I think most people are happy they spent the money on. Considering at this point you can reasonably run a CPU for 5+ years without really needing an upgrade, spending an extra $40 or so bucks on a cooler makes a lot of sense. Plus it was the nicest unboxing of a freaking CPU cooler I've ever experienced, and you cannot put a price on that.
|
# ? Dec 4, 2019 22:23 |
|
Lockback posted:The NH-D15 is also reasonably quiet. As long as you can fit it, it's one of those things that I think most people are happy they spent the money on. The black nhd15s is also good. I have that and a Morpheus 2 in itx.
|
# ? Dec 5, 2019 00:25 |
|
TheFluff posted:Most of the principles I'd say are accurate still, but the practical side of things has generally been streamlined quite a bit these days. For a daily overclock you don't really want to disable any of the power management features (C-states, etc) or set a fixed vcore voltage. Basically, you want to use adaptive vcore mode, and the knobs you can tweak are basically the frequency multiplier, the target vcore voltage at the max boost frequency, the power limit and the load line calibration (don't go too high with this, leave it at one of the middle settings). Don't gently caress with the bus clock (BCLK), leave it at 100MHz or you're likely to get all sorts of weird issues with PCIe devices and other peripherals. And make sure you are clear on the difference between VID (basically target Vcore) and actual Vcore and that you're reading the right values in HWinfo64. This owns. I was about to pull the trigger on some Noctua fans anyways so awesome. Glad I got a full tower now because that cooler looks thicc in the best way.
|
# ? Dec 5, 2019 05:35 |
|
tehinternet posted:This owns. It is.
|
# ? Dec 5, 2019 14:50 |
|
Do you peg the rear CHA fan to the CPU fans, or do you just run them at independent speeds?
|
# ? Dec 5, 2019 16:34 |
|
tehinternet posted:This owns. I seriously hesitated before installing my Noctua because I wanted to just sit it on my desk (without the fans) as just a sculpture/conversation piece.
|
# ? Dec 5, 2019 18:17 |
|
ZekeNY posted:I seriously hesitated before installing my Noctua because I wanted to just sit it on my desk (without the fans) as just a sculpture/conversation piece. You're going to like this build then https://www.youtube.com/watch?v=N-z9PidYH4E .
|
# ? Dec 5, 2019 19:12 |
|
Does that thing make coffee?
|
# ? Dec 5, 2019 19:16 |
|
ufarn posted:Do you peg the rear CHA fan to the CPU fans, or do you just run them at independent speeds? I found there is so much airflow it literally doesn't matter.
|
# ? Dec 5, 2019 19:35 |
|
Perplx posted:You're going to like this build then https://www.youtube.com/watch?v=N-z9PidYH4E . Oh man, want!
|
# ? Dec 8, 2019 01:45 |
|
TheFluff posted:Noctua NH-D15 is the gold standard of air cooling. Main downside is that it's big, heavy and sort of expensive, but it's a really solid buy and can almost certainly be carried over to your next build - the warranty for the fans is like 6 years IIRC and if your next build's CPU socket has an incompatible mounting system they've historically had a standing offer to ship you mounting brackets for new sockets for free as long as you have proof of original purchase. I am fairly certain you could kill a man with a NH-D15; that thing is massive. I still run a Thermalright Ultra 120 at home (mostly out of laziness); I am pretty sure they have sent me two sets of socket adapter kits / brackets to keep it working on newer sockets.
|
# ? Dec 19, 2019 19:06 |
|
B-Mac posted:Even deliding only nets you another 5C or so based on what I’ve seen. Like you said nearly everyone will run into temp issues before voltage becomes the problem, custom loops included. Can't recall if it was the 9900K or 8700K but JaysTwoCents actually did a test with just lapping the CPU & knocked temps down like a solid 8-10 degrees in testing. The IHS itself wasn't anywhere near flat, he was spinning it like a top on his bench
|
# ? Dec 19, 2019 20:21 |
|
BOOTY-ADE posted:Can't recall if it was the 9900K or 8700K but JaysTwoCents actually did a test with just lapping the CPU & knocked temps down like a solid 8-10 degrees in testing. The IHS itself wasn't anywhere near flat, he was spinning it like a top on his bench Yeah, I saw that one. I think he lapped and delidded it, but the conclusion was that because of just how bad the hotspot was because of the uneven IHS, that the lapping did way more for the heat than the delidding did.
|
# ? Dec 19, 2019 22:07 |
|
Delidded 5ghz 8700k can stay below 75c on a nh-d15 though And you dont have to deal with the hassle or having to worry about water stuff in water cooling either. Albeit the AIO systems are fairly good nowadays.
|
# ? Dec 20, 2019 08:11 |
|
I'm not overclocking at this point but this might be a better thread to ask- I want to use software to drive a PWM pin for my case fans based on either cpu or gpu activity. Any suggestions? currently running stock cooling on a 3800x instead of my nh-d14 because noctua snailmails the new mounting kits from austria and it's late
|
# ? Dec 26, 2019 05:27 |
|
A lot of motherboard fan settings in BIOS allow you to set fan curves for each header based on GPU temp. TBH it won't make a huge difference but you can usually find it there. I'm pretty sure I have this option on my Aurus Ultra. E: you better be overclocking your GPU, there's no downside!
|
# ? Dec 26, 2019 05:34 |
|
Something tells me I'll never get it past 115% tdp but I'll keep trying! Mostly I just want to waste time playing with the levers, and if I can bring temps down a little without making the noise worse I'll get another little dopamine hit
|
# ? Dec 26, 2019 05:55 |
|
poverty goat posted:I'm not overclocking at this point but this might be a better thread to ask- I want to use software to drive a PWM pin for my case fans based on either cpu or gpu activity. Any suggestions? Argus monitor.
|
# ? Dec 27, 2019 13:59 |
|
B-Mac posted:Argus monitor. Yup. Ever since Speedfan ceased to be updated, this paid alternative is the best thing I could find, too.
|
# ? Dec 29, 2019 01:50 |
|
Argus Monitor should be in the OP. It's weird and ugly but pretty easy to use, and has been very reliable for me.
|
# ? Dec 30, 2019 08:38 |
|
When using DRAM Ryzen Calculator, are those pretty hard limits? I'm running at CL14 3466 stable right now (by using the ryzen calculator timings, RAM was rated for CL15 3600) https://www.newegg.com/g-skill-16gb-288-pin-ddr4-sdram/p/N82E16820232306?Item=N82E16820232306 I'm just lazy and don't want to have to CMOS clear everything and re-type everything in if it won't boot trying to reach higher levels
|
# ? Dec 31, 2019 17:31 |
|
Thom P. Tiers posted:When using DRAM Ryzen Calculator, are those pretty hard limits? I'm running at CL14 3466 stable right now (by using the ryzen calculator timings, RAM was rated for CL15 3600) Most motherboards will completely reset the CMOS by itself if the RAM fails after a couple of boots.
|
# ? Dec 31, 2019 17:37 |
|
Thom P. Tiers posted:When using DRAM Ryzen Calculator, are those pretty hard limits? I'm running at CL14 3466 stable right now (by using the ryzen calculator timings, RAM was rated for CL15 3600) you can probably get higher clocks with slightly looser timings. In the ryzen-3000 series world you'll want to max out your infinity clock and worry about timings later.
|
# ? Dec 31, 2019 22:08 |
|
KillHour posted:It is. Has anyone researched this setup where the rear case pulling exhaust is so close to a pushing CPU heatsink cooler? What happens if the rear case fan pulls less air volume than the heatsink's chain of fans push? If they're mismatched it seems like it would cause a buffet effect rendering the last CPU heatsink fan a noise maker while air gets forced out of the sides of tower.
|
# ? Jan 1, 2020 03:45 |
|
I usually just take out the case fan in that situation.
|
# ? Jan 1, 2020 03:53 |
|
Fabulousity posted:Has anyone researched this setup where the rear case pulling exhaust is so close to a pushing CPU heatsink cooler? What happens if the rear case fan pulls less air volume than the heatsink's chain of fans push? If they're mismatched it seems like it would cause a buffet effect rendering the last CPU heatsink fan a noise maker while air gets forced out of the sides of tower. While it's not intuitive I think those fans are far enough from the case fan to leave it in and running to your benefit.
|
# ? Jan 1, 2020 06:11 |
|
Fabulousity posted:Has anyone researched this setup where the rear case pulling exhaust is so close to a pushing CPU heatsink cooler? What happens if the rear case fan pulls less air volume than the heatsink's chain of fans push? If they're mismatched it seems like it would cause a buffet effect rendering the last CPU heatsink fan a noise maker while air gets forced out of the sides of tower. You could try to visualize the flow with a stick of incense or something else that generates visible smoke, that would tell you for sure
|
# ? Jan 1, 2020 18:58 |
|
|
# ? Apr 24, 2024 13:18 |
|
I have a 6600K Skylake I never got around to overclocking until now. I have an ASrock Z170 pro4s and evo 212 with it. First I'm trying to get some baseline benchmarks.I started using RealBench and I noticed the image editing occasionally only puts ~50% load on the CPU for the first 30s or so which tanks the score. Anyone know the cause or if I should be using something else for short tests? What sort of temps are usually expected running stock under load? I have one core that is consistently 3-4C cooler than the rest which isn't a worry but I'm curious about. I feel like I am reading a lot of conflicting information from guides on how I should deal with voltage and which power options I should have on/off. Should the end state should be I have an adaptive offset but maybe I have a static vcore while testing but different places say different things. Similar for things like if or when to have things like speed step on or off.
|
# ? Jan 1, 2020 19:30 |