|
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 |
|
|
# ¿ May 16, 2024 02:01 |
|
Kerosene19 posted:I could also see the lights in my office start to flicker with the power draw... lol Call an electrician before you burn your house down.
|
# ¿ May 4, 2020 16:41 |
|
I've been working on getting a stable OC for a 10700K, and I have two prospective OCs that I might be able to use. The Vcore and temperature below were measured in Cinebench R20 over 11 renders. x52 | +0.115 | 1.385 reported Vcore | 96C Package x51 | "Auto" offset | 1.27 reported Vcore | 80C Package I've only done a little bit of OC in the past, so I don't trust my judgement. My initial thought is that my cooling is insufficient for x52, so x51 is my best bet, even if I'm pushing it at 80C. I'm running a 6 hour OCCT stress test as part of the x51 validation. Is there any reason I should be ignoring Cinebench (or P95 and other superheavy test) temperatures that would justify the x52 multiplier?
|
# ¿ May 31, 2020 10:12 |
|
VelociBacon posted:I wouldn't "pay" 16C for 0.1GHZ even if the cpu is rated to go to 100C or whatever (I haven't looked at this stuff for the new intel CPUs). It just means louder cooler even with a good cooler. I can't get x51 P95 stable at 1.27 Vcore. If I take it to x51 at 1.38 Vcore, it seems fine... but as you said it gets to 100C in about a minute. (I have the throttling limit set to 105C for testing purposes, and I manually end any test that hits 100C. Once I'm "done" it'll be going back to 100C) There might be some play in there somewhere to get P95 stable, but honestly if I can be stable on tonight AIDA64 at 1.27, I'd be happy with it. Dogen posted:They do thermal throttling as well, so you might be losing speed with that much of an increase. Maybe overclocking overrides that though? I havent seen much on OCing them yet. I think they're just a pain in the dick to get. I got mine on launch day, and the biggest thing for me right now is regretting not spending the extra $100 for a 10900K instead. edit: Also if you already have a recent 14nm Intel chip with SMT, there's not much of an upgrade. I was upgrading from Ivy Bridge. Most of the notes I've been using for decision-making are on the 9900k, which seems to be roughly equivalent to the 10700k in terms of performance and ability. Warmachine fucked around with this message at 01:11 on Jun 1, 2020 |
# ¿ Jun 1, 2020 01:04 |
|
VelociBacon posted:Sorry if this seems dumb I'm not parsing this maybe correctly. Going from 1.27v to 1.38v is skipping a number of steps - you should be incrementing the voltage up from 1.27v by 0.05v or so while testing to see where it gets stable (or is that how you arrived at 1.38v?). That's how I arrived there. I'm using offset voltages, incrementing anywhere between 5mV and 15mV steps, then dialing in when I hit a stable/unstable point to get as tight a voltage curve as I can. What I'm posting here is Vcore readings from HWiNFO when under load. I think I found a stable zone around 1.32v Vcore after fiddling with it some more (setting SVID to "typical" and using a negative offset of -0.07). It still hits 100C inside of three minutes in Prime95 with AVX2 instructions, but at lower voltage it fails before hitting thermal limits. I have one more hail mary to try and get better thermals, but I'm pretty sure this is the best I'm going to get using a stock Eisbaer LT. Dogen posted:I have a 6700k and it just doesnt seem like the 10x00 is quite worth the trouble, especially since the socket is going to last for this and I think the next chip. I was thinking about upgrading since I just got a high refresh monitor but I think Im going to keep waiting another year or two. I'm inclined to agree. Like I said, I went from Ivy Bridge 3570K to this. 10900K wouldn't be too bad--the binning is a lot tighter from what I am hearing--but I'm not going to say yeah definitely upgrade if you're already on 14nm silicon. And, honestly? Knowing nothing else about you I'd bet you're GPU bound before you're CPU bound on things where that refresh rate matters.
|
# ¿ Jun 1, 2020 03:15 |
|
VelociBacon posted:Oh, gotcha, cool. I personally don't aim for p95 stability, but I use my most demanding actual day to day usage as the stress test. Battlefield 5 is extremely taxing (AVX) and I would suggest you use that to test stability as it worked well for me. Nothing else I've played has come close, this is with RTX on but I don't believe that would affect CPU load. Yeah, I think I'm just going to ignore P95 stability, at least until I improve the cooling loop. My demanding games are all CPU bound anyway (poo poo like Rimworld and KSP), and workstation tasks would be light statistical work, which sound be roughly approximated by things like OCCT and AIDA64. I don't have BF5 so that test is out. For what it is worth, my computer is basically for single player gaming, entertainment media, and tinkering.
|
# ¿ Jun 1, 2020 03:35 |
|
Lockback posted:How are temps without avx2? While those instructions aren't rare it's not like anything is going to run then down that pipeline for minutes at a time, full throttle. Without AVX2 temps don't get above 80C. I'm running some AIDA64 right now. The -70mV offset wasn't stable after 4 hours of AIDA, so I'm trimming it back.
|
# ¿ Jun 1, 2020 16:20 |
|
VelociBacon posted:Sounds like it didn't stick! What does CPU-Z or hardwareinfo64 show? Read from these. I don't recall things like Cinebench ever reporting my overclocks. HwI64 is my go-to just because I can get as much or as little info as I want when I want it.
|
# ¿ Jul 20, 2020 15:06 |
|
Well, I beefed up my cooling system. I replaced the Alphacool AIO with a custom EK loop sporting an thicker radiator. So I'm back chasing overclocks on it, only I'm running into the same problem I had two months ago: wherever I go to see what people are getting for voltages on 10700k chips, I can't parse how they are claiming to get some absurdly low voltages. Especially when I go on Buildzoid's channel last night and see him doing a 10700k overclock needing 1.46 Vcore in the BIOS. I know silicon lottery is a thing, but something seems screwy here, and I'm starting to suspect I'm looking at the wrong numbers? So I go and look at various sensor software, and none of them agree. HWinfo produces a set of minimum and maximum values for Vcore as well as VID. CPU-Z as I understands reports VID, but doesn't match literally any reported results from HWinfo. HWMonitor agrees with CPU-Z (being the same company I expect it would), but that doesn't give me any confidence. So here are some numbers for the 5.1 I'm working on. BIOS Voltage: Manual, 1.46 HWiNFO 64 Min (At load): 1.359 Max (At idle): 1.447 VID Min: 1.35x VID Max: 1.434 CPU-Z: Idle: 1.304 Load: 1.200 Question 1: Which of these is the important one? My brain says Vcore Min and Max are the important ones, with the difference suggesting how much voltage could spike when experiencing a transient. Ergo, a ~90mV delta between them could give me a transient voltage of ~1.53, which is... not good. My brain also says this logic is probably wrong and bad. Question 2: When Joe Internet is reporting an overclock, what is he using for his voltage number? Because the poo poo I see out there doesn't make sense at all based on my observations, even if I'm discounting the use of p95 AVX as a stress test. Ultimately, conventional wisdom of a sub-1.4v Vcore doesn't seem possible for a daily driver to have perfect 5.1GHz stability on my chip, but I can't say that for certain until I figure out what the gently caress these voltages are.
|
# ¿ Aug 3, 2020 02:49 |
|
VelociBacon posted:Okay first off the min and max stuff, ignore that essentially. You want to be observing it while it's under load, look at the current value. You can't do testing and then come back to this after to look at it - you have to be watching it live. In some cases the maximum is going to be what you get under load but sometimes it isn't - you're seeing your voltage drop under load there in your top measurement but you didn't say which sensor that is from. The actual sensor is going to depend on your motherboard so google around trying to find out what sensor reported in hwinfo64 is the 'right one' for your motherboard. For me with a z390 Aorus Ultra (gigabyte) it's VR VOUT. This gives me the voltage from my voltage controller into the chip. My VID is reported at ~0.10v higher than my VR VOUT. A lot of this has to do with your LLC setting. Sorry, I thought about that after I posted but I wandered off to do some other research. It's an ASUS Z490-I, so it doesn't have the VR VOUT sensor the Gigabyte boards have. Thus that min-max are are Vcore as reported by (assuming this is how I'm supposed to interpret this) the Nuvotron NCT6798D.
|
# ¿ Aug 3, 2020 04:01 |
|
spunkshui posted:Ram overclocking adventures! If I want to crack-ping my computer, I try and do anything other than read sensors while running Prime 95 Small AVX.
|
# ¿ Sep 21, 2020 18:21 |
|
|
# ¿ May 16, 2024 02:01 |
|
VelociBacon posted:Nothing against a merge here. As a first-time user with my last build, I felt a lot better doing some prep beforehand. Since I was doing both the GPU and delidding a CPU, I practiced on my old 970, using the "fancy nail polish" TG sells to mask the SMDs around the die and get my application right. I also cut out a "surgical shroud" to drape over the rest of the card to ensure only the GPU socket/die was exposed. The last part may have been more security theater than anything, but it felt a sight bit more secure than moving an applicator across the entire exposed PCB.
|
# ¿ Jul 5, 2022 15:54 |