|
Allow up to 95C on your stress tests. You'll find you don't get anywhere near that in daily use. If clocks do increase temps to the extent the other poster was describing then my bad, I've never noticed that to be the case.
|
# ? Feb 20, 2019 04:05 |
|
|
# ? May 15, 2024 03:27 |
|
My G.Skill 4x8GB 4133MHz 17-17-17-37 1.4v posted with just XMP. I did not expect that in the slightest. Can't wait to see what they will do with 1.5-1.6v. The gigabyte T-topology on their top end boards is loving awesome, these speeds and timings were 1DPC only a gen ago.
|
# ? Feb 20, 2019 07:06 |
|
BurritoJustice posted:My G.Skill 4x8GB 4133MHz 17-17-17-37 1.4v posted with just XMP. I did not expect that in the slightest. Can't wait to see what they will do with 1.5-1.6v. Those timings are what they are garunteed to hit with XMP. E:sorry unless your 37 is higher than the advertised poo poo
|
# ? Feb 20, 2019 10:36 |
|
XMP is not a guarantee, dude. It's board and chip dependent.
|
# ? Feb 20, 2019 11:07 |
|
XMP generally is tested. When I got my memory I ran it at the default 2400-15 for about a year, but I read somewhere XMP is tested and not a guess and have been running it at 2666-14 for over 2 years now. I'm sure some cheap brands might not test it, but good brands should be validating their XMP settings.
|
# ? Feb 20, 2019 11:46 |
|
If you look at the QVLs for the really fast kits (like that 4133 kit) they cover only a handful of Intel boards (good luck getting those XMP timings to work on Ryzen). They can't reasonably test every board or account for variable silicon.
|
# ? Feb 20, 2019 12:09 |
|
VelociBacon posted:Those timings are what they are garunteed to hit with XMP. Anything past 3866 with 2DPC and it becomes extremely dependant on CPU IMC quality. Usually without fail kits like this need large increases to VCCSA and VCCIO and still can't hit their rating. It's pretty typical to get a kit like this and end up running them at 4000c19 or something, it's typically a much bigger issue with 2DPC quad rank situation than it is with 2x8GB kits. 4133c17 is pretty nutty as a timings to frequency thing, requires lots of luck to get that straight up from XMP. Checking Newegg reviews shows almost everyone downclocking to post. I haven't done any stress tests (loop is not plumbed, so just a short test boot), so it's not guaranteed stable. But I was surprised to even post XMP.
|
# ? Feb 20, 2019 13:35 |
|
Ran the x264 stress test overnight and no crashes! Temps never got above 84 on any core either so I think I can officially declare my OC to 4.9 Ghz @ 1.28V is stable! Now to debate if it's worth it to try for 5 Ghz... edit: CPU-Z link with details if y'all are interested. https://valid.x86.fr/j1bz0p edit 2: system crashed running RealBench with 5 Ghz and 1.35V with temp readings in the low 90s. I'm not comfortable going up higher on voltage because of temperature so I suppose I'm done. edit 3: huh, now my previous 4.9 Ghz @ 1.28V overclock crashes running Cinebench. Guess I need to up it to 1.29. Weird. axeil fucked around with this message at 01:20 on Feb 21, 2019 |
# ? Feb 20, 2019 14:37 |
|
I think that's volts not mV. Maybe in 45 years we can run our PC's on 1.3 mV.
|
# ? Feb 20, 2019 19:25 |
|
VelociBacon posted:I think that's volts not mV. Maybe in 45 years we can run our PC's on 1.3 mV. Oops! You're right. I had another weird thing happen. I overclocked the RX 580 I got and ran RealBench to confirm it was stable. It passed so I went in to Cinebench to get scores and the system crashed again (likely from insufficient voltage based on the error code). It's weird that Cinebench would cause a CPU crash when other more stressful tests passed. I boosted the CPU up to 1.30V and all seems well now. Just a strange situation I suppose.
|
# ? Feb 21, 2019 01:20 |
|
My experience with Realbench is it's just not stressful enough: it can run forever on a CPU without errors but yet the same CPU can still fail within the 1st min of P95.
|
# ? Feb 21, 2019 17:24 |
|
That depends on how you look at it. I'd put it the other way around - P95 is way too stressful and will cause failures that would never happen with any real workload. Seriously, nothing will ever sit and constantly issue AVX/FMA3 ops on every thread as fast as they can be retired. Not even other unrealistically heavy benchmarks come close to P95. It's only useful for stress testing cooling and power delivery. Use OCCT's small dataset test if you want find errors quickly with an unrealistically heavy workload.
|
# ? Feb 21, 2019 17:43 |
|
Realbench passes indefinitely on OCs that crash in demanding games so its basically worthless to me. Prime 95 is on the other side, even very demanding games will be stable on OCs that crash within 45min of AVX-enabled p95.
|
# ? Feb 21, 2019 18:36 |
|
Palladium posted:If your PC runs P95 and Furmark fine for 5 mins, and also succeeds in 4 passes of memtest86 block move pattern it's highly unlikely the CPU/GPU/RAM is the culprit. OK so I passed both of those for 5 minutes each as follows: P95 - v29.4b8 Small FFTs Furmark - v1.2.4.0 GPU stress test However I still got a random reboot. This was after 24 hours of Kingdom Come and a about 3 hours of Sandstorm Insurgency with no issues. Is that something to worry about? Note that I plan to OC my CPU soon as well so I'm keen to ensure my GPU is completely stable before I start that so it's easier to fault find.
|
# ? Feb 22, 2019 23:28 |
|
Try it at stock again.
|
# ? Feb 22, 2019 23:47 |
|
GRINDCORE MEGGIDO posted:Try it at stock again. P95 and Furmark? Longer than 5 minutes each maybe?
|
# ? Feb 23, 2019 05:54 |
|
I'd run it all at stock and just game on it.
|
# ? Feb 23, 2019 10:00 |
|
Crossposting from the OC thread: New build of MSI Afterburner (4.6.0) is out with some improvements. MSI posted:
e: Editing to say that it's allowing me now to increase the voltage for my EVGA 2080ti, something that I couldn't do before with afterburner. The way it refers to voltage is different than previously but it seems to work, same vcore measured with x1 and afterburner set up like the screenshot. Thank god I can finally get rid of precision X1. VelociBacon fucked around with this message at 20:21 on Mar 7, 2019 |
# ? Mar 7, 2019 19:44 |
|
This was a pretty interesting read (by an Asus engineer). Apparently, if you're using AVX offsets it seems beneficial to use a higher VID and a lower LLC (larger vdroop) because it improves transient response and avoids undershoot. Possibly true even without AVX offsets as well.
TheFluff fucked around with this message at 06:53 on Mar 8, 2019 |
# ? Mar 8, 2019 06:40 |
|
I used the DOCP (XMP) profile in BIOS for my Ryzen, which seems to set my RAM at their default 3200, but should I try increasing the speed of them just to see if they can pull it?
|
# ? Mar 8, 2019 23:31 |
|
I'm finally ready to jump in this game with my new setup. Got an i7-9700k on an Asus Prime Z390-A, and watched some basic overclocking guides for a first attempt which was mostly leave everything on Auto and manually adjust the cores up to 49 while only playing with VCore. I was surprised at how low the default voltage was. Not entirely sure what my voltage ranges or speed targets are supposed to be. Currently sitting at a Prime95-Unstable (but is totally fine for playing video games) OC of 4.9GHz @ 1.3v and this barely touches 70C at full load. Is this good, bad, needs more work?
|
# ? Mar 9, 2019 14:32 |
|
Fhqwhgads posted:I'm finally ready to jump in this game with my new setup. Got an i7-9700k on an Asus Prime Z390-A, and watched some basic overclocking guides for a first attempt which was mostly leave everything on Auto and manually adjust the cores up to 49 while only playing with VCore. I was surprised at how low the default voltage was. Not entirely sure what my voltage ranges or speed targets are supposed to be. Below 1.4v Vcore is generally considered "safe", but for a daily overclock in a system that's supposed to live for a long time I'd prefer to stay below 1.35v under load. Are you using adaptive mode or a fixed voltage? For daily use, adaptive is the way to go. Keep in mind though that because of how the load line works, you'll get somewhat lower voltages under the heaviest loads. Also, the VID will be ~0.02v higher than what you set for turbo voltage in adaptive mode when running AVX workloads (so if you set 1.34v in BIOS you'll get 1.34v VID when running non-AVX, but 1.36v when running AVX). For comparison, I've run my 8700K (which is very similar in power usage to a 9700K) at 4.9GHz with the following settings on an Asus Maximus X Hero for the last year: - 49x all-core multiplier - 45x cache (uncore) multiplier - adaptive mode vcore, with turbo voltage set to 1.36v, offset -0.02v - LLC 5 This gives around 1.35v vcore with memory-heavy AVX benchmarks like OCCT's large dataset tests, and around 1.34v with the most extreme tests (OCCT small, P95). OCCT small and P95 draw 170-180W, ish (according to HWinfo64, at least). Non-AVX loads like linpack with AVX disabled get around 1.31v vcore. I've delidded the CPU, so this stays below 70C at all times (staying just under 70C in P95/OCCT small). My CPU is a quite mediocre specimen; there are many that need much less voltage than mine. I can boot it at 5GHz, but getting it 100% stable there would require more voltage than I'd be comfortable with for daily use. Do note though that I think Asus improved voltage measurement accuracy on Z390, so your voltage numbers are most likely not directly comparable with mine. TheFluff fucked around with this message at 17:21 on Mar 9, 2019 |
# ? Mar 9, 2019 17:12 |
|
I'm running fixed voltage and no speedstep. The one I don't get is the VCCIO voltage. It's default is 0.96 but the guides had it upped to like 1.1 but I'm not sure what this really controls.
|
# ? Mar 9, 2019 22:26 |
|
Fhqwhgads posted:I'm running fixed voltage and no speedstep. The one I don't get is the VCCIO voltage. It's default is 0.96 but the guides had it upped to like 1.1 but I'm not sure what this really controls. Don't do that, use adaptive and don't disable any power saving (C-states) or speedstep/speedshift stuff or you'll run unnecessarily hot for no reason. Running manual with downclocking disabled is something you do on liquid nitrogen, not for something you're going to use every day. The stability benefits are completely negligible under normal circumstances. VCCIO is essentially the voltage for the CPU's memory controller. It's rarely beneficial to raise it above stock unless you're pushing your memory super hard (like trying to run DDR4-3800 or above), and even then it's sorta doubtful if it actually does anything. 1.1v is still considered safe territory though. If you've enabled XMP (and you should) the BIOS might've already made the auto setting 1.15v or something though and in that case it might be a better idea to just set it to 1.1v or below manually.
|
# ? Mar 10, 2019 03:32 |
|
I haven't played with vccio since skylake, but running xmp / 3400 ram or higher upped the vccio to 1.1v+ automatically and it wasn't stable otherwise.
GRINDCORE MEGGIDO fucked around with this message at 12:19 on Mar 10, 2019 |
# ? Mar 10, 2019 11:52 |
|
Auto mode on my gigabyte board sets 1.36v for VCCIO and VCCSA for 4133MHz C17. I manually set it down to 1.3v and it doesn't seem to have any issue. I think I read in the sponsored Gigabyte overclocking guide that 1.3-1.35v is the recommended max for daily clocks. Buildzoid agrees around there too. Supposedly auto on the Asus Maximus boards will run 1.6-1.7v for 4000+, which is pretty catastrophic.
|
# ? Mar 10, 2019 12:02 |
|
I had to take mine on a 9900k up to 1.3v to get my 3200 C14 XMP to work. 1.25v got a POST or two out of it, but was unstable and 1.2v wouldn't POST at all. IIRC my old Asus board with 3200 C16 never managed to actually work at 3200, I had to run it at 3000 and it automatically cranked vccio and vccsa on that 7700k to something like 1.33v to get there.
|
# ? Mar 10, 2019 13:01 |
|
My 2700x wouldn't run Samsung b die 4000 at anything over 3000, even with all subtimings set. They were fine at 3733 on Intel But it's fine with b die 3200 c14, different sticks. Must be some weird hidden timing thing? No idea
|
# ? Mar 10, 2019 13:35 |
|
Dang, seems I was misremembering the VCCIO stuff pretty badly.
|
# ? Mar 10, 2019 14:58 |
|
TheFluff posted:Dang, seems I was misremembering the VCCIO stuff pretty badly. Not necessarily, it just seems to be wildly variable plus XMP/Auto tends to send it flying all over the place and most people don't check to see where it lands at. I did quite a bit of hunting around in online forums from basically every major motherboard vendor when I was fighting my XMP settings and found mountains of conflicting information from anything over 1.2v will turn your CPU into a plasma torch that will burn a hole in the socket to auto 1.35v is fine for daily OC use. The only general consensus seemed to be that stock 0.9v Intel spec only works for the base JEDEC timings/frequency of the DIMMs and pretty much never works for XMP. Indiana_Krom fucked around with this message at 15:55 on Mar 10, 2019 |
# ? Mar 10, 2019 15:53 |
|
Trying to overclock my threadripper 2920x on an asus x399 taichi, and it's simply never stable. Under PBO, it gets a 4ghz max boost and runs that just fine, and rather cool. I can't even run a 3.8 manual overclock without either my system or my VMs crashing within a few minutes. I ran vCPU up to 1.35 before calling it quits and letting the chip do it's own. RAM is a QVL set (8gb x4 3333 cl16), which runs at the full 3333 under PBO. Ideas? e: 850 watt evga supernova g3 with another 200 watts of video card, but it crashes under CPU load with nothing taxing the GPU. And again, works fine under full system load with PBO.
|
# ? Mar 11, 2019 10:01 |
|
Have you tested for memory stability?
|
# ? Mar 11, 2019 10:10 |
|
Indiana_Krom posted:Not necessarily, it just seems to be wildly variable plus XMP/Auto tends to send it flying all over the place and most people don't check to see where it lands at. I did quite a bit of hunting around in online forums from basically every major motherboard vendor when I was fighting my XMP settings and found mountains of conflicting information from anything over 1.2v will turn your CPU into a plasma torch that will burn a hole in the socket to auto 1.35v is fine for daily OC use. The only general consensus seemed to be that stock 0.9v Intel spec only works for the base JEDEC timings/frequency of the DIMMs and pretty much never works for XMP. I managed 3000MHz 16-18-18-36 with 2x16GB Hynix sticks on Z370 stock volts.
|
# ? Mar 11, 2019 14:52 |
|
GRINDCORE MEGGIDO posted:My 2700x wouldn't run Samsung b die 4000 at anything over 3000, even with all subtimings set. They were fine at 3733 on Intel I guess I got lucky. I bought Samsung B-Die DDR4 4000 CL 19 (Kingston Hyper X Predator RGB) and I am currently running it at 3600 14-15-15-30 on my 2700X, using 1.46v. Ryzen DRAM Calculator was pretty much right on the money for me. Looks stable so far, it and the CPU has passed various passes of Cinebench R15/20, Realbench and multiple game benchmarks. Last thing to do is hit it up with an overnight Memtest86 to really check stability and I should be locked in. Anything past 3600 is gonna need both a big jump in voltage and timing slacking so I can't imagine going higher will be better. Might try the "Fast" subtimings that Ryzen DRAM Calculator recommend, currently using the "Safe" subtimings.
|
# ? Mar 11, 2019 15:40 |
|
GRINDCORE MEGGIDO posted:Have you tested for memory stability? Why would my RAM become unstable at a lower clock than PBO manages? It's at 3333 for PBO or manual. I've tried backing it down to 3000 and it's still the same - PBO stable, even a minor manual overclock crashes.
|
# ? Mar 11, 2019 18:31 |
|
Tools usually give you a good starting point and real world testing is still needed to dial in exactly what you need. There's only so many situations a tool can test, and you may have found a case it can't test for.
|
# ? Mar 11, 2019 18:58 |
|
craig588 posted:Tools usually give you a good starting point and real world testing is still needed to dial in exactly what you need. There's only so many situations a tool can test, and you may have found a case it can't test for. But the RAM is tested, real world, on my machine right now. It's running 3333, right now, during stress tests. It's been running for 24 hours with PBO giving me a 4ghz ceiling. The same tests that immediately fail when I turn off PBO and try to do even a 3.8 manual overclock. So either there's a weird PBO dependancy on memory stability, or I'm not tuning my voltages right.
|
# ? Mar 12, 2019 01:48 |
|
Sudden reboots with no errors in the log were, on my Ryzen at least, due to memory problems.. seems worth a go to test it.
|
# ? Mar 12, 2019 10:02 |
|
I finally got around to overclocking my i5-9600K. However I'm a little confused about voltages. I followed this guide for my Z390 Aorus Elite mobo. Why does it all say 1.200v for core voltage when I set it at 1.345v in the BIOS? Is this something to do with LLC? CPU-Z showed 1.2v like HWinfo. I tried up to 5Ghz at 1.345v but couldn't even get it POSTing until I dropped it back to 4.7Ghz. And my HWinfo details are below: I ran it for an hour using P95 v29.4b8 on Small FFTs and had no issues (all cores stayed at 100%) with a core max of 74C. That should be fine right?
|
# ? Mar 16, 2019 06:38 |
|
|
# ? May 15, 2024 03:27 |
|
I asked about this in the parts picking thread, long story short I want to run an AMD FX-9370 220W old rear end slow CPU because . What are some good options for air coolers that can handle this processor? I'm looking at like a Noctua D-14/15 (I assume the 14 is just a little older and cheaper but almost as good?), Deepcool Assassin or Lucifer, or a Thermalright Macho Rev.B (I just don't know if I can take a "macho" cooler seriously). Are there any other air coolers I might have overlooked? I'm also considering going with water, like a 120mm or dual 120/240mm radiator, but I think most of the big air coolers will cool almost as well as a water cooling setup. I don't care about having a silent fan, it would be nice if it was quiet at idle but hell I kinda like hearing the fans spin up when the CPU is under load.
|
# ? Mar 19, 2019 16:52 |