|
Lockback posted:Ok, figured it out. The extreme 4 Z730 has a MASSIVE default AVX offset, to the tune of 800-1.4GHZ MHZ clock drop. And it doesn't cleanly release it after the AVX test is done. Apparently setting it to 0 sets it to "Auto"(aka, 800MHZ+), and the lowest you can set it to is 100MHZ. I'm not surprised one bit. BIOS software design has never been a strong suit of the Taiwanese mobo makers.
|
#
?
Nov 3, 2017 03:43
|
|
|
|
| # ? May 15, 2024 12:28 |
|
|
Which factors mainly affect stability and errors in prime95? Asus Maximus X Hero, 8700K, 2x8gb g.skill 3600cl15. I've been running prime95 and it seems to run fine at 1.23V on blend, smallfft instantly gives an rounding error. Can it be ram related or cpu related? Like I have no idea where to start. Would it be better to run ram on stock 2666mhz settings first and find some good settings for cpu? And only then start adjusting ram settings? E: also when running blend, in some tests cpu voltage is reported as 1.25V, but during some tests when fan speeds and temps max out, the vcore drops to 1.16V? I wish I knew what I was doing  E2: maybe it's best to test memory first with stressapptest. There doesn't seem to be any working distros (since it must be run on linux, sigh) that I know of. Stresslinux has it included but stresslinux doesn't boot. Maybe it only supports older hardware? E3: maybe unetbootin usb tool works, and linux mint, time to give it a try! Ihmemies fucked around with this message at 09:59 on Nov 4, 2017 |
|
#
?
Nov 4, 2017 09:37
|
|
|
Ok I’ll try to help! Prime95 can test many different parts of the computer depending on the FFT size. If it crashes at 8K then it’s probably the CPU (Vcore too low and/or not enough cooling). http://overclocking.guide/stability-testing-with-prime-95/ as per comments: 1344K = Vcore 448K = Vrin/Input 512-576 = Cache/Uncore 672-720K = VTT 768K = Agent/IMC 800K = Vdimm/Timings 864K = All Ideally you want to run the whole „custom run“ - which loops through all FFTs - stable. I tend to run 1344K to get the Vcore right, then run small FFTs (without AVX) for a while and if that’s stable run the whole custom run. Be very careful with small FFTs (8k, 16k, etc) and AVX because they produce very extreme loads that can overheat and potentially even damage your CPU or VRMs. I’d recommend using an older version of Prime95 (i.e. 26.6) until you know what you are doing, that creates plenty of heat as it is. Keep an eye on the temperatures with Realtemp, Coffee Lake seems to become unstable around 85C. Vcore dropping with load is intended. It‘s called Vdroop and can be countered by Loadline Calibration (LLC) at the cost of some CPU lifetime: https://www.youtube.com/watch?v=_XGhpKHWYAg For Memtest people use HCI Memtest these days, it seems to be more thorough than the old memtest86 and can be run in the background. http://hcidesign.com/memtest/ You’ll have to open and start 6 instances with ~2000 MB each. 1000% progress without error is considered stable. eames fucked around with this message at 11:36 on Nov 4, 2017 |
|
#
?
Nov 4, 2017 11:33
|
|
|
Thanks a lot for that! I'll try the older version of p95. I suppose the errors come because of the lovely tim = high temps so they are cpu/temperature related. Even 240mm aio coolers get 80C on smallfft, I got ~86C on air which probably then causes the errors. Apparently avx prime 95 is more like a "power virus" ala furmark so it doesn't have much point in running it. I was too lazy to wait for 1000% hci memtest. That would take 5 hours? 1 loop of techpowerup's memtest doesn't take that much (equivalent to 100% hci I suppose?) time so I maybe 10 loops is good enough. I can run more later/overnight. I'll try the 26.6 prime next. I ordered aqua computing's dr. delid tool, conductonaut and some arctic silver 5 last week, they should reduce even the prime95 avx temps to 75-77 or so, at least that's what they advertise. Ihmemies fucked around with this message at 12:09 on Nov 4, 2017 |
|
#
?
Nov 4, 2017 12:07
|
|
|
Yeah, you're right that Prime95 AVX is a power virus, I've never seen anything else consume remotely as much power. Typical full load with my 8700K at 5 Ghz is ~170W, small FFTs w/ AVX and no offset would use 240-250W. Asus RealBench is my favorite all-round stability test. I've had it crash where *everything* else was stable because it stresses the whole system and the additional heat from the GPU caused the CPU to run warmer than usual. https://rog.asus.com/rog-pro/realbench-v2-leaderboard/
|
|
#
?
Nov 4, 2017 13:40
|
|
|
4,7GHz core 4,4GHz cache will run smallfft w/o AVX at 1.168V, 140W 75C. AVX adds only extra 30W, I suppose those "auto" settings in bios with maximum power consumption etc. limit how much juice the cpu can draw. Those 1334K fft run at 1.2V and around 70C. I have LLC disabled. When cpu runs at 40C with dota and 55C with pubg that is good enough for now. I'll test the stability later when I first get temps under control with delid and modern thermal grease (using intel TIM and arctic silver 3 for now..) Testing the stability with these temperature issues is wasting time, it's stable enough for now. Games gained 50-100% more fps compared to i7 920@3.5Ghz. Now if I had a 1080 or 1080ti...  I need to buy a wattmeter too. I have only a 500W passive PSU and I'm a bit scared about how much the cpu and gpu can use together while running stress tests.
|
|
#
?
Nov 4, 2017 14:11
|
|
|
84C shouldn't be causing errors in your CPU, that is well within the operating temp. If you're seeing errors immediately when running smallfft's in prime95 I think you have an issue what might be worth digging into. Userbenchmark does a CPU test using AVX that is a lot less intensive, it might be worth trying that: http://www.userbenchmark.com Otherwise I think your voltage is a little low. I'd start setting it at 1.3volts and maybe bring your RAM down to 2.133 and run the Prime95 small FFT test. Then start bringing your ram up first, then your voltage down.
|
|
#
?
Nov 4, 2017 15:02
|
|
|
Yea. Intel cpus can bang on 100c all day and should still work correctly though they will be throttling.
|
|
#
?
Nov 4, 2017 15:52
|
|
|
At stock clocks, yeah, because Intel bins for relatively large stability margins. (those are the margins we use for overclocking.  )My observed behavior with CFL aircooling only 15C below Tjmax: more heat requires more Vcore more Vcore increases heat output exponentially I‘m pretty sure I read that - assuming constant Vcore - decreasing CPU temps by 10C will lower power consumption (and thus heat output) by 4% or something along those lines. That’s why LN2 works so well. If the opposite is true then it makes sense that CPUs require more voltage at higher temperatures to run stable. Maybe it’s just my particular CPU/MB though, I can’t get it to run consistently stable above 90C no matter the voltage, not that I would want to.
|
|
#
?
Nov 4, 2017 16:21
|
|
|
eames posted:At stock clocks, yeah, because Intel bins for relatively large stability margins. (those are the margins we use for overclocking. You are conflating a few ideas. Heat generation is going to be a function of voltage and load. More heat doesn't mean more power, more power means more heat, because that energy has to go somewhere. Heat can cause instability, as can higher clock rates, which is why LN2 is how you get stable at insane clock rates. Voltage can help stabilize a CPU, but can create more heat, which is why you hit a ceiling that can only be fixed by better cooling. (This is oversimplified, but for what you're looking at here its a pretty decent model) The fact that you are having stability problems at relatively low temps (which 90C is relatively low for an 8000 series intel CPU) is telling me you have something else going on. I think you are probably keeping your voltage a little low, or else you are outrunning your memory. OR, you have a faulty CPU/MB/PSU which is why I'm suggesting testing now while stuff is under easy return windows. Prime95 is a power hug and not a "realistic" test, but more and more stuff will likely use AVX so if it is throwing up problems for you now it might be a browser in the future using parallelization or a game that causes instability in the future for you, or whatever problem that might exist might start showing up in non-AVX workloads.
|
|
#
?
Nov 4, 2017 20:05
|
|
|
Lockback posted:You are conflating a few ideas. Heat generation is going to be a function of voltage and load. More heat doesn't mean more power, more power means more heat, because that energy has to go somewhere. Heat can cause instability, as can higher clock rates, which is why LN2 is how you get stable at insane clock rates. Voltage can help stabilize a CPU, but can create more heat, which is why you hit a ceiling that can only be fixed by better cooling. (This is oversimplified, but for what you're looking at here its a pretty decent model) I don't have time to respond to the whole post right now but please read this, tia. https://forums.anandtech.com/threads/effect-of-temperature-on-power-consumption-with-the-i7-2600k.2200205/ anandtech poster posted:Take for example the 2GHz chip at 1.290V and a cool 47°C, it consumes 68W. But let those temps rise from 47°C to 96°C and the power-consumption rises from 68W to 91W.
|
|
#
?
Nov 4, 2017 20:12
|
|
|
eames posted:I don't have time to respond to the whole post right now but please read this, tia. EFB You can get a good crash course on it from undervolting an AMD graphics card. GRINDCORE MEGGIDO fucked around with this message at 22:58 on Nov 4, 2017 |
|
#
?
Nov 4, 2017 22:53
|
|
|
That's one of my favorite collections of community data. I've seen commercial reports that aren't as thorough or detailed in their methodology.
|
|
#
?
Nov 4, 2017 23:04
|
|
|
the effect is so significant that I'm wondering if a watercooled PC with all its pumps and extra fans would end up using less power than a regular air cooled setup. I think it probably would unless it idles all day.
|
|
#
?
Nov 5, 2017 00:17
|
|
|
eames posted:I don't have time to respond to the whole post right now but please read this, tia. None of that data remains so it's hard to understand the methodology here. There will be a difference in power consumption, but I am shocked it would be that high. Was he controlling for load across all the tests? Because what your CPU is doing matters quite a bit. GRINDCORE MEGGIDO posted:EFB Of course, lower voltage means less current and wattage over the same activity. There is current leak at higher temps, but for the 8700K you should only see a difference of 15-25% from 60C-85C. My point to Eames is if you are seeing stability problems at or below 90C, it's not because you are butting up against thermal limits or even power draw limits (unless your PSU isn't up to snuff). AVX is going to draw more power, but that's not because of the heat, that is because of the workload too. I can show you an example of a 8700k running stable at 90C pretty easily.
|
|
#
?
Nov 5, 2017 02:50
|
|
|
Lockback posted:None of that data remains so it's hard to understand the methodology here. Dead photobucket fix for Chrome For whatever reason car forums really liked using photobucket over everything else for build threads and someone came up with a fix when photobucket tried to become premium. craig588 fucked around with this message at 03:00 on Nov 5, 2017 |
|
#
?
Nov 5, 2017 02:56
|
|
|
craig588 posted:Dead photobucket fix for Chrome http://i272.photobucket.com/albums/jj163/idontcare_photo_bucket/Intel%20Core%20i7-2600K/TempvsPowerfor2GHzat1290V.png Oh, yeah for the same clockspeed you are talking about 10% power consumption from very low-to-very-high. Note that that is not TDP. That much power consumption is probably within the regular normal variance for desktop OC'ers. That is a pretty cool article, but I still think something is wrong in your 8700k setup if you are having stability problems at 85C.
|
|
#
?
Nov 5, 2017 03:45
|
|
|
Just an FYI but avx instructions are only not used because they're new. They're a new way of getting more performance, so expect things to use them. Handbrake already uses them for some encodes, so if you need to do video encoding consider a separate overclock or stock for that specific task.
|
|
#
?
Nov 5, 2017 10:15
|
|
|
But is avx handbrake equivalent to avx prime95?
|
|
#
?
Nov 5, 2017 10:52
|
|
|
AVX is for parallelization, I believe. That's different than multithread and not something that will be used universally. But it's used in some video and photo editing effects today, and we'll may see it in web browsers (which likely wouldn't tax a system) at some point in the future. Games probably won't use that instruction path in the short term but who knows?
|
|
#
?
Nov 5, 2017 14:55
|
|
|
Ihmemies posted:But is avx handbrake equivalent to avx prime95? old prime 95 > handbrake with AVX > prime95 with AVX There's a bigger gap between AVX and not and prime or not, at least on skylake.
|
|
#
?
Nov 6, 2017 02:09
|
|
|
So I contacted ASRock about the inability to set the AVX offset to anything less than 100 MHZ. They replied back with a custom bios for me enabling that. Pretty cool response to what was essentially a request for benchmarking. I'm going to keep AVX offset on during my day-to-day OC, but it's nice that I have a bit more fine control over it.
|
|
#
?
Nov 8, 2017 00:32
|
|
|
Lockback posted:So I contacted ASRock about the inability to set the AVX offset to anything less than 100 MHZ. They replied back with a custom bios for me enabling that. Pretty cool response to what was essentially a request for benchmarking. This is interesting. I think my next mobo will be asrock.
|
|
#
?
Nov 8, 2017 09:09
|
|
|
Completely anecdotal: I tried to replace Noctua NT-H1 paste with Thermal Grizzly Kryonaut since that seems to perform slightly better in most tests. I absolutely cannot reproduce the results of the Noctua paste even after 5 separate applications (3x spread, 1x grain, 1x line) of Kryonaut on my NH-D15S. The Noctua paste just oozes into all nooks and ridges of the heatsink, the Kryonaut just stays the way I left it and performs 4-7°C worse under load, even after 24 hours at full load. It's not the end of the world but if you own a Noctua cooler just save yourself the headache and use the boxed paste with the rice grain method. The upside is that I'm really fast at removing the fiddly fan clips now. 
|
|
#
?
Nov 8, 2017 11:10
|
|
|
Spread the kryonaut
|
|
#
?
Nov 8, 2017 17:40
|
|
|
As another anecdote, I used NT-H1 between the die and heatspreader of my 3770K, and it worked amazingly! For about 12 hours, and then the temps shot up. It suffers from a pump-out effect, apparently. I replaced it with Thermal Grizzly Kryonaut (the best one can get before delving into liquid metal, and I can't be bothered to risk accidentally getting that poo poo on something), and the temps have been stable for months. In a way my story agrees with eames's - NT-H1 will spread under heat load, Kryonaut is super stable. It needs spreading, it comes with a nice spreading attachment on the syringe. I used Kryonaut both under the heatspreader and on-top, under my Noctua NH-D15S.
|
|
#
?
Nov 8, 2017 17:56
|
|
|
Don Lapre posted:Spread the kryonaut I did, twice, in varying amounts. I think the machining marks in baseplate are just too deep for that paste. The best results were with kryonaut thinly spread on both HSF and IHS. I reapplied NH-T1 just now with the sloppy rice grain method and got significantly (7-10C) lower load temps than I ever achieved with the 5 different applications of Kryonaut. Maybe it’s a bad or old batch.
|
|
#
?
Nov 8, 2017 18:06
|
|
|
That or someone making fake batches. I'd expect 2-3 degree differences between the best and worst applications using the same mounting.
|
|
#
?
Nov 8, 2017 18:10
|
|
|
Well, NewEgg has the 8700k in stock finally, for $420 each, and SiliconLottery.com for $45 shipping included will delid the chip, replace the TIM with liquid metal, and put the IHS back on with the same type of silicone glue. I got a chip from NewEgg and mailed it straight to SiliconLottery, wish me luck... Edit: And yeah I know, forfeiting my warranty, but I absolutely need those lower temps because my case only has 45mm of clearance for heat sink plus fan.
|
|
#
?
Nov 9, 2017 23:10
|
|
|
Is there a CPU benchmark that's dynamic enough to test stability of overclocked turbo states? Just running Prime95 on a limited amount of cores is misleading because it doesn't test for edge cases like full load on all cores with heatsoaked HSF -> full load on only a single core.
|
|
#
?
Nov 10, 2017 15:55
|
|
|
Noctua finally listened and now sell Black fans, and heatsinks with them in various configurations, as well as colour coordinated cables etc.quote:you can now get black, white and colour swap Black heat sink covers for the top end Noctua Heat sinks, as well as their leading fans in black with a choice of colour coordinating anti vibration pads. Not only that, but you can top off the whole buiild with a choice of six colours for fan splitter and fan extension cables. We're not expecting the products in the range to be dirt cheap, but what we have learned is that with Noctua, you get the quality you pay for, so yeah, the fans may cost a few quid more than your average fan, but it is, after all, a Noctua fan. https://overclock3d.net/reviews/cases_cooling/new_noctua_chromax_range/1 GRINDCORE MEGGIDO fucked around with this message at 23:58 on Nov 10, 2017 |
|
#
?
Nov 10, 2017 23:55
|
|
|
just how powerful of an x370 or b350 mobo is needed to get a decent OC out of a ryzen 5? https://www.youtube.com/watch?v=vcqu-9Ejf60 watching this seems to say anything but the beefiest x370 is inadequate. Its Bullzoid though and anything that cant be used for extreme overclocking isnt enough. I just want a feature rich board, maybe with wifi that will take a non "x" 1600 to 4.0 on air.
|
|
#
?
Nov 22, 2017 06:54
|
|
|
I am overclocking for the first time and used der8aur's guide https://www.youtube.com/watch?v=CoUtA7DKXhU with a delidded 8700k on an Asus Strix Z370-E motherboard with a Noctua D15S. He does some weird stuff that I'm not sure about and wanted to know if it is a good idea. XMP on with DDR4-3200MHz Restart Bios Multicore Enhancement disabled AVX offset at 3 (I see people use 2 here) CPU Core Ratio to sync all cores Multiplier at 50 CPU LLC at 6 Long and Short Duration Package Power Limit to max values (4095) CPU Core/Cache Current limit to max (255.50) Max and Minimum CPU cache ratio to 42. BCLK Aware Adaptive Voltage disabled CPU Core/Cache Voltage to manual at 1.350 to start Save and Exit Bios Run Cinebench Run prime95 (I turned off AVX. Using version 29.4) If it crashes, go back to bios and increase CPU Core/Cache Voltage by 0.01 volts and run tests again. I'm currently running it at 4.8GHz with 1.35v but still tweaking. I'm mostly wondering about the bold settings. If I wanted my computer to last a long time, what voltages and temps should I aim for? Is the wear on the processor lifespan negligible as long as I stay under 1.4v and 80C? WarpZealot fucked around with this message at 07:47 on Nov 26, 2017 |
|
#
?
Nov 26, 2017 06:20
|
|
|
Cache clock speed increases can offer a very small performance increase. The fastest cache overclock is still worth less than 1 CPU multiplier, but once your CPU is maxed out it's a few percent of performance left. All the maxed out current things are to make sure it doesn't throttle. I never use manual voltage because it can be marginally more stable and I want to keep the power savings of adaptive or offset voltage so I don't see the point in stability testing a mode that uses more power and will perform differently. I like offset voltage the most, but adaptive has proven to be a reliable setting as well. Offset applies a fixed offset from the requested voltage, adaptive dynamically changes the offset to make the supplied voltage match whatever the requested voltage is while not in a low power state. I don't have a Z370, but I think multicore enhancement is ignored once you key in any overclock. LLC is for increasing the voltage under load to compensate for droop, most motherboards handle it slightly differently so you'll need to play with LLC to find exactly how much you need. I assume 6 on the Asus 370 makes it droop as close to zero as possible, at least that's the setting I'd recommend to start with if I was writing a guide. Once you get a stable overclock without AVX test it again with AVX to find out how much of a negative AVX offset you need to keep temperatures where you want them. (Or possibly to remain stable)
|
|
#
?
Nov 26, 2017 13:14
|
|
|
New 8700k system with a Gigabyte Aorus Z370 Gaming 7, aircooled with an old Hyper 212 Evo. It ran Prime95 overnight stable, at 4.9 GHz / -2 AVX 1.35V. Peak temp was 92 C, average about 90ish? Either way seems a bit high to me. Haven't tried to push it farther based on the temps. I have it running another test at 1.345V right now, but based on preliminary tests last night I'm pretty sure I can't drop it any lower without dropping the multiplier / AVX. I'm a little bit uncomfortable that I have to run it (aircooled) at this voltage / temperature for the speed I'm seeing. Am I off base here?
|
|
#
?
Nov 29, 2017 16:58
|
|
|
Don't use prime95 to judge temps
|
|
#
?
Nov 29, 2017 19:03
|
|
|
Or do a larger AVX offset. AVX instruction set stress tests are not going to lineup with what you want for everyday temp. Your numbers look correct for your situation.
|
|
#
?
Nov 29, 2017 21:21
|
|
|
Is there a way to increase the maximum overclock allowed in MSI Afterburner? I got a new laptop with a 1050ti and I've maxed the memory overclock bar at +1000. I was initially just undervolting the CPU and GPU and figured I'd see what kind of memory overclock I could get at the lower voltage settings but I've run out of slider.
|
|
#
?
Nov 29, 2017 21:42
|
|
|
Switched to adaptive vcore with automatic offset and a turbo voltage of 1.330v. I think I finally have an overclock that I'm happy with on my 8700k. 4.9ghz, 1.330v vcore, AVX -3 Offset ran Realbench for 8 hrs with a max cpu temp of 78C. Are there any other settings I need to look at and/or are any of my automatically set voltages too high? Is it worth it to reduce the vcore voltage by 0.001 until I find the lowest stable voltage? It was unstable at 1.32v. I have XMP on with DDR4-3200mhz running at 1.344v (auto voltage) CPU VCCIO 1.328v (auto) CPU System Agent 1.232v (auto). PCH Core 1.016v (auto) CPU Standby Voltage 1.351v (auto)
|
|
#
?
Dec 3, 2017 05:39
|
|
|
|
| # ? May 15, 2024 12:28 |
|
|
I split the voltage difference to find stability then add .001 when I find the lowest stable voltage because of weird edge cases stability testing might miss. Since 1.33 is stable and 1.32 isn't I'd start with testing 1.325. Your voltages also are low enough that if you don't want to mess with it they're nowhere near risky and you'll probably spend a week stability testing for a few dollars in electricity over the lifetime of the CPU.
|
|
#
?
Dec 3, 2017 15:33
|
|