|
VelociBacon posted:Where are you measuring the voltage? You should be making all these changes manually in bios and not using that Intel program. Your voltage could be bouncing around because your LLC is set to auto. Oh the intel program was really just for getting started. I've been using CPU-Z for in-windows monitoring and making changes with the BIOS. There seems to be a few LLC settings in my BIOS, should they all be turned to manual? From memory with one the choice is between Auto and +0.1V . Edit: After flipping all of the LLC settings to off, I'm still seeing the Voltage in BIOS fluctuate between .992V and 1V , while before it was between .992V and .984V. Another setting in DIGI + VRM > is "CPU Load-line Calibration" which I've changed from auto to 50%, is that a factor too? Xerxes17 fucked around with this message at 15:24 on Nov 4, 2020 |
# ? Nov 4, 2020 15:01 |
|
|
# ? Jun 18, 2024 05:05 |
|
Xerxes17 posted:Oh the intel program was really just for getting started. I've been using CPU-Z for in-windows monitoring and making changes with the BIOS. I would leave all the memory stuff auto, but your LLC should be manually set (look up a guide for your mobo but generally just somewhere in the middle is fine). You're measuring CPU voltage under max load too right?
|
# ? Nov 4, 2020 15:22 |
|
VelociBacon posted:I would leave all the memory stuff auto, but your LLC should be manually set (look up a guide for your mobo but generally just somewhere in the middle is fine). You're measuring CPU voltage under max load too right? Not yet, but I will. Thought I'd focus on getting the basic BIOS settings correct for now, I'll get googling for guides.
|
# ? Nov 4, 2020 15:28 |
|
1. Asus's voltage readers weren't great in that generation. I think you'll see a fair bit of fluctuation regardless. 2. The CPU will use more voltage under load, so if your load is fluctuating (a CPU that old using Windows 10 will probably be fluctuating just sitting at the desktop) then thats normal. Pushing the CPU to 100% then watching voltage is a better option (though see #1). 3. quote:Another setting in DIGI + VRM > is "CPU Load-line Calibration" which I've changed from auto to 50%, is that a factor too? Dunno if you caught VelociBacon but "Load-line Calibration" is LLC. Personally, a CPU that old I'd probably be pretty aggressive with the LLC if your seeing stability issues. With that cooler you should be able to handle some heat. I'd set that to 100% and get everything working, then decide if you want to back it off/lower your voltage. 4. Again, on a CPU that old I think you'll need to push more power to keep it stable if you intend to use it under appreciable loads. I'd feel ok with a 1.3v CPU core to if you say you're seeing crashes. Temp is probably not a big deal yet if it's getting over 80C, we can back off later once things are stable. Basically, I assume this isn't a daily driver so I'd take a more aggressive approach than I would have if this was a new CPU. My trusty 2500k can't hit the same numbers it did a decade ago either, so your target of 45x (which while reasonable) may be harder than it was.
|
# ? Nov 4, 2020 16:06 |
|
Also since I don't think you mentioned it make sure you're using a stress test of some kind to get that load to the CPU.
|
# ? Nov 4, 2020 16:13 |
|
Xerxes17 posted:Oh the intel program was really just for getting started. I've been using CPU-Z for in-windows monitoring and making changes with the BIOS. Definitely manual LLC. I was on auto and one day I realized it was going way higher then in the past under load. I get the best stability if I can get voltage to step up ever so slightly under full load vs a droop. I use manual voltage, not offset. I typically end up typing something close to Intel’s safe voltage max value in the bios. Edit: OCCT for 10 mins is how I check stability spunkshui fucked around with this message at 19:44 on Nov 4, 2020 |
# ? Nov 4, 2020 19:41 |
|
Manual voltage is absolutely fine provided you can still set it to idle properly. A lot of the time manual voltage automatically runs that voltage all the time which is something you don't want.
|
# ? Nov 4, 2020 22:07 |
|
VelociBacon posted:Manual voltage is absolutely fine provided you can still set it to idle properly. A lot of the time manual voltage automatically runs that voltage all the time which is something you don't want. If you’re worried about it for thermal reasons as long as the CPU isn’t really doing anything it wont make much heat. If you’re worried about it for CPU longevity I pretty much never hear of that being a problem. All of my and my wifes old overclocked CPUs are actually still functioning and deployed to my robotics team (no longer OCed). As long as your 24/7 voltage is at or under what Intel recommends there’s no reason why the CPU wont literally end up living longer than it’s needed. I’ve never had much luck getting C states to work during overclocking anyway and I don’t really care because stability over everything. Edit: if you’re able to get voltages to drop at idle and C states to work while being fully stable then great but I just consider them to be very low priority. spunkshui fucked around with this message at 00:29 on Nov 5, 2020 |
# ? Nov 5, 2020 00:25 |
|
Lockback posted:1. Asus's voltage readers weren't great in that generation. I think you'll see a fair bit of fluctuation regardless. Thanks for the further details, on point 3 I understood that but I included the full name just to be sure The desktop is my daily machine, so I don't want to be super aggressive on it. On the point of longevity, this system did sit totally unused for 3 years and I didn't overclock it before so it should be doing okay.
|
# ? Nov 5, 2020 00:41 |
|
spunkshui posted:If you’re worried about it for thermal reasons as long as the CPU isn’t really doing anything it wont make much heat. Really, I think the last few processors I've overclocked have had no issues throttling with c states. I know it used to be an issue with instability.
|
# ? Nov 5, 2020 00:46 |
|
VelociBacon posted:Really, I think the last few processors I've overclocked have had no issues throttling with c states. I know it used to be an issue with instability. I mean they just don't do it. I don't even know if its stable because it doesn't want to try for some reason. I enable "cstates" as much as I can in the bios but if I set the multi to 50 it just sits there.
|
# ? Nov 5, 2020 00:56 |
|
I've had C-states break on me when I'm pushing to the absolute limit but I've been mostly keeping my stuff at a "moderate" everyday OC and C-states work fine then. I agree its pretty low priority though. Xerxes17 posted:Thanks for the further details, on point 3 I understood that but I included the full name just to be sure The desktop is my daily machine, so I don't want to be super aggressive on it. On the point of longevity, this system did sit totally unused for 3 years and I didn't overclock it before so it should be doing okay. None of the advice is really going to hurt the processor. Basically, its lived this long an extra .1v isn't going to be what knocks it over the edge.
|
# ? Nov 5, 2020 01:01 |
|
Currently getting x45 @ 1.240V core, testing with prime95. However in my HWiNFO64 I see that cores 1/2 run 5 degrees hotter than cores 0/3, putting them past the thread recommended 73 ℃ average under load at around 78-80. Annnd as I was typing this on my laptop, prime 95 crashed, hmm. Another issue is that cstates seem to be not working. As far as I know I have the correct settings for it in the BIOS, or is it something that just doesn't happen past certain multipliers? Because if that's the case, I might as well just switch over to manually setting the voltage instead of using offsets. Xerxes17 fucked around with this message at 02:33 on Nov 5, 2020 |
# ? Nov 5, 2020 02:23 |
|
Xerxes17 posted:Currently getting x45 @ 1.240V core, testing with prime95. However in my HWiNFO64 I see that cores 1/2 run 5 degrees hotter than cores 0/3, putting them past the thread recommended 73 ℃ average under load at around 78-80. Annnd as I was typing this on my laptop, prime 95 crashed, hmm. Use manual anyways for now to make sure you're getting the right voltage and switch to offset later on once you know the values you want. Do not look at any averages with temps or voltages. You only care about peak temps and you only care about them under max load. Remove the average column from HWiNFO64. There are other things that turn idle on/off besides c-states so it's something to look at once you have a stable overclock.
|
# ? Nov 5, 2020 02:36 |
|
VelociBacon posted:Use manual anyways for now to make sure you're getting the right voltage and switch to offset later on once you know the values you want. Okay, so for example when I have it manually set to 1.280V in the BIOS, and then when I put it under Prime95 stress test, seeing a drop to 1.256V in CPU-Z is expected behavior, correct?
|
# ? Nov 5, 2020 02:52 |
|
Xerxes17 posted:Currently getting x45 @ 1.240V core, testing with prime95. However in my HWiNFO64 I see that cores 1/2 run 5 degrees hotter than cores 0/3, putting them past the thread recommended 73 ℃ average under load at around 78-80. Annnd as I was typing this on my laptop, prime 95 crashed, hmm. If you CPU paste is anywhere near as old as that cpu I would recommend spending 10 bucks on some thermal grizzly ceramic stuff. Then when you’re taking the CPU cooler off you can take the opportunity to wash it to really get all the dust off of it. The application process of the CPU paste doesn’t seem to matter a ton but age and quality of paste does matter. Your temperatures are not great, but the crashing is unacceptable. You also want to make sure that those temperatures don’t creep up overtime once it is stable. If your voltage stays fairly the same with load and without load then you might be looking at needing more of it. If it drops a sizable amount when you hit go I would move LLC up a step if you have any left.
|
# ? Nov 5, 2020 02:55 |
|
Xerxes17 posted:Okay, so for example when I have it manually set to 1.280V in the BIOS, and then when I put it under Prime95 stress test, seeing a drop to 1.256V in CPU-Z is expected behavior, correct? It’s expected behavior but it’s also behavior we want to try to make stop. Its why it crashed. If you have any more LLC levels if you move it up a step you might see that droop get smaller. The droop is actually something most stock CPUs want to do because it prevents him from getting too hot under full core loads. We dont want it, in fact one step up for me gave me a tiny voltage uplife under full load. 1.385 to 1.394 under load. I am on step 6 out of 7. But my wife is on step four out of four because her motherboard is lower quality so she gets a little bit of droop no matter what. I regret the bundle deal I got with her setup.
|
# ? Nov 5, 2020 03:03 |
|
spunkshui posted:If you CPU paste is anywhere near as old as that cpu I would recommend spending 10 bucks on some thermal grizzly ceramic stuff. Until yesterday I was using the stock cooler with the original paste, but I've replaced that with a 212X and fresh paste now. spunkshui posted:Its expected behavior but its also behavior we want to try to make stop. Its why it crashed. I had changed the LLC amount to 50%, but now I'll try with 100%. Edit: 100% would put the V up too much, but 75% does the job correctly. Xerxes17 fucked around with this message at 03:24 on Nov 5, 2020 |
# ? Nov 5, 2020 03:11 |
|
Xerxes17 posted:Until yesterday I was using the stock cooler with the original paste, but I've replaced that with a 212X and fresh paste now. Nice, now you want to pass like 10 mins of stress.
|
# ? Nov 5, 2020 03:34 |
|
spunkshui posted:Nice, now you want to pass like 10 mins of stress. Well at the moment I'm currently going through the voltage adjustment process for temperature as a BIOS set 1.280, 1.270 or 1.260 gets the CPU running too hot straight off the bat. Edit: that is to say, above the 73℃ target. Edit2: 1.250 and 1.240 both get crashes under quick stress tests. Considering that I'm getting core temp issues at even .260, I think I'll have to drop the multiplier to x44. Xerxes17 fucked around with this message at 03:47 on Nov 5, 2020 |
# ? Nov 5, 2020 03:36 |
|
Xerxes17 posted:Well at the moment I'm currently going through the voltage adjustment process for temperature as a BIOS set 1.280, 1.270 or 1.260 gets the CPU running too hot straight off the bat. 73C is like super ideal, 75 or 77 isn’t really the end of the world or even 8OC if that only happens after 10 mins. Your cooler is on the smaller side but games will run cooler then stress tests. If you have the ability to change your computers case fan speeds I would turn that all the way up to 100% just to see what that does. or just take off the side and point a fan at it. If it doesn’t change anything then you might just be looking at the limits of your cooler. You can run OCCT in the background and graph your temperature while you play games. Edit: occt off of course, just graphing spunkshui fucked around with this message at 04:22 on Nov 5, 2020 |
# ? Nov 5, 2020 03:59 |
|
Xerxes17 posted:Well at the moment I'm currently going through the voltage adjustment process for temperature as a BIOS set 1.280, 1.270 or 1.260 gets the CPU running too hot straight off the bat. Where did you see that 73C should be your target? I'm not super familiar with the ivy bridge stuff but that seems far too low. As mentioned above by Spunkshui you can go up to 80C without real issue and this guide states up to 90C. As spunk mentions, your real-world applications will be probably 20C cooler than what you're getting with synthetic tests so don't be afraid to let the temps get a little close to the max of what you're comfortable with when you're setting up the overclock, especially with p95 which is far more brutal than any real world process. e: For content's sake: I upgraded the stock H115i Pro 140mm fans and the 120mm rad fan on my 2080ti hybrid kit to Noctua versions. I then went in manually to see where I notice the fan noise from each fan and set custom fan curves so it's completely quiet when I'm not using it for games/rendering, and ramps up as you'd want it to for when you're using it. Overall the corsair fans were already pretty quiet but the 120mm EVGA fan must have been fairly loud because now I can run the rad fan at 40% and it's completely quiet (previously it would auto-off until the GPU temps went over 50C). VelociBacon fucked around with this message at 04:22 on Nov 5, 2020 |
# ? Nov 5, 2020 04:08 |
|
VelociBacon posted:I upgraded the stock H115i Pro 140mm fans and the 120mm rad fan on my 2080ti hybrid kit to Noctua versions. Team fans for sound, not RGB! Now with silent fans you might notice soft fishtankish sounds from the AIOs spunkshui fucked around with this message at 07:39 on Nov 5, 2020 |
# ? Nov 5, 2020 05:39 |
|
spunkshui posted:I mean they just don't do it. I don't even know if its stable because it doesn't want to try for some reason. I'm not sure if I'm misunderstanding you, but C-states don't have anything to do with downclocking the frequency multiplier. That's speedstep or whatever Intel calls it. C-states are just a way for the CPU to turn parts of itself off when not in use, reducing power consumption while idling. They can be in use at any multiplier.
|
# ? Nov 5, 2020 07:47 |
|
TheFluff posted:I'm not sure if I'm misunderstanding you, but C-states don't have anything to do with downclocking the frequency multiplier. That's speedstep or whatever Intel calls it. C-states are just a way for the CPU to turn parts of itself off when not in use, reducing power consumption while idling. They can be in use at any multiplier. Oh, well I learned something. Speed step seems to do nothing then. Ill have to investigate cstates when I can turn this thing off. Now that I googled what they are I kinda want to make sure they are off anyway since a quick google shows stability issues could exist. Idle power draw concerns are for laptops
|
# ? Nov 5, 2020 08:09 |
|
spunkshui posted:73C is like super ideal, 75 or 77 isn’t really the end of the world or even 8OC if that only happens after 10 mins. Your cooler is on the smaller side but games will run cooler then stress tests. VelociBacon posted:Where did you see that 73C should be your target? I'm not super familiar with the ivy bridge stuff but that seems far too low. As mentioned above by Spunkshui you can go up to 80C without real issue and this guide states up to 90C. As spunk mentions, your real-world applications will be probably 20C cooler than what you're getting with synthetic tests so don't be afraid to let the temps get a little close to the max of what you're comfortable with when you're setting up the overclock, especially with p95 which is far more brutal than any real world process. From the OP of this thread, or some other guide, I wrote down that the official Intel word for for sustained max temperature is 73℃ to avoid lifetime shortening/instability. But you are right, when stress testing like in P95, having it go up to even 90 is fine if the test isn't going for too long. And yes that "real world" temp from the test is indeed -20 from P95 bench. With the latest setup: x45 Multiplier, 1.285V core in BIOS, I ran OCCT for 15 minutes (core temps 77 and under), P95 for 15 minutes (81-90℃ core temps), and then a few rounds of War Thunder both air and ground battles with OCCT tracking in the background (70℃ max recorded temp) with no crashes, WHEA or BSOD. I'll test a few more games but I think the current setup is good for my purposes. Thank you both and Lockback for your help btw
|
# ? Nov 5, 2020 08:14 |
|
spunkshui posted:Oh, well I learned something. Speed step seems to do nothing then. Turning off c-states for stability used to be a thing but it's not anymore. It has no meaningful effect for an overclock on ambient cooling. All you're doing by disabling them is making more fan noise when you're in any sort of non CPU intensive task (basically any desktop app). Probably not a huge difference but just leave them enabled.
|
# ? Nov 5, 2020 08:21 |
|
TheFluff posted:Turning off c-states for stability used to be a thing but it's not anymore. It has no meaningful effect for an overclock on ambient cooling. All you're doing by disabling them is making more fan noise when you're in any sort of non CPU intensive task (basically any desktop app). Probably not a huge difference but just leave them enabled. Im idling under 40 thanks to water so they might already be on but im down to see how low we can get this baby to idle. Xerxes17 posted:From the OP of this thread, or some other guide, I wrote down that the official Intel word for for sustained max temperature is 73℃ to avoid lifetime shortening/instability. But you are right, when stress testing like in P95, having it go up to even 90 is fine if the test isn't going for too long. And yes that "real world" temp from the test is indeed -20 from P95 bench. I'm not an expert (clearly) but this all sounds perfect to me, enjoy!
|
# ? Nov 5, 2020 08:37 |
|
It seems I spoke too soon! Tarkov will crap itself on maps other than factory and cause system instability too. Presumably because whatever system they use for loading maps really taxes the CPU. I'll have to do another voltage increment I guess, or failing that, reduce the clock multiplier by one. Edit: it could also just be that Tarkov is crash happy at the best of times... Xerxes17 fucked around with this message at 11:55 on Nov 5, 2020 |
# ? Nov 5, 2020 10:06 |
|
I think Tarkov is also REALLY CPU intensive in general, and not particularly well optimized. Have you tried it on stock speeds? It may give you trouble there too. If it works in stock, yeah try bumping the voltage or LLC curve. Tarkov might be one of those outlier games where you need to switch a profile to play. It happens.
|
# ? Nov 5, 2020 15:08 |
|
I think at this point he should leave his LLC where it is and just adjust voltage. If he's near max on voltage and bumps his LLC up we don't know if it would be way too high at the CPU.
|
# ? Nov 5, 2020 15:12 |
|
VelociBacon posted:I think at this point he should leave his LLC where it is and just adjust voltage. If he's near max on voltage and bumps his LLC up we don't know if it would be way too high at the CPU. Sandybridge was pretty resilient. The 3000 a bit less so than the 2000 but you'd really have to try to go too far. I was under the impression he was still at 50%, but if he's at 75% then yeah I'd probably tweak the voltage or just downclock for Tarkov if everything else is working well.
|
# ? Nov 5, 2020 15:15 |
|
VelociBacon posted:I think at this point he should leave his LLC where it is and just adjust voltage. If he's near max on voltage and bumps his LLC up we don't know if it would be way too high at the CPU. /agree
|
# ? Nov 5, 2020 16:02 |
|
Yep, the LLC is right now at "Very High" which is 75%, so the CPU-Z core voltage I'm getting is 1.288-1.296. As it turns out, Tarkov really doesn't like it when you disable your page file! Turned it on again and bam, no issues. Play was noticeably smoother on reserve which used to chug a fair bit, and I also noticed that my game remained smooth and controllable in close-range mag dumps which had been a problem before. I'm still often at 100% CPU on that map and get some stutter, but that's to be expected. To what I've read with a x45 multiplier, I feel that I'm already at the upper limit of what air-cooled OC'ing can do with my chip (Tarkov pushed it to 68℃ max), and pushing it further would require way more voltage, so already at the diminishing returns point. So is it really worthwhile to adapt the setup to offset power, or should I just keep trucking with manual? The next topic I'm considering is RAM optimization. I've got a rather jank DDR3 setup with 24GB total: 2x4GB G.Skill F3-17000CL11-4GBXL and 2x 8GB Kingston blue HyperX KHX16000C10D3/8G. At the moment I have them running at 1600MHz because that's the listed spec of the HyperX sticks, but the G.Skill is advertised as being capable of 2133MHz and I've skimmed some mentions of the HyperX being pushed to 1833. On the one hand I'm loath to spend money on getting some matching G.Skill sticks because that'll be RAM that can't be used in a future build, but on the other hand I've heard that RAM OC'ing isn't usually worth the effort, is that true? Plus I think the G.Skill sticks in the nearest slot could have clearance issues with the 212X fan.
|
# ? Nov 6, 2020 01:53 |
|
Xerxes17 posted:Yep, the LLC is right now at "Very High" which is 75%, so the CPU-Z core voltage I'm getting is 1.288-1.296. Ram gains not worth it really but memtest86 for an hour is what I used when I lowed my 3600 CL18 ram to CL16. It found an error so I increased ram voltage by 1 step in thee bios and it went away. Maybe you can find some used DDR3 for ultra cheep in your area?
|
# ? Nov 6, 2020 02:51 |
|
TheFluff posted:Turning off c-states for stability used to be a thing but it's not anymore. It has no meaningful effect for an overclock on ambient cooling. All you're doing by disabling them is making more fan noise when you're in any sort of non CPU intensive task (basically any desktop app). Probably not a huge difference but just leave them enabled. I checked, they are all on. Interestingly enough some of them are set to "don't report" but c7 is set to report. All are "enabled."
|
# ? Nov 6, 2020 02:53 |
|
spunkshui posted:Ram gains not worth it really but memtest86 for an hour is what I used when I lowed my 3600 CL18 ram to CL16. Well to be honest I'm confused on the utility of OC'ing or buying some better RAM because I keep reading comments along the line of "There is hardly any difference between 1600 and 1833 or even 2000". So if that's the case then messing with it seems pointless?
|
# ? Nov 6, 2020 02:53 |
|
When it comes to OCing RAM I don't get it. The gains are so minimal and it's so much effort. If you want to see what little difference it makes, go into your BIOS and downclock your ram and benchmark it. Consider that probably 25% of people with XMP enabled RAM are actually running it at the non-XMP profile speed and don't know about it.
|
# ? Nov 6, 2020 02:59 |
|
Xerxes17 posted:Well to be honest I'm confused on the utility of OC'ing or buying some better RAM because I keep reading comments along the line of "There is hardly any difference between 1600 and 1833 or even 2000". So if that's the case then messing with it seems pointless? I meant mostly for boredom reasons. Lotta people upgrading computers you might get lucky and snag something nice for $20 since its DDR3 stuff. 100% agree the gains are basically in the placebo effect range vs what just happened to the CPU.
|
# ? Nov 6, 2020 03:13 |
|
|
# ? Jun 18, 2024 05:05 |
|
spunkshui posted:I meant mostly for boredom reasons. Well, searching in the thread I found a link to this article: https://www.eurogamer.net/articles/digitalfoundry-2016-is-it-finally-time-to-upgrade-your-core-i5-2500k And looking at the table of FPS results, you can in fact get some serious gains from upping RAM speed, from 5-10 FPS by memory alone.
|
# ? Nov 6, 2020 03:49 |