|
Factory Factory posted:Passing IBT gives it a presumption of stability, but you should still really do a day of Prime95 to be sure. Farts. It crashes hard on prime95 testing FPU after about 30 seconds. Temps are still sub-75C. Can I throw more volts at this thing or should I back off?
|
#
?
Feb 17, 2014 05:33
|
|
|
|
| # ? May 27, 2024 14:56 |
|
|
Personally, I would not add volts, but it's up to you. People seem to be willing to do 80 C on Haswell where I would not.
|
|
#
?
Feb 17, 2014 05:36
|
|
|
Something I read while doing some OC'ing research is the concept of NOT using synthetic stress-tests to "prove" a safe overclock, as they are things one only runs to do the stress-test and will never been seen in real life. There is a x264 encoding stress tests that I see recommended as something that could be run in a loop to test for a safe real-world overclock, as that's essentially the worst thing you will see besides running a synthetic. Is this a reasonable thought process or am I missing something? Edit: I suppose this is almost entirely academic in my case because I can't get this thing stable at 4.6ghz even pumping some voltage but 4.5ghz is stable as all hell. I guess I'll just have to deal with the loss of a few mhz. ShaneB fucked around with this message at 21:44 on Feb 18, 2014 |
|
#
?
Feb 18, 2014 19:14
|
|
|
ShaneB posted:Something I read while doing some OC'ing research is the concept of NOT using synthetic stress-tests to "prove" a safe overclock, as they are things one only runs to do the stress-test and will never been seen in real life. There is a x264 encoding stress tests that I see recommended as something that could be run in a loop to test for a safe real-world overclock, as that's essentially the worst thing you will see besides running a synthetic. Is this a reasonable thought process or am I missing something? Its all a matter of what you're happy with at the end of the day, if it runs fine with everything you chuck at it on a daily basis apart from extreme tests like IBT or prime 95 and you are happy to live with that then its your choice and money to waste if it burns out. Its pretty much impossible to guarantee an overclock as 100% stable as you are running it beyond its intended specifications and have no idea what's really going on inside there. I have had CPUs pass long stress tests but still crash in every day use now and again and CPUs fail long stress tests run quite happily with general daily use and gaming.
|
|
#
?
Feb 19, 2014 03:42
|
|
|
h.264 encoding is actually pretty drat intensive, but it's not guaranteed to stress all of your processor. Your encode settings non-trivially change how much of the encode is float math and how much is integer and what particular operations are done. It's an excellent benchmark of stability, in that if your system can do it, it's stable at a very intensive load, but it's not a proper test.
|
|
#
?
Feb 19, 2014 04:14
|
|
|
I dropped $160 at NewEgg to try out the EVGA 750ti ACX FTW and it is amazing on wattage but infuriating because it runs my games around 57-59 FPS with the settings I want. I was able to overclock modestly from 1189mhz to 1230mhz before things started to get dicey. Now, I know the OP says don't gently caress with the BIOS to overvolt. But I'm reading an article where they state that the cards are capped way below their listed TDP (I know it's for mining but I'm gaming so let's ignore that): http://cryptomining-blog.com/tag/gtx-750-ti-cudaminer/ Specifically they show this for a reference card:  The 38500 values they show in that image are like 55000 on my card, so does that really mean my TDP is capped to 55 watts? Surely I can raise them since the card is currently running at 30 Celsius on max load? It has a 6-pin and everything. I've never flashed a GPU before but I'm experienced with JTAG-ing assorted electronics so assuming I can handle that, could I gain enough stability to hit 60FPS?
|
|
#
?
Feb 25, 2014 22:39
|
|
|
Zero VGS posted:I dropped $160 at NewEgg to try out the EVGA 750ti ACX FTW and it is amazing on wattage but infuriating because it runs my games around 57-59 FPS with the settings I want. I was able to overclock modestly from 1189mhz to 1230mhz before things started to get dicey. I've bios modded my 660 TI and never looked back. Even at 1.212 (hard limited max) I never get above 66*C and ~40% fan duty. I was above to overclock to 1280 mhz and 1650 mhz memory. It might have been a no no in the past but there are massive threads of people that have been doing this for at least two years with no issues. That said this is a new card it's possible it's a different story (but I wouldn't know sorry). I delidded my it-4670k yesterday, after a bit of headache, and got these results Before  After  ~17-20* drop which is fantastic. This is after a bit of rough overclocking:  Just an hour of testing (Prime95, the XTU) and some gaming, but it passed so far. But the vcore is pretty high up there. Also I had to bump the ring voltage to 1.15 and the vrin to 2.1. The issue is the vcore though. Been doing research into "reasonable" high vcore for these processors and getting mixed results. A lot of people are saying 1.45 is simply too high, but not much more info than that. It's certainly believable of course. Then they are often countered with people who say they've been folding at 1.52 volts for 9 months straight (for example), but I also can't take that as proof that it's okay. The only directly reported failures at this vcore I've come across are always associated with much higher temperatures (some +100* lol), so far. Anybody have opinions on vcore this high? I'm still going to try and tweak it to lower it some but I'm not expecting much.
|
|
#
?
Feb 26, 2014 15:40
|
|
|
Ignoarints posted:Anybody have opinions on vcore this high? I'm still going to try and tweak it to lower it some but I'm not expecting much. What Prime95 are you using? Those temps seem low even for delidding. Small FFTs will make my de-lidded/CLP'd/H60'd 4670 hit low 80C's eventually, and that's at like 1.29Vcore, let alone 1.45! I keep the fan at a very moderate speed to avoid noise, however. I'm not expert enough to know why higher voltages with low enough temps would be dangerous, really, but I could be wrong. Edit: I think part of my dreams last night included de-lidding CPUs... kinda scary that that would happen. Ignoarints posted:Just noticed how much higher your cache voltage was than mine. Was that for stability reasons? Maybe I can increase my cache voltage to lower my vcore. Also the 80+ spikes were small EFT, but blend wasn't much better. All in the past now To answer this question from the Intel thread: I based my initial choices on voltages on this thread: http://www.overclock.net/t/1411077/haswell-overclocking-thread-with-statistics I care a lot more about non-synthetic stability, so I'm going to lower my voltages a little to see if I still can do things like x264, batch importing of photos into lightroom while making 1:1 previews, playing BF4 multiplayer, etc, without crashing. Right now it's stable at everything I throw at it, synthetic or not, and temps stay good, so I'm not too concerned. I don't have AC in my place, however, so I might have to in the summer. ShaneB fucked around with this message at 16:55 on Feb 26, 2014 |
|
#
?
Feb 26, 2014 16:31
|
|
. I ended up using TX4 (?) for the water block after seeing how annoying it was for the cpu die.
|
ShaneB posted:What Prime95 are you using? Those temps seem low even for delidding. Small FFTs will make my de-lidded/CLP'd/H60'd 4670 hit low 80C's eventually, and that's at like 1.29Vcore, let alone 1.45! I keep the fan at a very moderate speed to avoid noise, however. V279 I believe, that is with blend and small FFT after, although like I said I didn't have a chance to let it run for more than an hour total. Blend would go through a few cycles of highest recorded temperature of either blend or small FFT then go back down to ~65 or so for some other tests. I switched to small eft to figure out where to set my final vcore since it more readily crashed it. XTU actually got the average temperature the highest (76*). CoreTemp is latest version, and XTU had it's own temperature reading. Hopefully it's accurate. I was also surprised at the temperatures at 1.45 since the temps at around 1.35v were definitely lower but not more than 10*. And there was just a few degrees temperature difference between 1.29 and 1.35 after delidding. Where as before delidding even the jump to 1.30 would have a intolerable increase with spikes well past 80 degrees. I was at 1.29 vcore before delidding and never went above 80* even after hours though. I'm using a thermaltake performer 2.0, which is supposed to be pretty mediocre, I am just using it because air coolers that performed the same just didn't fit on the motherboard. I let the fans run at whatever they want the noise doesn't bother me much. I'd prefer a dual fan radiator but I have an old antec 300 case that wouldn't be able to fit it. Ignoarints fucked around with this message at 17:01 on Feb 26, 2014 |
|
#
?
Feb 26, 2014 16:51
|
|
|
Ignoarints posted:Anybody have opinions on vcore this high? I'm still going to try and tweak it to lower it some but I'm not expecting much. 1.45V will literally kill your chip in one year's worth of 100% load. If your machine is a weekend-warrior type thing, that'll last you a while, but otherwise that's an awfully short lifespan IMO.
|
|
#
?
Feb 26, 2014 18:02
|
|
|
Factory Factory posted:1.45V will literally kill your chip in one year's worth of 100% load. If your machine is a weekend-warrior type thing, that'll last you a while, but otherwise that's an awfully short lifespan IMO. I still don't understand how you are running at 1.45V and doing synthetics without huge watercooling. What does IBT do for you when you run it?
|
|
#
?
Feb 26, 2014 18:06
|
|
|
I will try IBT after work, along with lengthier stress testing in general. I will also try and raise my cache voltage and lower the vcore, I didn't realize that was you in the other thread ShaneB. I know this sounds bad but a year doesn't sound too terrible to me if that's for heavy use. The computer might be on for 30-35 hours a week and not stressed that whole time (although since these are manually set voltages I'm not sure how much that matters). Either way I'm going to work on dropping the vcore. If I can't, I'll bump it down to 4.6 which is way below 1.4 volts and runs cooler. All I use it for is games, and to play with overclocking for no really good reason, so it's not a critical computer but I obviously don't want it to burn out in a few months. I really don't think I did a good job on the delidding so I'm surprised at the results. I'm positive I used too much CLU for the cpu die, and I was putting so much force on the heat spreader I actually rolled the edges up a little (think like a rolled edge on a knife blade). But as far as cooling goes it is just a $60 watercooler. Are there any other ways I can check temperature to make sure these are good numbers? Also thanks for the link, looks like I can tank my uncore to get stability at a lower vcore and ill still come out on top performance wise. I was trying to keep it within ~600 mhz but it looks like that is totally unnecessary Ignoarints fucked around with this message at 18:51 on Feb 26, 2014 |
|
#
?
Feb 26, 2014 18:24
|
|
|
ShaneB posted:I still don't understand how you are running at 1.45V and doing synthetics without huge watercooling. What does IBT do for you when you run it? I turned up my ring voltage to 1.25 and returned my uncore to stock. I got my vcore as low as 1.43 before it became unstable at 1.42. Again, short tests with XTU and prime 95 but the temperatures never got above 70* and more typically in the 60-65 range, but what really shocked me was IBT. I never used IBT before and my temps immediately soared to ~83*. I did not like that at all. I tried to lower my vcore by really cranking up the ring bus but could not get IBT to run for even 5 seconds under 1.43. I turned it down to 4.6 and I'm working my way down in vcore, currently at 1.37 and IBT hits 79 degrees. Prime95 and XTU are comparatively cool in the 60's now. Unless I can figure something out I'm probably going to leave it at 4.6 and get my uncore back up after I find a stable low vcore. Even though I never saw a temperature during normal use ever get higher than P95 I'll work my way from 4.6 as time goes on. I'm still new to overclocking, perhaps I'll figure out some more stuff to try later.
|
|
#
?
Feb 27, 2014 00:51
|
|
|
Ignoarints posted:I turned up my ring voltage to 1.25 and returned my uncore to stock. I got my vcore as low as 1.43 before it became unstable at 1.42. Again, short tests with XTU and prime 95 but the temperatures never got above 70* and more typically in the 60-65 range, but what really shocked me was IBT. I never used IBT before and my temps immediately soared to ~83*. I did not like that at all. I tried to lower my vcore by really cranking up the ring bus but could not get IBT to run for even 5 seconds under 1.43. Yeah you aren't gonna hit 4.7 unless you have a magic chip. I was stable on IBT at 4.6 but would crash running P95 small FFTs after a bit, so I just dropped to 4.5 and am fully stable.
|
|
#
?
Feb 27, 2014 01:26
|
|
|
ShaneB posted:Yeah you aren't gonna hit 4.7 unless you have a magic chip. I was stable on IBT at 4.6 but would crash running P95 small FFTs after a bit, so I just dropped to 4.5 and am fully stable. I wasn't sure that it wasn't going to work, it was just too hot for my comfort at 1.43 volts. Maybe if I had a better cooler... I dunno. Maybe later. It wouldn't work at lower vcore than that though at 4.7 ghz. Strangely IBT runs just fine at 4.6 at 1.37 but now small FFT is harder to stabilize (when it would run at 4.7 @ 1.43v). I had to bump it to 1.39 for that to be happy for an hour. But now all the tests seem happy at 4.6 at this vcore. Edit: I crashed in BF4 in the most intense thing I could think to play in (64 player metro 24/7) although it seemed like a GPU crash since I've been raising my memory for a while, but I raised the vcore 0.1 anyways. It never got out of the high 60's after 2 hours so I am pretty pleased. I finally found a nice deal on another 660 TI. Well in fact two MSI 660 TI's, supposedly never overclocked. Probably going to sell my ASUS 660 TI instead of sli'ing a MSI to it, which makes me very sad. It's such a champ. I wonder if it loses value for being bios unlocked or if I should revert to stock. It runs crazy cool even bios unlocked but I imagine people don't like that sort of thing. From what I understand overclocking two cards in SLI is a totally different beast? Ignoarints fucked around with this message at 16:15 on Feb 27, 2014 |
|
#
?
Feb 27, 2014 02:12
|
|
|
Ignoarints posted:1.43 volts. On your chip that's really risky. I'd be really surprised if you make it past 8 months running at that vcore. 
|
|
#
?
Feb 27, 2014 23:21
|
|
|
cisco privilege posted:I've stayed just under 1.4V on my 2600K and I'd be wary running 1.43 for any long-term period even with the kind of excessive cooling I'm using.. I've been trying to find out if it's degradation from voltage that really kills these or is heat an important factor too. Even at 1.45 my heat was amazingly low all things considered - until I ran IBT at 1.43 volts and it hit mid 80's. But even that isn't as high as other people at way lower voltages. Either way, I've turned it down to 1.39 for now and I'd be happy at 1.37 or so @ 4.6. The temperatures are pretty phenomenal at 1.37 which before delidding it would have probably exploded. I couldn't make the jump from 1.29 to 1.30 before at 4.4ghz, much less 1.37, or 1.45 for that matter. Interestingly enough XTU and EFT blend temperatures seem to drop pretty much in concert with vcore, but small FFT is virtually unchanged and IBT dropped maybe 4 degrees even with a 0.6 volt drop (and 0.8 volt for small FFT). In fact I was having stability problems with small FFT 1.37 vcore (and 4.6ghz) than at 1.42 volts and a 100 mhz higher clock speed when it ran fine (this is why I'm at 1.39 right now), while IBT was too hot for me at 1.43 volts but totally unstable at 1.42, but seemingly completely stable at 1.37 and 100 mhz less.  Anyways still working on it Ignoarints fucked around with this message at 23:51 on Feb 27, 2014 |
|
#
?
Feb 27, 2014 23:45
|
|
|
I got my G10 finally, and got it all hooked up. I was bummed because I wanted to use the front radiator mount but the hoses on the H55 were too short. I guess the NZXT Kraken series uses longer hoses. Luckily I had a bottom radiator mount to use. It's keeping the GPU a lot cooler while mining, like 30C, and that's with a replacement Enermax Magma running at 7V. I have some cheap stick-on heatsinks on the RAM right now, but they wouldn't stick on the tiny VRM modules. I'm currently testing temps without anything on the VRMs at all... not really into that idea.
|
|
#
?
Feb 28, 2014 00:07
|
|
|
Ignoarints posted:I've been trying to find out if it's degradation from voltage that really kills these or is heat an important factor too. Even at 1.45 my heat was amazingly low all things considered - until I ran IBT at 1.43 volts and it hit mid 80's. But even that isn't as high as other people at way lower voltages. Heat is also a factor. Electromigration happens because electrons knock atoms around, eventually pushing them out out of the intended path and bridging with another one. Both increased voltage and increased heat give electrons more energy with which to mash things out of the way.
|
|
#
?
Feb 28, 2014 00:37
|
|
|
A ton of people reported degradation on overclocking forums after running 2500Ks at 1.5v for 6 to 8 months so it's definitely not something to be discounted, and Northwood chips used to die in droves due to SNDS. Heat will certainly exacerbate the problem though. Suicide runs and binning at high voltages run a higher risk of damaging chips, but long-term high vcore is just as risky. I've always gone slightly above recommendations without issues although it's always a possibility. ShaneB, for VRMs you'll probably want to get specifically made VRM heatsinks or something. For the ones you have now try using a pencil eraser on the first before trying to attach them. Also you can use a hairdryer to bond them faster. future ghost fucked around with this message at 01:32 on Feb 28, 2014 |
|
#
?
Feb 28, 2014 01:22
|
|
|
Cool I will try to keep it out of 1.4's in the end. I decided to start back down at 4.5 ghz to get a feel for how all the settings affect the stability. So far I feel like I have to pump up vrin a little higher than normal to get lower vcore stability, at least for 4.6. So I'm seeing how 4.5 does just for a baseline. Still interesting to me is that at 1.32 vcore 4.5 ghz, small FFT produces higher temperatures (~80* max) than it did at 1.43v at 4.7 ghz (~75*). The temperatures remained virtually unchanged throughout the 1.3x range at both 4.5 and 4.6 ghz. IBT had a similar characteristic, although started out a bit higher, it is now finally getting lower temps in the lower 1.3 range. P95 Blend and XTU are downright cold at 1.32, barely hitting 60*C now and sticking in the 50's mostly. This all seems weird to me. Small FFT seems to be the last thing to crash me as well. Another weird thing is I thought large FFT was going to be the hottest, but after 30 minutes or so it barely got above 65* and usually stayed near 60 (at 1.32, I haven't been testing it with the higher clocks yet) Also I feel pretty dumb, I forgot I had turbo mode on which was pushing my uncore up to 39 or something even when I had set to 34. It immediately stabilized some weirdness from before when I turned it off. Ignoarints fucked around with this message at 03:57 on Feb 28, 2014 |
|
#
?
Feb 28, 2014 03:53
|
|
|
I think I have a source of some strangeness figured out. Even though I turned off turbo boost and set a manual uncore multiplier of 34, my processor cache frequency (this is the same thing right?) is 4000 mhz according to the Intel Extreme Tuning utility and the log of Core Temp. I even tried enabling turbo boost and setting all those multipliers to 34 in the uncore menu of my bios but it also affects my clock multiplier so I can't do it even if I set the clock multiplier to what I want (in this case, 46). In the Intel utility I am able to manually change the processor cache frequency and it seems to stick, however it actually turns my computer off within 5 seconds of any stress testing of any kind. Then when I go into the bios everything is totally messed up (all the settings are there, but nothing applies and nothing I do can change it until I reset the bios) I reset the bios and updated the firmware but nothing has changed. It is a Gigabyte Z87X-UD4H firmware F8. There are even specific overclocking guides for this motherboard and I haven't seen anybody with this problem yet. I am able to get stable this way but only with very high bus voltage, vrin, and vcore. To the point of off the charts compared to even the worst case scenarios for other people. Ignoarints fucked around with this message at 05:50 on Mar 2, 2014 |
|
#
?
Mar 2, 2014 05:48
|
|
|
So the layout of the 270x is different than most other cards, and the power management stuff is on the left side of the card rather than the right. This means that the fan on the G10 bracket isn't cooling the VRMs, which is bad. During the most intense benchmarking the VRMs were showing 90C in GPU-Z, which was, you know, bad. I had an idea to use the heatsink that was built into the stock cooling system, which required disassembling it and getting my dremel and cutoff wheel out: This allowed me to use the screws that mounted the cooler to the card in the first place, and still put the whole thing back together if I need to move the G10 to another card and return the stock cooling system to the 270x. Unfortunately I didn't see any temperature changes. The heatsink just got heat saturated and still was in the high 80s. I then got an idea:  Is there nothing zip ties can't do? I zip tied a super quiet (even at 12v) Enermax fan to a case card slot cover and around part of the G10 bracket. Temperatures on the VRMs are now about 18C lower with no added noise. Success.
|
|
#
?
Mar 2, 2014 20:59
|
|
|
ShaneB posted:Is there nothing zip ties can't do? Many years ago I pointed at them as being an example of an almost perfect invention. Cheap, simple, incredibly versatile, longer lasting than fixes like tape, and quicker and usually tighter than string. The fact you can simply attach more and more together to make any length is icing. One of my all-time favourite items. Which is an odd thing to say about a humble nylon ratchet strap.
|
|
#
?
Mar 2, 2014 21:07
|
|
|
You have me curious now that I have two cards almost touching. I can't find that temp in GPU-Z though?
|
|
#
?
Mar 2, 2014 21:51
|
|
|
Ignoarints posted:You have me curious now that I have two cards almost touching. I can't find that temp in GPU-Z though?
|
|
#
?
Mar 2, 2014 21:57
|
|
|
Weird I don't have it. Maybe there is no sensor for it
|
|
#
?
Mar 2, 2014 22:14
|
|
|
Ignoarints posted:Weird I don't have it. Maybe there is no sensor for it NVidia cards don't have them.
|
|
#
?
Mar 2, 2014 22:17
|
|
|
ShaneB posted:NVidia cards don't have them. Oh okay. They're basically stock probably wont even worry about it. My overclocking adventure is finally leveling out. Apparently if you set uncore to 3.4 it automatically goes to 4.0 (without much indication) but it will in fact stick to 3.5 ghz. Finally getting closer to average vcore per ghz now. Small FFT still the only thing that will crash me when other things wont. I'm trying AIDA64 now along with IBT, XTU, P95 Small FFT/blend but that seems to be as hard on the processor as P95 blend, however I really love the info it gives.
|
|
#
?
Mar 2, 2014 23:02
|
|
|
I cannot figure out how to lower my vcore per ghz. I'm about to just say live with it, but just in case I'm doing something wrong I'll ask one last question. My stable vcore for 4.5 ghz is a whopping 1.35 vcore, 1.9 vrin, and 1.2 cache voltage with uncore at 3.5 ghz. Cache voltage *may* be unnecessarily high for this but even with the hardest stress testing I could find it never goes above 75 degrees. 4.6ghz (and it might actually be higher than this no very long testing yet) is 1.40v vcore, 2.1v vrin, and 1.2v cache voltage. I raised the cache voltage all the way up to 1.3 to lower the vcore even 0.1 and it was unstable. The hardest stress testing I can do (Small FFT, AIDA FPU only, IBT) can put this into the mid 80's and I'm slightly uncomfortable doing long testing on this. Virtually all other stress tests are 60's though. I've never seen it go above 65 or so during actual computer use. I'm just unsure if it's stable. 4.7 is 1.45+ vcore and I've decided not to shoot for 4.7 anymore. I know Haswell cpu's can vary greatly, but this seems off the charts compared to anyone else I've seen. Fortunately my actual temperatures seem to be on the good side on the spectrum. But I'd like to know if there is some setting somewhere that needs to be changed that is requiring my vcore to be this high. If I can lower my vcore even 0.3 for 4.6ghz it would easily be under 80 degrees for the hardest stress testing. For overclocking purposes all C states are off, EIST is off, LLC is "Extreme", PWM Phase is "Extreme", and I believe that's it for settings I've changes. I plan to have c states back on after I'm done. I've seen some vague ranges like "you may need 1.25-1.35 for 4.4-4.6 ghz" but after looking at user submitted results I never actually see anyone who needs a vcore as high as mine, per ghz. It's a Gigabyte Z87X-UD4H and i5-4670k. I have never overclocked ram before because the last time I looked into it there was no benefit for me, but this time around I turned on XMP and it's running 8-8-8-24 at 1600 mhz, and "more performance" is set whatever that really means. I don't know if that affects stability, but for the majority of the overclocking I've done it was just bone stock (no XMP). I know little about RAM overclocking.
|
|
#
?
Mar 5, 2014 19:18
|
|
|
I'm looking for a quick sanity check. I installed a Havik Dual 140mm fan for my 4770k (Z87-A Asus), and set it to 4.5ghz and 1.25v. At 1.20v it is only stable up to 4.4 but 1.25v gets 4.5. Running Prime95 at 4.5ghz puts the CPU at a max temp of 90c as reported by CPUID. That's all reasonable right? If I'm only very occasionally maxing CPU? This is my first time overclocking a CPU that wasn't on an Android. I only game but some of them are CPU bound so I figured why not try it out. From the sounds of if I can't push much further than this without consequences.
|
|
#
?
Mar 11, 2014 06:35
|
|
|
The general rule of thumb is to aim for ~70c in Prime95, and definitely no higher than 80c. Around 90c you're going to start running into thermal throttling, and it's just not healthy for your chip to be that hot regardless.
|
|
#
?
Mar 11, 2014 13:50
|
|
|
Hace posted:The general rule of thumb is to aim for ~70c in Prime95, and definitely no higher than 80c. Around 90c you're going to start running into thermal throttling, and it's just not healthy for your chip to be that hot regardless. Throttling is a state of the CPU that can be seen in software that reports it. It happens when you are hitting 100C, not prior. If you are seeing 90s and aren't throttling in intel burn test (which could be seen by your mflops going down even with a higher CPU speed), you aren't going to come close to seeing that in any non-synthetic benchmark. Basically: go as high as you can without thermal throttling, because you won't see anything within, say, 20C of that in any real application. This is my experience, at least. There just isn't any reason to be worried about synthetics pushing your CPU into the 90s unless you are very conservative, or constantly run synthetics just for kicks.
|
|
#
?
Mar 11, 2014 15:17
|
|
|
Zero VGS posted:I'm looking for a quick sanity check. I installed a Havik Dual 140mm fan for my 4770k (Z87-A Asus), and set it to 4.5ghz and 1.25v. At 1.20v it is only stable up to 4.4 but 1.25v gets 4.5. Running Prime95 at 4.5ghz puts the CPU at a max temp of 90c as reported by CPUID. jealous of your drat vcore  my temperatures at 1.25 are downright cold but I can't stabilize poo poo with it. Man if I had that kind of stability I could probably hit 4.8 Just out of curiosity are you using blend or something else (small fft) with prime95, and also how long did it take to get to those temperatures? I used to use blend, and my temperatures are crazy good with it but I found out pretty quickly it's not a very good indication of stability or a good load that makes heat, unless it runs for a long time like overnight. On the other hand, blend has been a good indication of how hot I'll get under real heavy CPU use temperature wise (***for a short term test, the longer blend goes the more "unrealistically" hot it gets). If you are using blend and hitting 90's quickly (think 30 minutes), don't try small FFT or intel burn test as it will very likely push you to 100 almost instantaneously. Another thing is if your motherboard has any sort of auto-voltage settings on it can really make the heat spike. Although you should see it in CPU ID As far as the reality check goes, those temperatures seem pretty high. Are the other factors (case air flow) adequate? I ran cooler than that with an Evo 212 at almost 1.3 volts, before delidding. I feel like I should beg you to trade processors because I want your vcore stability, sounds like delidding would be really awesome with that one. Edit: Also did you use stock paste? I had at least a 5 degree+ drop switching from artic silver 5 (? the common one) to tuniq tx-4 or xigmatek PTI-G4512, and if you are using the paste that came with it there would likely be a total of 10 degrees or more on the table. Ignoarints fucked around with this message at 19:16 on Mar 11, 2014 |
|
#
?
Mar 11, 2014 19:10
|
|
|
Ignoarints posted:
Wha, you shouldn't be seeing a 5+ temp drop unless you had a problem with your installation. http://www.tomshardware.com/reviews/thermal-paste-performance-benchmark,3616-18.html
|
|
#
?
Mar 11, 2014 19:59
|
|
|
Don Lapre posted:Wha, you shouldn't be seeing a 5+ temp drop unless you had a problem with your installation. er wow, I probably just hosed it up then. There are other charts that show a ~3* difference between AS5 and say, TX-4. But at the very least if he was using the paste that came with it he could probably do way better. I don't think that would account for those temps entirely though. I might have been using Zalman paste now that I'm trying to remember. It sucked, whatever it was, I only had to use it because I kept changing coolers to find something that worked and I was low on pastes. I definitely saw a 10 degree drop for my video card, which reminds me I need to do that on my new ones Edit: Also, if I did do a bad job originally, that's something to consider for the post above too Ignoarints fucked around with this message at 20:29 on Mar 11, 2014 |
|
#
?
Mar 11, 2014 20:22
|
|
|
Ignoarints posted:er wow, I probably just hosed it up then. There are other charts that show a ~3* difference between AS5 and say, TX-4. But at the very least if he was using the paste that came with it he could probably do way better. I don't think that would account for those temps entirely though. They actually test video cards too and paste makes a lot more difference. http://www.tomshardware.com/reviews/thermal-paste-performance-benchmark,3616-20.html
|
|
#
?
Mar 11, 2014 20:34
|
|
|
Don Lapre posted:They actually test video cards too and paste makes a lot more difference. Yeah and the stock was just utter junk, it reduced up to 10 actual degrees after I was done for just paste. Was amazed Ignoarints fucked around with this message at 21:04 on Mar 11, 2014 |
|
#
?
Mar 11, 2014 20:59
|
|
|
Ignoarints posted:jealous of your drat vcore I tried all the different Prime95 options but they kind of all hit the same 90c in about 30 seconds of 100% CPU utilization, but it didn't go any higher. I'm fine with that since this is only for rare occasions when I need a fuckoff compute blast. I just felt guilty having bought a 4770k when it was on sale at Microcenter for $200 with a free FFXIV download voucher I sold for another $25, and I got them to apply the $30 mobo discount to an already open-box discounted $110 Z87-A. You can't get all that and then not overclock. I used Arctic Silver 5 since I had some at work. I used a suuuper thin layer, so thin that after a couple minutes I had 95% covered but then the index card I was spreading with was revealing bits of the heatsink metal instead of covering it up 100%. I plopped on the cooler from there, and it was surprisingly slip-sliding all over the place until I lined up the crossbar perfectly and got it clamped down.
|
|
#
?
Mar 12, 2014 00:41
|
|
|
|
| # ? May 27, 2024 14:56 |
|
|
Zero VGS posted:but then the index card I was spreading with was revealing bits of the heatsink metal instead of covering it up 100%. I plopped on the cooler from there, and it was surprisingly slip-sliding all over the place until I lined up the crossbar perfectly and got it clamped down. Don't spread thermal paste manually.
|
|
#
?
Mar 12, 2014 01:53
|
|