|
Sir Unimaginative posted:Should I be getting the same temperatures with the current Linpack at 4096MB stress level as I do with Prime95 Small FFT? (Both 72~73c on 2500K@4.2 GHz) Small FFT, for sure?
|
#
¿
Aug 3, 2012 21:13
|
|
|
|
| # ¿ May 16, 2024 14:18 |
|
|
Hold up, should be idling closer to 33ºC regardless of load. Wait, are you using the stock cooler or an aftermarket heat sink?
|
|
#
¿
Aug 3, 2012 22:24
|
|
|
Let me rewind that last statement as well, ah, how exactly are you doing the overclocking? It could be idling normal if you're going with a constant voltage. If, however you're using the offset method it should be undervolting to .9V and downclocking to 1600mhz and idling cooler than that.
|
|
#
¿
Aug 3, 2012 22:30
|
|
|
Sir Unimaginative posted:I set it to 42 and stuck with default voltages and apparently it only undervolts so low as 1.000. So, one-button overclock, and it is doing its power saving, that's good - but your temps aren't. If it's not underclocking to 16x (1600mhz) when idling, nearly 40ºC is a high idle, especially for the type of cooler that you're using, and could indicate some issue with its mounting (or, as you note, maybe just really bad airflow - that's the thing with the Xigmatek iirc, it isn't really optimized well as far as fin placement goes so a lack of airflow could mean that it needs a higher temperature delta in order to increase the efficiency of heat removal). Either way, those temps are 100% safe with Sandy Bridge, if a bit on the high side, and you can feel free to file all this away as academic if you'd like. You can run that setup for as long as you have the computer and expect no issues, I'd imagine.
|
|
#
¿
Aug 3, 2012 22:47
|
|
|
Sir Unimaginative posted:Not a one-button overclock, and yeah it's underclocking at idle. I must have thought that goes without saying; sorry. Well, basic physics of taking heat from one thing and moving it to another thing is that the greater the difference in temperature, the quicker it happens. So while you're idling high, the performance under heavy loads is good. Living where you do, with high ambient temperature and expensive as poo poo electricity, maybe you're just gonna have to deal with high idle, it's not going to hurt anything, it's just a little weird. The load temperature is fine in Linpack and a bit high for Prime95, common advice here would be that you've got 10ºC of wiggle room with the multiplier and voltage with IBT, but I'm concerned about the Prime95 score. I don't get DRAMATICALLY different Prime temperatures, but they're consistently at least 6ºC lower than anything Linpack-based, and that's pretty important when you want to push your hardware as safely as possible. Well, if you want to go farther, I'd say the first thing to do is to get away from default voltage, if by that you mean "I changed the multiplier to 42 and didn't manually adjust the voltage," which it sounds like is what happened? Hell, your temps could just be due to an overzealous automatic voltage - what is your vcore under load as measured by HWiNFO64? Careful not to take the first values you see that relate to voltage, as those are VID in that monitoring utility, you need to scroll a bit to see vCore. 24/7 safe voltage is 1.38V for Sandy Bridge, but that assumes operational temps within safe parameters too, which Intel floats as 72.5ºC*. What's your motherboard? You're getting to the fun part of overclocking now.  *I think that's bullshit for a few reasons, not the least of which is they say the same thing for Ivy Bridge and there's no way the more sensitive and heat-issue-prone 3D transistors in IVB exhibit the same resistive, capacitive, or inductive properties, or experience electromigration exactly like Sandy Bridge, but they're being conservative and that's their prerogative, I'd estimate the real safe long-term temperature at 75-76ºC with infrequent excursions to 80ºC under extraordinarily heavy loads, within safe voltage parameters... Well, that's neither here nor there.
|
|
#
¿
Aug 4, 2012 00:27
|
|
|
n0n0 posted:So... I've got a 2500k overclocked to 4.3ghz (planning on pushing it to 4.5) in a nice, roomy case with plenty of fans and pretty good cable management. The heatsink is a Hyper 212. I used the thermal paste that came with the Hyper 212. While compatibility can vary by motherboard, mainly, we don't trust CoreTemp. We do trust RealTemp (run as Admin) if all you want to see is momentary clock, load%, and temperature in a very resource-light kinda way. HWiNFO64 (run as Admin) is the most accurate utility for all-around sensor detection and reporting. Also, make sure to run any and all stress tests as Admin, too, or they may be limited and not giving you good results. E.g. Prime95, running with user privileges only, it'll make a guess at hyperthreaded cores and the relationship between physical and logical cores; in admin mode, it knows and assigns workloads with 100% accuracy as a result. IBT also seems to be more effective at stressing the CPU in admin mode.
|
|
#
¿
Aug 12, 2012 05:43
|
|
|
Dubplate Fire posted:I am concerned about CPU temps on my i5-3570k. I replaced the stock heat sink with a Hyper Cooler 212 Evo I noticed that temperatures did not really change. When the CPU is running stock speeds it can hit up to 65 degrees on full load, which is worrying me. When I overclock to 4.2 ghz, it can go up 75 degrees. I tried reseating the heat sink, which did not help. I am going to reapply the thermal grease to see if that makes any difference. Should I be worried about these temps? If nothing changes, should I consider getting a water cooler? Those temperatures aren't sky-high by any means, but they could indicate an issue. 1. Incomplete installation of the heat sink, slightly (did you tighten it down fully?) 2. Misapplication of the thermal paste (too much and it's an insulator, and with the type of heat pipe cooler that the Evo is, you need to observe some best practices that you'll need to adjust only very slightly to make up for the non-bracketed design of the Evo's heat pipes) 3. Some minor malfunction of the heat sink that would be a warranty issue (if it were a major malfunction you could expect substantially worse temperatures) 4. Really poor air-flow inside the case preventing it from exhausting properly, but it would have to be bad enough to basically trap air inside the heat sink's fins, like, "your rear 120mm fan is backwards and fighting the Evo's fan" bad. Do not consider getting a water cooler. It's an inefficient technology, and while it can give you better temperatures under extraordinary usage conditions, you're looking at dramatically higher noise, power draw, etc. in order to get remotely the same level of cooling relative to the amount of radiator space. Heat pipe and fin designs are really efficient, even the 6mm non-optimal ones in the Evo and most other inexpensive heat pipe coolers are, for surface area, far, far more efficient than liquid cooling. There are many heat sink benches out there if you'd like to examine this yourself, liquid cooling loops are noisy, inefficient, and the failure scenarios for water cooling can actually destroy your components whereas if a fan goes out on your heat pipe + fins cooler, it'll probably still have enough passive radiation to allow a very graceful transition to a new fan. You might try a push-pull setup, that was good for something like 6-7ºC of additional cooling on my Hyper 212+ that's cooling my old/backup comp with a Q9550 that has an overclock from 2.8GHz to 3.8GHz.
|
|
#
¿
Aug 30, 2012 17:43
|
|
|
Dubplate Fire posted:I just reapplied the thermal grease and I am getting 5-10 degree lower temperatures for now. i'll run a 24 hour stress test for 24 hours and see what happens. Highly recommend you do Prime95 for that one, IBT for that long risks damage. You probably already know that, but just a FYI in case.
|
|
#
¿
Aug 31, 2012 00:53
|
|
|
You could just have kind of a hot-running chip. We've certainly seen them before in IVB, thanks to the construction of it. You can either not worry about it, or get [H] and use a razor to remove the IHS, reapply a superior thermal paste, re-seat the IHS, re-seat your cooler, and maybe drop 5-7ºC if that is the cause.
|
|
#
¿
Aug 31, 2012 15:23
|
|
|
Lot of it is luck of the draw. The Q9550 could take quite a lot, but close to 1.6V is pushing even that. For reference I'm at 3.8GHz on mine at 1.38V. It'd do 3.4GHz on stock voltage. My DDR2 is Geil 1100MHz and requires an even higher voltage than that, 2.2V. DDR2 was when ram heatsinks got really silly in the first place, since it was never standardized past 800MHz yet for obvious reasons people wanted to push the clock as the lowest multiple might exceed 800MHz depending on your CPU overclock. So they just binned DDR2 modules that would run at whatever voltage. For what it's worth, the RAM has worked since 2008 like that and today still gives no memtest86/+ issues. The CPU voltage seems really high, very much "just turning it to 11 and hoping for the best," as you put it. Noctua hasn't made an outright lovely cooler, though their earlier parallel mounted model (NH-C14) doesn't offer the kind of performance you get out of the dual-tower perpendicularly mounted NH-D14. NH-C14 was a champ at quiet performance when it came out, and for a non-silly overvolt it should still cool an LGA775 CPU just fine. Obviously not ideal to have a poo poo-ton of thermal paste everywhere acting as an insulator. Your max temp looks fine; they idle higher because their voltage is always on and they don't downclock as much. That said I'm using a Hyper 212+ in a push/pull config on mine and it idles lower than that, around 37ºC iirc. Pretty much the same max temperature, though, which is what matters more for longevity of an already old chip, and I'm fairly sure that's quite safe. Gigabyte's always had a bit of weirdness to their overclocking from my perspective, I never had to mess with anything but the CPU and DDR2 voltages to get my OC Linpack/Prime95 stable. But could just be pure luck of the draw. You didn't say what your voltage was, exactly, though.
|
|
#
¿
Sep 17, 2012 13:06
|
|
|
GPUs are overclocked from within the OS environment. However, any card in the HD 3800 series is painfully dated at this point. HD 3870 vs. HD 5870 to establish a baseline for comparison, then look at how the 5870 stacks up against the current-gen price:performance cards from each company: nVidia: 660Ti AMD/ATI: HD 7850 All the overclocking in the world isn't going to bridge that performance gap and let you play current games at respectable settings if your resolution isn't really, really low. You're also boned on CPU overclocking  Usually brand names use proprietary OEM'd motherboards with stripped down BIOS options that disallow voltage control, etc. that you need to properly overclock. Though that's not your biggest problem. Even if you did find a way to overclock it, say with an OS tool which specializes in letting you overclock processors in OEM setups, the fact is even if you got it up to a higher clock you're still kind of screwed. An Athlon II X2 265 at 3.3GHz compares poorly to a previous-generation (Sandy Bridge i3-2100) Intel dual core CPU at 3.1GHz. If you want a comparison to a standard recommendation for a gaming computer today, Anand seems a little behind with most of their CPU benches in the i7s of Ivy Bridge, so instead compare the Athlon II X2 265 at 3.3GHz vs. the Sandy Bridge i5-2500K at stock clocks. Ivy Bridge is actually faster at the same price point, but they don't have an i5 for a budget-friendly comparison, just a bunch of i7s that have features you don't need, features that only marginally improve games by a few FPS at most, usually within the margin of error for benching. Going to need to upgrade at some point soon if you intend to play any current games at decent framerates, I'm afraid.
|
|
#
¿
Sep 18, 2012 03:30
|
|
|
Farecoal posted:Yeah, I've got a 1600x900. Thanks for the advice though, I was just wondering if I could get a little more life out of this computer. 1600x900 is just too many pixels for a card that old to provide a good experience with where modern games are concerned. You might be able to drop down to 720p to get a little smoother experience at the cost of some fidelity, just make sure to let the card do the scaling. Even quite good monitors have crap scaling compared to video cards, and it's basically free in terms of performance, netting you a lot of benefits so long as you stay at the same aspect ratio. If you're relying on your monitor to scale, it'll probably just be a garbage nearest-neighbor afterthought scaler or something like that. You'll still run into the CPU bottleneck on a lot of games, I'm afraid; we're finally at a point where if you want to have smooth gameplay, you really need to have at least a 2008-era quad core at a decent clock speed, or a modern dual core that can perform like a 2008-era quad. Anything lower than that is going to choke when anything somewhat sophisticated comes down the execution pipeline, and games are getting better at utilizing more cores more efficiently (less "game runs on Core 0, some easily synced stuff runs on Core 1 but barely loads it" and more actual multi-threaded games). Agreed fucked around with this message at 03:56 on Sep 18, 2012 |
|
#
¿
Sep 18, 2012 03:54
|
|
|
There is really no reason to set the power target below its maximum value. The card won't exceed what it can safely run at regardless, that's controlled in hardware. You can test and observe that yourself by firing up OCCT or Furmark and watching it hit the max power draw you set, then gradually back off in your measurements as the card throttles to protect itself and stay within hardware parameters. Remember that all GK104 chips are pretty tight with their bandwidth out the gate, and that overclocking just the GPU and cores will actually run you up against their bandwidth limitation. nVidia has really nice integrated memory controllers this go-around so your overclock needs to be balanced between the GDDR5 and the GPU/shaders. Luckily most manufacturers seem to be using fast fast fast RAM and a "middling" VRAM overclock is like 6400MHz, or a solid 100MHz of real overclock. That was tough to get out of Fermi cards. Many are successfully running VRAM clocks at or above 6700MHz. It all factors into the total power draw that you can have, so at some point, and where it is exactly depends on the luck of the draw, the quality of your specific sample, robustness of the design, parts picking choices... But at some point, you might run into your TDP limit and have to decide in favor of increasing the core clock OR the speed of your GDDR5. Assuming you're not kneecapping it (it'll use the full bandwidth of PCI-e 2.0 16x, or PCI-e 3.0 8x), generally you want to find your maximum memory clock first, then back off from there to feed the core 'til the core becomes unstable. Things that help in my experience: 1. Set the voltage as high as it will let you. The card will do it on its own, but it has to make the jump. It's almost but not quite instantaneous in doing so, and that little tiny moment where it's not at the full voltage required might be enough to cause instability and a driver crash. Otherwise, with a high overclock, you'll see it bounce around a bit between the top three available voltage steppings but in anything remotely demanding, spend *almost* all of its time maxed. So just do it a favor and give it that to start with, it's not going to hurt it (again, it does it on its own anyway, you're just stepping in and saying "don't decrease it when under load conditions!" so that you're never screwed for lack of voltage. 2. In games that you know will be demanding on the card for one reason or another, set it to Prefer Maximum Performance rather than Adaptive. This is basically an extension of the above recommendation, but applied to everything about the card, especially how aggressive it is with its turbo. If something's on the edge of stability, telling the card to give it everything it's got has absolutely been the difference between getting a driver-not-responding (aka "your overclock is not stable, buttface  ") versus uninterrupted, fine gameplay.3. Aggressive fan profile. If you want your card to perform, keep it under 70ºC. I have a reference-ish blower (it has some sound dampening stuff on it) and I've got my profile set to kick the fan into full operation at 65ºC. The card will automatically throttle in 13MHz increments at 70, 80, and 90 degrees Celsius, but it should never get to those temperatures in the first place as in my experience a cool card is a happy card when it comes to overclocking. This just amounts to helping it work like it should anyway; Kepler is a cool running design and even the reference cooler is great at moving large amounts of heat away quickly, since it's using a nice vapor chamber design. Finally, just as a general point, a rock-solid power supply is key to having a stable overclock of any component in your system. General instability is usually the motherboard or the power supply. Specific instability is more likely to be specific components. Not every card is a massive overclocker, just like not every CPU is, not all RAM hits unnecessarily fast speeds, and so on.
|
|
#
¿
Sep 24, 2012 00:27
|
|
|
craig588 posted:I finally got around to messing around with the power targets in my 670s bios. My peak boosted overclock is unchanged, but now under load the core clock stays within 50Mhz of the peak clock. Before it'd swing as far as 200Mhz below the peak, depending on whatever was happening in the game. (OCCT would even let it go from 1300mhz to 950mhz before, now it only drops to 1200mhz) Temperatures are up by about 7 degrees, but nothing to have me worried. That's quite a lot of core voltage, any stability issues in different test cases? My jealousy is showing! I was at 1296 but Metro in DX10 didn't like it and Starcraft II in DX9 didn't like it. But if I had more voltage... Realistically, I don't know if I'd void a warranty on my EVGA card with the whole warranty and advance RMA crap for 20mhz, just tempting... The clocks, the CLOCKS! 
|
|
#
¿
Oct 8, 2012 00:40
|
|
|
Chose wisely. Performance for noise is hard to beat, I love this gigantic mother of a CPU cooler. It's kind of hilarious if you think of the ratio of radiator space to the size of the transistors themselves. Buncha 22nm transistors makin' lotsa heat in a tiny package? No problem, hook this car radiator son of a bitch up to it, welcome to (among) the best and quietest overclock(s) you can get on air.
|
|
#
¿
Nov 6, 2012 17:54
|
|
|
SocketSeven posted:I played with the Asus auto overclock utility. I'd do a few quick Extreme linpack (well, IBT) runs first just to establish basic stability and then never use that again. I was worried that you were doing this to a Haswell processor, but Prime95 and IBT do a pretty solid job of stability testing Sandy Bridge. If it'll Prime95 for a few days, you'll be solid, I reckon. It's absurd though that you find yourself heat limited at that clock with such low voltage; that's the "pretty much every chip can get this" clock range for the 2500K/2600K/2700K, what precisely is the cooler you're using? It kind of sounds like it is not doing its job very well. I remember when I first OC'd my 2600K, which I got before they segmented it further into a 2700K and a 2600K in the SB i7 line... It was trivially easy to get to 4.5GHz. Landed on 4.7 after the hard work. But I'm using one of the gorilla coolers. Nonetheless, I find it really peculiar that your 2700K, binned for performance, and performing well at a relatively low voltage, is somehow putting off too much heat for your cooler. Frankly makes me wonder if the cooler is a good choice 
|
|
#
¿
Jun 5, 2013 08:09
|
|
|
I'm leaning that way, while it is an unorthodox design, it's still packing a solid compliment of heat pipes (6mm, not ideal, but that's the usual size for coolers like the Hyper series too so I don't think that ought to be held against it). The biggest problem it seems to have is just that regardless of heat pipe wicking capability, there isn't enough radiator area. The fins aren't long enough, it's not put together in a way that allows for the kind of extremely rapid heat transfer we can expect from modern coolers. Still, it should be doing a hell of a lot better than it is. Try reapplying with a thin bead of TIM straight down the middle of the IHS, or maybe a thin X. I've had good results from either a thin X or a thin + on Sandy Bridge, for what it's worth. While the basic design of the cooler seems married to form over function, it is nonetheless - within that constraint - a pretty solid cooler going by specs. I mean, have a look at the heat pipe and fin setup on it - it's not what we're used to seeing but they're using the same kinds of effective technology and doing so in what I think are efficient ways, given that their first objective is it still has to look like a Zalman cooler instead of yet another tower, and it's lower profile. Edit: Delidding a SB processor would probably kill it. It's soldered on. It's with IVB they started loving around with TIM under the IHS like big jerks, and have too much space between the CPU and the IHS at that
|
|
#
¿
Jun 5, 2013 08:50
|
|
|
SocketSeven posted:Yeah, I was joking about that since I've hosed up a few things this week, I might as well go for breaking something expensive. 2mm is a bit thicker than I'd go. Width of the nozzle and no wider. You're filling microscopic gaps, anything more than that turns the TIM into an insulator and overapplication is an extremely common problem. I hosed up the first install of my NH-D14 when I built this computer, uuuuugh. But I got a solid 6-7ºC or so (hard to remember, it was June 2011) just from reseating it with a proper, spare application of TIM. I use Noctua's TIM, by the way, but the difference between generic goop and the best poo poo on the market is about 2-3ºC, being generous, so it comes down more to what you prefer - something viscous, something goopy, blah blah blah. Arctic Silver 5 is a little long in the tooth these days if only because it has a problem with dessication that many TIMs don't now, but that just means you "get" to reapply it in a couple years, woo. Lapping isn't bullshit if you're going for THE ULTIMATE OVERCLOCK, but for our purposes (getting to and past 4.5GHz at comfy temps) it should be completely unnecessary. If you're aiming for 5GHz, then it's time to get out the oh-so-fine sandpaper, amigo. 
|
|
#
¿
Jun 5, 2013 09:14
|
|
|
I have three fans on mine  Keeping it big in '13
|
|
#
¿
Jun 15, 2013 22:59
|
|
|
For what it's worth I did enjoy watching you do a proper rundown of your OC experience with the chip. And, hey, if your last OC was in 2003, lapping etc. ought to feel right at home Sorry you couldn't get a better result, but that is some useful information to have. Screw the totally offhand way they're treating overclockers since IVB. Remove unnecessarily and detrimentally large glue pad around IHS, save 20ºC. Like THAT'S not a big deal. Probably enhance the lifespan of the chip in heavy usage scenarios, they're putting the screws to more than just us desktop users here in the big picture with that crazy decision.
|
|
#
¿
Jun 17, 2013 00:52
|
|
|
Guni posted:Are water-cooling setups more recommendable for big-overclockers now that Haswell is out? IIRC Sandy and Ivy Bridge didn't really gain anything by having a water-cooler, as top-end air setups were about as big as you really needed to go. Is this not the case anymore? If you delid the IHS and remove the unnecessarily (detrimentally) thick glue, it makes vastly superior contact with the CPU, as it is supposed to. For that scenario, you don't need nor will you particularly benefit from watercooling in any closed loop format. The gentleman above went direct-to-die. No IHS at all, so he's mounting a water block directly to the chip. That's a whole different can of worms, because while the IHS is engineered such that with the help of heat sink mounting brackets, it can withstand a VERY solid contact - 60-80lbs of mounting pressure in pounds per square inch - a naked CPU is basically extraordinarily porous glass and would be crushed by the weight of mounting it without massive changes to the whole setup. This is not nearly as much of a concern with a water block mounting bracket setup, when properly shimmed. Delidding and fixing their engineering don't-give-a-gently caress to nab 20ºC for your effort is probably going to become a more common practice with Haswell. Going direct-to-die will probably remain a really niche thing because of the risks involved. Either way, water loops are still inefficient and noisy and the NH-D14 can outperform most of the non-absurd ones with their dumb loud fans on high and noisy as hell, while remaining whisper quiet. Edit: To be clear, if you don't go to the trouble to de-lid the processor and clean the overuse of glue (not thermal paste, though you'll need to reapply TIM too and why not use the opportunity to do it right with your favorite TIM  )? Yeah, you'll be heat limited. Basically, Haswell is relying almost entirely on the TIM to transfer heat between it and the IHS, and that puts a limit on how fast anything can remove heat. You have to ensure it's making proper thermal contact by delidding it, removing the glue, replacing the TIM, and re-lidding it before mounting your cooler of choice if you expect to beat the thermal wall that many "average" chips will experience. There may be some EXCEPTIONALLY good chips that require very little voltage to hit higher clocks and so work without any of the delid/relid  , but it's not an OC friendly generation in that regard. Worse than IVB by quite a lot in fact.
Agreed fucked around with this message at 07:20 on Jun 17, 2013 |
|
#
¿
Jun 17, 2013 07:14
|
|
|
deimos posted:Well this is certainly a thing: http://www.guru3d.com/news_story/amd_a10_6800k_reaches_8ghz_mark_on_all_cores.html 2V and -186ºC: AMD, ladies and gentlemen! Awesome that he did it with an APU, iirc he's a prominent suicide runner and that's pretty loving funny to see. Maybe next he'll get out a Pentium M and see what he can do
|
|
#
¿
Jun 18, 2013 19:26
|
|
|
Those aren't temps I would worry about, personally, no. I sometimes think that in the pursuit of cool operating temps we sometimes forget that Intel engineers these things to be capable of running at their maximum package temperature, at their maximum clocks 24/7 for the operating life of the processor. I guess if you run F@H or other distributed computing that might be more applicable to you, but I don't think most usage scenarios, even overclocked and overvolted, are going to do anything more than reduce the useful lifetime of the processor some. And who knows how long that might be?
|
|
#
¿
Jun 19, 2013 23:47
|
|
|
cat doter posted:So I'm poo poo outta luck? I probably should have checked motherboard specific stuff before buying it. Is there a way around it or do I just have to stick to default clocks? You're SOL on that motherboard. It's generally better to measure twice, cut once when it comes to things like this, but see if you can return the motherboard and get an overclocking capable model?
|
|
#
¿
Jun 21, 2013 11:41
|
|
|
cat doter posted:It's weird, I'm usually the type to over prepare, but I didn't this time. Gotta gently caress up some time. Yeah, even in its stock form it'll make your Phenom 955 look like an rear end in a top hat, but you bought the -K variant for a reason, and that reason is to make the stock unit look a tad bit slow. Make sure you've got a good cooler, IVB runs hot thanks to Intel's "We Don't Give A gently caress, You Fix It" approach to IHS construction after they stopped soldering them (post Sandy Bridge, basically). It isn't Haswell hot, now that's a feat of engineering idiocy w/r/t the heatspreader and the glue used that makes IVB look like a frosty chip by comparison. If you can exchange, do so. Then go apeshit with a good cooler, because why not? Stable high clocks are kinda cool if you do anything that can use them. I feel like I straight up stole performance with my Sandy Bridge by pushing it to 4.7GHz fully stable, it drastically outperforms a stock 2500K or 2600K. Like, twice as fast in some things compared to a stock 2500K. Going from turbo-by-core with SB and IVB to "turbo on all cores," an overclocking feature, is a huge boost to anything multithreaded because it tells the processor to ignore TDP limitations and keep all cores at the same speed. So instead of one core boosting to whatever its Turbo bin is, you get all four cores boosting to your new overclocked speed. Do the math on it, it's not entirely linear because of the gestalt of overclocking being a bit more complicated than that, but it does provide very substantial improvements to anything using more than one or two cores at a time.
|
|
#
¿
Jun 21, 2013 12:46
|
|
|
unpronounceable posted:This is an interesting development. Apparently ASRock has figured out how to overclock K-series Haswell CPUs on B85 and H87 mobos. I didn't notice any manual settings for the overclock though. I wonder if there will be any developments for non-k CPUs, like setting them to the max turbo bin as you could for SNB and IVB. This seems like something Intel might want to squash. And it is within their power to do so, a chipset revision to fix whatever loop-hole ASRock has figured out and we're right back to original market segmentation. In the meantime though, loving good on them for pushing the envelope and doing something unexpected. It's not nearly common enough anymore for motherboard makers to get tricky and find ways of doing things that are outside the rules set up by Intel, ASRock seems like they're both pretty damned clever and pretty damned plucky for figuring out some means to bypass the non-Z lockout. Rock on, ASRock 
|
|
#
¿
Jun 27, 2013 17:20
|
|
|
Stew Man Chew posted:So this might be slightly out of place here but I have a overclocking related question. How far out of normal voltage did you have to run it? If it wasn't too far from nominal, I'd sell it in plenty good conscience, no reason to expect it to die noticeably sooner than any other processor. Remember that Intel's 24/7 numbers are no-poo poo intended for what the chips are capable of being run at 100% 24/7 (e.g. in a server rack, or if you were devoting all of your free cycles to Folding@home), so if your clock cycles haven't been that level of usage, other stuff is almost certainly going to croak before the CPU regardless. They're hardy. On to my own OC thing... I am thinking that with the new and improved airflow I may be able to run 4.8GHz on the processor, for a "oooh that's nice" showier overclock. IBT below Very High has trouble getting my processor over 72ºC now, and that's with the side fan dead but left mounted, slightly impeding and certainly not helping airflow. Once I get it moved up to the Spectre Pro, think things should be amenable to a change. And my vcore is close to 1.4V anyway, if I can get it to 1.4V I have zero problem running that as I've got that "overclockers' special warranty" thing for a no-questions-asked replacement officially. Does Intel still sell that, or was I part of a pilot program that didn't become a bigger thing? One thing I will do is replace the front and middle NF-P12/P14 fans on the NH-D14 with their re-engineered counterparts, though. Those NF-F12/A15 are awesome fans by any measure and the NF-F12 in particular is a spectacular engineering achievement, specifically designed to put a lot of static pressure through a heat sink or a radiator and excelling in that role to the extent of several degrees cooler at the same inaudible noise level - and only that role. It makes a mediocre case fan and works very poorly as a pull fan in a push/pull (or push/pull/pull, as it were) setup. The NF-P12 remains the best third pull fan in the lineup, though, no benefit at all to replacing it with anything else, especially with the kickass Prolimatech PRO-USV14 working right along with it to pull the column and exhaust it. Could not be happier with that case fan, seriously, if you've got a 120mm mounting hole and you need a fan that is silent, long-lived, and moves about 100CFM, check that bad motherfucker out. I think if I give the cooler these last little tweaks, finish the airflow setup with the Spectre Pro coming in today, and maybe remount it to ensure best possible contact, I can get an extra 100MHz/core at roughly 1.4V. Tribute to the power of air cooling that it's just .004v away from that already and Linpack can't get it past 72ºC, I think. Water cooling may be able to brute force its way to better gross performance, but this remains an extremely high-performance cooler and it seems to have just enough room to improve yet that I might be able to squeeze the 48x multiplier out of it that I've wanted ever since I put the system together. I might also see about bumping up the BCLK from 100.00MHz to 103-104MHz, which might not cause instability if I'm careful, and would get me drat near 5GHz.  I dunno, though, We'll see how the attempt at 48X goes before shooting for the moon.
Agreed fucked around with this message at 16:31 on Jul 2, 2013 |
|
#
¿
Jul 2, 2013 16:24
|
|
|
Factory Factory posted:It looks like the Intel Performance Tuning Protection Plan is still kicking, with entries for the 4770K and 4670K. Oh so tenuous. I mean poo poo they'd probably send me a 2700K as a replacement since they retired the 2600K but I dunno, it's not like this is a one in a million chip to begin with - it has done well as far as not walling out at the multiplier but it runs kinda hot and takes some real voltage to get where it's going, compared to the really high percentile chips. It's a good one and if I melt it I will be very sorry, but it's not top tier. I'm some kind of awful miser because I don't run F@H so a solid amount of my processor cycles are just the idle loop process, and it'd be pretty neat to have it running at or near 5GHz just for the fun of it fully stable once I get the final touches put on my cooling setup (final? well, I did start in June of 2011, not my fault Noctua takes their time making better fans ).Sandy Bridge: On air, no delidding required, rock and loving ROLL 
|
|
#
¿
Jul 2, 2013 21:59
|
|
|
Well, the difference in performance between the Bitfenix Spectre Pro and the Xigmatek XLF 200mm fans is absolutely nuts. Right at twice as much airflow, and now I have a nice, balanced intake vs. output, nudged slightly in the negative pressure direction but with magnetic DEMCiflex filtering to prevent dust build-up. Dropped temps across the board by a whole hell of a lot. My GTX 780 is idling at 22ºC. Woah. Balance and making the air go where you want it is the key, and the tricky part, and the rewarding part. Case fans are done now, I think. Final profile is all fans operating at full voltage, still inaudible since I'm only using one high-CFM 200mm fan (if I had three Bitfenix Spectre Pros, I imagine I could hear them - that's where the Xigmateks are better, but with roughly half the airflow, they damned well ought to be inaudible by comparison). The side Xigmatek failing is a bit of a blessing in disguise as it made me reconsider my options and this is clearly a better choice for the side fan and sole intake. The rest of the fans can split the air it brings in, it supplies plenty of fresh cool air to everything and when that air becomes hot it can leave the case. It may sound odd to have a front exhaust, since you wouldn't think air would work that way, but there's a reason. I've got two graphics cards which both exhaust inside the case and to their back ends, which, with the addition of cool air underneath them and the removal of the top HDD cage, plus the airflow draw of the front Xigmatek XLF exhaust, moves the hot air safely over the top of the (sole) bottom fan cage (which thanks to slight convection/shear gets cool air from the bottom of the case running up through it. keeping the HDDs within a safe temperature range). Meanwhile the Noctua NH-D14's three fans and the Prolimatech rear exhaust plus the top Xigmatek XLF exhaust work together to keep the warmer air (already moved toward the front exhaust) out the top before it can be taken into the CPU cooler's push fan. Unconventional setup, but it works really well. If I had blower GPUs I'd change the front to an intake for reasons that are probably clear. All of this works primarily based on the GPU both physically and with regard to convection splitting the case airflow into three sections - the first being from the GPU to the front, the second being from the GPU to the top, the third being from the side intake and ample side panel holes to the CPU cooler. Minimal turbulence thanks to the specific choices, probably significantly less now that the side fan is pushing more air and splitting into the three areas it needs to be routed to more easily and as intended. The XLF didn't have the oomph when it was still working (for all of a day) to maximize the benefit of the idea; the Spectre Pro is just in a different performance class altogether. Diggin' it. Now to finish optimizing the NH-D14 with the more modern offerings from Noctua and I'll be ready to shoot for 48x, and if that works, 5GHz via BCLK. 
|
|
#
¿
Jul 3, 2013 00:11
|
|
|
Well, project supaclocka  is on hold as it looks like I will be upgrading to a Haswell system after all in the near future. Just swapping out the motherboard and getting a new CPU, I will be buying the OC protection and safely delidding to fix the distance issue and make sure TIM is applied properly internally. So it's back burnered, I'm getting the motherboard from an SA-mart sale soon, will be a bit longer for the processor but it's the exact motherboard I would have purchased anyway.Really looking forward to the architectural improvements, power savings, more SATA, AVX2 (etc.), and PCI-e 3.0 lanes. Shaocaholica posted:Well I was running a CPU intensive app (think folding but not quite) on my OC'd 4770K and now one of the cores gets way hotter under load than the rest. Oh, dear. No. That's very very bad. Is it possible at all that the temperature sensing utility you're using is misreporting? Otherwise, it sounds like you might have done the chip in on accident. That's not normal under any circumstances at all, even extreme ones. The only other thing would be if despite all efforts to the contrary it's somehow not making contact with the heat transfer block, and it's mainly showing up only on that core. Agreed fucked around with this message at 18:17 on Jul 3, 2013 |
|
#
¿
Jul 3, 2013 18:12
|
|
|
Delidding is perfectly safe if you're handy and make sure to space your razor beforehand; direct-to-die is some daredevil poo poo that, well, doesn't work out for a lot of reasons I've got two scenarios for my incoming 4770K based build: one is I luck out like one of our posters did and end up with a chip that'll go to 4.5GHz or higher (performance parity in games, superior performance in applications to my 4.7GHz Sandy Bridge i7). That's regular install, nothing fancy except about $120 worth of heat sink and another $125 in fans for the case If it won't hit that and I'm temperature walled, I'll delid it (safely) and reseat the IHS properly to grab that extra 10-20ºC and go from there. So long as I can achieve performance parity with my 2600K at 4.7GHz in games and benefit from the additional features in applications and chipset improvements, I'm solid on the upgrade. A 2600K is not a necessary upgrade for most uses right now at all, but I am looking to expand my audio work and I need more PCI-e bandwidth in addition to more SATA lanes, so I'm doing an overhaul that will have the benefit of giving my wife a badass computer she can make lesson plans with etc. without having to fuss with a rather pokey laptop.Regardless - no way in hell am I going direct to die in this day and age, with all the crap you have to do to make that even work, let alone make it safe, no offense intended it's just drat that's way more extreme than necessary for standard "high clocks" stable overclocking - that's more into the realm of, y'know, at least dry ice, maybe liquid nitrogen cooling suicide runs. Bad way to kill a good chip if you're not immaculately careful and/or have an effectively bottomless budget (like the Gigabyte in-house OC team, with their busted Titan PCBs laying around makin' people sad). Agreed fucked around with this message at 05:32 on Jul 18, 2013 |
|
#
¿
Jul 18, 2013 05:29
|
|
|
Shaocaholica posted:I'm 80% confident in saying that my die crack was due to the shittiness of the H110 block. I would totally direct die again with a premium waterblock. I'm not sure if that's on par with LN2 but for me, its a lot easier to watercool these days than in the 'good old days' that I personally remember. I say this with love in my heart, so please don't take it the wrong way, but while that may very well be true (it was quite a lot of effort to try to get it to mount correctly) - I am 100% sure that it's cracked because of your decision to go direct-die in the first place
|
|
#
¿
Jul 18, 2013 18:44
|
|
|
go3 posted:Schrodinger's CPU This cat is definitely dead though GOD drat IT DOGEN
|
|
#
¿
Jul 19, 2013 17:09
|
|
|
Shaocaholica posted:All is fixed now. Heres the lovely part. CPU AND mobo were dead. In addition, the dead mobo would eat working CPUs so I ended up burning a replacement 4770K testing in the bad mobo. Good thing Frys had it covered. Oh fun times. You're KILLING them, man, my god, you're killing them 
|
|
#
¿
Jul 19, 2013 17:42
|
|
|
Gonkish posted:
If you're not careful someone from [H] is going to lurk and find this thread and start a vendetta man, watch your back. Heavy is the head that wears the crown. For what it's worth those are dangerously high temps and I wouldn't expect that chip to live a healthy life, but for now I guess you've got some pretty wicked clocks... Just, you know, take care with that level of heat, 22nm 3d transistors are TINY TINY TINY and fragile, and resistance increases as a function of heat and power, so you're basically asking for trouble running anything at those kind of temps. That's dangerously hot for a GPU, and they're made to run at 80ºC+, you might want to look into a better cooling solution. Agreed fucked around with this message at 00:12 on Aug 4, 2013 |
|
#
¿
Aug 4, 2013 00:09
|
|
|
Gonkish posted:Of course, mine still shits out 100C temps in IBT even with the H100i, but one of my standoffs was bent and I sort of fixed it, but I'm in the process of getting new mounting hardware from Corsair to fix that issue. This is eventually going to kill your chip, quit stress testing it and don't regularly use applications that get it to that temperature. If you haven't delidded, you are ultimately limited by the thermal transmission of the TIM under the integrated heatspreader. It becomes the bottleneck at the highest power draw scenarios (and while your voltage is stellar, it's still gonna be sipping plenty of juice at 4.8GHz, as evidenced by the fact that it gets hot as gently caress and can't disperse the heat within acceptable parameters). Be more careful with this, you're endangering the golden goose of a chip here by the stress testing you're doing and not delidding it for a more permanent solution to worst-case thermal envelope issues. 20ºC delta between factory IHS + TIM + too-large TIM glue and delidding it to fix that goof. Consider it, especially if you're going to keep doing things that cause your EXTREMELY loving DELICATE CPU to reach 100ºC. 
|
|
#
¿
Sep 7, 2013 05:56
|
|
|
Well, had to reduce my 2600K down to 4.6GHz. Might take it down another 100MHz and lower voltages accordingly, a 2600K remains "a really fast loving processor" even at stock behavior, it'll do fine with a couple hundred MHz off the top... But for anyone curious, here's how it went. Here's a quick post-mortem on 4.7GHz on my 2600K - ran great with a NH-D14 for cooling and an Asus Sabertooth P67 from June of 2011 through late August of 2013. Manually adjusted settings were 1.38V vcore, set 350 VRM switching frequency, RAM at 1.525V. Spread Spectrum disabled. Voltage adjustments made using positive offsets and Asus LLC adjustments (used High for the majority of the period). Other settings included manually enabling all phases, assigning T-Probe priority, and allowing a 20% overcurrent if necessary for (this allowed less voltage to hit 4.7GHz, but may also be the ultimate reason it croaked - voltage, current, resistance... it all adds up). All other relevant factors, including PLL overvolting, VCCSA, VCCIO, long-Turbo voltage behavior, etc. all managed by the motherboard and BIOS. Hypothesized possible causes for current instability include 1. Microarchitectural failure due to long-term threshold of safety voltage. I'm not sure this is it, because it should just generally be unstable if this is the case rather than unstable at 4.7GHz but stable at lower clocks... Still, see 3 for why this is hard to call. 2. Concurrent BIOS update that may have changed automatic handling of PLL, VCCSA, VCCIO behavior - I haven't hosed around with this as I no longer need that level of overclock anyway. With my back down, not exactly getting a lot of work, so getting it back up past 4.6GHz 3. Never 100% stable to begin with - this can never really be ruled out, our testing tools only tell us what they tell us. Low level instability that doesn't affect anything visible until it eventually does is absolutely something that can happen. It was a good run, and the chip still works as far as I can tell (passes stability testing, no unusual stuff in event viewer, no application errors, plays CPU-intensive games fine) at 4.6GHz, but as mentioned I might drop it down to an even easier to achieve 4.5GHz just because it would reduce the required voltage by a tremendous amount compared to the 100MHz/core. Hell of a chip! I've got overclocker's insurance from Intel so if it ends up being a u-arch failure due to electromigration I can get it replaced.
|
|
#
¿
Sep 17, 2013 01:02
|
|
|
I'm actually kinda looking forward to overclocking going away. Just getting a beefier, purpose-built SKU and mating it to appropriate hardware as far as the chipset and RAM are concerned are way, way easier, and frankly Intel is one hell of a lot better at validating their hardware than we are with our imperfect tools and vastly out of spec operation and suppositions. We really don't have a means of saying, for sure, that a given overclock is actually low-level stable, because the tools we have just don't... test with that kind of integrity. We're basically trying to validate with a bunch of variables inaccessible; while I've been overclocking and doing enthusiast poo poo since I got into building computers, more and more I'm starting to really appreciate the idea of purpose-built processors doing what they do rather than giving a user base the choice to roll the dice and hope for a good processor in a big batch of pricey processors. They're just so fast, and incredibly efficient with power and heat, it seems like so much effort is expended for really nebulous returns. I've been thinking more and more on it and I really feel like I could do more with an -E sku and just let that high clocked, Xeon "lite" do its thing rather than pushing performance to the bleeding edge and trading in chip lifetime for the privilege. Overclocking heresy committed, but when they get to the point where they have achieved even more extraordinary power efficiency, scheduling improvements, and in general are making kickass processors that run like a bat out of hell (even more so than today), I'll be kind of glad to be done with the whole thing. Redoing my overclock has been a pain in the rear end and I don't think the opportunity cost makes sense given my root objection that we can't really determine stability with nearly the same level of accuracy that Intel (or AMD, to be fair to the "competition"...  ) can.
|
|
#
¿
Sep 20, 2013 06:50
|
|
|
When I say "going away," I mean I genuinely think overclocking is going to not be possible in the relatively near future. I think chips are going to be validated for their stock configuration and that's how they'll run. Designs have moved further and further in the direction of efficiency on all fronts, and overclocking is deciding to axe all gains in efficiency to run the part out of spec on purpose because it's supposedly worth it. And yeah, my 2600K hung in at 4.7GHz as stable as I could tell for about two years, but who knows what happened to the microarchitecture during that period? As more and more of the things we rely on to move the chip outside of its nominal operational parameters are centralized, and as the design principle of efficiency over anything else begins to further negate the possible gains to be had from overclocking anyway, and - let's be real, here - as CPUs and GPUs and related stuff just gets faster and faster anyway, what will then be the point of trying to snatch a few hundred MHz? I'm very wary of lithographic shrinks and the 3D lithography required for really small and really high density transistors even being able to sustain overclocks like older CPUs could. Especially with so many OC-specific things moved on-die. I don't think overclocking is gonna be around that much longer and I'm just saying I'm okay with that. Overclocking per se started as market segmentation, but it's ending more out of necessity due to the specific nature of progress in design. Or such is my take, anyway.
|
|
#
¿
Sep 20, 2013 20:49
|
|
|
|
| # ¿ May 16, 2024 14:18 |
|
|
My EVGA 2070 non-super does not artifact in 3DMark Time Spy, but if too heavily overclocked will just not start the demo at all. If it can load it, it doesn't seem to have an issue the whole time. Too-ambitious overclock causes some game engines to crash for me too. But I have not yet seen artifacting like I would have expected from it, just go-until-it-crashes behavior in games 'til I bumped the clocks down a bit to their current values.
Agreed fucked around with this message at 14:51 on Aug 9, 2019 |
|
#
¿
Aug 9, 2019 14:49
|
|