|
24 hour Prime95 turned out pretty well: On my original settings Prime95 ran for about 5 hours before it crashed. The computer itself didn't lock up or error or anything, it just crashed in Windows, which was different from how IBT would lock up the whole thing, but whatever. I bumped up the offset to +0.1v with LLC at 25%, and it went for the whole day without a hitch with minor change in temp/fan speed. Those max temps hitting 73C shouldn't be a problem as long as the average for the whole period is substantially lower, right? I'm going to play with turning down my case fans a bit more, and I want to make sure I have plenty of overhead to do so.
|
#
?
Jun 19, 2013 23:38
|
|
|
|
| # ? May 10, 2024 14:30 |
|
|
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
|
|
|
AnandTech's Ian Cutress got a tour of Gigabyte HQ's Overclocking Lab and did a write-up about that and competitive/exotic overclocking.Ian Cutress posted:Every so often I came across a dead Titan PCB and wept softly. Link
|
|
#
?
Jun 20, 2013 02:11
|
|
|
Haha Windows XP 4eva. Why not just boot tinycore linux  $ cat /proc/cpuinfo [submit]
|
|
#
?
Jun 20, 2013 03:09
|
|
|
They do, if it fits the benchmark right. v v These aren't usable systems. Hell, by the end of a session they're usually non-functional piles of silicon detritus.
|
|
#
?
Jun 20, 2013 03:11
|
|
|
I was joking but actually there's a linux build/source of superPi. Not sure how well optimized it would be compared to the windows version and it would be apples to oranges still but tinycore might boot faster and be more 'stable' than XP but who knows.
|
|
#
?
Jun 20, 2013 03:13
|
|
|
These guys have taken the time and effort to strip XP to the bones; I'm sure they've looked into it.
|
|
#
?
Jun 20, 2013 03:15
|
|
|
Most HWBot submissions require verification with Windows-specific tools, so there's probably some issue with restrictions when you throw linux into it since everything would have to run in compatibility layers (like wine I guess). It's probably more effort than it's worth given the number of submissions they go through.
|
|
#
?
Jun 20, 2013 04:48
|
|
|
I haven't overclocked anything since I had a q6600 so I'm kinda confused. I upped my turbo ratio to get 4.5ghz and I didn't change anything else, I ran intelburntest and cpu-z says my core voltage is at 1.432v, which the OP says is too high. How is it going up that high if I didn't change any voltages in the bios?
|
|
#
?
Jun 20, 2013 13:38
|
|
|
Because it's defaulting to Auto behavior, which means the manufacturer's suggested tweaks are active. You have to set Offset +0.000 (or as close as you can get) to keep the stock behavior. That's why suggested multiplier-only overclocks use a low multiplier; auto settings are crazy.
|
|
#
?
Jun 20, 2013 15:16
|
|
|
Can confirm. Use automatic settings as a last resort only. Bumping up the Vcore offset first, to get it into a close range of what the Vdiv wanted got me booting, then turning on LLC to 25% got me stable at 4.5 Ghz. Before, when just trying to use higher LLC's and no offset, (because I'm an idiot who can't read the OP) limited me to 4.4ghz, and excessive heat. I'm currently testing for stability. 15 hours of small FFT's on prime 95 and no hiccups. I suppose after 24, I'll switch it over to the large FFT's, and then follow that up with a blend test just to be sure. My max Vcore tops out at V1.302, and only 1 core has hit over 72 in 16 hours, and only for a couple minutes. Average core max is 68.5C after 16 hours of testing. Using 100% of my processor now requires the use of an Air Conditioner on warm days. 
|
|
#
?
Jun 20, 2013 16:41
|
|
|
I picked up a new motherboard and CPU today since I'm currently on a phenom II x4 955 and just wasn't happy with the performance. I ended up getting the intel i5 3570k (of course) and a gigabyte B75M-D3H, which probably isn't great for overclocking but whatever. I'm planning to overclock since I've often heard these chips have a lot higher potential (usually) than their stock clocks. On the other hand, I only have the default air cooler and not so fancy thermal paste. I'm not looking to pry off the heatspreader or anything crazy like that, and I'm not looking to hit insane speeds like 4.7 GHz. I'd be happy with 4GHz. Can I get this sort of result easily with what I have? Has anyone used a similar gigabyte board to overclock something?
|
|
#
?
Jun 21, 2013 10:05
|
|
|
You need a Z77 board to overclock that CPU at all.
|
|
#
?
Jun 21, 2013 10:10
|
|
|
Is it a compatibility issue or is it just not supported on this board?
|
|
#
?
Jun 21, 2013 10:20
|
|
|
B75 and H77 aren't designed to support overclocking. There's nothing weird about a 3570k or that specific board, it's just that a Z77 board is needed to overclock it and all other unlocked Ivy Bridge chips.
|
|
#
?
Jun 21, 2013 10:38
|
|
|
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?
|
|
#
?
Jun 21, 2013 10:44
|
|
|
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
|
|
|
Agreed posted: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? It's weird, I'm usually the type to over prepare, but I didn't this time. Gotta gently caress up some time. I probably can't return it, I got it from a store that's notorious for its great prices but lovely customer service (cheaper than most internet places usually). Refusing refunds for faulty models, that sort of thing. They usually cave when you remind them it's against the law however. But hey, upside is that it's probably a fair bit faster than my phenom 955, right? ...Right?  I'll ask them if they can do an exchange or something.
|
|
#
?
Jun 21, 2013 11:45
|
|
|
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
|
|
|
By the reviews of Haswell motherboards we've so far, have additional power phases really made any sort of noticeable difference in average overclocks, or is everything still just gently caress-all random thanks to the stupid heat wall on those chips?
|
|
#
?
Jun 21, 2013 15:20
|
|
|
Sidesaddle Cavalry posted:By the reviews of Haswell motherboards we've so far, have additional power phases really made any sort of noticeable difference in average overclocks, or is everything still just gently caress-all random thanks to the stupid heat wall on those chips? Haswell's FIVR does all the switching work, so your clock limits are based on walls and heat, yeah. That said, if you run a high overclock on four cheap phases, you're running a shitload of amps through components that may not be able to handle the load. Stick with 8 phases or 6 heavy duty ones, but there's no need for more.
|
|
#
?
Jun 21, 2013 20:59
|
|
|
I see then--I was a bit surprised when ASRock put 12 phases on their top Micro ATX board this generation compared to the Maximus VI Gene/Z87 Gryphon's eight(+two), but I'm guessing if I were pumping enough amps for that to make a difference, I'd already be baking the Haswell inside its TIM and glue anyways. Meanwhile MSI puts out three different ATX boards that require liquid nitrogen to make any real use out of their power configurations. 
|
|
#
?
Jun 22, 2013 08:20
|
|
|
More phases = higher numbers = higher price tag. People will buy it because it's flagship. For what it's worth, there's more to this than number of phases. You also have to take the amperage capacity of each phase into account. Besides the absolute capacity, heat matters - more phases or phases with a higher efficiency/capacity will produce less heat for the same work. This helps keep up the VRM's stability and ability to move power without setting itself on fire. The Intel reference 4-phase design for 95W CPUs typically handles 100-120A in 4 phases, plus or minus, or about 25-30A per phase. Using higher-quality components (typically 40A phases) gives you up to 160A before things catch fire. At nominal Vccin, Haswell can draw, eh, ~100A on a fully nut-busted stable overclock (~180W), or more like 80A with the voltage sanity stop engaged and the chip delidded. Gigabyte made a big deal of switching from 40A phases to 60A on its top-end -UP boards. It seems like MSI has picked up these up as its standard Ultra-Durable 4 parts, and as a result its 4-phase design has the current delivery of last gen's 6-phase design (MSI loves doing stuff like this - last gen's -GD hardware had doubled/integrated FETs to double the amperage per phase compared to standard stuff). Now, you might be tempted to think "Oh, 80A on a 100A VRM for a top overclock? Why even a low-end Z87 motherboard can handle that!" There's a popular car analogy with overclocking: You might go 60 in a Dodge Stratus, but a Porsche is a lot better suited for it. The idea is that with sufficient safety margin and overengineering, there's less risk of something going wrong, less waste heat being spewed out (likely without a heatsink) in a part of the board with suboptimal airflow, etc. etc. So get more than the basic. There's a limit to how useful increasing the power delivery over what's necessary can be, of course. With Haswell, it seems like 8 fairly standard phases really is where you stop seeing increases. No need to get something dumb like Gigabyte's Z77X-UP7, with 32x60A phases spec'd for up to 2000W.
|
|
#
?
Jun 22, 2013 08:58
|
|
|
Factory Factory posted:There's a limit to how useful increasing the power delivery over what's necessary can be, of course. With Haswell, it seems like 8 fairly standard phases really is where you stop seeing increases. No need to get something dumb like Gigabyte's Z77X-UP7, with 32x60A phases spec'd for up to 2000W. I saw that board last year and my eyes turned into ferrite chokes.
|
|
#
?
Jun 22, 2013 10:33
|
|
|
The problems never end. Usually caused by my stupidity. So I've been load testing at 4.5 ghz. Thermals are good, it's prime stable, Intel burn test doesn't blow anything up. Vcore tends to hover around 1.33 maximum. Today I was fiddling around with GPU settings, having confirmed that everything with the processor seem stable when I noticed my hardware monitor was suddenly showing a vcore max value of 2.03  I cannot seem to duplicate it while I'm watching it. It must not stay there long. Could this possibly be a hardware glitch, or is something horribly wrong?
|
|
#
?
Jun 24, 2013 16:34
|
|
|
Do you have LLC turned on or anything? I don't know that it would cause a spike that high, or if the spikes it causes in vcore are perceptible, just thinking aloud.
|
|
#
?
Jun 24, 2013 16:39
|
|
|
LLC is turned to 25%. The minimum. When LLC is turned off, Vcore dips too low under load with offset and is unstable. Under mild load, the offset seems to pull the Vcore too high? I'm a bit fuzzy on my process, I'm terrible at documenting things. If you make a list of suspect things, I'll happily report back. I understand that LLC can cause Vcore to jump during midrange loads, but I believe that I had accounted for all that, and as I said, I couldn't get it stable with just vcore offset without hitting my voltage limit of 1.35v Is there any app I can use to test how high the Vcore will jump? Prime95 and burntest are very good at loading the chip way down, but not so much for this 50% load I need to try and diagnose this vcore issue. The spike does not crash my machine. I just look over at the monitor and yelp in surprise.
|
|
#
?
Jun 24, 2013 16:50
|
|
|
Can you bump it down to 4.4 and turn off LLC and call it a day, or is still not stable without it at that speed? I totally understand the impulse to fiddle, I kept shooting for 4.5ghz stable for a loooooong time before I finally pulled it off with some magic voodoo seasoning mix of settings that I really have no idea what in the hell made it work, but if it's really spiking your vcore that high, and you don't know how long it might be staying there, that ain't good.
|
|
#
?
Jun 24, 2013 16:52
|
|
|
I'm pretty sure I tried that, and it was unstable at 4.4 as well. It was stable with no vcore issues for some time at 4.4 with LLC on though. I just want that nice number. 45 is a good number. Edit: I lean towards sensor glitch. I'm positive it does not stay there long. Normally the absolute max Vcore I've ever seen out of it is 1.352 (I'll take that risk since it never settles there). SocketSeven fucked around with this message at 16:59 on Jun 24, 2013 |
|
#
?
Jun 24, 2013 16:54
|
|
|
It's quite possible. I assume you're using hwinfo64. Are you using the most up to date version? It gets updated frequently with better support for newer boards.
|
|
#
?
Jun 24, 2013 16:59
|
|
|
The 32 bit version actually. I'm going to drop down to 4.4 again, and see if I can get it stable with no LLC, then slowly work my way back towards 4.5, documenting what the hell I'm doing this time. Maybe I'll succeed.
|
|
#
?
Jun 24, 2013 17:21
|
|
|
All right, now that I've got my i5-2500K transplanted to a P8Z77-I Deluxe and a Bitfenix Prodigy, it's time to clock it back up again. Unfortunately, it's not going to be as simple as plugging in the same offset and multiplier - apparently the P8Z77-I has a lot of Vdroop. As in, at stock clocks (plus "multicore enhancement" so uniform 37x), it's calling a VID of 1.366V - same as the P8P67 Pro when I was overclocked - but only getting a Vcore of 1.176V. Kinda...  here. But okay. Looks like I gotta kick up the LLC a smidge and live with spikes, otherwise my light-loading Vcore is gonna be insane.E: Just increasing the multiplier from 37 to 38 bumps Vcore up almost 0.060V.
Factory Factory fucked around with this message at 18:04 on Jun 24, 2013 |
|
#
?
Jun 24, 2013 17:49
|
|
|
Well, that was an adventure full of lovely mysteries and disappointment. First off, the behavior was because I forgot to shut off the EPU, whoopsie.Second, I've had to settle at 44x (was 46x) at the same voltage as before (which, conveniently, comes from the same offset). For whatever reason, the P8Z77-I wanted 1.33V to push 46x, and it's too Goddamn hot in here for a Hyper 212+ to support that. Seriously, it's like 95 F/35C ambient. Makes me sad that going from an older 8-phase board to a newer 12-phase board meant worse voltages by that much, even with the same Digi+ VRM design. Also, Agreed finally sold me on LLC as useful (practically indistinguishable from bumping up the offset, except it doesn't apply to idle voltages whereas offset does), so when I revisit this when it's cooler and I can tolerate this bullshit again, I'll see about going for the minimum-positive-offset overclock.
|
|
#
?
Jun 24, 2013 19:42
|
|
|
If my board won't let me do an offset of 0.000, only -0.005 or +0.005, which one should I lean toward in general? I'm guessing negative, if it is stable?
|
|
#
?
Jun 24, 2013 20:25
|
|
|
It barely makes a difference. I'd got with +0.005 just because a negative offset lowers the idle voltage, as well, and that could create frustrating-to-diagnose idle time crashes. Of course, working with just 5 thousandths of a volt, this is pretty unlikely, but still.
|
|
#
?
Jun 24, 2013 20:34
|
|
|
SocketSeven posted:The problems never end. Usually caused by my stupidity. If it happened coming out of standby/sleep mode it's more than likely a sensor being misreported or a glitch of some kind, since 2.0V being fed to the CPU is incredibly unlikely unless something is very, very wrong with your motherboard.
|
|
#
?
Jun 24, 2013 20:46
|
|
|
LCD Deathpanel posted:I've seen something similar happen in HWiNFO, where the fields for my videocard (6970) will occasionally report impossible voltage minimums and maxes coming out of standby (like, voltages that would fry the VRM chips immediately if they were actually happening, and not a sensor read malfunction). Yeah HWiNFO seems glitchy sometimes. The past nearly a year I've lived with a 7950 pulling around 20000-30000W after a bios flash. Sometimes it fixes itself, sometimes it shows those values.
|
|
#
?
Jun 24, 2013 23:02
|
|
|
That's a lotta watts. Personally, I had fun with Asus AI Suite trying to set my BCLK to -66.5 MHz earlier today.
|
|
#
?
Jun 24, 2013 23:06
|
|
|
Factory Factory posted:That's a lotta watts. Did anything fun happen or did it just not boot?
|
|
#
?
Jun 25, 2013 00:08
|
|
|
|
| # ? May 10, 2024 14:30 |
|
|
It's purely a monitoring glitch. The board defaults to 100.5 MHz (and honestly it doesn't look like there's a BIOS option to change that, wtf?), and moving the slider shows 80 to 120 MHz. But as far as AI Suite is concerned... I've got a hell of an underclock. 
|
|
#
?
Jun 25, 2013 00:11
|
|