I completed my first build approximately 5 days ago, installed Windows 7 RC, flashed the motherboard BIOS to F5, and finished updating all the drivers. I then proceeded to attempt overclocking for the very first time, and fell flat on my face as my system failed to even POST Undeterred, after manually clearing the CMOS, I chose to use EasyTune's Quick Boost to see if it would make any difference, and fortunately I managed an easy 3.2GHz, albeit with soaring temperatures whilst running Prime95 (80+°C). Deciding to risk manual editing in the BIOS once again, I ended up with these settings: BCLK: 167 CPU Multiplier: x20 (3.34GHz) Memory Multiplier: x8 (1336MHz) Uncore Multiplier: x16 (2672MHz) QPI Link Speed: x36 (6.012 G/T) CPU VCore: 1.225 QPI/Vtt: 1.24 DRAM: 1.64 Memory Timings (Manual): 8-8-8-24 Everything else on [Auto]. Turbo Boost disabled. With those settings I completed a 10 hour torture test in Prime95 using Blend mode, and the highest recorded temperature according to Real Temp was 67°C on Core #0. I've now decided to lower the VCore again, down to 1.21875, and short preliminary tests have shown this to be stable. I would go lower, but I'm afraid I'll end up with non-POSTing system again. I have to consider the coming Summer months, the AS5 break-in period (200 hours?), and the fact that I won't be running anything near as stressful as Prime95 normally. Are my temperatures OK? My cores are idling at 33-40°C. Should I risk lowering the VCore further, and raising the memory multiplier to 10?
well since the default Vcore of an i7 920 is around 1.20-1.25 i wouldnt mines @ 3.95 and at 1.35V also just built a system with new D0 stepping and its @ 3.8Ghz @ 1.31V 33-40 arent too bad and seem reasonable for a warm flat. mine is around 30-40 depending on room temp your uncore value according to posts ive seen on xtreme forums should be 2x mem speed +1 multiplier. so yours shud be x17. is your mem 1333 if so then no need to raise mem multi to x10 unless you want to overclock your mem as well.
I think Uncore should be 2x or 2x+1, depending on what values are available. My memory is rated at 1600MHz, and I don't have a x9 multiplier available, so I'll be forced to 1670MHz.
Your temps look fine. Try running LinX as a stress test instead of Prime95 - it hits the CPU, RAM and memory controller very hard so picks up any slight instability. It also gives higher temps than Prime95 does. By a good few degrees. Prime95 won't make my Core i7 break 69*C @ 3.8GHz, while LinX pushes it to 73-74*C. I would be tempted to drop Vcore another notch and see if it's still stable - mine was very nearly stable @ 1.2v @ 3.4GHz and the D0 stepping chips seem to be great at lower voltages - it's only really after 3.8GHz that they start needing the Vcore increased a lot. And from an experiment I tried, even a fairly small bump in Vcore means quite large temp jumps. I'm basically running the exact same stuff as you, but with the UD5 board rather than the UD3R and Corsair RAM. Screenies here and here. 1.275v set in BIOS, 1.248v idle (CPUz) 1.184v load (CPUz). Real values (taken with a DMM) are a tiny touch lower. Ignore most of what Hardware monitor says - I was only using it to keep an eye on chipset temps. Is it just me... or do all of the Gigabyte X58 boards take ages to run through POST routines?
yep they do slow start issue. build a couple recently and although great boards start up is slower than it should be as they seem to do nothing as they start up for about 10 sec then start post sequence
Ah, so it's not just mine. Whenever I restarted after tweaking in the BIOS, I thought the OC was gonna fail 'cause it takes so long to get itself going. Then again, looking at the LED POST code display... if it's gonna fail an OC directly, it had a tendency to get stuck at 'b8'. But I'm impressed with the failed OC recovery on the UD5. Didn't have to clear the CMOS once. More than can be said for my old Asus board, where even a slightly wrong setting meant popping the battery for 20 minutes... Still, aside from that (which I can live with, even if I'm not totally keen on it) and the no-sleep-mode-when-overclocked issue... I quite like the UD5.
You have pretty much the same set up as myself. I have the extreme version of that board and have managed to get mine up to 4,494mhz (214x21). When I tried overclocking mine at first I encountered several BSOD's. What I found was that in my bios CIA2 was enabled. You need to disable this. I have managed to get mine up to this overclock on 1.38v with HT enabled. I run it lower than this for 24/7 (185x21 = 3,885mhz). I can run this with only a small increase to the default voltage too. My system is watercooled, but this overclock would be easy on air. If you wish, I'd be happy to make a note of my bios settings and post them for you. Might help you get a little extra out of yours.
hey chris why the use of lower QPI for 3.8Ghz. is it a RAM speed issue you could have 3.8Ghz with 200QPI and x19 this should provide better performance, although how much in the real world im not sure.
I realise that, to be honest, I run it at this for browsing the net etc. I saved this as one of my initial overclocks and never got around to changing it. When I'm gaming or encoding etc I boot up with another profile (202x21=4242mhz). It's a valid point though. When I have time I'll look at changing it. The 4494mhz is for beching only. I have PC12800 ram. Wish I'd bought faster. I am having to run my ram much slower than its top speed at the high overclocks. From memory, I'm having to use the x6 multiplier. It either has to run quite some way below its max speed or it goes quite some way past its top speed. It's still new to me (only had it for a month). I haven't really had chance to mess around much for the last few weeks.
Before I can go further, I did encounter a strange problem. I was running Memtest86+ and completed an error free pass, after which I foolishly decided to fiddle around with the configuration and chose "BIOS - All". I ended up facing a bunch of red errors with a count of over 5000 and I was seemingly frozen in Memtest with no keys responding, so I performed a hard reset. At this point the system attempted to boot, but failed and immediately rebooted itself, whereupon I was greeted with a familiar message informing me of boot failure due to overclocking/voltage changes. Now, this boot failure occurred when my voltage was set at 1.21875 in the BIOS, with C1E and EIST enabled. As I mentioned previously, prior to this I had completed a 10 hour Blend test with the voltage at 1.225 (C1E and EIST enabled, Turbo disabled). I'm really not too sure what went wrong, but I'll try and describe my boot sequence. Each time my system initially boots, it's extremely loud. It takes a few seconds, but the noise eventually settles and the system POSTs. When I initially used EasyTune's Quick Boost to 3.2GHz, I experienced the usual loudness upon rebooting; however, subsequent reboots after manual BIOS tweaking didn't generate any of the usual noise. I continued lowering the voltage and (quietly) rebooting until I performed the Blend test using Prime95, and after that was successful I lowered the voltage further to 1.21875 (the reboot was quiet). That was when I decided to use Memtest, and it wasn't until the hard reset that I ended up experiencing the loud reboot sequence, and the subsequent boot failure. I reapplied the same BIOS settings, including the 1.21875 VCore, but chose a x17 Uncore, disabled C1E and EIST (Turbo was already disabled previously), and enabled Load Line Calibration. I can't remember whether the reboot after was loud or quiet, but I managed to login to Windows nevertheless, and decided to shut down the computer. I wanted to ensure a full power cycle, including the initial loudness, so after several minutes I powered the system back on, and it booted without a problem. I've just completed an 8 hour stress test, and I'm about to do some further tweaking and testing. I think I'll leave my RAM at 1336 and concentrate on lowering the VCore a bit more if possible. C.I.A.2 is disabled. I'm really not sure why my idle temps are so high either. When I first powered up the system after building it (before the motherboard BIOS flash), I could have sworn that EasyTune's HW Monitor was showing 19-22°C at idle for the CPU (I hadn't yet installed Core/Real Temp). After flashing the BIOS and installing/updating drivers and software, although still at stock speeds, it was showing 33-34°C.
maybe because the chips warmed up, if you have had it on a while your idle temps will be higher than when you first turned it on. also use coretemp as not sure how accurate ET HW monitor is. 33-34C on idle is perfectly normal. there is no way your temps are gunna be the same as room temp or lower. @chris if your RAM is underclocked at your settings using a higher QPI should help raise your MEM frequency while keeping your CPU at 3.8ghz by lowering the multi. altohugh even on a x6 and 200 QPI your mem is still only 1200mhz.
I've reverted to stock settings because I've just experienced another boot failure for some reason, again after a hard reset, this time due to crashing from LinX stress testing. It doesn't matter if I can go through a full power cycle, boot to Windows, and stress test for over 8 hours in Prime95; I still end up having a random boot failure sometime afterwards, during a hard reset. I'm now deterred from overclocking, as I hate the fear of my system failing to boot during a restart/reboot. I do want want the system to run smoothly at at least 3.2GHz which in theory really shouldn't be a problem, but I can't understand these boot failures. Flashing my bios from F2 to F5 is the reason for the increase in temperatures according to ET HW Monitor, although I don't trust its temperature readings anyway.
if i remember there is a bios bug about soft restarts (i.e. windows restart) it nearly always fails. try avoiding soft restarts and use shutdown and power on instead. you cud also try a beta bios this may help fix some issues. I run F6a on my UD5 and it works a treat. the ud3r is upto F6b now link: http://forums.tweaktown.com/f69/gigabyte-latest-bios-28441/
I've never actually had a problem with soft restarts, at least for now. Twice now I've had boot failures after performing hard resets. I'm wondering whether the problem lies with the overclock or some other factor. I know stress testing for 10 hours isn't indication of perfect stability, but surely it wouldn't be unstable to the point of boot failure.
I've never had a problem with soft restarts on my UD5 with BIOS F5, but all BIOSes are not created equal, so... *shrug* and the BIOS doesn't read the CPU temps right. For example, my board reckons my chip idles at 16*C (about four degrees below ambient right now) when at stock and idles at 22*C when at 3.8GHz. RealTemp tells me otherwise - 35*C or so at stock, and about 41*C at 3.8GHz. I'd trust CoreTemp or RealTemp. My temps are a little higher than they could be because of the case and fans I'm running. Low noise fans means less air movement. Boot failures on hard resets are often because the BIOS thinks that the system reset due to a failed overclock (I've had two boards in the past that gave me "Overclock failed!" messages upon reboot after a hard reset or the mains power being out for long enough to discharge the board...) If you will humour me, let's go back to basics for a moment. Load BIOS Optimised defaults. Run Memtest for two or three runs. Leave it at its default settings. If it passes (which it should do) go into Windows and run LinX for, say, 30 runs using 4096MB of RAM. It'll take a while - it takes an hour twenty @ 3.8GHz... have RealTemp running so you know what temps are getting to. LinX hits a system hard... don't be too surprised if screen updates get a bit sluggish at times. Next, a question... does that OCZ RAM support XMP memory profiles, or does it make you do all the groundwork yourself? If it supports XMP, just set it to the XMP profile in the BIOS (should be to 1600MHz, 8-8-8-24-2T, DRAM 1.65v, QPI 1.35v or something like that...) then boot and hammer with LinX again. Then set the BCLK to 160 or so. Check that the RAM is running at or below it's rated speed. Don't run it over spec for now. The fewer points of potential OC failure, the better. Lower the RAM multiplier if necessary. Check that the QPI (uncore) multi is still 2x+1 of the RAM multi. Feed the CPU a little more juice. Setting it to 1.25v will probably be overkill, but rather a little too much than too little right now. Go back into Windows and hammer with LinX again. I've never had LinX hard lock my system... although I've had it BSOD a fair number of times. Rule of thumb is: BSOD "Clock interval not received on secondary processor"/hard lock within a couple of minutes = not enough Vcore. BSOD "general hardware failure"/hard lock in 10-20 minutes = not enough QPI/uncore volts. You shouldn't have to reset to stop LinX - there's a stop button right on the program. Unless it's locked up, of course. Now, I don't run my system 24/7, so I don't worry too much about Priming or LinX for 10+ hours. If I was video encoding and it had a problem, I'd know to back off the OC. Avoid setting loadline calibration. Intel builds in Vdroop for a reason. Loadline calibration negates this, but can mean that during transition from a fully loaded to idle state that the CPU Vcore spikes. Now, at 1.25v this spike isn't going to push over 1.35v, which is still within Intels (generous) safety specs, but seriously... why risk frying your new CPU? Of course... I might have grabbed the wrong end of the stick here. If all this is unnecessary, ignore my waffle.
The system was stable whilst running Prime95 (10+ hours in Blend), but about 30 minutes into LinX my system froze (I couldn't even move the mouse). I did choose the "All" option for problem size and memory usage though. The first hard reset was due to me fiddling with Memtest86+. That allays my fears somewhat. The boot failures just didn't make much sense to me considering the stress testing I had performed. What if one encounters a software related error that ends up freezing the system? Is it better to power off and on again rather than perform a hard reset? Unfortunately it doesn't support XMP, and at the moment on [Auto] settings it's running at completely different frequencies and voltages (7-7-7-19 and 1.5v). That's precisely the reason why I left the multiplier at x8.0 with the memory running at 1336MHz, below its rated speed. I left Memtest running overnight with the aforementioned settings and all passes were error free. Hmm, so hard locks are usually due to insufficient voltages? Well my system froze after 29 minutes and 30 seconds of running LinX, so I guess that was probably due to the low 1.24 QPI/Vtt voltage, but after experiencing the boot failure after the hard reset I decided to set the VCore to 1.225 (from 1.21875) and the QPI/Vtt voltage to 1.3. I thought was a good enough initial increase to get me going, but then I ended up with a system that was stuck in a boot-POST loop, and I had to clear the CMOS settings again. Strangely that also ended up reverting my BIOS back to F2, and after POST I had I think a checksum error and it went onto Auto recovery.
I don't use Prime blend - it's not really enough stress, IMO. However, it is a fairly good indication of what temps you'll get with the system loaded using 'normal' programs at 100%. Also, I don't set LinX to 'all' RAM, either - sometimes Windows'll suddenly decide it needs more RAM, and if LinX is chewing it all up, there's no room to breath and Windows goes belly up. On a 6GB system, 4096MB is about right. Leaves me with about 1-1.2GB of RAM unused during the stress test. I had a few moments in LinX testing that I thought it was going to lock up, because the computer was so sluggish. With older computers with 'hard off' switches, I'd say that powering off did less harm than a reset. Now, though, it's a debatable point. Some cases don't come with reset switches now (example: Silverstone TJ07) but it's really personal preference. If my computers crashed/locked, I usually hard power off for 10 seconds or so. Most overclocking BIOSes now just assume that if you've had a reset or power off not initiated by the OS, that the OC failed somehow. It's safer for the board that way. Presuming that's one of the JEDEC standards in the RAM chip, that's fine. Excellent. So it's not a RAM issue, then. Still setting it to its max rated and trying that again wouldn't go amiss. It's another box ticked in the 'this isn't the problem' column. Yeah, sounds like the Gigabyte dual BIOS saved your skin there. The POST loop is weird... increasing voltages shouldn't be doing that. I'm running 1.65v (DRAM) 1.35v (QPI/uncore)... the only change on my CPU from 1.275v to 1.35v is that temps go up about 12*C. Is that LinX lock on the overclock, or just at stock? Stock QPI is 1.2v, IIRC. That 1600MHz RAM really shouldn't be needing the QPI much above stock until it's running at it's max rated, at which point it'll probably need it running at Intel's recommended max of 1.35v... (which probably isn't available unless running via an XMP profile - my UD5 offered 1.34v or 1.36v, not 1.35v. 1.34v would probably do it, but if not, trying 1.36v isn't likely to harm the chip, given the settings I've seen some people run at to get high OCs. ... Further thought... leave everything on Auto... and put the BCLK to 150. See what the voltages do. They'll likely be a little higher than necessary if you set everything manually, but it'll probably boot. If it boots and passes LinX for an hour or so, try setting it to 160. But going too far with voltages on Auto isn't really ideal, as most boards have a tendency to be a little overzealous with automatic voltage adjustments. My UD5 fed my CPU 1.34v at 3.2GHz using 'auto' Vcore, which it certainly didn't need given that it's doing 3.8 @1.25v. But providing it works, it'll give you a point to start tweaking voltages down from. There is a very important lesson from i7; you can have exactly the same kit as someone else, and their OC settings almost certainly won't work for you.
Would that be using the case power switch or the one on the power supply? I'm sure as hell avoiding the reset button from now on. The LinX hard lock was on overclock (167 BLCK), at 1.2125 Vcore, 1.24 QPI/Vtt, 1.64 DRAM. This was with the "All" setting however, so I'm guessing using too much RAM probably lead to the hard lock as opposed to insufficient voltages? That's exactly how I started this whole endeavour. I used EasyTune's Quick Boost, which as far as I can tell merely changes the BLCK. I opted for 160 for 3.2GHz and my [Auto] Vcore was 1.34v resulting in temperatures hitting 81°C. That's when I started lowering the voltage step by step, Blend tested for 10 hours at 1.225v, ran Memtest, foolishly fiddled with it (BIOS - All) to the point where I performed a hard reset and ended up with the "boot failure".
