Originally Posted by Cyro999
Hey Sin, 1.33vcore fine 24/7 (gaming, idle, some video encoding) loads for 2-4 years?
IBT without avx support/enabled gives me an absolute max of ~87c but it's quite a bit cooler for my normal usage. Maybe 80c with x264, 40's-60's gaming, somewhere between those for live encoding at decent settings. I've got a silver arrow, ~15c ambients so with a bit of work on airflow in the case i think that gives manageable thermals
Nice guide, thanks. The VRIN LLC to extreme really helped, i was having no luck at all scaling clocks, cinebenched on 0.05v less than i managed before (~1.28v) though x264 seems really really temperamental with haswell, i need a ton more vcore for it to stay stable than other stuff, like i pass endless runs of IBT with 7000MB RAM below 1.3v, prime for hours at 1.31 but at 1.32, x264 will consistently fail in like a minute or two. Any idea why?
Wow 1.33v, there is no way to tell if it will last 2-4 years, i would guess it would. You can always increase the VCore if the CPu degrades, and if you haven't delidded your CPU then you will be okay b/c Intel has an OC warranty plan, but I am unsure if they will extend that for haswell. Keep under 90C, most people like AIDA64 b/c it gives lower max temps, it s a bit more realistic.
Originally Posted by Go Gators!
I am using a DMM at the motherboard test points and along the separate 12v rails. This PSU (sig) is supposedly of decent quality from when I was looking up actual reviews (like disassembling to check the caps them and measuring ripple) but I've been told that the Rosewill Capstone PSUs are very reliable and have minimal ripple... the 650W variant is $80 @ newegg right now so that's where the $80 figure came from. Like I said though, I measured my voltages with a DMM and they are always within spec... maybe a little on the high side, like 12.6 or so. Is a Capstone or Fortress a bad idea? I've heard the opposite.
Did memtest for an hour the first day out of desperation. Made it through a few passes in an hour or so with 0 errors. I can re-test though, I just wanted to look for anything glaring.
This is the second fresh install of Win8 to my new SSD since the build. I'll clear CMOS just in case and maybe flash a BIOS as a last option before I start RMAing parts or sorting out a new PSU.
OKay try just reflashing your BIOS, that would be the best course of action. On P67 and above a clear CMOS wouldn't always totally clear all the CMOS data, however flashing the BIOS an doing a power cycle will clear everything and re-flash your ME firmware which can be updated if you update your BIOS. I would then test it again at stock and see if you get any BDOSes. It might not be your PSU then, I just personally don't like OCZ(their older PSUs like the 520W power stream were really nice tho0, don't bother with a new PSU if you are testing with a DMM and everything looks fine.
Originally Posted by Forceman
I'll have to check the VID when I get home. I think my next move is going to be flashing the F6 beta BIOS and start from scratch. I tried clearing the CMOS, but I haven't reflashed the BIOS. I'll try your other tips also, but I think I already tried seting the uncore to 35 and it didn't help.
The annoying thing is not knowing what the crashes indicate. Using Win 8 for testing and I get WHEA Uncorrectable Error (which I think straight Vcore will fix) but also Clock Watchdog Error (something like that) which I have no idea what causes. It's the Clock Watchdog's that don't seem to make any sense. Settings will work fine, then I'll change something completely non-relevant in the BIOS (like disabling hot plug for a SATA drive) and then it'll start failing in minutes. Which is why I was leaning toward BIOS problems - although I never thought of the corruption/reflash angle.
Yea list the VID of both your CPUs, the procedure is in the guide under"what to expect" Basically load optimized defaults and save and exit and then go back into the BIOS and look in the CPU voltage menu near the VCore setting. The grayed out voltage is the VID.
Originally Posted by ABAD1DEA
Sin i now know which caused the intel burn test crash.
can you give me good settings for them? if i leave it on auto i just cant run intel burn test for longer than 10 seconds.
Really the CPu IO Analog and CPU IO Digital are designed to change on the fly with Auto, infact it is best to set them to auto, but if you want to set them manually then you can try +0.2 on both of them maybe +0.3 if you are going above 2400 on air and aren't scared to try that high on air.
System Agent Voltage is the main voltage to change for the IMC, on some boards other things are named differently, however on GBT boards the SA is directly tied to the IMC. CPu IO Digital and Analog used to be called, get this, the CPU PLL Voltage on Z77 for instance. Intel has divided them into separate domains and a while ago they used to help with BCLK on the early BIOSes, however now that BCLk has been refined they only help a bit with memory OC. System Agent however has a pretty big impact on memory OC, so try +0.2v or +0.3v, there are upper limits in the table in the guide....
Originally Posted by rancor
Coming from X58 overclocking 124 errors are related to memory and memory controller stability. From my experience with my 4770K 124 errors are either uncore(cache) or memory controller/ memory. Try lowering your uncore frequency, lowering your memory frequency, increasing your system agent voltage, or increasing your ring voltage.
Sin I love the guide, one thing I think would be good to add and I miss from the x58 guide is a BSOD guide. Just a small section saying what is unstable /needs more voltage for a given BSOD. As more people start to overclock Haswell it should be easier to build if we get some input on what made an OC stable.
yea I might do that.
Originally Posted by BlockABoots
These are my system specs:
Crucial Ballistix Tactical 16GB 1600MHz
MSI HD7950 3GB XFIRE
|64GB SAMSUNG 830 SSD
256GB CRUCIAL M4 SSD
WD 3TB GREEN
CORSAIR 650D CASE
YAMAKASI CATLEAP Q270 LED
I have the latest BIOS for the mobo, F6i. I tried lowering the uncore to 42x, i also upped the vcore to 1.270 and CPU Ring to 1.2v, and the PC lasted about minute before BSOD whilst video encoding with RipBot. I cant understand how it can pass 30 mins of OCCT on the most stressful settings and then 3hrs 40mins of ADIA64, but fall over in under a minute when video encoding!?. Any other suggestions?
Most likely because of AVX, and the fact that OCCT is very outdated. You should try AIDA64 it is the best and what Intel recommends. it can be frustrating to be totally stable on one benchmark and not on another, however AIDA64 tests the uncore an everything, IMO lower the uncore even more until you stabilize the CPU OC, then raise it slowly along with vRing.
Originally Posted by Forceman
I thought VCCIOD was the memory controller.
No, on Z77 it was, however here it is a totally different voltage, Intel renames things and thinks people should forget what they used to be. They are mess up the OC on their CPUs and there is no consistency, even if both your CPUs are from the same batch no one can tell how they will clock, however there is a small correlation between batch # and speed, but I have seen even in the same batch totally different clocks.
Originally Posted by FtW 420
I though memory controller was the ring voltage.
I guess I need to give this guide a good read, thanks for posting it sin!
It is the system agent now, the vring is the ring bus voltage(on asus they call it the cache voltage). Here is the system agent voltage is like it is on X79, it is the one that correlates most closely to the IMC.
TO Almost everyone: STRAIGHT UP:
Someone put in the manual and in the GBT BIOS that the Uncore should be equal or higher than the CPU Ratio, that is NOT, i repeat that is NOT going to help you in any way. Lower it in fact, lower it to 35x or even 30x to be safe, then OC the CPU and then start raising it once the CPU is stable. Most of you can probably boot into windows at 5ghz and in windows OC the multiplier with GTL to 52x or even higher, I can, but the issue is I can't increase my uncore over 4.7ghz on air at all. That is a 500mhz less OC I can do on the Uncore than I can on the CPU ratio. The Uncore helps performance because it controls the speed of pat of the cache, however what I stated in my guide and what others don't really realizes is that the cache has double the bandwidth of sandy bridge and ivy bridge caches, which means that there is no reason to match 1:1, I recommend 300-500mhz lower on the uncore to unbottleneck it on air/water, on Ln2 it is better to keep it closer or on benching use the 300mhz if you can cool it. This uncore is not the same as it was on X58, it has a HUGE impact on system stability so watch out. Also only the recent GBT BIOSes can actually change the uncore(raise it over 39x) so if you have been setting high uncore before it wasn't actually setting that uncore. Thank you for reading this, now I don't have to repeat myself 20 times.