Overclock.net › Forums › Intel › Intel CPUs › Haswell Overclocking Guide [With Statistics]
New Posts  All Forums:Forum Nav:

Haswell Overclocking Guide [With Statistics] - Page 489

post #4881 of 19539
Quote:
Originally Posted by Darkwizzie View Post

'The point you made is that vRING isn't necessary to stability if I'm not mistaken. It's crucial to overclocking, and by extension the guide as well.'
vRing of course is nessesary for stability. If you're overclocking ring bus obviously we need to up the voltage, right? But that wouldn't give me any reason to add voltage to a ring bus that's at stock. I believe you said ring bus at stock may require more than stock voltages to be stable. In some cases, stock uncore stock vring is fine up until you overclock it.

'I'm all up for investigating this. And I've mentioned multiple times that it improves stability. Even someone who tested earlier mentioned that his prime ran for longer with a vRING boost.'
I'm not doubting the fact that you claim it improves stability.
You don't get to point to that guy for statistical reliablility man. First you go on about how my giant charts after charts of evidence of the uselessness of uncore impact on performance is potentially flawed because I didn't do enough tests (which isn't true), then you pick the guy who did one quick run of Prime? At least wait for the guy to do the tests he needs. So as it stands I have 1 quick test of Prime from one guy. That's not enough for me to take full on as fact. And I admit we might need to recommend something like, set Vring to 1.2v, ring bus to stock while overclocking core. I'm not ruling out that possibility. The only thing I actually full on disagree with you is your notion of uncore providing tangible, observable real-world increase in performance. oh, and the bottleneck.

'A few tests where you get lower results with a higher uncore isn't consistent with how the uncore should behave IMO.'
That's probably correct. But then there's margin of error. On chess for example, having a much lower uncore did register a visible change in the nodes calculated.

'It is possible that you had outliers while testing which affected averages.'
It's impossible for  a few outliers to change the entire verdict because a whole battery of tests were done. For example, if I got a ridiculous outlier for say, Cinebench, that will not affect my average result for Skyrim or Chess. If I had an outlier for each test, then that's less of a freak incident and more of just a low result.

'And judging by the fact that you just dismiss Linpack and such, like what I used in the post earlier just means that you think only your tests are right.'
Because my tests at least somewhat resembles a real world workload. I may be wrong on that. But you don't have to agree. If somebody manages to find multiple games where uncore observably impacts FPS, and this can be replicated by multiple people, then we've got something interesting going on. Games work too. People play games.

'I'm quite sure if BF4 ( as they state on the internet ) is affected by memory speeds, then it'll be affected by uncore as well. When there are simple tests and results that I show you ( not my own, because I'm a little short on time ) you want to dismiss them as overly sensitive. Then where is the progress?'
Ok, first do you agree with me that just using the benchmarks I chose at the time, you would get a result similar to mine? Or do you still think my results are hooey? Beyond that, something like encoding tests. x264 is also a benchmark. We can use that, and that is basically a real-world-usage IIRC. Also as far as I know, Linpack/SuperPi are really intensive stressers that are just about as synthetic and artificial as you can get.

About the BF4 memory speeds: The entire point is to conduct tests, not assume things. Granted what I quoted just now does not have you saying for sure you're right on the BF4 claim but until evidence is brought forward it's not worth anything, it's only a hunch. I'd test it myself but I don't have BF4.

Bear in mind, claims and speculations made without any actual testing is what spawned the 1:1 myth. So you see why I'm so skeptical.

'And also, if there's better performance to be had from uncore, what is the added benefit to overclocking the CPU alone if it is restricted by the uncore ( like in Linpack and such ). <-- This is merely a question I'm asking, not a point I'm trying to make.'
From everything I know and have tested, my evidence and conclusions have led me to this: The core clock is the #1 most important factor for CPU performance. Any uncore performance you can gleen will not be as large as an extra multiplier up for core. I also know that setting too high of an uncore often leads to a stifled core overclock. Therefore it makes sense to overclock core, then uncore. I picked up not major difference in performance improvements from uncore regardless of how much slower your uncore is to your core. 

There is absolutely no 'bottleneck'. Bottleneck is a strong word. Bottleneck doesn't mean having a higher uncore leads to higher performance. Bottleneck would mean something closer to, until you raise your uncore, you can set you core to 50000ghz and you won't get an ounce of improvement. Or barely any improvement. This can easily be proven; simply run stock uncore. Bench 4.0ghz @ Cinebench. Bench 4.3. Bench 4.5. Bench 4.7. Bench 5.0. You will see a noticeable performance increase even though uncore has not been touched. Bottleneck disproved.
Quote:
Originally Posted by Shanenanigans View Post

Either way, whether or not there's a bottleneck, increasing vRING does add stability whether you like it or not. And for your charts, ideally, you should run the tests at least 4 times ( like on review sites ) but 10 times to be sure, and to kick out any outliers.

For reference

http://www.xtremesystems.org/forums/showthread.php?266589-The-OverClockers-BSOD-code-list

A 124 BSOD is usually related to vcore or QPI/VTT ( which for haswell, would be IMC voltage and L3 cache voltage, which is vRING )

So basically, you've ignored my post above completely. Not to mention my other post where I say there is a difference when you change the uncore. Like I said, whether the difference is practical or not, a difference is a difference. In whatever benchmark. See, when I show proof, much like you did in your charts, you just simply discount them. And then you say I'm calling out your testing methods in YOUR thread. I don't mind your benchmarks at all. I do believe uncore is app-dependent. However, the whole basis of this was that you may need some additional vRING voltage to get your vCore stable if you get something similar to a 124 BSOD.

Also, just as an example, having a super high CPU speed with stock uncore to me, is like having a C63 AMG Black. There's so much power, but if it can't be transferred to the wheels which can grip the road ( or better still, use AWD ) it's really just wasted spinning the wheels and not really getting anywhere.

Also, I don't particularly care about the 1:1 myth. Mostly because I've always hit much higher speeds on the CPUs and I simply couldn't clock the NB to match.

Right now I'm trying to figure out why my temperatures are so high while stressing. And I can't seem to figure out why whichever LLC I set to Extreme/Turbo increases the voltages beyond what I set it to.
post #4882 of 19539
I'll disagree with one point you're making there dude, i don't find vring to have any effect on core stability, regardless of core multi. Unless it just doesn't have an effect on my chip? I've only O/C'd the one chip so far so might be different with others? I do find 124 uncore related though and for me it's either vrin/vring/SA/IOD/IOA. ( i know others may disagree (i'll rephrase - find different for their chip) and for them 124 is vcore)

With LLC are you referring to vrin? Take a multimeter to the board and test each LLC setting. You'l find that somewhere around 80% (or what ever you're equivalent is) you'll probably be just over 1:1. Anything more will increase voltage. There's a review on the net somewhere that tested this, i've not got round to doing it myself but it's on the list of things to do.

EDIT - the point which Wiz made about uncore (which i think you are both actually agreeing on) is that it can make a minimal/insignificant difference in day to day running/gaming etc and there's no point pulling hair trying to get it higher. It does make a difference when benching but again this is minimal in the grand scheme of things but can be quite significant if chasing scores.

With my core at x50-x52 changing uncore from x43 to x49 can make a difference in some benches but i would never notice the difference in general running. Point being x43 uncore is fine for me and i'd never notice a difference increasing that when pissing about on the rig. The good thing with Haswell if the NB is already plenty fast.
Edited by Doug2507 - 11/2/13 at 3:51am
post #4883 of 19539
Quote:
Originally Posted by Doug2507 View Post

This is what i end up with whilst testing stability.....


Wait so sorry what are you stable with? interesting shots
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
post #4884 of 19539
Bottom one was 50x43 which was x264 stable, XTU bench, 9hr5min XTU stress stable. (haven't sorted out the folder apart from deleting a load of test shots already).

Top one is work in progress upping uncore to x44. x264 stable, xtu bench, 4hrs+ XTU stress stable. It may be perfectly fine for gaming but i've just got a 9hr rule. The stock shot was changing no voltage and upping uncore to x44. I always run this 1st before changing anything.

Bare in mind i could pass x264 with up to x47 uncore (can't remember if i checked x48). Just shows what i've been doing using x264 to do the ground work and use XTU for the next stage.

I've also got all of my x264 runs saved in another folder.
Edited by Doug2507 - 11/2/13 at 3:59am
post #4885 of 19539
Ok ty, What vrin, vrin llc and ring volts is that? You're listing offsets but i have no idea what kind of volts they'd give you
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
post #4886 of 19539




These are x42/x43 stable.

Offsets or increase in voltage for x44 is all on top of x43 stable settings (should get most of them from the SS's, SA was +.01v iirc for x43). Vdroop 100%.

Note, the only change i made to get x43 stable was to drop SA/IOD from +.015v to +.01v. The +.015v for x42 was just a random guess/starting point as i couldn't get it stable with just vring/vrin.
Edited by Doug2507 - 11/2/13 at 4:20am
post #4887 of 19539
Thanks a ton
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
post #4888 of 19539
Quote:
Originally Posted by Doug2507 View Post

I'll disagree with one point you're making there dude, i don't find vring to have any effect on core stability, regardless of core multi. Unless it just doesn't have an effect on my chip? I've only O/C'd the one chip so far so might be different with others? I do find 124 uncore related though and for me it's either vrin/vring/SA/IOD/IOA. ( i know others may disagree and for them 124 is vcore)

With LLC are you referring to vrin? Take a multimeter to the board and test each LLC setting. You'l find that somewhere around 80% (or what ever you're equivalent is) you'll probably be just over 1:1. Anything more will increase voltage. There's a review on the net somewhere that tested this, i've not got round to doing it myself but it's on the list of things to do.

EDIT - the point which Wiz made about uncore (which i think you are both actually agreeing on) is that it can make a minimal/insignificant difference in day to day running/gaming etc and there's no point pulling hair trying to get it higher. It does make a difference when benching but again this is minimal in the grand scheme of things but can be quite significant if chasing scores.

With my core at x50-x52 changing uncore from x43 to x49 can make a difference in some benches but i would never notice the difference in general running. Point being x43 uncore is fine for me and i'd never notice a difference increasing that when pissing about on the rig. The good thing with Haswell if the NB is already plenty fast.

Ah, I don't have a multimeter at home. I'm using the software readings. My VID is set to 1.132v but still goes up to 1.152v while stressing. Which, on the stock Intel cooler, really causes problems. I'm just ordering a hyper 212x since I can't find the Noctua U12S or the D14 anywhere online near me.




I'm just rearranging my table. I'll respond more after that.
post #4889 of 19539
I think that's perfectly normal. VID is always lower. HWMon is quite good at showing this with VID and IA (actual). I'll have for example 1.35v set in BIOS (VID), CPU-Z reads this (slightly under), with HWM showing 1.35vVID and 1.38vIA. Actual off the board being 1.376v.
post #4890 of 19539
Quote:
Originally Posted by Doug2507 View Post





These are x42/x43 stable.

Offsets or increase in voltage for x44 is all on top of x43 stable settings (should get most of them from the SS's, SA was +.01v iirc for x43). Vdroop 100%.

Note, the only change i made to get x43 stable was to drop SA/IOD from +.015v to +.01v. The +.015v for x42 was just a random guess/starting point as i couldn't get it stable with just vring/vrin.

I'm really not sure what to do to stabilize 4.8

Crazy unstable at load, am lucky to finish a cinebench run, 0x0101 bluescreens

just did tests @1.4vid, 2.1vrin. I don't need that much vcore i think - i've passed cinebench more consistently than this with 0.12vcore less only 100mhz down on the core

tried 1.9-2.1vrin

tried quite a few combinations of sa/iod - stuff like +0.2+0.2, +0.1+0.2, +0.05+0.2, +0.1+0.1 etc, also +0.2+0.2+0.2 on sa/iod/ioa but didn't get anything

ram @800mhz, uncore 4000 with 0.05 ring more than i use for 4000@4.7ghz though i tried more and less

etcetc

I poked at a lot of stuff, it's completely beyond me how to get it remotely load stable. Getting 124's to go away is pretty easy (and i think i can guess vcores i need for it etc based on mainly that) but even with a bunch more vcore etc i can't do anything to the 101's


It just now occurs to me that changing RAM frequency changes IO voltages and maybe SA. I'm not sure entirely how or why, i actually have no way to measure them as far as i know unless i were to own a multimeter, as they can only be adjusted in offsets. Maybe that's the cause of confusion
Edited by Cyro999 - 11/2/13 at 4:46am
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Haswell Overclocking Guide [With Statistics]