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

Haswell Overclocking Guide [With Statistics] - Page 1067

post #10661 of 19539
Quote:
Originally Posted by Caos View Post

Use only for testing prime 28.3 .. I try with 27.9?

do not touch the settings then?

dont use v28.3 its insane v27.9 is the better one for testing and i would drop the uncore to 33x at 1.15v just to take more things out of the equation. you can bring it back up and lower voltage once you hit a good oc.
post #10662 of 19539
thanks for the replys..

then remains.

uncore x33
ring cpu voltage: override
cpu ring voltage: 1.15
ring cpu voltage offset mode: auto
ring cpu offset voltage: auto

right?
post #10663 of 19539
Hmm. Running into whea_uncorrectable_error now. Sounds like it's time to pull back from 4.7...
post #10664 of 19539
Quote:
Originally Posted by mk16 View Post

dont use v28.3 its insane v27.9 is the better one for testing and i would drop the uncore to 33x at 1.15v just to take more things out of the equation. you can bring it back up and lower voltage once you hit a good oc.

The newer FMA3/AVX2 instructions require more volts than AVX or Integer instructions to be stable. That is the reason Intel's on-chip Power Control Unit bumps the voltage a bit when encountering them (dirty little secret). If you don't test against the new FMA3/AVX2 you may not be stable in applications that use them.

You can test with custom FFT length of 1344k in P95 v28.3 for stability. It doesn't generate high temperatures - in fact, less than P95 v27.9 does.

P95 28.4 is out. Just more efficient loops I think.
post #10665 of 19539
So here is my take on testing. It all clicked with me the other night. I was banging my head against the wall because I could be very stable with a vcore of 1.184 (4.2 GHz - I don't have a great chip), but as soon as I tried P95 28.3, I would crash until I increased the voltage significantly. Then it clicked. This voltage boost that Intel adds to adaptive or offset mode when AVX instructions are detected is there for a purpose - they don't do things like that capriciously. Their newer FMA3/AVX2 instructions must require more voltage to run than other instructions so they boost the core voltage to compensate.

This put a new light for me on how to overclock the Haswelll. Instead of avoiding stress testing with adaptive mode and AVX instructions, one should actually stress test with them and adaptive mode. You just need to make sure that the maximum core voltage attained is reasonable. So I set out to do this with adaptive mode and P95 V28.3. My stability test was P95 V28.3 1344k 1344k - this test does not generate high temps and as usual is a good indicator of stability. I started with offsets that I thought I had a stable (1.184V) vcore = +.01V. I ended up having to bump up the offset to +.015 to get 2 hours 1334k. The end result was a vcore, under FMA3 1344k stress, of 1.216V. Stress testing with non-AVX gave me a vcore of 1.2V (no boost). So the boost for my chip was only 0.016V.

It seems to me that the advantages of this approach are:
1. Full stability against all instructions
2. Lower vcore at non-AVX loads and the needed higher vcore on the AVX2/FMA3 instructions
3. No need to worry about accidentally running AVX on adaptive mode
4. No need to translate fixed volts to adaptive offset volts.

I note that the 27.9 AVX are stable at the lower voltage, I infer that this mean they really don't need a boost.

There are drawbacks to using fixed voltage to determine your overclock that this addresses:
1. You could do fixed and translate to offset and may or may not be stable for the FMA3/AVX2
2. If you do fixed against p95 v28.x, then translate this to an adaptive offset, when encountering AVX, the boost may overshoot your targeted vcore.

Some caveats:
1. I don't know if this "boost" intel gives varies from processor to processor, you would need to be careful
2. It may not be feasible (the boost) for higher overclocks. This just may mean higher overclocks aren't fully stable
post #10666 of 19539
If you have to use 0.02 more vcore anyway for stuff that uses avx2 instructions like x264, what's the point in lowering it by 0.02 for other stuff?
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 #10667 of 19539
Quote:
Originally Posted by Cyro999 View Post

Throw 1.5v at it on same timings and speeds if you feel like it to see if anything changes. I really doubt it though. What's your highest solid stable for everything profile?
those 4,4 if you do not count the crash. Though everything else was really ok (i might used another more friendly settings with a bit higher core (1,295 and Input Voltage 1,9 or 2,0 - not sure now) Uncore was at 33-33 at 1,204.
Edited by Wakizashis - 3/3/14 at 12:39pm
post #10668 of 19539
Quote:
Originally Posted by Cyro999 View Post

If you have to use 0.02 more vcore anyway for stuff that uses avx2 instructions like x264, what's the point in lowering it by 0.02 for other stuff?

Less wear on the CPU. less heat. If you are talking about manual instead, with manual Vring doesn't drop with EIST or C-states. That apparently only happens with offset or adaptive,

In any case, there doesn't seem to be much point in avoiding testing AVX in adaptive mose
post #10669 of 19539
Quote:
Originally Posted by GeneO View Post

Less wear on the CPU. less heat. If you are talking about manual instead, with manual Vring doesn't drop with EIST or C-states. That apparently only happens with offset or adaptive,

In any case, there doesn't seem to be much point in avoiding testing AVX in adaptive mose

I have my CPU Cache Voltage set to Manual (1.15v) and it drops to .003v at idle (measured with HWInfo).

Asus Maximus VI Hero, 1301 UEFI
Min CPU Cache Ratio 8, Max CPU Cache Ratio 42
EIST and all C-States enabled
Balanced Windows 7 Power Plan
My System
(20 items)
 
  
CPUMotherboardGraphicsRAM
Intel i7-4770K @ 4.5GHz ASUS Maximus VI Hero Z87 Gigabyte GTX 980 Ti G1 Gaming 16GB (2 x 8GB) G.Skill Trident X DDR3-2400 
Hard DriveHard DriveHard DriveHard Drive
750GB Samsung 840EVO 256GB Crucial M4 150GB Velociraptor 1TB Samsung F1 
Optical DriveCoolingOSMonitor
Asus DRW-24B1ST Phanteks PH-TC14PE Win 10 64-bit 27" Qnix QX2710 96Hz 
KeyboardPowerCaseMouse
Cooler Master QuickFire Ultimate Cooler Master V1000 1000W 80+ Gold Fractal Design Arc Midi R2 Logitech G500s 
Mouse PadAudioOtherOther
Func Surface 1030 Asus Xonar Essence STX M-Audia AV 40 Polk PSW10 Sub 
  hide details  
Reply
My System
(20 items)
 
  
CPUMotherboardGraphicsRAM
Intel i7-4770K @ 4.5GHz ASUS Maximus VI Hero Z87 Gigabyte GTX 980 Ti G1 Gaming 16GB (2 x 8GB) G.Skill Trident X DDR3-2400 
Hard DriveHard DriveHard DriveHard Drive
750GB Samsung 840EVO 256GB Crucial M4 150GB Velociraptor 1TB Samsung F1 
Optical DriveCoolingOSMonitor
Asus DRW-24B1ST Phanteks PH-TC14PE Win 10 64-bit 27" Qnix QX2710 96Hz 
KeyboardPowerCaseMouse
Cooler Master QuickFire Ultimate Cooler Master V1000 1000W 80+ Gold Fractal Design Arc Midi R2 Logitech G500s 
Mouse PadAudioOtherOther
Func Surface 1030 Asus Xonar Essence STX M-Audia AV 40 Polk PSW10 Sub 
  hide details  
Reply
post #10670 of 19539
Quote:
Originally Posted by GeneO View Post

Less wear on the CPU. less heat. If you are talking about manual instead, with manual Vring doesn't drop with EIST or C-states. That apparently only happens with offset or adaptive,

In any case, there doesn't seem to be much point in avoiding testing AVX in adaptive mose

Adaptive mode is pretty loosely defined though, and controls vary mobo to mobo. It doesn't exist on some, such as giga. Likewise, avx, avx2, fma3 etc can be used without the volts rising sometimes but not all of the time. It's not very well understood by anyone AFAIK
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]