Haswell overclocking discussion.

[24601]

94°F ambient?

No AC = I sad.

 

[crashtech]

At those temps, a small A/C unit for your computer room might be considered a vital component of your PC cooling system.

 

[24601]

At those temps, a small A/C unit for your computer room might be considered a vital component of your PC cooling system.

Not at home right now, so installing an A/C unit isn't really an option ._.;;

 

[allenk09]

This is my overclock. I originally had a 4770k that wouldn't go above 4.2Ghz and wasn't stable stock on Prime 95 with different sticks of RAM, memtest, everything. I RMA'd it and got one that now won't go above 4.5Ghz, but is completely stable, which I am OK with, though I really dislike odd numbers and wish it would hit 4.6Ghz at temperatures I could handle.

 

[BallaTheFeared]

Chip looks good, delidding is an option for gaining some additional Hz! Your frequency to voltage looks good so there is probably a bit more left in the tank

 

[NFarnzy]

I have been working to lower my Vcore. I got it down to 1.21 from 1.26 .. Better temps and it is stable. I am liking this. 4.4ghz with uncore 40x, for adaptive voltage.. I set my additional turbo cache volt to 1.21v .. running stable!

 

[BrightCandle]

... now won't go above 4.5Ghz, but is completely stable, which I am OK with, though I really dislike odd numbers and wish it would hit 4.6Ghz ....

Both 4500 and 4600 are divisible by 2 without producing a remainder. Both 4.5 and 4.6 fail to be divisible by 2. Thus either both or neither are technically "odd" numbers in the mathematical sense. I suspect if you asked most people they would say 4500 (4.5) is a more round number than 4600 (4.6). I understand the sentiment that you want more speed, I have always wanted more from my SB-E, it has never gone above 4.4Ghz reliably at a voltage I am willing to push through it.

 

[Splenyi]

Ok, here's what I did to my 4670k (not all at once, but this is what it is now)

- cpu ratio is 44
- uncore is 3.8Ghz
- vcore is 1.275
- turned XMP on (9 9 9 24 1600Mhz)
- turned turbo off

That runs 94°C at the hottest while running Prime95 for at least 10 minutes. Tried cranking the ratio up to 45, but that crashed a few seconds into Prime95.

Either I have a terrible piece of silicon, or Prime95 is amazingly taxing (probably a bit of both).

Any recommendations on what I could change with it? My board is a Z87X-D3H and cooler is a H100i with 2 SP120 performance (in quiet mode on Corsair Link) fans (well, not mine, this is a friends PC I'm tuning)?

 

[BallaTheFeared]

If you want to go any higher in prime95 you'll have to delid first.

The two main BBCodes I've noticed, are 124 and 101, maybe they'll help someone out.

124 - More core voltage
101 - More uncore (ring bus) voltage


Also some memory codes D)


Currently at 4.6GHz core 1.155v, 4.5GHz Uncore 1.140v, 2400MHz 13-13-13-37 1T on the ram with 1.7v (lulz). Using the Intel optimized Linpack, actually finding Crysis 3 to be a better overall test than anything else, it will pick up on core, ram, and uncore it's amazing... lol

 

[Arachnotronic]

Both 4500 and 4600 are divisible by 2 without producing a remainder. Both 4.5 and 4.6 fail to be divisible by 2. Thus either both or neither are technically "odd" numbers in the mathematical sense. I suspect if you asked most people they would say 4500 (4.5) is a more round number than 4600 (4.6). I understand the sentiment that you want more speed, I have always wanted more from my SB-E, it has never gone above 4.4Ghz reliably at a voltage I am willing to push through it.

Odd mod 100?

 

[CakeMonster]

- cpu ratio is 44
- uncore is 3.8Ghz
- vcore is 1.275
- turned XMP on (9 9 9 24 1600Mhz)
- turned turbo off

Which "turbo" setting is that?

 

[Splenyi]

If you want to go any higher in prime95 you'll have to delid first.

The two main BBCodes I've noticed, are 124 and 101, maybe they'll help someone out.

124 - More core voltage
101 - More uncore (ring bus) voltage


Also some memory codes ()


Currently at 4.6GHz core 1.155v, 4.5GHz Uncore 1.140v, 2400MHz 13-13-13-37 1T on the ram with 1.7v (lulz). Using the Intel optimized Linpack, actually finding Crysis 3 to be a better overall test than anything else, it will pick up on core, ram, and uncore it's amazing... lol

I don't know how to do those codes
1.155v for 4.6Ghz seems soooooo low. I think that overclock would crumble in Prime95 Or you have some silicon crafted by the gods.

Which "turbo" setting is that?

Intel Turbo Boost.

I changed my overclock, because 94°C was just too hot. It's 4.2Ghz on 1.225v now, peeking at 81°C on Prime95. That's a 200Mhz trade off for 0.05v less and 13°C cooler, I'm happy with that.

 

[ehume]

I've bought an i7 4770k and put it on a GA-Z87X-UD4H with 8GB Crucial Ballistix DDR3-1600. I'm running the CPU at 4.7GHz on a Vcore of Normal + 0.222v, RAM at 1600MHz. I tried tweaking some other Voltages, and speeding up the uncore, but gave it up.

Heatsink is my long-suffering NH-D14 with two NF-A15's. Testing environment is a basement where the ambient runs around 20c.

Observations:


The heatsink puts out mild warmth, not real heat. I could heat the room with the efflux from my i7 860 at a mere 4GHz. That implies that the IHS is a thermos bottle for this Haswell.

Early on in LinX, before it crashed, the 1st core was at 85c with a Vcore of normal+0.20v. At the same place in LinX with a Vcore of normal+0.22v, the 1st core is 90c. So it is clearly past the inflection point: an increase of 0.02v causes a 5c rise in core temp.

The chip is stable for 20 reps of LinX at N+0.22v, unstable on the 4th rep of LinX at N+0.20v. So I set it to N+0.222v just to be on the safe side. I am waiting to hear from Gigabyte tech support as to whether Normal + X is automatically Adaptive Voltage.

GFlops seems to depend on Voltage. With fixed Voltage of around 1.27-1.28 Volts, GFlops was high 68GFlops to low 69. On N+0.222v it's 69.6+ GFlops.

AIDA64 runs much cooler than LinX on this setup.

 

[Phantom9309]

What brand heatsink are folks using?

 

[sxr7171]

What brand heatsink are folks using?

With those temps I'd be inclined to go with prepackaged water. Some of them cost about the same as high end air.

 

[Splenyi]

Yep, got a H100i here

 

[JoeRambo]

GFlops seems to depend on Voltage. With fixed Voltage of around 1.27-1.28 Volts, GFlops was high 68GFlops to low 69. On N+0.222v it's 69.6+ GFlops.

The bad news is that your Linx is outdated. GFLOPS with AVX enabled Linx (google for it) should be ~130GFlops @ your clock. But be warned that wattage output is ridiculous at that rate.

 

[ehume]

The bad news is that your Linx is outdated. GFLOPS with AVX enabled Linx (google for it) should be ~130GFlops @ your clock. But be warned that wattage output is ridiculous at that rate.

You are so right. I tried one and it crashed my system. Since I couldn't find a DL I trusted, I've put the LinX project on the back burner.

I did manage to boot in 5.0GHz. Once:

 

[nurturedhate]

4670k at 4.7 at 1.21v at 78c. On an gigabyte z87x-ud3h and a thermalright macho rev.a

 

[BallaTheFeared]

Grats, nice chip!

 

[fastamdman]

Just picked up a haswell chip 4770k and the asrock z87 extreme6. Currently running 4.2/4.2 with 1.17 voltage on both. Stable.

I have a pretty good chip but this evo 212+ isn't keeping it cool enough. With only 1.17 volts I am at 85-87c with AVX linx. I might reseat / reapply thermal paste.

I am realllllly thinking about going with a nzxt kraken x60 and rocking 4.5ghz

 

[nurturedhate]

I'm having great success with a thermalright macho rev.a. Running a 4670k at 4.7 at 1.21v at 78c. You running a 4670k or 4770k?

 

[fastamdman]

4770k forgot to mention that. I noticed the difference from non AVX linx to AVX is a huge jump in temps, at the same voltage level.

Running 43/43 at 1.65 on both in linx avx currently, temps of 83 max. So far so good. I really wanna go watercooling for this chip. I am 100% confident I could get 4.6 stable, probably even closer to 5ghz with enough cooling.

 

[nurturedhate]

1.65v? that seems VERY high.

 

[NTMBK]

I noticed the difference from non AVX linx to AVX is a huge jump in temps, at the same voltage level.

That's what happens when you use the other half of the vector units.