Overclocking 4770k

[freinando]

Hi guys,


I've spent two days testing my new build. Overclocking I get 4.5ghz on 1.32v. Temps hit the mid 80s with an h110 in pull push set up. Wondering if this is normal. Cant go beyond 4.5 otherwise I have to increase vcore and the cpu hits the thermal wall. What do u think?

 

[masteryoda34]

Sounds pretty normal. Probably need to delid to get better temps.

 

[blackened23]

80C at 1.32V? That actually sounds low for that voltage, but there's no issue at all with that temp - if you get that during prime95/IBT, then you have a golden chip my friend. There is no issue at all with up to 90C on Haswell or IVB, although i'm not sure if you get that during normal use or stress testing?

1.32V does sound a bit higher than desirable. But if you get mid 80s during prime95, you are good to go.

 

[freinando]

Update: its high 80s now

Using Aida64 stress test

Ram clock at 1866mhz

 

[blackened23]

There's no issue with those temps on Haswell or IVB, that is normal. They do get hot at maximum load, but they have a high temperature tolerance (higher than SB). I would try to stay under 90C at all times.

If you're able, you may want to dial voltage down a bit in small increments. 1.25V or lower seems to be preferred, but if you're good to go at 1.3V that works as well.

 

[freinando]

I would be happy with a 4.6ghz - 2100mhz ram but cant get pass the 4.5 mark. Cant be a golden chip

 

[blackened23]

I would be happy with a 4.6ghz - 2100mhz ram but cant get pass the 4.5 mark. Cant be a golden chip

4.5ghz is a great speed with Haswell - consider that a 4.5ghz haswell is faster than a 5ghz 3770k. Anyway, the higher your RAM speed on Haswell, the lower your CPU overclock. I'd stick with 1600-1866. Don't bother with 2100. I'd be really happy with 4.5 with the 4770k, personally - there are some 3770k's that begin to struggle around that area (4.6) as well..

 

[2is]

Keep in mind, Haswell has a LOWER heat threshold than IVB. 100C for haswell vs 105C for Ivy Bridge. I don't remember what it was for Sandy Bridge.

 

[blackened23]

Perhaps, but if he's at 88C during prime 95 then he will be around 65C during normal tasks and games. That's how it is on my 3770k - I never get past 65-67C , even though my load temps approach 90C. So I think he's good to go. I do have to double check what the throttle threshold of the 4770k is, I think 95C perhaps? Anyway, Prime95 temps are far, far higher than what you will ever encounter during normal use.

 

[tential]

4.5ghz is a great speed with Haswell - consider that a 4.5ghz haswell is faster than a 5ghz 3770k. Anyway, the higher your RAM speed on Haswell, the lower your CPU overclock. I'd stick with 1600-1866. Don't bother with 2100. I'd be really happy with 4.5 with the 4770k, personally - there are some 3770k's that begin to struggle around that area (4.6) as well..

For some reason, I feel like people are directly comparing Haswell and Ivybridge clockspeeds and ignoring that Haswell is clock for clock faster than Ivy.

I dunno the exact clockspeeds that are equal to what, but if 4.5 is actually as fast as 5 ghz on ivy, that's blazing fast....

 

[blackened23]

Totally agreed, that's true. It also happened when IVB was released - a lot of folks only looked at overall clockspeeds while ignoring the sizable IPC increase which IVB had. I myself had a 2600k prior to my 3770k and even though my SB could do 5ghz (with a TON of voltage) the 3770k is still faster even at 4.6ghz.
The same theme really applies to the Haswell.

Granted, it's not really noticeable much in real world use, but all synthetics such as cinebench show the lower clocked IVB to be faster by a good margin - despite the 400mhz deficit. Now, i'm not saying upgrading from SB to IB or Haswell is worth it, but it's a good point to bring up nonetheless - the IPC difference makes up for any potentially lower overclock. So with that being the case, when I see someone on the market for a new PC and questioning whether they should get IVB over Haswell, i'm left scratching my head. I don't think getting IVB over Haswell is ever worth it, not unless you get a HUGE price discount.

 

[freinando]

Ok guys, everything ok so far with these settings. I will try later to lower the vcore a little.

Earlier today something really weird happened though,

Today I completed the pull push setup with the two noctua af14 flx that arrived. Yesterday I was cooling the radiator with just one fan because of missing an adaptor for the other.

I was doing stress testing today with a setting of 1.35v core and 4.6ghz, 35 mins no problem in aida64, temps in the low 80s I stopped the test.

Repeated the same test then crashed in the 2 min mark.

Restarted, set 1.36v core and 4.7ghz. Booted up, tested and BOOM!, it hit 100C under load. BSOD

Only one increment of .01v made the cpu temps raise 15-20 degrees.

Restarted, booted up, set the old 4.6 1.35v setting, which was supposed to yield temps in the 80s like before and guess what, BSOD, 100C under load.

Strange isnt it?

Since then I lowered the clock and the voltage to the settings I have now, however the temps never were quite like they were before. I cant understand this.

Im overclocking in asrock extreme6.

Temps idling are in the 30s. But this sudden change under load iis inexplicable to me.

 

[bradly1101]

But this sudden change under load iis inexplicable to me.

CPU damage because of overvolting? Those are tight tolerances in there. The insulating layer(?) may be breaking down.

 

[freinando]

Overvolting? Whats the voltage tolerance of this cpu? Also it happened from 1.35 to 1.36.

 

[2is]

Ok guys, everything ok so far with these settings. I will try later to lower the vcore a little.

Earlier today something really weird happened though,

Today I completed the pull push setup with the two noctua af14 flx that arrived. Yesterday I was cooling the radiator with just one fan because of missing an adaptor for the other.

I was doing stress testing today with a setting of 1.35v core and 4.6ghz, 35 mins no problem in aida64, temps in the low 80s I stopped the test.

Repeated the same test then crashed in the 2 min mark.

Restarted, set 1.36v core and 4.7ghz. Booted up, tested and BOOM!, it hit 100C under load. BSOD

Only one increment of .01v made the cpu temps raise 15-20 degrees.

Restarted, booted up, set the old 4.6 1.35v setting, which was supposed to yield temps in the 80s like before and guess what, BSOD, 100C under load.

Strange isnt it?

Since then I lowered the clock and the voltage to the settings I have now, however the temps never were quite like they were before. I cant understand this.

Im overclocking in asrock extreme6.

Temps idling are in the 30s. But this sudden change under load iis inexplicable to me.

If you were unstable and crashed within 2 minutes at 1.35 at 4.6GHz what makes you think adding .01 volts and ADDING 100MHz would be better????

You're trying to hit speeds that are hard enough to hit with haswell as it is, and your solution was to go higher? Your technique needs some tweaking.

 

[freinando]

If you were unstable and crashed within 2 minutes at 1.35 at 4.6GHz what makes you think adding .01 volts and ADDING 100MHz would be better????

You're trying to hit speeds that are hard enough to hit with haswell as it is, and your solution was to go higher? Your technique needs some tweaking.

I ran the same test for 35 min without crashing. Then ran it again and crashed in 2 mins. Why?

Also I did this to see how quick it would crash with the 4.7ghz clock. What I didnt expect was to have the temps soaring that much to tj max only with .01v when previously they were in the 80s

 

[2is]

Because you were unstable. That's what unstable means, you can't predict instability, it can happen in 2 minutes it can happen in 2 hours, it can happen with only a handful of applications.

1.35 is a lot of volts to push through a 22nm processor, and your processor was already telling you something at 4.6GHz and your solution was to push it even harder? I was dumbfounded reading that and though I had to be misunderstanding because it made no sense at all.

Anyway, your best bet now is to reset everything to default and see what happens. Cross your figures that you didn't fry your chip

 

[BallaTheFeared]

Haswell overclocking discussion, as well as various links and my own personal spin on baby step overclocking.
http://forums.anandtech.com/showthread.php?t=2326066

 

[JimmiG]

Because you were unstable. That's what unstable means, you can't predict instability, it can happen in 2 minutes it can happen in 2 hours, it can happen with only a handful of applications.

1.35 is a lot of volts to push through a 22nm processor, and your processor was already telling you something at 4.6GHz and your solution was to push it even harder? I was dumbfounded reading that and though I had to be misunderstanding because it made no sense at all.

Anyway, your best bet now is to reset everything to default and see what happens. Cross your figures that you didn't fry your chip

This. 35 minutes is not enough to determine stability. You need to run a 24-hour test to be sure. While testing overclocks, I usually go with 1-hour runs. Then when I've achieved an overclock that *seems* stable I run a 24-hour Prime95 test (plus various other tests and games, but that's the main one). Then I may have to up the VCore slightly or maybe even lower the multi by 1x.

I think Haswell has another temperature threshold at 130C or something, where it will shut down completely to protect it. Doubt you did any permanent damage to the chip, but 1.35V is not good for the processor in the long term. I'm at 1.214V with mine.

BTW to test for temps in Aida64, remember to run the test with only "FPU" checked. The CPU will actually run hotter than it does with with CPU+FPU.

 

[crashtech]

4.5ghz is a great speed with Haswell - consider that a 4.5ghz haswell is faster than a 5ghz 3770k...

For this to be true, Haswell IPC must be >10% than IB, which I have not seen in benchmarks. I'd really like to see some evidence to confirm your assertion.

More realistically, 4.5GHz Haswell = 4.7GHz Ivy, 4.8 at most.

 

[blackened23]

For this to be true, Haswell IPC must be >10% than IB, which I have not seen in benchmarks. I'd really like to see some evidence to confirm your assertion.

More realistically, 4.5GHz Haswell = 4.7GHz Ivy, 4.8 at most.

I can try to find concrete benchmarks, but I am extrapolating from my experience of my former 2600k which was able to do 5ghz, and my current 3770k. I did all sorts of benchmarks a year ago when I swapped all of my components out, and my 3770k at 4.6GHz convincingly beat the 5ghz SB in every benchmark.

Every single benchmark was a win for the 3770k with a 400mhz deficit. Now, the performance difference certainly isn't noticeable and only synthetics will pick it up, but seeing as the IVB beats SB even with a 400mhz deficit I'm sure the same applies to Haswell. Absolutely sure - the IPC difference between the 4770k and 3770k is greater than that of the 2600k and 3770k IIRC.

Also, since you specifically mention it, 4.7 / 4.8 ghz is not a realistic overclock for the average 3770k. I'm only able to get 4.7ghz with a TON of voltage with temps that pass 90C during extreme stress testing, I consider my chip to be "golden", if you will - IVB is not that different than Haswell in terms of overclockability / temperatures. Anyone who states otherwise clearly has selective memory. Anyway, the point remains that a Haswell with a clockspeed deficit, even 400mhz, will still pass IVB.

 

[crashtech]

I can try to find concrete benchmarks, but I am extrapolating from my experience of my former 2600k which was able to do 5ghz, and my current 3770k. I did all sorts of benchmarks a year ago when I swapped all of my components out, and my 3770k at 4.6GHz convincingly beat the 5ghz SB in every benchmark.

Every single benchmark was a win for the 3770k with a 400mhz deficit. Now, the performance difference certainly isn't noticeable and only synthetics will pick it up, but seeing as the IVB beats SB even with a 400mhz deficit I'm sure the same applies to Haswell. Absolutely sure - the IPC difference between the 4770k and 3770k is greater than that of the 2600k and 3770k IIRC.

Also, since you specifically mention it, 4.7 / 4.8 ghz is not a realistic overclock for the average 3770k. I'm only able to get 4.7ghz with a TON of voltage with temps that pass 90C during extreme stress testing, I consider my chip to be "golden", if you will - IVB is not that different than Haswell in terms of overclockability / temperatures. Anyone who states otherwise clearly has selective memory. Anyway, the point remains that a Haswell with a clockspeed deficit, even 400mhz, will still pass IVB.

The (very limited) data I'm seeing has the average OC Haswell beating the average OC Ivy by 2%, that may improve with time as new steppings are released and the average enthusiast figures out these new chips, but I think your claims are... optimistic. Keep in mind I'm not in the crowd that considers Haswell a failure. I do appreciate accuracy, though.

And I was not the one that brought up a hypothetically unrealistic overclock first. I thought we understood it is by way of comparison only.

 

[JimmiG]

It depends a lot on which app you compare:
http://anandtech.com/bench/Product/551?vs=836

Anywhere from just over 4% (~180 MHz at 4.5 GHz) up to over a 20% difference (~900 MHz @ 4.5 GHz), not counting that E-Learning standout.

 

[boxleitnerb]

On average 9% in games and 10% in apps with the same clocks:
http://www.hardware.fr/articles/897-25/gains-moyennes-cpu.html

 

[crashtech]

Assuming 10% IPC increase on average, it takes a 4.1 Haswell to beat a 4.5 IB, 4.2 to beat 4.6, 4.3 to beat 4.7, etc. Since the average OC'ed Haswell doesn't make it quite as high as the average OC'ed Ivy (about 200 MHz less on average), the difference between average OC CPUS should actually be more like 5%, in theory. The 2% result I saw is likely because Haswell OCing hasn't quite been perfected yet.