Help with overclocking a 4790k?

[Ed1]

the newer version of prime95 are harder on system, they changed the length of time each FFT runs . I use 27.9 mainly cause that was one i started with so my data is comparable, it does support AVX .

 

[UaVaj]

there is absolutely nothing wrong with that 4790k. it is a binned 4770k with factory overclock. then rebadge to be called a 4790k.

 

[Rvenger]

The only thing that was a dud here was Intel's marketing of the 4790K. I've seen no evidence at all that it can clock higher than a 4770K, and for both, 4.7GHz is above average.

The only think that Devil's Canyon has going for it is lower temps, but that doesn't necessarily mean higher overclocks.

And I hear the L3 batches are real duds and run hot as well. The beefier input caps they put on the CPU didn't do much in the way of overclocking either... so I kept my trusty ol' 4770k.

Next time I grab my hands on a stock 4ghz CPU I will probably quit overclocking unless 5ghz is guaranteed.


I would not run Prime 95 version 28.5, it will throttle your CPU at +4.5ghz clocks.

 

[tolis626]

And I hear the L3 batches are real duds and run hot as well. The beefier input caps they put on the CPU didn't do much in the way of overclocking either... so I kept my trusty ol' 4770k.

Next time I grab my hands on a stock 4ghz CPU I will probably quit overclocking unless 5ghz is guaranteed.


I would not run Prime 95 version 28.5, it will throttle your CPU at +4.5ghz clocks.

Well, overclocking from stock (practically 4.2GHz on all 4 cores) to 4.7Ghz on the core and 4.5GHz on the uncore gave a nice performance boost. Nothing groundbreaking by any means, but not insignificant either. I'd swear I could perceive a difference too, but it's probably placebo. I don't think anyone could tell the difference in general usage.

Prime95 usually overwhelms my H110 within 30 minutes or less of testing. At that point, there is a sudden spike in temperature and I never let it continue past that point. I will risk having my system crash on me some day, but I won't risk frying it to avoid any crashing. Although I don't think crashing will be an issue when OCCT and x264 are fully stable.

 

[Rvenger]

Honestly, I used to run my uncore high and you really don't need to. It can create unneeded extra heat and stability issues. Plus if you are running a cache voltage of over 1.3 you can run the risk of damaging the CPU. I backed off mine and set it at 3.7ghz at 1.08v. If you read the Haswell guide on OCN it provides a ton of useful information regarding uncore and voltages.

For the 2% performance increase with a high uncore (cache voltage, or ring voltage it is also called depending on mobo brand) you will see you are at a much higher risk for degradation and stability issues down the road.

If you want to use Prime95 use the older 27.9 version.

 

[UaVaj]

I would not run Prime 95 version 28.5, it will throttle your CPU at +4.5ghz clocks.

is that a feature or bug in 28.5?

 

[Ed1]

is that a feature or bug in 28.5?

Its a feature to give better performance but on downside it harder on the CPU . read the file whatsnew.txt in zip.

""New features in Version 28.5 of prime95.exe
-------------------------------------------

1) Changed the output to the worker windows during LL and PRP tests. The new output includes
the estimated time to complete the test. There are two new options described in undoc.txt:
ClassicOutput and OutputRoundoff.

2) Added some new options described in undoc.txt: ScaleOutputFrequency, TitleOutputFrequency,
and SilentVictoryPRP.

3) Benchmarking on hyperthreaded machines now times only the most common cases. Specifically,
hyperthreading is used only in the one cpu and all cpu cases.

4) Benchmarking trial factoring is now off by default. Prime95 should not be used for trial
factoring. GPUs are about 100 times more efficient at that task.

5) On multi-core machines, benchmarks are now run on multiple workers. This measures the effect
of memory bandwidth during testing and helps you select the setup that gives you the most throughput.

6) There are many new options described in undoc.txt to customize the benchmarking process.

7) In build 2, added a preference to bypass some of Mac OS X Mavericks' power saving features.


New features in Version 28.4 of prime95.exe
-------------------------------------------

1) Reduced memory usage. This may make some single-thread benchmarks slower, but when
running several workers on machines where memory is a bottleneck there should be a
small performance increase.


New features in Version 28.3 of prime95.exe
-------------------------------------------

1) Supports Intel's new for fused multiply add instruction introduced with the Haswell CPU.
This results in faster FFTs. Note that performance on many Haswell systems is memory-bandwidth
limited. This means that when running workers on all cores performance gains will be small.
2) Some minor optimizations may give a very small performance boost for AVX CPUs.
3) All new test torture test data for AVX CPUs. The new data runs more interations, thus more time
is spent torturing the CPU rather than initializing the FFT routines.
4) Information added to result lines containing "has a factor". This information may be used
by the server's manual web page to give proper TF / P-1 / ECM cpu credit at a future date.""

 

[tolis626]

Honestly, I used to run my uncore high and you really don't need to. It can create unneeded extra heat and stability issues. Plus if you are running a cache voltage of over 1.3 you can run the risk of damaging the CPU. I backed off mine and set it at 3.7ghz at 1.08v. If you read the Haswell guide on OCN it provides a ton of useful information regarding uncore and voltages.

For the 2% performance increase with a high uncore (cache voltage, or ring voltage it is also called depending on mobo brand) you will see you are at a much higher risk for degradation and stability issues down the road.

If you want to use Prime95 use the older 27.9 version.

Well, for 4.5GHz uncore I use 1.25V, and that's because I'm too lazy for now to test lower, but I can probably go lower than that. I will get to testing during the weekend. Overclocking the uncore really didn't increase performance by too much, but it did increase it nevertheless. I may not keep my uncore overclocked (see why below), but for now I'm keeping it this way because I feel good getting the max out of my system. I'll get over it in a while and think more logically, but for now it's "MOAR MEGAHERTZZZZ!!!!". Can't help it! :biggrin:

Now, I read the Haswell overclocking guide on OCN (Excellent write up by the way) a couple of weeks ago, and I tried to follow whatever they said. So I went on and underclocked/undervolted my uncore to 3.7GHz/1.125V and managed to get 4.7GHz stable at 1.305V on the core. That's 0.01V down. However, no higher frequencies stabilized at sane voltages with the uncore underclocked, so I figured what the hell, that extra performance, albeit little, may be worth the extra 0.01V. Or it may not be, but thing is, now I have a choice. The performance hit is small, the voltage reduction is small, and my uncore's settings weren't extreme anyway. So I don't know.

PS : I have no idea how adaptive voltage works for the uncore. It seems to give 0.03V more than what it's set at when under load. Is there any point in running adaptive on the uncore, anyway? Or setting a different min and max ratio? Things have gotten more complicated than they needed to, IMO.

 

[tolis626]

Oops, double post

 

[Ketchup]

tolis626, what temps are you seeing on your 4790? The cores on mine are down around 30 at idle (not overclocking), but it does get pretty warn under load. I'll have to update the high when I am back at my PC tonight, but I wanted to ask you before I forgot.

 

[Rvenger]

Ketchup and Tolis, what batches did you guys get?

 

[tolis626]

tolis626, what temps are you seeing on your 4790? The cores on mine are down around 30 at idle (not overclocking), but it does get pretty warn under load. I'll have to update the high when I am back at my PC tonight, but I wanted to ask you before I forgot.

At idle my temps are bouncing between 27 and 35 C with a 24C ambient. AC is on. Under 100% load at 4.7GHz/1.315V it will get up to 85-90C during OCCT, but these are temperature spikes that happen momentarily when fans spin down. Generally I get about 80C during OCCT/x264 with fans at full speed and about 85-90C during Prime95 small FFT. I've also played some Borderlands 2 on the iGPU to kill some time, and temps never exceeded 75C, even with the iGPU overclocked to 1600MHz with 1.2V.

If you are concerned about your CPU getting "warm", don't be. Haswells are hot, there's no going around that unfortunately.

Ketchup and Tolis, what batches did you guys get?

Mine is L331C516. It was bought about a month ago in Greece, if this sort of thing matters.

Considering what you said about L3 batches being duds that overheat... Maybe that's true. Wanting an easy 4.8GHz was probably stupid on my part, but I expected temperatures to be quite better. Granted, they aren't 4770k bad, but from what I've seen in reviews, they should be better, especially when there's a 280mm AIO doing the cooling.

PS : After almost messing up big time, I got a little shaken and decided to stop being paranoid about the last bit of performance. My CPU is now running happily at 4.7GHz with 1.305V and an uncore of 3.8GHz with 1.12V (Although HWMonitor says that ring voltage is 1.147V under load, but it's still low and ok). I can always bump those easily for benchmarking, but for daily use there's no point in torturing a 300+$ CPU.

 

[Ketchup]

Thanks tolis626. Just got her put together a couple days ago (only the CPU, motherboard, and RAM were upgraded) and haven't overclocked yet. I will probably do one run just to see what she will do.

Idle temps today are 24-24-26-24. Load temps are 68-69-66-61. This is while running the latest 3DMark a couple times. I was actually surprised, my 2500k with the same cooler was hitting 60 at load, so I was glad to see this one isn't getting too[/] much hotter.

Rvenger, I didn't think to look at the stepping, sorry. If I need to take the EVO off for some reason, I will look.

 

[tolis626]

Thanks tolis626. Just got her put together a couple days ago (only the CPU, motherboard, and RAM were upgraded) and haven't overclocked yet. I will probably do one run just to see what she will do.

Idle temps today are 24-24-26-24. Load temps are 68-69-66-61. This is while running the latest 3DMark a couple times. I was actually surprised, my 2500k with the same cooler was hitting 60 at load, so I was glad to see this one isn't getting too[/] much hotter.

Rvenger, I didn't think to look at the stepping, sorry. If I need to take the EVO off for some reason, I will look.

No problem. Glad I helped.

I think what Rvenger wants is the batch number that is written next to the serial number on the CPU's box, not on the CPU itself.

 

[Rvenger]

L3 stands for a manufacture date of 2013. The L4 (2014) batches are known to run cooler and at less volts. The batch number is located on the box as well. No need to pull the heatsink.

 

[Majic 7]

Here are some basically stock readings. Turbos to 4.4 on all cores under load. 1.160 voltage, tops out at 91c on prime small ffts. 75F ambient. The most intense game I run right now is Skyrim with RealVision ENB and 65 other mods. Maxes out the GPU but the CPU temps are 40 to 50C. Idles at around 30C, + - 2. L329C241. Wanted a beast at stock and I think I'm pretty close.

 

[tolis626]

Here are some basically stock readings. Turbos to 4.4 on all cores under load. 1.160 voltage, tops out at 91c on prime small ffts. 75F ambient. The most intense game I run right now is Skyrim with RealVision ENB and 65 other mods. Maxes out the GPU but the CPU temps are 40 to 50C. Idles at around 30C, + - 2. L329C241. Wanted a beast at stock and I think I'm pretty close.

For an H70, I think that's pretty awesome. I think I've seen 4770k's that the H80i couldn't cool at 4.3GHz during Prime95. Plus, 1.16V for 4.4GHz is very good, even for a 4790k. Mine was 1.195V at stock. I think you should slap a better cooler and overclock that sucker!

 

[Galatian]

I had my i7-4790K delidded as well and it did drop temperatures significantly (using CLU). I can now run it at 4,8 GHz Core @ 1,35 V and Cache at 4,6 GHz. Temperatures go no higher then 80 degrees Celsius in my custom Water loop in it's "quiet mode" (fans and pump at low speed).

I was hoping for 5,0 GHz but even above 1,45 V I could not get it stable and that was reaching a point where I would feel really uncomfortable ;-)

I do think the sweet spot is/was at 4,6 GHz though but a 200 MHz overclock didn't feel like much.

A few things I noticed:

1. Devil Canyon needs a lot less input Voltage. Mine currently runs at 1.7 V

2. There is no obvious difference between override and adaptive voltage. Both still idle at the same amount and both are stable when you use the the same voltage. The difference is in the details: AVX2/FMA3 is using more voltage at full load. If you are running Prime 28.5 you probably fare better using adaptive. If you are using the older prime versions to get your chip stable and just don't care about AVX2/FMA3 you might as well use the override voltage and you can probably even save 0.07 V. I also checked with Handbrake which uses AVX2/FMA3 and it still is stable; Prime 28.5 is therefore applying an unreal load.

3. The new TIM is somewhat better, delidding is still king.

4. This is probably obvious: I needed to increase Voltage to the Cache to get it stable again after I increased the core clock. Sometimes lowering the Cache is preferable to get higher Core Clocks (Core is King ;-) )

 

[blackened23]

I don't even care what Prime95 does to temps, because Haswell automatically increases the voltage when prime95/IBT are used by .1V but that never happens in real world apps. Ever. Obviously when voltage increases by .1V, yeah, temps will be bad.

My real world temps even at their most demanding, with near 100% threads in use, are 20-25C lower than prime95. But prime95 will shoot straight to 90C thanks to the auto increase in voltage. Even though prime95 is generally stable despite the high temps, screw it. I haven't used Prime95 for some time as a stress test and my system has been rock solid. Aida 64 for stress testing has worked great and playing crysis 3 and doing real world apps for hours on end has shown the system stable.

 

[tolis626]

I had my i7-4790K delidded as well and it did drop temperatures significantly (using CLU). I can now run it at 4,8 GHz Core @ 1,35 V and Cache at 4,6 GHz. Temperatures go no higher then 80 degrees Celsius in my custom Water loop in it's "quiet mode" (fans and pump at low speed).

I was hoping for 5,0 GHz but even above 1,45 V I could not get it stable and that was reaching a point where I would feel really uncomfortable ;-)

I do think the sweet spot is/was at 4,6 GHz though but a 200 MHz overclock didn't feel like much.

A few things I noticed:

1. Devil Canyon needs a lot less input Voltage. Mine currently runs at 1.7 V

2. There is no obvious difference between override and adaptive voltage. Both still idle at the same amount and both are stable when you use the the same voltage. The difference is in the details: AVX2/FMA3 is using more voltage at full load. If you are running Prime 28.5 you probably fare better using adaptive. If you are using the older prime versions to get your chip stable and just don't care about AVX2/FMA3 you might as well use the override voltage and you can probably even save 0.07 V. I also checked with Handbrake which uses AVX2/FMA3 and it still is stable; Prime 28.5 is therefore applying an unreal load.

3. The new TIM is somewhat better, delidding is still king.

4. This is probably obvious: I needed to increase Voltage to the Cache to get it stable again after I increased the core clock. Sometimes lowering the Cache is preferable to get higher Core Clocks (Core is King ;-) )

First off, may I ask what your WC loop consists of? 80C does seem kinda high for a custom loop, but not everyone likes overkill. Although, now that I think of it, 80C for 4.8GHz at 1.35V isn't bad at all.

Also, my input voltage is set to 1.95. Guess I should tone it down a bit.

Underclocking cache does do miracles for Haswell it seems. Mine made a more modest difference, "just' 0.01V for 4.7GHz (still, more than I hoped for). I suggested the same to a friend who has a 4670k and his voltage for 4.5GHz dropped from 1.28V to 1.255V. Now that's good.

I don't even care what Prime95 does to temps, because Haswell automatically increases the voltage when prime95/IBT are used by .1V but that never happens in real world apps. Ever. Obviously when voltage increases by .1V, yeah, temps will be bad.

My real world temps even at their most demanding, with near 100% threads in use, are 20-25C lower than prime95. But prime95 will shoot straight to 90C thanks to the auto increase in voltage. Even though prime95 is generally stable despite the high temps, screw it. I haven't used Prime95 for some time as a stress test and my system has been rock solid. Aida 64 for stress testing has worked great and playing crysis 3 and doing real world apps for hours on end has shown the system stable.

I totally agree. The only reason I still use Prime95 is to do quick and dirty stability testing. Most of the time, if Prime95 is to crash, it will do so in the first 10 minutes. I usually do 20 minutes of Prime95 before I deem something almost stable. Then the hours of Aida64 (awesome app by the way, I'm considering buying when the trial ends), OCCT and x264 testing usually verify that it's stable. Also Prime95 is better at finding cache instabilities it seems. But yeah, it's bad. Really bad. :thumbsdown:

 

[Galatian]

It's pulling almost 200W under load in Prime, so the temperatures in my loop are fine, water is at 32 degrees. There is only so much you can do, if you can't transport the heat fast enough put of the silicon. Bare die might be a solution though.

That being said I'm pretty happy with my Devil Canyon and the ASRock Extreme9. Just waiting for new GPUs and a NVMe M.2 SSD to come out ;-)

 

[tolis626]

It's pulling almost 200W under load in Prime, so the temperatures in my loop are fine, water is at 32 degrees. There is only so much you can do, if you can't transport the heat fast enough put of the silicon. Bare die might be a solution though.

That being said I'm pretty happy with my Devil Canyon and the ASRock Extreme9. Just waiting for new GPUs and a NVMe M.2 SSD to come out ;-)

200W eh? 80C is fine then.

I'm always tempted to delid and try cooling without the IHS. But, as I said, I'm afraid to do it. Really, I'm just stupidly clumsy even at things I'm comfortable with doing. For instance, I scratched my mobo a little when I was trying to install the H110. I almost cried until I got it to boot. The scratch was too small and right next to the screwhole, but still, it was enough to scare me quite a lot.

I probably haven't mentioned that, for personal usage at home, I had a laptop with a Core2Duo P8600 at 2.4GHz and some 4GB of DDR2. I could underclock the 4790k and still be more than happy. I'm just enthusiastic about overclocking. It's definitely not that I don't have enough performance.

 

[Dave3000]

Here are some basically stock readings. Turbos to 4.4 on all cores under load. 1.160 voltage, tops out at 91c on prime small ffts. 75F ambient. The most intense game I run right now is Skyrim with RealVision ENB and 65 other mods. Maxes out the GPU but the CPU temps are 40 to 50C. Idles at around 30C, + - 2. L329C241. Wanted a beast at stock and I think I'm pretty close.

The stock all-core turbo frequency is 4.2 GHz on the 4790k.

http://www.intel.com/support/processors/corei7/sb/CS-032279.htm

 

[tolis626]

The stock all-core turbo frequency is 4.2 GHz on the 4790k.

http://www.intel.com/support/processors/corei7/sb/CS-032279.htm

Depends on the motherboard, really. By default, my Asus Maximus VII Formula (and I guess many other mobos too) had multicore enhancement on and turboed up to 4.4GHz on all cores.

 

[Dave3000]

That does not change the stock specs of the processor. My motherboard does not use the 1-core turbo frequency on all cores when the BIOS is set to default settings. It only does so when I change the overclock profile to XMP or Manual, or change the memory frequency manually. Are you sure that the fail-safe default settings on your motherboard turn on MCE automatically while leaving the memory frequency on Auto and not using XMP?