Need Advice on Overclocking 5820k

[Xerin7]

I'm fairly new to overclocking and could use some advice. Just built my 5820k rig with an Asus X99-A board. I used the 5-Way Optimization feature to overclock my cpu, which landed at 4.609. I've tried a couple of games and everything has been solid. Ran 3D Mark and my pc crashed due to cpu overheating.

Should I scale back the overclock due to this? If it runs fine in my games and only crashes on a stress test like 3D Mark is that ok or should I be concerned? If I need to reduce the overclock, what would you recommend? Should I just reduce the ratio for my 6 cores from 46 to something lower and keep everything else the same? Currently it is:

BCLK: 100.2
Ratio: 46
CPU Cache Ratio: 30x

CPU Core Voltage
Offset Voltage: .075
OC Voltage: 1.225

 

[FiLeZz]

3dmark not that stressful compared to other CPU tests.

That for sure is not a stable CPU..

 

[Xerin7]

So given that, how should I change it? Should I just reduce the ratio and if so to what? Do I need to change the voltage or anything else?

 

[Burpo]

Drop multi to 45..

 

[BonzaiDuck]

The "Auto-overclocking" features are pretty much good for one thing -- in my [humble] opinion: To "give you an idea about how far it will OC easily." At least, that's the way it seemed with my skt-1155 processor. The voltage set by these features is likely to be excessive. Even so, this is a new generation of motherboard and processor.

I pretty much "knew" the safe voltage limit for my SB-K: it had the same lithography as the i7 Nehalem in its last release. I might have been able to "guess" the upper limit of a "safe range" for Ivy Bridge, or I thought I saw folks discussing a level around 1.30V. I couldn't be sure about Haswell, and my guess -- you can only extrapolate based on the lithography of 22nm.

Recent Anandtech article on the 5960X shows voltages above 1.2 and heading toward 1.25V before encountering "the wall." In that range for the 8-core processor, it looked as though it was capable of 4.3 to 4.4 -- and might be pushed to 4.5.

For the 6-core processors, you might be able to get a little further.

.. . But I still have to wonder about the voltage . . .

 

[biostud]

I would lower the multiplier. I had same problem with crashing in 3Dmark when I started overclocking my 5820K. I've now settled with 4.4Ghz @ 1.275V. But I might be able to run at lower voltages, as I started with 1.3V and run it for a couple of weeks, then lower voltage by 0.025V. If I can get 4.4Ghz @ 1.2V that's perfect for me.

 

[BonzaiDuck]

I would lower the multiplier. I had same problem with crashing in 3Dmark when I started overclocking my 5820K. I've now settled with 4.4Ghz @ 1.275V. But I might be able to run at lower voltages, as I started with 1.3V and run it for a couple of weeks, then lower voltage by 0.025V. If I can get 4.4Ghz @ 1.2V that's perfect for me.

I'm puzzled as to how or why you started at a "higher" voltage, intending to move "down the scale." I could understand this in context with my own previous remarks about using the "auto-overclock" features as a preliminary measure.

What sort of temperatures are you getting (average-of-cores and average-of-core-maximums) with your H100 cooler? More specifically, using a stress-test like OCCT "CPU:LinPack," IBT ("all threads") or LinX for all threads or affinitized for four threads?

It would be good to see some of our "veterans" and "old-hands" begin a discussion of OC'ing these beasts. But they would all need to jump on the Haswell-E bandwagon to do it. As I said -- I haven't; I plan to do so; and I'm waiting as I gather more info and plan a cooling strategy.

BY THE WAY: That ASRock "Killer" motherboard looks like a winner. I'm too enamored of ASUS, and it's hard to change old habits. I looked at the TweakTown review of the X99X: The BIOS screens look manageable and promising.

 

[biostud]

I'm puzzled as to how or why you started at a "higher" voltage, intending to move "down the scale." I could understand this in context with my own previous remarks about using the "auto-overclock" features as a preliminary measure.

What sort of temperatures are you getting (average-of-cores and average-of-core-maximums) with your H100 cooler? More specifically, using a stress-test like OCCT "CPU:LinPack," IBT ("all threads") or LinX for all threads or affinitized for four threads?

It would be good to see some of our "veterans" and "old-hands" begin a discussion of OC'ing these beasts. But they would all need to jump on the Haswell-E bandwagon to do it. As I said -- I haven't; I plan to do so; and I'm waiting as I gather more info and plan a cooling strategy.

I've dropped LinX with AVX as a stress test because it reaches +95C with H100i and get a BSOD when I run it. So my system is not LinX stable, but it's "stable" when I use it for normal computing and gaming, and that's fine enough for me. I don't do anything mission critical on this machine so if I run into 1-2 crashes due to oc/year that's not a problem for me. Under "normal" stress load like ASUS realbench it reaches around 75C.

I won't say that this is the best way to do oc, but it works for me. When I continue my lowering of voltages and if I experience crashes I will probably use ASUS realbench to test the CPU, or simply add 0.025V and be done with it.

 

[exar333]

Not sure about the 5820k, but I had a lot of success using the 1.25x strap on my SB-E. I will definitely be trying that out as well in combination with the straight multi adjustment. Some CPUs respond better to different methods.

I also second the fact that your voltage might be too high, only raise it when absolutely necessary. Usually that means starting from stock and making 100-200mhz jumps and stabilizing at each step. This helps you (1) find the 'sweet spot' of your chip but also identify the point where diminishing returns occur.

Example:

-CPU is able to OC to 4.0ghz at stock
-CPU needs an addtl .15v to hit 4.3ghz
-4.4ghz+ needs .2v

This would tel you that you could go balls-against-the-wall and do 4.4-4.5ghz, but at a LOT higher cooling. You could also do 4ghz and keep the voltage stock, or go in the middle with moderate voltage adjustments.

You need to spend the time and chart the speeds and voltage and that should help you decide. You might need a few days of adjusting, validating and tweaking to get there. The alternative is an unstable machine with issues that could creep in down the road.

Unfortunately I don't have any specific advice for the 5820k yet, but I should in about a month.

 

[GLeeM]

Drop multi to 45..

Yes and by doing this you may also be able to lower the OC voltage.

You can lower temps by lowering Mghz or lowering voltage or lowering both.

 

[RaistlinZ]

Why is the CPU Cache ratio so low? Shouldn't it be run at the same speed as the CPU clock, ideally?

 

[biostud]

Why is the CPU Cache ratio so low? Shouldn't it be run at the same speed as the CPU clock, ideally?

From what I've read, cache ratio is not that important. I run mine @ x35.

 

[BonzaiDuck]

I've dropped LinX with AVX as a stress test because it reaches +95C with H100i and get a BSOD when I run it. So my system is not LinX stable, but it's "stable" when I use it for normal computing and gaming, and that's fine enough for me. I don't do anything mission critical on this machine so if I run into 1-2 crashes due to oc/year that's not a problem for me. Under "normal" stress load like ASUS realbench it reaches around 75C.

I won't say that this is the best way to do oc, but it works for me. When I continue my lowering of voltages and if I experience crashes I will probably use ASUS realbench to test the CPU, or simply add 0.025V and be done with it.

This seems to be an emerging issue for stress-testing current-gen processors and overclocking them.

I've discovered that OCCT's "CPU" test generates temperatures about 5+C lower than IBT or LinX -- or even its own embedded "CPU: Linpack" test option. So far, the only remedy I contemplate for this is to use radiators with bigger cooling capacity than a single H110 AiO, and I'm still trying to match a choice of cases and radiators to get what I want.

The author of OCCT observes that a test can fail because it detects problems with the over clock settings, or merely from thermal causes. And I think our colleagues here have confirmed that a drastic lowering of CPU temperatures (drastic in the eye of the beholder) will allow stable clocks at slightly lower volts. With the 22nm lithography, there is more electrical leakage at higher temperatures, meaning more power consumption and (therefore) less efficient power consumption.

 

[Xerin7]

I've dropped LinX with AVX as a stress test because it reaches +95C with H100i and get a BSOD when I run it. So my system is not LinX stable, but it's "stable" when I use it for normal computing and gaming, and that's fine enough for me.

As a newb to overclocking, this is the question I'm struggling with. I'm finding that I can run stable at much higher clocks for my regular pc use, but if I stress test my pc in Prime95 the temps get very high.

Is it better to run at the highest speed possible as long as your pc doesn't crash very often if at all? Or is it better to go with a more moderate overclock that will keep your cpu cooler and theoretically give it a longer life span? Does running your cpu at the highest clock possible really matter outside of benchmarks, especially as a 100% gamer where most of my fps is determined by gpu performance?

Even at 4.0 (1.168V), Prime95 brought my cpu up to 90C after 10 mins. The sweet spot for me may be 3.9 (1.136V), where I can keep the cpu at 72-75C even after 25 mins of Prime95.

 

[biostud]

As a newb to overclocking, this is the question I'm struggling with. I'm finding that I can run stable at much higher clocks for my regular pc use, but if I stress test my pc in Prime95 the temps get very high.



Is it better to run at the highest speed possible as long as your pc doesn't crash very often if at all? Or is it better to go with a more moderate overclock that will keep your cpu cooler and theoretically give it a longer life span? Does running your cpu at the highest clock possible really matter outside of benchmarks, especially as a 100% gamer where most of my fps is determined by gpu performance?



Even at 4.0 (1.168V), Prime95 brought my cpu up to 90C after 10 mins. The sweet spot for me may be 3.9 (1.136V), where I can keep the cpu at 72-75C even after 25 mins of Prime95.

Basically you should go with what is ok for you. If you want most stable you should probably be able to run prime95/linx stable. If you want stable but not necessarily prime95/linx stable use occt/asus realbench to test for stability. How much difference there is is really hard to measure. But if you find your computer crashing more than you are willing to accept, then turn the oc down a notch.

 

[biostud]

This seems to be an emerging issue for stress-testing current-gen processors and overclocking them.

I've discovered that OCCT's "CPU" test generates temperatures about 5+C lower than IBT or LinX -- or even its own embedded "CPU: Linpack" test option. So far, the only remedy I contemplate for this is to use radiators with bigger cooling capacity than a single H110 AiO, and I'm still trying to match a choice of cases and radiators to get what I want.

The author of OCCT observes that a test can fail because it detects problems with the over clock settings, or merely from thermal causes. And I think our colleagues here have confirmed that a drastic lowering of CPU temperatures (drastic in the eye of the beholder) will allow stable clocks at slightly lower volts. With the 22nm lithography, there is more electrical leakage at higher temperatures, meaning more power consumption and (therefore) less efficient power consumption.

Since the high temperatures are reached within minutes I don't think the radiator is the problem in the cooling setup, but rather how fast the heat can be transferred from the cpu, since the cooling liquid is still way cooler than the cpu.

I'm no expert in this though.

 

[Xerin7]

Thanks all

 

[BonzaiDuck]

Since the high temperatures are reached within minutes I don't think the radiator is the problem in the cooling setup, but rather how fast the heat can be transferred from the cpu, since the cooling liquid is still way cooler than the cpu.

I'm no expert in this though.

I'm still mystified by this. I REALLY (really!) need to find all the articles I've seen so far with 5960X, 5930K and 5820K tests. I also vaguely recall that some of these tests were run with milder software like AIDA-64's stress tests. Even OCCT would be milder than the old stand-bys.

It was also likely that I saw the tests run with an H110 cooler, but this wouldn't eclipse your H100 by that much. But I never saw temperatures in those tests reaching that level.

Was there ever any issue with the pump on these AiO coolers configured as PWM? I don't even see how that would be a problem, unless the pump isn't functioning properly.

As for "expertise." EVERY-body here is more likely to have more experience than I do with water-cooling. I'm hoping to "read myself" up to speed. My first thought about it: I've got my ducted Noctua D14 air-cooler keeping the temperatures even with another member's H110 using the same stress test and approximately the same voltage. The respective i7 processors are different, but both SB-K.

So I'm guessing I want 1.5 to 2x the radiator cooling capacity of an AiO like the H110. And like I said about your temperatures, I'm still a bit mystified by it all.

 

[Satn]

From what I've read, cache ratio is not that important. I run mine @ x35.

hello mister i see same post where you help same ppl to oc the new processor 5820k i bought this cpu last day and i want to know more about the oc, i'm running on asus extreme v, and corsair lpx 2666 can you help me? want talk a bit about the oc?

 

[Satn]

anyone can post same about the test the test and bios... make a guide all togheter for this wonderfull cpu!