Are 5.0GHz+ overclocks on Coffee Lake really typical?

[IllogicalGlory]

If 4.6ghz gets 80fps then 5.0 might get 86fps assuming perfect scaling with clock speed. Not really a big dead IMO but still. People want those chips for the YUGE CLOCKS.

It isn't that big a deal, no, but I'm not so much concerned about games that perform well and scale with cores. I'm more concerned about games that have really repulsive performance profiles and about emulation. For example, the PC version of Final Fantasy XIII (which I do enjoy, along with games like Skyrim and Oblivion that are single-thread bound) won't operate at frame rates other than 60 or 30, and any CPU headroom that allows me to stay at 60 FPS is going to make a substantial difference. Emulated games are similar.

One the chief reasons I wanted this CPU was for its single core performance advantage over the 4790K; if I can't even get it to the same clock speed, I'm not going to be all that happy. Still, it's very good for games like BF1 even at 4.6GHz, and is an improvement over a 4.7GHz 4790K in single core, just not as much as I'd hope.

 

[Excessi0n]

The obvious answer is that Silicon Lottery's "rigorous testing" probably involves getting the thing to post and that's about it. That would make sense, because it would take a lot of time and labor to actually test 1000s of CPUs with prime95, superPi, etc.

Your obvious answer is wrong, because they do use Prime95 and Linpack for testing. Binning is time consuming but not actually difficult if you have a bunch of testing rigs set up for that purpose.

 

[Kenmitch]

Your obvious answer is wrong, because they do use Prime95 and Linpack for testing. Binning is time consuming but not actually difficult if you have a bunch of testing rigs set up for that purpose.

It's probably not that hard in the end. I'd imagine they for the most part exhibit a similr power profile in the end. Probably just a matter of fine tuning the voltages is the only thing that takes up some time.

 

[IllogicalGlory]

After delidding and applying liquid metal (about a week ago), I seem to be able to hold 4.9GHz stable at about 1.35-1.38V; I haven't noticed any of the old symptoms yet, however 5GHz BSODs in OCCT at the same voltage. I'm happy with the improvement, just wish I could have gotten these results out of the box.

 

[krumme]

After delidding and applying liquid metal (about a week ago), I seem to be able to hold 4.9GHz stable at about 1.35-1.38V; I haven't noticed any of the old symptoms yet, however 5GHz BSODs in OCCT at the same voltage. I'm happy with the improvement, just wish I could have gotten these results out of the box.

My take on it is its fine. Good job.
Even if it ends up at 4.8. Lets see.
Those high clocks 8700k we have seen is perhaps to a certain degree the need for this to absolutely break the 5 barrier after the 7700k. Its just a bit overrated and nobody want to tell a bad story.
Anyway. Those old dx9 derived games will slowly die. It just takes a solid handfull of years to get the new engines in. Then you are set to go even for 144 in far more demanding games.

 

[IRobot23]

YES!

i7 8700K is best overclocker out there even silicon lottery said really good numbers above 5GHz. What you need to do is to delid and have good cooler and really good MB.

Silicon lottery: (https://siliconlottery.com/collections/coffeelake/intel-i7-8700k)
As of 3/22/18, the top 99% of tested 8700Ks were able to hit 4.9GHz or greater.
As of 3/22/18, the top 88% of tested 8700Ks were able to hit 5.0GHz or greater.
As of 3/22/18, the top 54% of tested 8700Ks were able to hit 5.1GHz or greater.
As of 3/22/18, the top 22% of tested 8700Ks were able to hit 5.2GHz or greater.
They also had 5,3GHz for ~900$
As of 3/22/18, the top ~3% of tested 8700Ks were able to hit 5.2GHz or greater.

I think on ALL -2 AVX

I found something very interesting for delid CPUs :

Ryzen 2 APU: 7°C to 15°C
Ivy Bridge: 10°C to 25°C
Haswell: 10°C to 25°C
Devil's Canyon: 7°C to 15°C
Broadwell: 8°C to 18°C
Skylake: 8°C to 18°C
Kaby Lake: 12°C to 25°C
Skylake-X: 10°C to 20°C
Kaby Lake-X: 12°C to 25°C
Coffee Lake: 12°C to 25°C

 

[Campy]

After delidding and applying liquid metal (about a week ago), I seem to be able to hold 4.9GHz stable at about 1.35-1.38V; I haven't noticed any of the old symptoms yet, however 5GHz BSODs in OCCT at the same voltage. I'm happy with the improvement, just wish I could have gotten these results out of the box.

OCCT uses AVX instructions. Are you using a negative AVX offset?

 

[IllogicalGlory]

OCCT uses AVX instructions. Are you using a negative AVX offset?

I've found that doing so seems to throttle the CPU across the board, even in CPU-Z, Cinebench and games, so no. It's stable at 4.9GHz with no AVX offset.

 

[StinkyPinky]

I've found that doing so seems to throttle the CPU across the board, even in CPU-Z, Cinebench and games, so no. It's stable at 4.9GHz with no AVX offset.

It shouldn't be doing that. Must be a bug with the bios.

 

[epsilon84]

Depends on a lot of factors, but generally speaking 5GHz+ requires an AIO cooler and if you want 5.1 to 5.2 then delidding is almost mandatory as these chips run hot at 1.4V.

I would say the Silicon Lottery stats are a good reference point, though delidding probably adds an extra 100 - 200MHz of headroom.

 

[Arioch13]

I had terrible trouble with BF1 and my 8700k

Until I delidded it and used some Conductaunaut liquid metal. The temps were terrible in BF1 in particular. I mean, it really really stood out going almost 20 degrees higher than anything else. However, since delidding I have been having a great time with this cpu.

The problems you describe in BF were identical to mine if the cpu got to around 80 degrees and seemed to be triggered by heat spikes. The cpu just couldnt shed heat. I was considering getting rid of the whole platform when I found myself gaming on my guest pc instead. WHen anything was happening, even when loading a new level or end of road in UI the heat would spike for a second or two on some cores.

After delidding I found it a different beast entirely. No wonder. The intel job inside was an abomination. I have pretty elaborate watercooling and the loop has 2 overclocked 1080ti in it. Yet the loop was never changing temp. The cards go 37-40 degrees in BF1. The cpu just couldnt shed heat.

Since delidding, I am getting 5hz with 1.32 with total stability (playing BF1 etc.
I am currently playng BF1 at 5.1 though I can hit it at 5.2 with this cheap when benching (the most I tried so far) but I noticed it seems less than totally happy even at 5.1 My temps are 50-55 but the extra power required for 5.1 or 5.2 is higher than I would think and when the amount you increase is way less than linear, if gaming is your aim. Stop. Step back 100hz and reduce the voltage.

Unless you dont care if you replace it in short order. Or worse, have to deal with intermittent redacted until you figure out that it has become unstable.

On the upside it now stays sensibly cool but I never go above 1.375 set manually. Sometimes I will get some gear and do that and see it as the price of playing about for the weekend but I do that when purely benching all weekend so I already consider it written off. For gaming, the small gains make absolutely no difference to any app or game so its more for the fun of it.

The extra 100 or 200 megahertz beyond the sweet spot are almost always really pushing it. Often damage done can be really annoyingly intermitting and just enough so you are not sure whats up for a while. Dont want to sound patronising or like the old fool I am but be clear what you want. If you want your PC to play games reliably and still be doing it the same way in 6 months, step it back 100 hz an down the voltage slightly. If you're benching of course... go nuts ) Just be clear what your objectives are. Replacing stuff, instability, having to clock memory lower than it should be and the frustration of crashing out of a game because you did just enough cumulative damage for it to be annoying may not be worth it.

I just find the sweet spot and hold it with a good chip. The amount of chips I screwed up testing to destruction isnt funny and you sort of get a feel for when it is happy and when you are pushing your luck.

I think I got lucky with this chip as it does 5 ghz consistently all the time with a relatively low voltage from what I see people using. The delidding worked better than I dared hope but that just speaks for the terrible quality of intels work and the bad choices they made.

Using an Asus Maximus X code motherboard.

No profanity in the tech areas.

AT Mod Usandthem

 

[ehume]

I cannot comment about the results Silicon Lottery gets. However, I routinely overclock my 8700k to 5GHz without an AVX offset. I did find I could not get to 5GHz until my Vcore was 1.37v static (not adaptive and not normal + volts). The system was not as stable as I would like until I dropped my cache frequency to 4.5GHz. When I tried overclocking to 5.1 GHz I was uncomfortable with the voltages needed and the system was never really stable despite the increased Vcore.

Indeed OCCT uses Linpack with AVX, but that runs too cool. So I use LinX 0.6.5, which runs Linpack with AVX2. I routinely push the best heatsinks to their thermal limits on this system.

Based on the stats from Silicon Lottery most 8700k's can OC to 5GHz. If you have to stop short, you have to stop short. If you are rich or you do something to your CPU, after a year Intel will be more experienced on the process and you would likely get a better chip.

 

[Fir]

I thought pretested CPU biz died out with socket 462 and Pentium 4 days?
I'd never trust them. All the so-called pretested CPUs were NEVER truly stable at the so-called guaranteed speeds when I tested them. Looping a game or 3D Bench overnight or running Prime95 isn't really a decent stability test. Perhaps if you're a gamer or benchmark freak. Maybe.

If your system can run 8+ hours of Real Bench it's pretty stable. It uses GPUs as well so it maxes out the power supply.

Most of the folks bragging about 5+ GHz being 100% stable are simply lying. Or they are not actually stable and think they are because it can pass on things they are running. Which is like member of the band that's out of tune but doesn't realize it!

I'm not saying it's impossible to have such high clocks and be stable. It is doable but not with out of the box hardware. And if you have a chip that does, aka the golden egg, congrats! I know they are out there.

 

[JoeRambo]

Without delid thermals are definitely a problem. When delidded, I am certain 99% of chips can do 5Ghz with 1.4V or less. And calling people liers just because they don't torture their systems with 8 hours of RB or latest Linpack ??? There has to be a line drawn somewhere, as when testing for long enough time there is a chance for cosmic rays to flip bits in RAM, should that make everyone buy ECC?

 

[Fir]

Delidding is a the next step. Especially on the 2066 chips. My 7980XE had to be delidded to prevent throttling at highest clocks. It's amazing how much endurance (as far as stability is concerned) these chips have. And yes this could very well be the reason a good thermal bond between die and heat spreader has taken a backseat as if you look at stock performance temps are perfectly fine. But once the multiplier gets ratcheted up along with the necessary core voltage increases, temps certainly skyrocket. Delidding certainly isn't as scary as it used to be with the tools on the market now. As a matter of fact you have a better chance of killing your cpu with a careless application of conductive liquid metal TIM than using one of these tools to delid.

And perhaps lying is a strong word but if someone holds back because, well their system works for them at 5+ GHz but won't pass a series of grueling tests which are fine either at lower clocks or stock, who's lying?

Things will get interesting in the next month or two when the 32 core TRs arrive. Lots of VRMs aren't going to be able to handle the loads presented to them when their owners crank up the heat. Dual 8 pin gonna be a must for those.

 

[Markfw]

Delidding is a the next step. Especially on the 2066 chips. My 7980XE had to be delidded to prevent throttling at highest clocks. It's amazing how much endurance (as far as stability is concerned) these chips have. And yes this could very well be the reason a good thermal bond between die and heat spreader has taken a backseat as if you look at stock performance temps are perfectly fine. But once the multiplier gets ratcheted up along with the necessary core voltage increases, temps certainly skyrocket. Delidding certainly isn't as scary as it used to be with the tools on the market now. As a matter of fact you have a better chance of killing your cpu with a careless application of conductive liquid metal TIM than using one of these tools to delid.

And perhaps lying is a strong word but if someone holds back because, well their system works for them at 5+ GHz but won't pass a series of grueling tests which are fine either at lower clocks or stock, who's lying?

Things will get interesting in the next month or two when the 32 core TRs arrive. Lots of VRMs aren't going to be able to handle the loads presented to them when their owners crank up the heat. Dual 8 pin gonna be a must for those.

Not to derail this thread, just a quick comment about the 32 core TR's since you brought this up.. People who can afford them, most likely already have dual 8-pin X399 boards like my Taichi. And I will be getting a new motherboard for it anyway.

Now back to the 5 ghz discussion

 

[scannall]

5 Ghz on air isn't happening without a golden chip and a solid cooler. For me anyway, delidding and water coolers aren't somewhere I'm willing to go.

 

[ehume]

5 Ghz on air isn't happening without a golden chip and a solid cooler. For me anyway, delidding and water coolers aren't somewhere I'm willing to go.

Solid cooler I'll buy. I just finished three sessions of Linpack with AVX2 at 5GHz. The heatsink was an EOL NH-D14 with two NF-A14 2000 fans. In a 21.5C basement I got an average of 79.7C for the AVX plateaus for a net core temp of slightly more than 58C. The chip has never been delidded and it was gotten at the MC outside Philly. IOW I doubt it is a golden chip.

For a lot of people theiri7 8700k will reach 5GHz with a solid heatsink. You don't have to go delidding or use water cooling.

 

[epsilon84]

5 Ghz on air isn't happening without a golden chip and a solid cooler. For me anyway, delidding and water coolers aren't somewhere I'm willing to go.

My non delidded 8700K runs at 5.0GHz with a Hyper 212. Temps peak at around 80 - 85 degrees under stress testing with OCCT, but more typical workloads (for me anyway) such as gaming don't see temps go much over 60 degrees, which I'm more than comfortable with.

It's definitely not a golden chip either, needs about 1.38V to be fully stable at 5.0, golden samples can do 5.0 below 1.30V

 

[cmdrdredd]

My non delidded 8700K runs at 5.0GHz with a Hyper 212. Temps peak at around 80 - 85 degrees under stress testing with OCCT, but more typical workloads (for me anyway) such as gaming don't see temps go much over 60 degrees, which I'm more than comfortable with.

It's definitely not a golden chip either, needs about 1.38V to be fully stable at 5.0, golden samples can do 5.0 below 1.30V

1.38v is definitely not safe long term. It's your CPU though.

 

[Anubis]

1.38v is definitely not safe long term. It's your CPU though.

i'm running 5 on mine at 1.4V
temps not an issue and 1.4 is fine for 24/7

 

[cmdrdredd]

Everything I’ve read says 1.3 max for 24/7 long term. Whatever you are good with and gives you piece of mind is cool though. Not trying to sway opinion.

 

[epsilon84]

1.38v is definitely not safe long term. It's your CPU though.

Most 8700Ks need around this voltage for 5.0, check out Silicon Lottery... But hey what do I know.

The chip is coming on 9 months old now and no signs of it dieing so far.