Question Real temps for an 8700K

[beatle]

I finally upgraded last night from my 3 year old 6600K @ 4.5GHz to an 8700K. The new chip is running well @ 5.0GHz with 1.27v on a Gigabyte Z390 Aorus Pro. I'm still rocking my Thermalright Ultra-120 from October of 2007! Trying to figure out what the real temperatures are though.

Under load:
MSI Afterburner - 93
HWMonitor - 94 (package, highest core)
Gigabyte EasyTune - 78
AIDA64 - 94 (highest core)

Idle the temps are closer, maybe spread by about 10C. Mid 30s and 40s. As the system is (currently) stable while stress testing, I'm inclined to leave it alone. Games are the only thing that really stress the system these days, and even HWMonitor doesn't show the cores getting past the mid 70s in that use case, so this is truly the worst case that my chip should see for a sustained period. Based on the consensus, my guess is that the chip is actually in the low-mid 90s, but Gigabyte's own utility is sort of a wildcard. I'd also expect some instability that I've yet to experience so far, though I have not stress tested for more than an hour at this point.

 

[ehume]

Try Intel's Extreme Tuning Unit (XTU) with monitoring turned on, measuring every 1000ms. Look at "package" temps.

Or Real Temp will give you the max temps, core by core. Or both. I used to use both.

You can use LinX -- it is a Linpack front-end with AVX2, so it stresses the CPU to its max.

 

[JRzoid]

Delided and Liquid Metal about the same stuff..probably 80-85cish at 5ghz.
what my 7700k does with either the Sycthe or the loop/Block.

Point was to stop it going to 90-100c....dow corning paste is badddie stuff. It's like 3-4$ tub grease stuff. And putting that on 300-400$ Cpu's and stuff....

They Cheaped out...Some of the Generic Tim's that are like 5$ aren't any better much either tho...For Example I have Genereic tube of Tensan AS5/MX4 like stuff and the Scythe Paste beat it slightly. But Yeah it will Work.

Why I supported and Support Ryzen Still.

 

[frozentundra123456]

Delided and Liquid Metal about the same stuff..probably 80-85cish at 5ghz.
what my 7700k does with either the Sycthe or the loop/Block.

Point was to stop it going to 90-100c....dow corning paste is badddie stuff. It's like 3-4$ tub grease stuff. And putting that on 300-400$ Cpu's and stuff....

They Cheaped out...Some of the Generic Tim's that are like 5$ aren't any better much either tho...For Example I have Genereic tube of Tensan AS5/MX4 like stuff and the Scythe Paste beat it slightly. But Yeah it will Work.

Why I supported and Support Ryzen Still.

Yea, you certainly dont have to worry about temps at 5ghz!!!

 

[frozentundra123456]

I have an 8700k at 4.7. Under stress test (prime 95), temps can reach in the 80's, but in "real life" i.e. gaming, temps are in the 70s and package power is well under the TDP, so I am happy with it.

 

[JRzoid]

I have an 8700k at 4.7. Under stress test (prime 95), temps can reach in the 80's, but in "real life" i.e. gaming, temps are in the 70s and package power is well under the TDP, so I am happy with it.

Still HOT and not Delidded….and it sounds like you don't have that great of a cooler on it. I stay in the 40s or 60 degree's Celsius ALOT because of the Liquid Metal and Probably a better seat to the bottom of the Lid Both.

It's absolutely worth it. I am thinking about saving the Scythe Fan though. Put on just the cheap 1500rpm Arctic F12. Good Cheap Fan and it's also Quiet.

 

[beatle]

Well LinX and Prime95 both smoked the CPU (not literally). LinX fails after just 7 seconds and Prime95 brought a BSOD in a similar timeframe. I'll see what I can do with an AVX offset.

RealTemp also shows temps in the 90s. These things shoot up quickly! Strange that Gigabyte's own tool is giving a different reading.

 

[beatle]

Well LinX and Prime95 both smoked the CPU (not literally). LinX fails after just 7 seconds and Prime95 brought a BSOD in a similar timeframe. I'll see what I can do with an AVX offset.

RealTemp also shows temps in the 90s. These things shoot up quickly! Strange that Gigabyte's own tool is giving a different reading.

 

[ehume]

Well LinX and Prime95 both smoked the CPU (not literally). LinX fails after just 7 seconds and Prime95 brought a BSOD in a similar timeframe. I'll see what I can do with an AVX offset.

RealTemp also shows temps in the 90s. These things shoot up quickly! Strange that Gigabyte's own tool is giving a different reading.

I found LinX to be the best tool for assessing the stability of an overclock. Your core voltage probably does not support 5 GHz, for example. My i7 8700k needed 1.37v to get that.

If you are running in the 90's with a Vcore of 1.27 then your cooling setup cannot handle the OC. I found, for example, that only heatsinks with 6+ heatpipes would keep my hotrod under 100 in the spikes (AVX has spikes; I graphically showed them in my last reviews) (in the last one I reviewed the NH-D14 wearing new fans). Only the best air cooling would keep the system at 80-81c, and this was with an ambient of 20c.

Motherboard temp readings are infamous. This is why I relied on Real Temp.

BTW -- Core Temp and Real Temp seem to read the same sensors, along with Speedfan, Open Hardware Monitor, etc.

Beware looking at IHS or "package" temps -- they seem to be 10c less than the max core temps.

 

[beatle]

I found LinX to be the best tool for assessing the stability of an overclock. Your core voltage probably does not support 5 GHz, for example. My i7 8700k needed 1.37v to get that.

If you are running in the 90's with a Vcore of 1.27 then your cooling setup cannot handle the OC. I found, for example, that only heatsinks with 6+ heatpipes would keep my hotrod under 100 in the spikes (AVX has spikes; I graphically showed them in my last reviews) (in the last one I reviewed the NH-D14 wearing new fans). Only the best air cooling would keep the system at 80-81c, and this was with an ambient of 20c.

Motherboard temp readings are infamous. This is why I relied on Real Temp.

BTW -- Core Temp and Real Temp seem to read the same sensors, along with Speedfan, Open Hardware Monitor, etc.

Beware looking at IHS or "package" temps -- they seem to be 10c less than the max core temps.

I would definitely agree with you on that now that I've started to get some failures with AVX workloads. I was surprised to see the system pretty "stable" at such a low relative voltage. I started by running @ 1.37v and things were okay. Then I tried dropping voltage to reduce heat and noticed that I got crashes. A second look at real voltage showed I got approximately 0.1v of drop so I added "turbo" LLC. Now that voltage is a lot more stable, I could run 1.27v. I am now at 1.29v and doing a long AIDA64 (non-AVX) test at work.

I may leave it at 5GHz and 1.29v anyway unless I start seeing real world crashes.

Perhaps I could do better with a heatsink made in the past decade, though the Ultra-120 does have 6 heatpipes and a nice Noctua fan. Based on sheer mass, however, it is dwarfed by some of the BeQuiet and Noctua offerings though. We'll see how bad the overclocking bug bites me and whether I end up going that route.

 

[ehume]

I would definitely agree with you on that now that I've started to get some failures with AVX workloads. I was surprised to see the system pretty "stable" at such a low relative voltage. I started by running @ 1.37v and things were okay. Then I tried dropping voltage to reduce heat and noticed that I got crashes. A second look at real voltage showed I got approximately 0.1v of drop so I added "turbo" LLC. Now that voltage is a lot more stable, I could run 1.27v. I am now at 1.29v and doing a long AIDA64 (non-AVX) test at work.

I may leave it at 5GHz and 1.29v anyway unless I start seeing real world crashes.

Perhaps I could do better with a heatsink made in the past decade, though the Ultra-120 does have 6 heatpipes and a nice Noctua fan. Based on sheer mass, however, it is dwarfed by some of the BeQuiet and Noctua offerings though. We'll see how bad the overclocking bug bites me and whether I end up going that route.

Since my goal was to test heatsinks, I did not touch the LLC control. However, using it to get a Vcore of 1.29 supporting an OC of 5 GHz is fabulous. This is also in line with improvements to the 14nm++ process a year in. You should be happy with your setup.

Since I gave my 8700k system to my daughter I scaled back the OC to 4.8 GHz. I was able to support that with a Vcore of 1.24. It should be editing pictures for years.

BTW -- in many ways OCCT is a good overclocking app. But its AVX is not AVX2, and with AVX it runs about 10c cooler than LinX. Since my goal was to test heatsinks I did not use it but it tracks a number of functions simultaneously.

BTW2 -- IIRC the Ultra-12 has a convex contact plate. I would recommend sticking with that, or if you change heatsinks you will get best results if you stick with Thermalright.

 

[beatle]

Well, I don't want to pat myself (or the chip) on the back too hard yet. AIDA64 is all I've been using so far, and it may have already crashed since I started it up this morning. My luck with LinX and the latest version of Prime95 (small FFTs) shows I am clearly not "stable." Still, it might be good enough for the girls I go out with. I wanted to get to the "magical" 5Ghz, but I might only have 4.9. The chip was, however, initially stable with a 200mhz AVX offset, at least for a few minutes. I am not sure how much AVX is actually used in day to day applications other than encoding, but I hear the instruction set's use is growing, even in games.

 

[ehume]

I do not generally recommend offsets until you have settled what your CPU and cooling can manage.

A couple of minor matters:

I recommend LinX 0.6.5, which has AVX2.

AVX and AVX2 hardly matter for day-to-day use, but Linpack with AVX2 (LinX 0.6.5) maximally stresses the CPU. If you can get it stable with that, you will never see the system fail.

 

[beatle]

Well, though it didn't crash, I am noticing very brief throttling with AIDA64. HWMonitor reports that a few cores have hit @ 100C and the others are not far behind. I tried small FFTs with the latest version of Prime95 and BSOD came right away. LinX 0.6.5 ran a bit longer before erroring (1m 42s) but did not BSOD the machine. Still have a ways to go though. This is still @ 1.29v with LLC.

Looks like I should probably look for a way to cool this thing better. Though seemingly popular and relatively inexpensive for some of the potential benefits, I don't think delidding is in the cards for me. I'm using MX-4 as my TIM which seems okay, but honestly didn't perform any better on my 6600K than the old Ceramique that I was using.

I might need to finally retire my $40 Ultra-120. The Noctua NH-D15S looks promising and flexible enough to accommodate most any setup. I have a Fractal R5 BTW. I liked that I was able to keep using my Thermalright on a new socket just by buying a $10 bracket. I'd like to think Noctua would provide similar functionality? I could justify the cost if I consider it will last me a while. Not really interested in water.

 

[ehume]

The problem comes with your use of a Thermalright heatsink. They are known for their convex contact surfaces, and your CPU has already felt the imprint. One approach is to look at reviews and get the best TR you can find. I like Noctua heatsinks, in part because they are only a little convex. But since your IHS has been pressured (dented?) this is not an option for you . . . unless you delid.

With delidding, you can avoid having to get a new heatsink and keep your OC at 5GHz. You might have to increase your Vcore, but you should be able to keep it under 1.4v. I haven't done any delidding myself, but the reports on it range from good to very good.

Recall that the i7 8700k was in the last generation of unsoldered CPU's. Thus, it is the poster child for delidding. Get a tool, though. Slow beats fast.

And if you delid, you should be able to get that Noctua. I prefer the NH-D15 to the NH-D15S, but that's me. They're basically the same heatsink. If a friend has an old NH-D14, I'd say to buy new fans and go with that.

 

[JRzoid]

https://www.amazon.com/Mugen-Rev-CP...qid=1549322935&sr=8-1&keywords=scythe+mugen+5

Buy the Sycthe and save a few bucks. What I have...Yeah I can do 5ghz at 1.34v on the 7700k. Delidded. There's room to spare it hit like 80c. If you would buy the delid tool and do it, you would get good temps it is a 6 heat pipe cooler.

Like what CoolerMaster use to do put that Block in the middle to Help.

But I have not much point and just run the mild overclock at 4.7 and 1.24v on the EVGA.

 

[scannall]

Give yourself a little headroom. Back it down to 4.7 or 4.8. You wont notice a difference in actually using it, only in benchmarks. And those lose their shine fairly fast.

 

[beatle]

That's interesting (and unfortunate) about the TR heatsink. They were one of the defacto standards back in the day.

My attempt to prove people wrong by simply running a real world app (handbrake) and testing stability failed. Temps were in the 90s and then BSOD'd at some point while I was running an errand. I'll try 4.9 @ 1.29v and see what happens.

This happens almost every time I buy a new chip. I push to what I feel is an appropriate overclock and then end up having to back down or live with instability. I think I'm smart enough for the former. I suppose I could always throw more money at the problem to fix it (and still might) but the point of diminishing returns is on a pretty gradual slope at this point... but how are those cheap delidding tools on ebay? It seems like a simple enough tool.

 

[ehume]

That's interesting (and unfortunate) about the TR heatsink. They were one of the defacto standards back in the day.

They were and still are a de facto standard. Lots of people use and love TR heatsinks. But most users put their heatsinks down once, where I was comparing multiple heatsinks.

What occurs is that once you have put a really convex heatsink down, you should continue to use that heatsink, or another convex one, unless you change your system. By the time I was done with my 4790k system, the NH-D15 was 7-8c hotter than it had been when the system was new. Then I changed to an 8700k. At that point I was very careful about what heatsinks I allowed on it. A yer after I put the system in service, the D15 was still cooling the sane as it had when I started.

Bottom line: unless you delid, go with your TRUE, or get another TR heatsink.

On another subject: the suggestion you got about backing off your OC was good advice. In my own case, I needed to goose the Vcore to 1.37 to get the 8700k to run at 5 GHz. But before I gave it to my daughter, I reduced the OC to 4.8 GHz. The same chip that needed 1.37 volts to hit 5.0 GHz only needed 1.24v to run stably at 4.8.

 

[beatle]

I transcoded a 4K MKV last night at 4.9v without incident. Took about 3 hours. I'm trying LinX now and it seems pretty good (for all of 5 minutes).

I may leave the system as is for now and see if it's stable in daily use. I thought about delidding last night, and the lack of a warranty isn't what bugs me - it's the resale value of the chip. I'm sure some people will still buy a delidded CPU, but the value will surely be depressed, and the potential gains aren't really worth it. Games don't tax CPUs nearly as much as they did back in the day, and I rarely transcode. Heck, I may even try a 100mhz AVX offset and see how things are after a week and see how things are back at 5.0GHz.

 

[mopardude87]

This thread really opened up my eyes to how bad off my motherboard overvolted my i7 8700 non k,setting up a 4.3Ghz all cores turbo boost on auto voltage was putting the chip at 1.39v! With my CM212 i was hitting 85 cel in Prime95 and it was only a few seconds before it hit that but it stayed about those temps.I do use AS5 as TIM which perhaps is a bit dated but have been using it going back to 2007.Before i figured out the voltage issue i thought i had seated the heatsink wrong.My build is in a wind tunnel known as a Antec 900.

Not sure were 2 front 120mm fans and a rear 120mm in a Fractal Meshify c mini will put me when i get that case today.

 

[ehume]

The CM212 is not adequate cooling for either the 8700k or the 8086k. Also, you might try absolute Vcores, instead of +x Vcores. Essentially all MB's use "adaptive voltage" when supplying +x to Vcore.

As for cases, the one you picked should be adequate. If you go air, get a heatsink with at least 6 heatpipes. All of the heatsinks that kept a lid on my non-delidded 8700k at 5 GHz (and they barely did) had 6 heatpipes.

 

[mopardude87]

The CM212 is not adequate cooling for either the 8700k or the 8086k.

For a non k series 8700 its been fine. In a toasty room i have hit only 77cel under Prime 95 at 1.2195v. Not sure if its even worth adding in a second fan for a push/pull config.

 

[beatle]

A little update: I've been cruising at 5 GHz since Feb 6th. I had one crash in the Division (probably a dozen hours played) so I bumped the voltage to 1.295v and it has been great since then. Since games do not really stress the CPU that much compared to encoding, temps are not bad, hovering in the low to mid 80s when gaming. This is good enough for my use case. I experimented with the AVX offset, but I found that it was constantly on for some reason, even when not actively encoding something. I'm guessing there is some application I'm running, perhaps in the background, that uses AVX, and it is constantly kicking in the offset, so I disabled it.