Should I "trust" my E8400's temperatures? (and some OC questions)

[stealthycow]

Hey guys, I've been googling around this problem for a while now but I'm still not too sure and most information I've found has been pretty dated, so I figured I'd just make a new thread.


To get started, relevant info I guess, I have the 45nm e8400 and an abit ip-35e mobo with the latest BIOS (had to update to install e8400). All temps are consistent across realtemp, coretemp, speedfan, etc.

I picked up a used E8400 a little while ago on the forums, about a month back. I'm not sure how new it was, I don't think he was actually using it, was selling a few of them, etc...

So anyways, I installed it, put a CM Geminii S heatsink/fan on it, booted up...and was surprised. My old processor was an E6600, and I only had the stock cooler, and my temperatures were around the low 30s idle. I read the E8400s were supposed to be really cool, but my idle temps were showing 38C, hotter than I was hoping to see.

But the temperature doesn't fluctuate at all. The heatsink's fan was too noisy for me, so I changed it out for something quieter, no change. I didn't have any AS5 handy when I got the sink, so I just used the stuff it came with and later replaced it with AS5...no change. I went from 3.0ghz to 3.2ghz...no change.

It does change, however, during stress testing and gaming and the like. I run prime95 and the temperatures get right unstuck and start climbing as I'd expect them to, and these temperatures seem relatively normal to me (maybe a BIT high). What I find weird, though, is, I can be in the middle of a prime95 test, and press "stop", and it'll INSTANTLY go from wherever it was, say 50c, right back down to 38. I'm not too experienced with processors or anything, but...I don't know, losing 20c of heat instantaneously seems a little odd...This happens the other direction, too, as soon as I press the start button temperature goes from 38 to 44 instantly.

So. I guess my point is, what should I do? I have a fancy semi-expensive cooler and a reportedly chilly processor and good airflow and I want to use it, dammit! But I feel like I'm going in blind, hoping that temperatures reported are reliable, at least the stress-test temperatures. If my idle temps are stuck at 38c seemingly no matter what I do, though, can I really "trust" that the temperatures are accurate? If not, is there anything I can do to overclock safely?

 

[BababooeyHTJ]

Both my e8400 and q9650 had at least one stuck sensor that would not drop below a certain temperature. Its pretty common, don't worry about it. Idle temps aren't really that important anyways.

 

[toyota]

I idle about 40 C with my E8500 and as soon as anything uses cpu power at all it will shoot up 4-6 degrees.

 

[stealthycow]

Alright, well, I'm at 3.4ghz now. The temperatures seem to be...I dunno, less sticky now. They're moving up and down a lot easier, though only by 1-2c. Still won't ever get below 38, but whatever I guess. I'm sure it'll be fine, as you guys said.

I'm new to overclocking, by the way, and I'm just a little confused. So I have DDR2-800 RAM right now. The guide says not to let your RAM speed pass your FSB speed, which I actually learned myself the hard way (computer wouldn't boot, had to reset cmos jumper, heart skipped a beat when it happened). But...well, alright, fair enough. So...I gradually upped my FSB to 400mhz, and that makes my RAMs @ 800mhz...so I'm done? That's all I can do? I'm not saying that going from 3.0 to 3.6 ghz isn't a big deal, and I know I can still try to lower vcore and stuff to get lower temps (although my mobo's lowest vcore setting did seem kind of high, I'll have to learn more about this so I understand better) but that's it? That's the whole overclocking process?

I always viewed overclocking at this big daunting task and a couple hours later I'm stress testing at the highest clock I can get to without buying new components? And it's not like I'm about to go throw away money on faster DDR2 memory, no more old technology for me thank you.

Sorry, I feel like I sound ungrateful to the overclock. I'm not, though!

Edit: Oh, and after running a prime95 test from realtemp the highest temperature I'm seeing is 55c. That's good, right? I'm kind of a quiet freak, so my CPU heatsink fan isn't the most powerful, but 55c at 3.6ghz with a prime95 ftt test isn't bad, right?

 

[Deanodarlo]

Forget about temps - those internal core DTS sensors are not accurate until they approach the throttle temp, 100C.

They have huge slope errors at lower temps, 38C may not really be 38C, it can be anywhere between 5-15C off depending on a given chip. Some even have stuck sensors until you get above 50C or even higher.

Bottom line, the DTS sensors are for throttling only, so get the program realtemp and as long as your CPU is 20-25C away from the throttle temp (100C), eg 75-80C, pay no attention to the reading; it is not accurate in anyway! Just a rough indication that your cooler is on ok. Even then, these chips always protect themselves from thermal damage.

38C idle is completely normal reading for that chip by the way. 55C load is nothing, I run chips up to 70-75C prime95 full load for a quieter fan and that's well within Intel specs for max operational (which is 72.4C case/heatspreader/surface temp, so about 85-90 core temp).

 

[stealthycow]

Will keep in mind. I was using realtemp, it was showing the same temperatures as speedfan and coretemp. I know the BIOS temps aren't accurate, I was never looking at those though.

I had a little mishap this morning. I was running prime95 for about 5 hours last night after I got up to 3.6ghz, no errors or anything like that, seemed stable enough.

Shut down my computer, turned it back on when I woke up, wouldn't boot. It showed the abit screen (press DEL to enter bios screen, tab to show POST) but I couldn't press any buttons. This happened earlier when I accidentally set my FSB too high and wasn't paying attention to RAM divider/speed, the default was above 1:1 for some reason and it wouldn't boot, cleared CMOS with jumper cap and everything was all better.

But this time, that didn't help. I cleared CMOS with the jumper cap, and it still wouldn't get past the screen. So I started taking out RAM sticks thinking maybe I messed one up somehow...I tested them all in pretty much every one of my RAM slots, and I got no consistent results. Occasionally it would take 5 minutes to get passed the abit screen and give me some weird errors (CPU has been changed/unworkable, cmos checksum error, etc) and once or twice it actually got to Windows.

Don't know how it fixed itself...I put in all 4 RAM sticks and just hoped it'd work and it did, except my keyboard/mouse wouldn't work...and then it started giving me an error on bootup that no keyboard was connected

I ended up unplugging all the power cables from the motherboard, removed the CMOS battery for 20 minutes and then put it all back together, booted up just fine, keyboard working, etc...

I guess it's working fine now. I'm back where I started, no overclocks and I'm honestly really afraid to do it again. Well, quite honestly I'm afraid to shut down my computer at all...although I have successfully restarted a few times. Does anyone have any idea what caused all this? I mean I assume it's related to overclocking since it all happened right after I OCed.

 

[Deanodarlo]

Sounds scary. I have had boot issues as you said when raising the FSB, but a CMOS clear always got rid of it straight away unless it was the PSU dying.

Going back to my last post, these DTS sensors I'm talking about are sensors in the actual core of the CPU - they INSTANTLY change, as you described, unlike the old external thermistors for older CPU's inside motherboard sockets, which tried to read and estimate surface/heatspreader temp. They (core/dts temps) will read much higher values in general.

3.6Ghz should be fairly comfortable for a E8400, I remember it being easy on a P45 chipset, depending on your motherboard. Shouldn't require too much voltage either, a small bump perhaps. If you are worried about not pushing too hard, I'd stop there. If you want to go higher, do some research as you said you were new. It's fun!

 

[stealthycow]

I have an abit IP35e motherboard.

I actually haven't touched the voltage yet. My motherboard's lowest voltage it allows is the one that is set by default, 1.2250. When I look in programs like CPU-Z/Speedfan I see it fluctuate between something like 1.120 and 1.168, depending on what I'm doing (1.168 is with prime95 going). So I don't really understand what's going on there, although I think it has to do with the SpeedStep option in the BIOS, which I just left enabled. Still don't really understand why/how it's below the minimum voltage I'm allowed to set and it never goes over 1.168. If I turned speedstep off I'm thinking it might go straight to 1.2250 which would appear to be entirely unnecessary based on what I've seen so far.

 

[Deanodarlo]

Good old Abit eh? Are they still in business these days, always loved their boards. I had a quick look and that's a P35, so it should overclock well. There is LOADS of info around the net for P35 overclocking and your CPU.

Yes those voltage fluctuations are normal for EIST and C1E power saving features and VDROP on motherboards differs. I use a P45 on a MSI P45 Platinum, so I'll give you a few pointers but not sure how similar it is to your motherboard:

* You need to set the ram to 1:1 ratio (400FSB = 400Mhz RAM (DDR2 2x400 = 800Mhz). Most 800Mhz RAM will do up to 850Mhz (425FSB) on default voltage, but bump it up to 1.9v (max 2.1v) to be safe if going beyond.

* 266, 333 and 400 FSB have different chipset timings. Make sure you are running on the 400FSB chipset timing or you will run into stability problems - your motherboard probably does that for you automatically.

* The max voltage I'd run for that CPU is 1.35v, Intel states 1.3625v. Default is 1.225v, but the VID of the CPU (varies between E8400's) may allow the motherboard to set it lower. VDROP may result in values lower than what you set, this is normal.

* At FSB above 400, you may need to bump up the Northbridge MCH a small amount depending on how much RAM you run.

* If your happy with 400 and 3.6Ghz, then default voltage for everything may well be enough. I would perhaps bump the northbridge up a tiny amount, especially if you're running 8GB or 4 mem sticks. May not be necessary however.

* Voltages set to Auto on some motherboards will result in them being set higher as you raise the FSB beyond an official speed (ie overclocked) - usually much higher than actually needed.

Just a few pointers, but you will get so much more info just reading the web. I bet there are loads of people who've overclocked that CPU on a P35. Good luck!

 

[Binky]

You should define "latest bios." There were/are a lot of temp reading issues on those newer (at the time) CPU's on the IP35 boards. I think the last bios was version 18, but I could be wrong on the IP35 (non-pro). If I remember correctly, the last official bios may have been lower than 18 and you had to get the 18 version from forums, etc.

 

[john3850]

The early non E0 rev e8400 always had higher temps.

 

[Dadofamunky]

If you stop the stress test and see an immediate drop in temps, that's good - it indicates the cooling is working. Fact is, the Wolfdales almost never ran as cool as most people expected, including myself. You do need high-end air to really get the most out of those chips. The funny thing is, I rebuilt our Wolfdale machine for my wife, with a new stock Hyper 212+ cooler (the same as on my Sandy Bridge) and the temps are the best they've ever been - average 35-36C idle. And yeah, any PC software reporting temps is a crapshoot at best. If all comparable programs report the same values I'd leave it at that.

 

[john3850]

Gave my daughter my E3110 and also bought a new stock Hyper 212+ for it and had to add a second fan to keep temps below 40c.
I never knew the Wolfdales ran hot so thanks for that info.
When I saw everyone using the 212+ on the sandy I believed it was a bad idea for cooling and never realized that the 8400 ran hot cause I always had water setups and never used air.
The 3 8400 I had were all slightly discolored on base side and vcore was never over 1.4v.

 

[BonzaiDuck]

A bit (ha! pun!) scary on the post failure after you mis-set the FSB (or whatever . . )

This should be a fairly easy but satisfying over-clock project for you. Your 3.6 Ghz is a 20% OC -- not extreme at all for that processor (or any in its immediate family).

On the VCore setting, endeavor to install CPU-Z, and compare its reported Vcore under Windows with the value reported by BIOS (the actual value reported in the screen of sensor readings). Compare those to the "set" value -- the value you chose in the VCore setting pop-up usually specified to 0.00001th of a volt.

The maximum "safe" voltage (as stated by one of our colleagues already) is between 1.36V and 1.37. My E8600 is set somewhere around 1.38V, but shows in CPU-Z as about 1.336V. That's probably as high as I want to take it until the day I decide to flirt with destroying a stellar over-clocked system.

On the temperature sensors -- yes -- also as stated by others, the tJunction sensors for each core appear to be "stuck" on a wide, wide sample of these Wolfdales. Not just limited to the E8x00 CPUs, I just installed the last Wolfdale released spring, 2010 in the computer of a family member -- the E6700 (a Wolfie -- to be distinguished from the old Conroe core with the same model designation). It has at least one stuck sensor. My E8600 pegs at 51C at idle!! But Intel had issued a disclaimer, noting that the sensors were not meant to provide accurate idle-temperature values.

A "reasonable" test of the temperatures under your Over-clock load would be the Small-FFT test in PRIME95. Once you have that stable, see what the temps register for 10 iterations of LinPack or IntelBurnTest.

My memory seems a bit shaky, but I think the throttling temperature value for these cores was just above 80C. If LinPack stays under that value, then you should be "good to go." Also, if you want an extra reference for temperatures, you should be able to find some utility (perhaps that which came with the mobo) reporting the tCase temperature -- the old legacy sensor available before they also imbedded a sensor in each processor-core.