How to use Prime95 properly?

[j03h4gLund]

So, I am looking to do a bit of overclocking with my new build(see sig) and have downloaded prime95, cpu-z & coretemp and am ready to start testing.(I think)

What is the "danger" temperature that will start causing damage to the hardware? I ran prime95 without changing anything and realized the core temperatures reached 95+ degree's so I close it down and watched the temperatures plummet down to <30 degrees at idle.

What are some tips I can get to help me test this? I've never used Prime95. Should I overclock things, then run prime95 or should i run prime95 first at stock, then increase small increments while testing prime95? How long should I let prime95 run at and at which settings? Will my system detect "danger" temperatures and protect by shutting down?

I really dont wanna fry my new rig or deal with RMA'ing. Any tips are appreciated - thanks!

 

[Kenmitch]

What is the "danger" temperature that will start causing damage to the hardware? I ran prime95 without changing anything and realized the core temperatures reached 95+ degree's so I close it down and watched the temperatures plummet down to <30 degrees at idle.

Are you shure you got the stock cooler installed correctly? The stock cooler sucks but running at stock speed I wouldn't think your temps would be that high. You are talking in Celcius for the temps? You might wanna take a look and make sure it's properly installed if so.

If your viewing cpu-z and you start prime what speed is your cpu running at? Seems like either your cooler isn't correctly installed or possibly your already somewhat overclocked.

 

[dma0991]

1. I don't trust CoreTemp. Get RealTemp or HWMonitor.
2. Refrain from doing any overclocking without an aftermarket heatsink. If you can afford for that rig, you can definitely chip in a little for a CM Hyper212 Evo.

 

[janas19]

RealTemp is basically the BIOS temo monitor run in a window...

Is that correct?

 

[Maximilian]

1. I don't trust CoreTemp. Get RealTemp or HWMonitor.
2. Refrain from doing any overclocking without an aftermarket heatsink. If you can afford for that rig, you can definitely chip in a little for a CM Hyper212 Evo.

Whats wrong with core temp?

 

[dma0991]

Whats wrong with core temp?

I used to get readings that are 15C higher than it should be. It was a long time ago and they might have fixed the issue but I still doubt it's readings. I don't quite trust Speedfan either. I might consider trying them out again to see if they have improved somehow. Till that happens, I'll stick to RealTemp or HWMonitor.

 

[Towermax]

I've been using Speedfan, Core Temp, and HW Monitor for several years on various systems.

They always read the same for me and it's always the same as the BIOS readings . . .

YMMV

 

[Maximilian]

I used to get readings that are 15C higher than it should be. It was a long time ago and they might have fixed the issue but I still doubt it's readings. I don't quite trust Speedfan either. I might consider trying them out again to see if they have improved somehow. Till that happens, I'll stick to RealTemp or HWMonitor.

Pretty sure they've changed it, ive tried realtemp and core temp, they both spit out the same readings for my 2500k, same with my old i7 as well. Speedfan on the other hand was always 10 degrees cooler.

 

[Charlie98]

1. I don't trust CoreTemp. Get RealTemp or HWMonitor.
2. Refrain from doing any overclocking without an aftermarket heatsink. If you can afford for that rig, you can definitely chip in a little for a CM Hyper212 Evo.

I use RealTemp as well, (never used CoreTemp, so I don't know about that.) I would seriously look at an aftermarket CPU cooler, I'm running a 212+ with a push/pull fan setup and was hitting the mid 80C's with stock (auto) vCore at 4.2GHz, testing with LinX.

I actually use a combination of utilities when testing... RealTemp, CPUz (for frequency activity,) and the Gigabyte EasyTune (for voltages.) RealTemp also gives VID and wattage.

 

[Ben90]

1. I don't trust CoreTemp. Get RealTemp or HWMonitor.

I'm not exactly sure how you can say Core Temp and Real Temp give different temperature values and that faith is better put into other programs. Last time I checked, they were the only two programs that let you see the Distance to TJMax directly. You just simply can't get more accurate than 100%.

RealTemp is basically the BIOS temo monitor run in a window...

Is that correct?

Temperature monitoring utilities all basically work the same way. Just add the value reported by PECI with the TJMax. Good programs let you offset the TJMax number since it is subject to change. Its easier in my opinion to just use the PECI value since there is no guesswork.

 

[dma0991]

I'm not exactly sure how you can say Core Temp and Real Temp give different temperature values and that faith is better put into other programs. Last time I checked, they were the only two programs that let you see the Distance to TJMax directly. You just simply can't get more accurate than 100%.

I tried several temp monitoring software simultaneously(CoreTemp, RealTemp, Everest, HWMonitor, SpeedFan) and if a bigger majority gives me the same result then the minority must not be accurate. RealTemp was inaccurate for my case as it was reading my TJmax inaccurately. Its TJmax reading somehow did not match with the actual TJmax of my CPU.

That may not be true for all especially with the possibility that CoreTemp has been fixed since then. I will consider rerunning the test again if I have the time to see if CoreTemp has improved.

 

[Ben90]

I tried several temp monitoring software simultaneously(CoreTemp, RealTemp, Everest, HWMonitor, SpeedFan) and if a bigger majority gives me the same result then the minority must not be accurate.

Impeccable reasoning. Can't argue with that.

 

[dma0991]

Impeccable reasoning. Can't argue with that.

I can't help but feel there is a slight sarcasm somewhere. :hmm: My methods are not 100% accurate and can be flawed but it is the closest I could get to an accurate reading without resorting to putting a temp probe directly on my heatspreader.

 

[Ben90]

Something that is 100% accurate is setting the TJMax in Real/CoreTemp to your processors TJMax.

 

[LagunaX]

I first run linX - a simplified linpack stress test) to see what is the maximum available memory in my system, open linX and click "all" next to memory.
http://www.softpedia.com/get/System/Benchmarks/LinX-benchmark.shtml

Then in Prime95 blend I select "custom" and change the memory to whatever linX says is your maximum available memory to use.

For example, I am running 8gb (2x4gb) on Windows 7 Pro x64 and linX says I have 6.7gb available so I enter than number (6700 mb) under the "custom' option for Prime95. Otherwise Prime95 defaults to 1600mb (1.6gb) and you are not really testing the majority of your ram - I guess it was designed for a 2gb system default.

Of course I run realtemp and cpu-z while Prime95'ing also to see how my voltage and temps are.

EDIT: WTF are you thinking overclocking on the stock Intel cooler?

Get the COOLER MASTER Hyper 212 EVO for $35 shipped from NewEgg and then start overclocking!
http://www.newegg.com/Product/Product.aspx?Item=N82E16835103099

Actually you might be fine at 4ghz on the stock cooler - most will do 4ghz stock volts no changes other than setting for 4ghz.

Just stay under 1.38v load (you can see on cpu-z under stress testing) and under 80c as a general rule for air cooling.

 

[Concillian]

Step 1) Determine Tj_max for your CPU.

Step 2) put that into the appropriate temperature software.

Alternate method:
Use a program that reports 'distance to Tj_max' and don't use an abolute temperature, because all that really matters is distance to Tj_max.

you want max temps to be ~20c lower than Tj_max. Some use 25C lower, and others use 30C lower. Depends on your comfort zone and desired margin. I'd use 30C for Prime95, 20C for LinX / / Intel Burn Test.

Tj_max is when the CPU will begin auto-throttling. You want a bit of margin before that because the sensor isn't necessarily at the hottest point in the silicon.

If you don't care about the absolute temperature, you can just read distance to Tj_max directly. This is the number that really matters, and what the CPU reports. The reason some temperature software is different from other software is that different CPUs have different Tj_max absolute temperatures, and the software has to have an assumption for Tj_max in order to display an absolute temperature. There are many CPUs and the Tj_max is different, so you can see 5, 10, 15 degrees different between software. The only thing that matters is distance to Tj_max and you either need to make SURE you know your Tj_max or you just use distance to Tj_max instead of an absolute temperature.

 

[Seven]

LinX is meh. Maybe it does stress well under load, but how about idle? I've had crashes under idle more than under load IMO.

 

[Concillian]

LinX is meh. Maybe it does stress well under load, but how about idle? I've had crashes under idle more than under load IMO.

Do you have crashes at idle when you pass 6 hours of load stress with linx and don't exceed 20C from Tj?

 

[Ben90]

LinX is meh. Maybe it does stress well under load, but how about idle? I've had crashes under idle more than under load IMO.

This is most likely an effect of LLC or similar.

 

[Seven]

Do you have crashes at idle when you pass 6 hours of load stress with linx and don't exceed 20C from Tj?

Yes, I can pass LinX no problem. However, I gave up passing prime blend. Fixed the idle crashing by increasing the vcore and it's been rock stable for 24/7, but still can't pass blend..

The SB chips are very different from the others. They operate differently and the current software cannot test them properly. It probably took 5 months to get a fully stable rig under offset mode. I would pass prime large FFT, LinX, and would get a BSOD after 10 days for going 24/7 in middle of the night.