IntelBurnTest : A new stress testing utility for your overcooked CPU

[Max_87]

For years, overclockers who really care about stability of their system have been stress testing their overclocked systems using Prime95. Prime95 remains to be the most popular CPU stress testing program, even after OCCT was released, because OCCT doesn't stress the system as hard as Prime95.

So your system is Prime95 stable for 8 hours? 12 hours? or even 24 hours? It should be completely stable and shouldn't produce any errors during processing right? No, not really, at least with IntelBurnTest.

IntelBurnTest:
So what exactly is this IntelBurnTest? It's a new program, but not exactly new Let me explain, IntelBurnTest is actually an easy to use (new) user interface, slapped onto the Linpack software library and Linpack has been around for many years.


And you call this a new program?
It's easy to use, so who cares? Those of you who hate CLI, you have to thank AgentGOD for creating IntelBurnTest.


So what's the big deal with IntelBurnTest? I already have Prime95!
Now, onto the important question: Is IntelBurnTest better than Prime95? The answer is YES. It stresses the system much harder than Prime95, even if your system is Prime95 stable for 24 hours, it still has a high chance to fail in IntelBurnTest. I'm serious. Also, the best thing about IntelBurnTest: you don't need to spend many hours just for stress testing your CPU.

Many users reported that they had to increase their CPU's voltage as much as 0.04v just to get it to pass IntelBurnTest at the same overclocked speed. The load temperature gets even higher than Prime95, usually between 10-20C difference. This heavily depends on CPU, cooling and how much voltage you pump to your CPU though. The program author himself claims that the load temperature is 22C higher than when stress testing with Prime95. For my watercooled E8400, it's roughly 7-10C difference.

Here's my own experience with IntelBurnTest. My E8400 was Prime95 stable for 12 hours at 4.1GHz with 1.29-1.30v. After I learned about IntelBurnTest, I decided to give it a try, expecting that my system should pass IntelBurnTest without any problems.

To my surprise, my E8400 failed instantly in the first test Producing calculation error. I had to increase the voltage to 1.32-1.33v in order to pass IntelBurnTest. One thing worth mentioning, the vdroop is higher with IntelBurnTest, this shows that IntelBurnTest stresses your system even more intense than Prime95. For comparison, my E8400 voltage stays at 1.33-1.34v when running Prime95 and 1.32-1.33v when running IntelBurnTest.


How to use IntelBurnTest?
IntelBurnTest is straight forward and really easy to use. The options are pretty much self-explanatory. First, you select whether to enable the error detection or not. (Doesn't really make much difference, except the program will notify you at the end of the tests when there is an error with calculation if you enable error detection.) Next, you select the stress level, the higher the stress level, the higher amount of available memory will be used. More is better if you want to stress your memory, memory controller as well as CPU at the same time. If you just want to test your CPU, "Half-stress" or "Low-stress" would suffice. Finally, you specify the how many times do you want the test to run. The author recommends 5 or more times. I personally recommend at least 10 to 20 times, it doesn't take much time to complete the stress testing anyway. For my system, it only took about 22 minutes to complete 20 tests.

After the tests are completed, the values under "Residual<norm>" column should be matching. If one of them shows different value, it means there is a calculation error.



32-bit or 64-bit?
According to the author, it is better to run IntelBurnTest under a 64-bit OS as it gives more accurate results. This is most probably because certain parts of the CPU are inactive during 32-bit operations.

Also, under 32-bit Windows, you won't be able to let IntelBurnTest use more than ~2GB memory because Windows (32-bit XP AFAIK) does not allow allocation of more than 2GB memory to a program.

Nevertheless, IntelBurnTest is still a great stress testing program, even if you run it under 32-bit OS. After all, it's still more accurate and much "faster" than Prime95.


Can I Run This on AMD Processors?
Yes! Since version 1.8!


Conclusion:
When a system is stable and doesn't crash, it doesn't necessarily mean the system is not producing any errors. For example, Prime95 detects error with your CPU while your system never crashes during 24/7 operation. In this case, it's 24/7 stable, Prime95 stable, but fail IntelBurnTest.

It's good to see a new stress testing program that stresses the system even harder and detects errors MUCH faster than Prime95. You no longer need to spend hours just to stress test your overclocked CPUs. Just imagine if your system failed in Prime95 right after 12 or 24 hours of testing when you are just about to take a screenshot for bragging purposes or it failed within a few minutes in Prime95 while you leave the system for many hours to do other stuffs I believe all of us experienced this before.

Remember, there is no bulletproof stress testing program, but IntelBurnTest is the best thing for testing your overcooked CPUs now.

So what are you waiting for? Torture your CPUs NOW! I dare you!

[Unixlord]

Let there be a lite edition.

[The_YongGrand]

Wow, I didn't know they use linear algebra to stress-test the CPU.

I'm wondering do they do actual LU-Decomposition over there?

And... I guess using Finite Element Method to stress-test CPU is also a cool idea.

[Chai]

Tested, and passed.

[PsYkHoTiK]

So how long does the test last?

Is it like Prime95 where you just leave it running for however long?

Is it just CPU or does it do ram as well?


Definitely be trying this when I go home...

[Dashken]

Originally Posted by PsYkHoTiK

So how long does the test last?

Is it like Prime95 where you just leave it running for however long?

Is it just CPU or does it do ram as well?


Definitely be trying this when I go home...

From Max's post...

1) For my system, it only took about 22 minutes to complete 20 tests.

2) Oi... For my system, it only took about 22 minutes to complete 20 tests.

3) IntelBurnTest is straight forward and really easy to use. The options are pretty much self-explanatory. First, you select whether to enable the error detection or not. (Doesn't really make much difference, except the program will notify you at the end of the tests when there is an error with calculation if you enable error detection.) Next, you select the stress level, the higher the stress level, the higher amount of available memory will be used. More is better if you want to stress your memory, memory controller as well as CPU at the same time. If you just want to test your CPU, "Half-stress" or "Low-stress" would suffice. Finally, you specify the how many times do you want the test to run. The author recommends 5 or more times. I personally recommend at least 10 to 20 times, it doesn't take much time to complete the stress testing anyway. For my system, it only took about 22 minutes to complete 20 tests.

[Unixlord]

This confirms my theory that a lot of people actually post without properly reading.

[Chai]

I did 100 tests...looks simple.

[Max_87]

Originally Posted by Dashken

From Max's post...

1) For my system, it only took about 22 minutes to complete 20 tests.

2) Oi... For my system, it only took about 22 minutes to complete 20 tests.

3) IntelBurnTest is straight forward and really easy to use. The options are pretty much self-explanatory. First, you select whether to enable the error detection or not. (Doesn't really make much difference, except the program will notify you at the end of the tests when there is an error with calculation if you enable error detection.) Next, you select the stress level, the higher the stress level, the higher amount of available memory will be used. More is better if you want to stress your memory, memory controller as well as CPU at the same time. If you just want to test your CPU, "Half-stress" or "Low-stress" would suffice. Finally, you specify the how many times do you want the test to run. The author recommends 5 or more times. I personally recommend at least 10 to 20 times, it doesn't take much time to complete the stress testing anyway. For my system, it only took about 22 minutes to complete 20 tests.

LMAO!

That's the main advantage of IntelBurnTest, it detects error really fast, usually within the first 5-10 tests. Prime95 needs a few hours, or it doesn't detect it even after 12 hours.

If you have a habit of minimizing vcore as much as possible (when testing with Prime95), you are more likely to get errors in IntelBurnTest.

[Max_87]

More tests on my E8400

At 4.0GHz:
Prime95 12 hours: 1.25V
IntelBurnTest 10 minutes: 1.26V
Voltage increase: +0.01V

At 4.1GHz:
Prime95 12 hours: 1.29V
IntelBurnTest 10 minutes: 1.32V
Voltage increase: +0.03V

Seems like the higher you go, more voltage is needed just to be error free in IntelBurnTest, while in Prime95 it can still be stable for many hours.