Intel Burn Test and Linpack are worthless

[ShawnD1]

I know I already posted some numbers about Prime95 consuming more power than Intel Burn Test (Linpack) in that other thread I made, but I think this one deserves a separate thread. I even made a video to demonstrate this

http://www.youtube.com/watch?v=9rVkcUiNS-w

Prime95 small FFT - 318 watts
Intel Burn Test (Linpack) - 253 watts

I'm sharing this because everyone says the exact opposite. If you have OCCT installed, put your mouse over "CPU:LINPACK" and read the description.

(OCCT's description of Linpack, the library used by Intel Burn Test)
Linpack CPU Test, based on a library provided by Intel. Similar to IntelBurnTest. CAUTION, it is VERY extreme and will make your CPU real hot

Maybe it is extreme, but it's about 65W less extreme than Prime95

 

[Rifter]

most UPS's use pretty much the same circuit as a kill-a-watt to measure power usage and they are not very accurate.

 

[mav451]

Are your real-world loads going to be primarily CPU/RAM (no GPU)? If so then yeah - I suppose this is could be useful to know. But yeah, I dunno about correlating power usage to the intensity (or efficacy) of a stress tool. IBT stresses the CPU differently than say Prime or OCCT.

 

[ShawnD1]

most UPS's use pretty much the same circuit as a kill-a-watt to measure power usage and they are not very accurate.

The only way it could have huge calculation errors is if Linpack was causing a severely lagging power factor and very high harmonics. We know that isn't true because the PSU has active power factor correction. Most power supplies have power factor correction.

The other way we know Prime95 consumes more power is by looking at the CPU temperature. Prime95 small FFT makes the CPU quite a bit hotter than any other test. Large FFT comes in at second place. Blend test is far behind that. Linpack is about the same as Blend.

I'll take a screenshot of the temperature shown in OCCT while running Prime right now. I'll stop Prime, run Intel Burn Test for 10 minutes, then post another screenshot. This should be fun.

edit:
Prime95 just failed on me. This might take a bit longer to do because I need to increase my system voltage and get another Prime95 shot.

 

[Jovec]

most UPS's use pretty much the same circuit as a kill-a-watt to measure power usage and they are not very accurate.

If this is true then I'd imagine they are consistently inaccurate, so they should at least be representative of the difference between the various stressing apps.

 

[ShawnD1]

Alrighty, results are in. Prime95 is 4 degrees hotter on the first core and 4 degrees hotter on the second core. What seems weird is how Prime consuming a lot more power (apparently) is only 4 degrees warmer. Where does the rest of the heat go?

 

[Rubycon]

First of all you should NEVER run one while the other is running.
Second, if Linpack is set to use a large amount of system memory it has to load into memory which can take more than the time you have allowed. Linpack certainly stresses the CPU more than Prime95 or OCCT cpu tests.

Go back and test again.

 

[ShawnD1]

First of all you should NEVER run one while the other is running.
Second, if Linpack is set to use a large amount of system memory it has to load into memory which can take more than the time you have allowed. Linpack certainly stresses the CPU more than Prime95 or OCCT cpu tests.

Go back and test again.

I was running it for about 10 minutes and it still consumes only about 250-260W. I'll try running it on my Phenom 9600 and see if that shows anything different since every component in that computer is completely different.

 

[Rubycon]

Don't run both at the same time!

Note the power consumption when using Prime95. (write it down)
Make sure it's completely shut down (cpu utilization @0).
Then run IBT/Linpack and note power draw when CPU utilization is 100%.

 

[ShawnD1]

Don't run both at the same time!

Note the power consumption when using Prime95. (write it down)
Make sure it's completely shut down (cpu utilization @0).
Then run IBT/Linpack and note power draw when CPU utilization is 100%.

Prime crashed again so I dropped the FSB; the new power consumption amounts won't be the same as the first post. I gave Intel Burn Test a lot of time on this one.

E6600 @ 2800MHz:
Prime95 Small FFT - 307 watts
Intel Burn Test, "maximum" test, "auto" number of threads - 243 watts

Prime still consumes 64W more.


Maybe Intel Burn Test's version of Linpack is broken, so I tried a new test on my Phenom 9600. I'm using OCCT's version of Linpack (max data set) against OCCT "OCCT" with a small data set (resembles small FFT and takes almost no memory) .

Phenom 9600 @ 2600MHz:
OCCT OCCT, small data set - 54.5 to 55 degrees on all cores
OCCT Linpack, maximum data set - 52.5 to 53 degrees on all cores


OCCT OCCT on Phenom 9600:

OCCT Linpack on Phenom 9600:

Linpack is a nice test and what not, but it seems to consistently draw less power and generate less heat.

 

[Rubycon]

I've never seen a system do this before. Sounds like throttling. Also if Prime is crashing your system is not stable! I've had systems that could run Prime95 for weeks error free reboot or blue screen within a minute of Linpack. It produces much more heat - up to 10C per core in some instances.
Linpack on OCCT is broken. Try LinX 0.6.4 and set it to use ALL free memory.

 

[alyarb]

linx +1

 

[andy5174]

I've never seen a system do this before. Sounds like throttling. Also if Prime is crashing your system is not stable! I've had systems that could run Prime95 for weeks error free reboot or blue screen within a minute of Linpack. It produces much more heat - up to 10C per core in some instances.
Linpack on OCCT is broken. Try LinX 0.6.4 and set it to use ALL free memory.

Really?!!! It will save people a lot of time in testing system stability by using LinX if this is true.

 

[alyarb]

his point was that there is no single algorithm that determines universal stability. a synthetic integer load will fill up your pipeline, but when it comes to 21st century x86 there's a whole lot more to the CPU than that. you've got to hit it hard from all sides but it is best to not run tests simultaneously.

 

[Rubycon]

Really?!!! It will save people a lot of time in testing system stability by using LinX if this is true.

Even more interesting as I've had systems that passed every "stability test" out there yet produced bad data when rendering. The overclocks had to be reduced considerably before the output files were clean.

 

[andy5174]

Good to know this. Thanks!

 

[toyota]

intel burn maximum 156 watts
prime95 small fft 145 watts
prime95 large fft 151watts

 

[VirtualLarry]

OCCT's Linpack implementation is broke. I suspect cache alignment issues with the memory allocator.

When I was testing my Q6600 @ 3.6, I found that some runs of OCCT:Linpack would cause temps of nearly 90C, and some would only cause like 65-70C. I could never track down why there was such a difference from different runs. It didn't make logical sense. Unless, of course, certain times the cache were aligned perfectly, so the algorithms ran faster, less waiting for RAM, etc., and thus the CPU got hotter.

So I believe this is what the OP is seeing. In my experience, with a (good) run of OCCT, the temps do get a bit hotter than Prime95.

 

[krose]

I downloaded the Linpack math kernel library from Intel. In the documentation its says that hyperthreading should be disabled. From section 11 of the user's guide:

"Intel Optimized LINPACK Benchmark is threaded to effectively use multiple processors. So, in multi-processor systems, best performance will be obtained with Hyper-Threading technology turned off, which ensures that the operating system assigns threads to physical processors only."

So how do Linx or Intel Burn Test effectively test with hyperthreading enabled?

Edit: I guess that is for best performance, i.e., higher GFLOPS. Doesn't mean you can't use all available threads. But it is interesting that it is meant to be a benchmark, not strictly a stress test.

 

[Ben90]

Really?!!! It will save people a lot of time in testing system stability by using LinX if this is true.

I concur with this, ive primed for hours and hours only to fail IBT within seconds. Ive completely stopped using prime as three passes in IBT basically means your OC is stable, and 20 passes is rock solid stability. My computer crashes IBT always before 3 passes, yet its been on 24/7 and I have yet to have any crashes of anykind.

Data integrety may be jepardized, but since I format quite regularly, it doesn't bother me if my internets cache of porn gets corrupted.

For people looking for maximum data protection, but still need to overclock, you are going to have to look into using multiple test programs to stress every part of every component, and thats still no guarentee.

 

[EarthwormJim]

I downloaded the Linpack math kernel library from Intel. In the documentation its says that hyperthreading should be disabled. From section 11 of the user's guide:

"Intel Optimized LINPACK Benchmark is threaded to effectively use multiple processors. So, in multi-processor systems, best performance will be obtained with Hyper-Threading technology turned off, which ensures that the operating system assigns threads to physical processors only."

So how do Linx or Intel Burn Test effectively test with hyperthreading enabled?

Edit: I guess that is for best performance, i.e., higher GFLOPS. Doesn't mean you can't use all available threads. But it is interesting that it is meant to be a benchmark, not strictly a stress test.

The benchmarking is how you determine whether you are properly stress testing your CPU or not. If you're getting abnormally low GFLOPS, you're not really stressing your CPU.

 

[Gillbot]

Really?!!! It will save people a lot of time in testing system stability by using LinX if this is true.

My system passed prime95 easily, crashed within seconds upon launching linx. It took me a week or more to re-dial in the settings to get my system linx stable.

 

[dguy6789]

I love Linpack. Makes temps and power draw reach new levels. I run it for 12 hours and if it doesn't crash, then I'll keep the current settings. I still run Prime 95 afterwards just out of old habit however.

 

[andy5174]

My system passed prime95 easily, crashed within seconds upon launching linx. It took me a week or more to re-dial in the settings to get my system linx stable.

Did you use the latest Prime95 v25.9? What is the official site of Linx?

 

[MadScientist]

The post is dated 9-08 but the author keeps it updated with his latest LinX version.
http://www.xtremesystems.org/forums/showthread.php?t=201670

On my i5-750 build I run only LinX.