IntelBurnTest vs. SP 2004

[Rob Williams]

This is a test I've been meaning to do for a while, and since it's freezing outside (10°F), I figured today was as good a day as any to put the benchmarking machine to good use. The quandary's simple. Which stress-testing tool is more effective in determining a stable overclock: SP 2004 (or Orthos) or IntelBurnTest. I've been using SP 2004 for quite a while, but after hearing a few good things about IntelBurnTest a couple months ago (thanks Kougar), I decided to make it a regular part of our overclocking regimen.

I've heard in the past that IntelBurnTest heats up the CPU better than SP 2004's Small FFT test, so that's one thing I wanted to examine. As far as I'm concerned, if one stress-tester is able to heat up the CPU better than another, then it's obviously doing a better job. It's pushing the CPU in ways the others aren't, and that seems rather important.

For the test, I used a QX9770 at its stock speeds of 3.20GHz. Because the wind outside was ridiculous, I left the windows closed so that no cool air would invalidate the test results. Before each run, I noted the room temperature, and then boot the machine up and let it sit idle for five minutes. After that point, I started either of the stress-testers I was using for that particular run and then proceeded to sweat to death.

In the graphs, the top-limit is 120°C, and the duration is "361", which is equivalent to 361x5 seconds (~30 minutes). Although the stress was run for an hour with each tool, I had to cut the latter half of the results off in order to have the graphs look nice (they are 700px wide, as is).

Sadly, with SP 2004, the CPU got even hotter after the cut-off, it's peak being 87°C on the CPU as a whole, and 67° for Core #0. What's interesting to note is that there is virtually no fluctuation overall in the temperatures, due to the Small FFT's apparently linear nature. How did IntelBurnTest fare?

I think the results speak for themselves. At its peak, the CPU was pushed hard enough to hit 106°C, and 85°C on Core #0. That is a full 19°C hotter than SP 2004 on the CPU, and 18°C hotter on Core #0. What's interesting, though, is that while the Small FFT test doesn't fluctuate in its design at all, no matter how long it runs (it just runs with higher byte sizes), IntelBurnTest pushes more variability at the CPU, as shown in the above graph.

So, there's the data. Make of it what you will. Personally, I'm going to continue using IntelBurnTest, because there hasn't been another stress-tester that I've found to push a CPU anywhere near as hard as this, and I think anyone who reads my performance-related articles pretty-much knows how much I hate fake "stable" overclocks. Of course, using IntelBurnTest could also be depressing... if I find out that some of my seemingly stable overclocks aren't so stable ;-)

How about some more fun info? I've uploaded the Everest logs used during testing, so you can see every inch of detail, if you so desire. The first file for each stress-tester is the full-blown look, while the second is where all the averages can be found.

SP 2004 Full-Blown Log
SP 2004 Summary
IntelBurnTest Full-Blown Log
IntelBurnTest Summary

Here are the overall averages for the run, which includes the five minutes at the beginning where the PC was left idle:

IntelBurnTest
CPU: 90.56ºC
Core #0: 70.92ºC
Core #1: 70.95ºC
Core #2: 53.80ºC
Core #3: 53.50ºC

SP 2004
CPU: 79.14ºC
Core #0: 58.91ºC
Core #1: 58.96ºC
Core #2: 45.97ºC
Core #3: 45.39ºC

[Kougar]

Ahah, I was wondering when you would mention this program again... I noticed a certain person was given thanks in AgentGod's readme for helping make IntelBurn Core i7 compatible with their new 1.9 verison. So, thank you Rob! I appreciate you helping with his software, especially for whenever I start using a Nehalem system. And thanks for the mention, you're welcome! No point in keeping all the good software tools to myself. Especially as I share your views about those false "stable" overclocks people like to have...

Of course, using IntelBurnTest could also be depressing... if I find out that some of my seemingly stable overclocks aren't so stable ;-)

Heh, all my overclocks were 20-26 hours Prime95 stable, and IntelBurn managed to prove half of them had about 1% of instability, or made a few outright bluescreen. 1.200v @ 3.2GHz was great, 26 hours Prime stable using any of their tests... but it wasn't stable when using IntelBurn.

Just a silly question, but to make sure did ya use a 64bit OS for your testing?

Thanks for the interesting results, I did a very short test myself out of curiosity but yours is significantly more extensive and useful. Nice graphs!

In all fairness, I have balanced out my views a bit after gaining more experience with the program. IntelBurn is not as good at detecting memory instability unfortunately, so I still keep Prime95 around to stress test the memory system and IntelBurn to test the CPU. In my opinion they work great together, because if Prime95 detects an error in a memory test and IntelBurn is error free, then its pretty much guaranteed to be a memory related issue.

[Rob Williams]

Originally Posted by Kougar

I noticed a certain person was given thanks in AgentGod's readme for helping make IntelBurn Core i7 compatible with their new 1.9 verison.

Haha, I didn't even notice that, despite using the latest version for these tests. That was nice of him. Sadly, I haven't been able to test out the 1.9 on i7, due to benchmarking on the Core 2 machine for the past few days, but I should be going back to i7 in the coming days, so I'll give it a good test then.

Originally Posted by Kougar

Heh, all my overclocks were 20-26 hours Prime95 stable, and IntelBurn managed to prove half of them had about 1% of instability, or made a few outright bluescreen. 1.200v @ 3.2GHz was great, 26 hours Prime stable using any of their tests... but it wasn't stable when using IntelBurn.

1.2v at 3.2GHz is rather incredible if you ask me. I'm not even sure the QX9770 here could do that. I agree though... I love knowing that a CPU is 100% stable. It's fine if someone is pleased with their CPU being 99% stable, but to me, that's still not completely stable. But past that, the Small FFT test doesn't offer much variation in the testing, which to me, is its largest fault. You wouldn't determine GPU stability by running the same timedemo over and over... you need variation to make sure it's stable all-around.

As for using a 64-bit OS... yes, definitely. The only thing we ever used 32-bit Vista for was SYSmark 2007, but that's since been updated to work with 64-bit... so 32-bit, no more.

Originally Posted by Kougar

IntelBurn is not as good at detecting memory instability unfortunately, so I still keep Prime95 around to stress test the memory system and IntelBurn to test the CPU. In my opinion they work great together, because if Prime95 detects an error in a memory test and IntelBurn is error free, then its pretty much guaranteed to be a memory related issue.

Oh, I agree completely there. But the Small FFT test isn't supposed to detect memory faults, either, is it? For memory, I tend to stick to both MemTest 2.01 (just for a quick scan before heading into Windows), and then I use MemTest for Windows, which I find to be superb in detecting any faults in memory. Memory overclocking truly bores me though, so I touch it as little as I have to, and usually just deal with it if I'm forced to clock the modules up in order to hit a higher CPU overclock.

[Kougar]

Well, 1.20V @ 3.2GHz on my Q6600 seemed completely stable. In fact I used that overclock for almost a year before IntelBurn even came out. Nothing seemed to indicate the system wasn't fully stable and I loved the low temperatures/power consumption... but then IntelBurn had to go and ruin things. Admittedly, in hindsight I do have to wonder just how much that instability added to the incessent Vista problems I encountered.

You are correct, Small-FFT doesn't do anything for RAM errors but using Blend mode will pick those up. Large-FFT model also seems to work, but I stick to Blend as I believe it spawns a better mix of threads.

Due to the incredible amounts of time involved in OCing then turning memory I avoid it as well since the benefits are so small, but even during CPU-only overclocks I've noticed stock RAM settings can suddenly go unstable. My outdated sig info is a great example... 800MHz 4-4-4-12 RAM rated for 2.10v... if I use four modules and a 400FSB 1:1 ratio to run at stock settings, I discovered it wasn't stable with Blend mode, even though IntelBurn was 20 runs stable. So now I end up feeding the RAM 2.2v and it hasn't given another error since. I'm not sure why because those settings used to be stable at just 2.10v though.

[Rob Williams]

Originally Posted by Kougar

Admittedly, in hindsight I do have to wonder just how much that instability added to the incessent Vista problems I encountered.

That's a good point, but I'd imagine the chances of that playing a role is low. Hard to say though, really.

Originally Posted by Kougar

but even during CPU-only overclocks I've noticed stock RAM settings can suddenly go unstable.

Yup, I've noticed the same exact thing, haha.

[Kougar]

Originally Posted by Rob Williams

That's a good point, but I'd imagine the chances of that playing a role is low. Hard to say though, really.

Well, I still get incessent VIsta problems anyway, so I figured it only played a minor role in them... For example Tuesday I rebuilt the PC outside the case to test air coolers.... changed the PSU, GPU, CPU coolers, and CPU overclock... Halways through testing Vista booted up and said the hardware had changed and that it had deactivated windows. And that I'd have to reactivate in 3 days or else. The software licensing service was also chewing up my hard drive... wasn't the indexing service at all. Later, I had everything reinstalled back into the case with the original hardware, and upon boot it suddenly is "fine" again.

I wish stuff like that would happen regularly to the people that give the stamp of approval to DRM/licensing software. That'd make it all worth it...

Yup, I've noticed the same exact thing, haha.

Interesting to know... glad I'm not the only one?

[On_Wisconsin]

I thought winter techincally started when it dips below zero (fahrenheit)?

[Rob Williams]

Originally Posted by On_Wisconsin

I thought winter techincally started when it dips below zero (fahrenheit)?

I'd assume not, but it would depend on where you live, and what you consider winter. The other night here, it was 7°F, and that, to me, is FREAKING cold. To be honest, I consider anything under 15°C (59°F) too cool, so if it's at least November and it's >10°C outside... it's winter ;-)

[On_Wisconsin]

I didn't notice you were from Atlantic Canada...so I assume it's warmer than further away (or at least applies for WI during winter)

[Merlin]

Originally Posted by On_Wisconsin

I thought winter techincally started when it dips below zero (fahrenheit)?

December 22 when you have the shortest day of the year. Or Solstice. And ends on the Vernal Equinox ( when night and day are the same length )
But, I guess it is when you want it to be, like it is really cold in Autumn, so it might as well be winter.