Ivy bridge strange benchmark results.

[Hugo Drax]

3770K VS 2600K

Both stock speeds. 2600K on Asus Gene-Z Maximus IV board. 1333 7-7-7-21 XMP memory. Using videocard 6770 ATI

Other on Intel DH77DF board 3770K using HD4000 and Gskill 1600 7-8-8-24 XMP

On the 3770k I get 84.2733 Gflops average (Intel burntest)
On the 2600k I get 93.
Gflops average

But on Sisoft I get 85 Gflops for the 2600K
and I get 93.32 Gflops for the 3770K (expected 9.7% faster performance over 2600K)

Ram performance.
I get 17.8 GB/s for the 2600K
I get 21.09 GB/s for the 3770K

So its odd that the Intel burn test shows lower performance for the 3770K, it turbos to 3690Mhz VS 3500 Mhz on the 2600K so why the lower gflops if it is running at higher turbo and faster memory. But sisoft shows 3770K faster in the Gflops benchmark by close to 10% as expected over 2600K.

Is this because I am using the HD4000 built in GPU?


BTW IB runs hotter for sure 86c with IBT vs 60s on 2600K

 

[ShintaiDK]

Your numbers shows the IB is quite abit faster. I dont see anything strange? And lower performance on the 3770K? Your numbers says its higher.

 

[Hugo Drax]

Your numbers shows the IB is quite abit faster. I dont see anything strange? And lower performance on the 3770K? Your numbers says its higher.

I edited my post for the intelburn results. They 3770k got higher scores in everything except intelburn test. That's what is so strange.

 

[cmdrdredd]

Gflops don't really matter on IBT that much. Test with like PCMark or 3DMark AIDA etc

 

[ShintaiDK]

I edited my post for the intelburn results. They 3770k got higher scores in everything except intelburn test. That's what is so strange.

Sure it runs the same IBT and with the same memory?

My 3570K stock gets over 100Gflops. Mainly due to memory amount.

 

[2is]

I edited my post for the intelburn results. They 3770k got higher scores in everything except intelburn test. That's what is so strange.

I've seen the same phenomonons under IBT. My 2500k even scores higher than my 3770k

 

[Ferzerp]

For one thing, you have to use 4 threads for a quad core HT chip to get the highest it can. Go back and do that and tell us what you get. 1 thread per core is faster than 2 threads per core in linpack.

Also, to the guy who is getting a higher score with a 2500k, that is why.

 

[Lepton87]

The number of threads does not seem like a good explanation, those CPUs are both 4c/8t so I assume IBT was run with the same number of threads. I guess throttling could explain that but you already stated that 3770k does not go higher than 86c. Did you just change the CPU without reinstalling the OS? But I guess 84Gfps is way too high without AVX so that's also not a good explanation.

 

[Ferzerp]

The number of threads does not seem like a good explanation, those CPUs are both 4c/8t so I assume IBT was run with the same number of threads. I guess throttling could explain that but you already stated that 3770k does not go higher than 86c. Did you just change the CPU without reinstalling the OS? But I guess 84Gfps is way too high without AVX so that's also not a good explanation.

Depends. Sometimes I see some odd scores with threads set to "all". Also, if you don't have it set to "Xtreme Stress Mode", any background tasks may make you get much lower scores.

Really, the only way to run it is to make sure as few things as possible are running, use a thread count matching your real core count, and turn on the "Xtreme Stress Mode" to raise the priority.

For example, mine at 4.4Ghz, not doing all those things, and just running it default, only get 93GFLOPs, but if I raise the priority, and set the thread count to 4, I get 115ish.

 

[2is]

For one thing, you have to use 4 threads for a quad core HT chip to get the highest it can. Go back and do that and tell us what you get. 1 thread per core is faster than 2 threads per core in linpack.

Also, to the guy who is getting a higher score with a 2500k, that is why.

That's a negative on that one. I did 4 threads on my 3770k just now and scored about 10gflps less than doing "all" or "8" threads

 

[Ferzerp]

You're doing something else wrong then. Sorry. Your OC may be unstable due to too low voltage and the internal error checking is catching it (which the more loading 4 threads would show, as opposed to using 8), but costing you work done.

What speed and what scores?

 

[inf64]

Maybe thermal throttling?

 

[Idontcare]

That's a negative on that one. I did 4 threads on my 3770k just now and scored about 10gflps less than doing "all" or "8" threads

Something is definitely setup unoptimal on your system.

At the same clockspeed and on the same system (same mobo/ram/vcard/ssd), my 3770k scores higher gflops than my 2600k, and both score higher gflops (and higher temps) when set to use 4 threads instead of 8 threads.

The fact you are seeing higher gflops despite the losses in thread contention when using HT with IBT versus 4 threads tells you that something is really amiss, and thus it is not surprising that it carries over into lower scores than your SB chip.

It must be doing some throttling of some kind. A quick test of this would be to manually set the multiplier for both the IB and the SB to 16x (1.6GHz) where thermals should definitely not be an issue and then see of the GFlops of the IB are better than the SB chip.

 

[Hugo Drax]

Depends. Sometimes I see some odd scores with threads set to "all". Also, if you don't have it set to "Xtreme Stress Mode", any background tasks may make you get much lower scores.

Really, the only way to run it is to make sure as few things as possible are running, use a thread count matching your real core count, and turn on the "Xtreme Stress Mode" to raise the priority.

For example, mine at 4.4Ghz, not doing all those things, and just running it default, only get 93GFLOPs, but if I raise the priority, and set the thread count to 4, I get 115ish.

Looks like that was the problem, the 2600K was set to auto, and the 3770K set to all.

now I got 101Gflops on max. Stock speed.

 

[2is]

Something is definitely setup unoptimal on your system.

At the same clockspeed and on the same system (same mobo/ram/vcard/ssd), my 3770k scores higher gflops than my 2600k, and both score higher gflops (and higher temps) when set to use 4 threads instead of 8 threads.

The fact you are seeing higher gflops despite the losses in thread contention when using HT with IBT versus 4 threads tells you that something is really amiss, and thus it is not surprising that it carries over into lower scores than your SB chip.

It must be doing some throttling of some kind. A quick test of this would be to manually set the multiplier for both the IB and the SB to 16x (1.6GHz) where thermals should definitely not be an issue and then see of the GFlops of the IB are better than the SB chip.

Doubt it. I've seen screenshots of other IB's running ibt and I'm right there with them. All my other benchmarks score right where they should be as well. Temps are well below TJ Max. IBT is essentially the only "benchmark" that the 2500k scores higher in. My IB is on average 17% faster than my 2500k (ran a series of tests a couple days ago) so I don't think there's anything wrong with it.

Come to think of it. I'm running 2 sticks of 1866 on my SB and 4 sticks of 1600 on my IB. I'm not sure how much difference that would make. The difference in IBT performance is about 10% though

Also, I don't see a "xtreme test mode" option. Maybe I'm using an old version?

EDIT: Googled and found the xtreme stress mode option. Gflops went from around 95-97 to 106-107 with peaks up to 109 and under that mode, selecting 4 vs 8 threads DID make a difference. Interesting

 

[Idontcare]

Doubt it. I've seen screenshots of other IB's running ibt and I'm right there with them. All my other benchmarks score right where they should be as well. Temps are well below TJ Max. IBT is essentially the only "benchmark" that the 2500k scores higher in. My IB is on average 17% faster than my 2500k (ran a series of tests a couple days ago) so I don't think there's anything wrong with it.

Come to think of it. I'm running 2 sticks of 1866 on my SB and 4 sticks of 1600 on my IB. I'm not sure how much difference that would make. The difference in IBT performance is about 10% though

If you aren't worried about it then there is no more worrying about it to be had.

It would bother me though, knowing that something is clearly not right in the situation. Its not that the symptom (lower IBT score) is disconcerting, but that you have yet to identify the root-cause and as such you really should have no confidence that same root-cause issue is not going to rear its ugly head in some other computing instance in ways that you might not appreciate.

Some coughs you can safely ignore, some are indications of pneumonia or worse. If you are confident you can safely ignore this anomaly in your IB rig then you are the only person that you need convince.

 

[ShintaiDK]

The last version of IBT is 2.53? Then its around 3 years old.

 

[2is]

If you aren't worried about it then there is no more worrying about it to be had.

It would bother me though, knowing that something is clearly not right in the situation. Its not that the symptom (lower IBT score) is disconcerting, but that you have yet to identify the root-cause and as such you really should have no confidence that same root-cause issue is not going to rear its ugly head in some other computing instance in ways that you might not appreciate.

Some coughs you can safely ignore, some are indications of pneumonia or worse. If you are confident you can safely ignore this anomaly in your IB rig then you are the only person that you need convince.

It was the xtreme option I was missing (see last edit) just tried it once more using more memory and was hitting 112s which is identical to the 2500k at the same IBT settings, same clock speeds and same memory speeds and about 2gflops lower than the 2500k at 1866MHz memory speed. Far better than the 10% difference before. My SB system still running 2 vs 4 sticks of ram thats on my IB which likely allows the IMC to run at more aggressive timings. Both are on Z77 chipsets.

 

[Ferzerp]

Looks like that was the problem, the 2600K was set to auto, and the 3770K set to all.

now I got 101Gflops on max. Stock speed.

:thumbsup: Great. Glad nothing is broken for you.

 

[Ferzerp]

It was the xtreme option I was missing (see last edit) just tried it once more using more memory and was hitting 112s which is identical to the 2500k at the same IBT settings, same clock speeds and same memory speeds and about 2gflops lower than the 2500k at 1866MHz memory speed. Far better than the 10% difference before. My SB system still running 2 vs 4 sticks of ram thats on my IB which likely allows the IMC to run at more aggressive timings. Both are on Z77 chipsets.

IBT is surprisingly more complex to get accurate linpack results than at first glance, eh?

 

[tweakboy]

There is NO noticeable speed difference in real world usage between a 2600k and Ivy Bridge top model. You will not see a difference. Maybe if you time your video rendering Ivy might be a tiny bit faster not enough. going from a 2500k or 2600k to Ivy Bridge Don't expect anything special to happen. You will not have difference....... Your sidegrade is no good. Talk about waitseing money on a CPU. Instead brake the bottle with a SSD... and your set whichever chip you use,,,,,,, same shizz.. unless you time and do useless benchmarks.. gl


Haswell will show big gains, save money for that. its new generation CPU with 6 cores and 12 thread models. ....

 

[2is]

IBT is surprisingly more complex to get accurate linpack results than at first glance, eh?

Apparently so. I was just finishing up a little "mini review" I was doing between my 2500k and 3770k for a friend who's looking to upgrade from his Q6600 and had to re-run some of my tests based on this new information.