lower clocked i5 faster than higher clocked i7 with 280x?

[Mussels]

Heres a conundrum for you guys, while testing my new 280x compared to my 7970, i put it in my i5 system... and the 280x is faster in heaven DX11 on the slower system.


While not the most thorough of tests, one thing did show up consistently - the i5 system is about 10% faster in heaven, getting my score from 2000, to 2200. This makes no sense based on the specs and same hardware.

Heaven was confirmed at the same settings in both PC's, test was re-ran a few times - accidentally deleted the screenshots however.


I7 2600 @ 4Ghz (4.2GHz turbo)
Asrock Z68 Extreme 3 Gen 3
16x PCI-E 2.0 (confirmed via GPU-Z)
16GB DDR3 1600Mhz 9-10-10-30
280X 3GB (1.1GHz GPU, 1500Mhz GDDR5)

1950-2000 in heaven

I5 2400 @ 3.4Ghz (no turbo)
Gigabyte GA-Z68A-D3-B3
16x PCI-E 2.0 (confirmed via GPU-Z)
16GB DDR3 1333Mhz 9-9-9-24
280X 3GB (1.1GHz GPU, 1500Mhz GDDR5)

2200 in heaven.


why is the far inferior system giving me higher scores? should i be switching mobos or something here?

 

[xfia]

http://www.cpu-world.com/Compare/415/Intel_Core_i5_i5-2400_vs_Intel_Core_i7_i7-2600.html

the systems are pretty much the same but i5's have no hyperthreading. the i5 is a little stronger while using 1 core but the i7 takes over in threaded apps. sometimes people disable hyperthreading or cores to get a higher clock if they dont need more threads.
the gpu isnt at max rendering output on 1 core.. so heaven is really only good as a bottleneck reference not vindictive of aaa games and newer programing methods.

 

[FordGT90Concept]

Maybe the overclock isn't stable.

 

[Aquinus]

Are you using the same GPU drivers and have you set your power settings to "Performance"?

Also is this the same memory? I'm seeing 1333 on the i5 and 1600 on the i7. I think the tRAS is a little high for 1600 memory, are you using BIOS defaults?

Lastly: How hard are you pushing the BCLK? Your specs say you're doing 1680 which means you're hitting the bus which can negatively impact PCI-E performance if it goes too far from 100Mhz, 5Mhz or more is "the danger zone."

Edit: If I did my math right, you're pushing the bclk to 105Mhz which is the low end of where hardware starts to not be very happy with the overclock on the PCI-E and DMA buses. I recommend going back to 100Mhz and working just with boost to see if it makes a difference. Increasing VTT, VCCSA, or the PCH voltage may get you stable, but it's not something you want to mess with too much IMHO.

Edit: Edit: My experience is from my own overclocking of the bclk on my 3820. It doesn't really like to go 5Mhz over the current BCLK strap. In fact I usually didn't do much more than 3Mhz for the sake of stability (which is important because the PCH handles my RAID.)

 

[xfia]

isnt @Mussels pro like 6 970's strapped together.. trying to go to the moon on gpu power lol

the overclock is fine its just the overclock cant completely sway the single thread performance.

 

[MrGenius]

I'd really like to know how you can score that high with a 280X. As in, besides DX11, what is everything else set to? CCC and benchmark settings if you would please. And driver version too I suppose.

I'd like to compare to my 280X with identical clocks on a system MUCH slower than either of yours. Maybe I'll find something of interest. Specifically, just how GPU bound is Heaven anyway?

 

[Ferrum Master]

Usually it is the driver... remember to disable core parking... it causes stupid things in benchmarks.

btw what are the settings... I have very similar system to yours... my 7970 uses a 280X bios now(it has higher voltage )

For the lulz...

Disable that CG miner

 

[SK-1]

Maybe the overclock isn't stable.

you ninja

 

[Kickus_assius]

xfia as per your own link the I72600 is 12% faster in single-thread tasks, since on stock settings it turbos up to 3.8ghz. Clock for clock for single core they are both the same Sandy Bridge core, so it should be the same performance. When overclocked above 3.4, the I7 should start to pull away. I would guess the lower score is from driver inconsistencies or possibly the overclock is not 100% stable. If the benchmark is more gpu-bound, then the same score would be achieved by both systems, otherwise the I7 score would always be higher (clocked at 4.2GHz vs. 3.4Ghz in the I5).

 

[m&m's]

Is your BCLK running at 100? I'm pretty sure it will solve the conundrum.

 

[xfia]

xfia as per your own link the I72600 is 12% faster in single-thread tasks, since on stock settings it turbos up to 3.8ghz. Clock for clock for single core they are both the same Sandy Bridge core, so it should be the same performance. When overclocked above 3.4, the I7 should start to pull away. I would guess the lower score is from driver inconsistencies or possibly the overclock is not 100% stable. If the benchmark is more gpu-bound, then the same score would be achieved by both systems, otherwise the I7 score would always be higher (clocked at 4.2GHz vs. 3.4Ghz in the I5).

haha i seen the no advantage part and got the colors mixed up at the bottom.
well first of all the mobo's are 3.0..
possibly throttling and the turbo isnt going to kick on at a certain temp.
dont really how its the driver because its the same gpu on the same chipset.
this is just some kind of a test.. i guess mussels wants to see how many problems go misdiagnosed. to really get to the heart of it with no doubt there is several diagnostic tools that need to be run and everything needs to be reporting right.
from what people say there has been games that hate hyperthreading and you get better performance with it off. your honestly better off going off using the same settings as a game benchmark if your just gauging performance and not trying to stress. could just speculate all day but when you dont have all the information you need thats what happens.

 

[m0nt3]

Tried disabling hyper-threading on the i7?

 

[XSI]

as said: hyperthreading, drivers, unstable oc,
maybe some virus/malware or process in the background taking a performance tall.

 

[Mussels]

as said: hyperthreading, drivers, unstable oc,
maybe some virus/malware or process in the background taking a performance tall.

background process is possible, but with the amount of CPU power an i7 has i cant see whats running that would cause it.

http://www.cpu-world.com/Compare/415/Intel_Core_i5_i5-2400_vs_Intel_Core_i7_i7-2600.html

the systems are pretty much the same but i5's have no hyperthreading. the i5 is a little stronger while using 1 core but the i7 takes over in threaded apps. sometimes people disable hyperthreading or cores to get a higher clock if they dont need more threads.
the gpu isnt at max rendering output on 1 core.. so heaven is really only good as a bottleneck reference not vindictive of aaa games and newer programing methods.

you ignored the overclock on the i7, which makes the speed comparison heavily favour the i7.

Maybe the overclock isn't stable.

completely stable - its been crash free for a long time at 103 Bclk with these settings, and i lowered it to 100 when testing to rule that out.

Are you using the same GPU drivers and have you set your power settings to "Performance"?

Also is this the same memory? I'm seeing 1333 on the i5 and 1600 on the i7. I think the tRAS is a little high for 1600 memory, are you using BIOS defaults?

Lastly: How hard are you pushing the BCLK? Your specs say you're doing 1680 which means you're hitting the bus which can negatively impact PCI-E performance if it goes too far from 100Mhz, 5Mhz or more is "the danger zone."

Edit: If I did my math right, you're pushing the bclk to 105Mhz which is the low end of where hardware starts to not be very happy with the overclock on the PCI-E and DMA buses. I recommend going back to 100Mhz and working just with boost to see if it makes a difference. Increasing VTT, VCCSA, or the PCH voltage may get you stable, but it's not something you want to mess with too much IMHO.

Edit: Edit: My experience is from my own overclocking of the bclk on my 3820. It doesn't really like to go 5Mhz over the current BCLK strap. In fact I usually didn't do much more than 3Mhz for the sake of stability (which is important because the PCH handles my RAID.)

I did have the Bclk upto 105, its at 100 now while i was testing this (hence the numbers in the OP differing vs my specs)

isnt @Mussels pro like 6 970's strapped together.. trying to go to the moon on gpu power lol

the overclock is fine its just the overclock cant completely sway the single thread performance.

overclock is stable, i had to rebuild my PC to make the cooling keep up with the planned crossfire tho.

I'd really like to know how you can score that high with a 280X. As in, besides DX11, what is everything else set to? CCC and benchmark settings if you would please. And driver version too I suppose.

I'd like to compare to my 280X with identical clocks on a system MUCH slower than either of yours. Maybe I'll find something of interest. Specifically, just how GPU bound is Heaven anyway?

i'm not running at max settings. DX11, 1080p, no tesselation or AA.

Usually it is the driver... remember to disable core parking... it causes stupid things in benchmarks.

btw what are the settings... I have very similar system to yours... my 7970 uses a 280X bios now(it has higher voltage )

For the lulz...

Disable that CG miner

core parking is disabled on both systems, but definitely a good idea.

you ninja

Your mums a ninja.

Is your BCLK running at 100? I'm pretty sure it will solve the conundrum.

It was for these tests on both systems, after the initial test confused me.

haha i seen the no advantage part and got the colors mixed up at the bottom.
well first of all the mobo's are 3.0..
possibly throttling and the turbo isnt going to kick on at a certain temp.
dont really how its the driver because its the same gpu on the same chipset.
this is just some kind of a test.. i guess mussels wants to see how many problems go misdiagnosed. to really get to the heart of it with no doubt there is several diagnostic tools that need to be run and everything needs to be reporting right.
from what people say there has been games that hate hyperthreading and you get better performance with it off. your honestly better off going off using the same settings as a game benchmark if your just gauging performance and not trying to stress. could just speculate all day but when you dont have all the information you need thats what happens.

mobos are PCI-E 3.0, CPU's are 2.0. not throttling, and turbo only does 200Mhz due to the way its OC'd so 4GHz on 8 threads should be powerful enough.

Tried disabling hyper-threading on the i7?

normally that helps people due to core parking, so no i havent.

 

[m&m's]

What I would do is:
Track the GPU load with GPU-Z while the bench runs.
If it stays at 99%, I would do a factory reset in the BIOS and try again.
If it fluctuates, I would do a Selective Startup via msconfig by only checking "Load System Services" and try again.

 

[Steevo]

Simple enough, the i5 had its veggemite.

 

[AsRock]

If it bugged me enough i would try setting them both at default clocks to see what the difference was, then if same kind of results were happening i would swap the mobo's around but it would have to be bugging me a lot and as far as benchmarks go they are pretty lame anyways.

Could always try 2 clean installs make sure every thing is identical.

EDIT: other wise try other benchmark program like 3D Mark.

 

[Mussels]

i guess i'll try 3dmark after work.

 

[xfia]

i feel like your joshing here with everything you said.. you could literally try 10 benchmark tools and end up with a even greater margin of error the more times you run them that is actually around 5 percent normally when drivers and tests are working right not 10 but as you can see it can possibly be more from the 7zip test and others that just have more error margin. not what you need to do.

 

[Mussels]

View attachment 65823
View attachment 65824
View attachment 65825
i feel like your joshing here with everything you said.. you could literally try 10 benchmark tools and end up with a even greater margin of error the more times you run them that is actually around 5 percent normally when drivers and tests are working right not 10 but as you can see it can possibly be more from the 7zip test and others that just have more error margin. not what you need to do.

who was that post aimed at? I'm not sure who you're talking to or what about, or what benchmarks for unrelated hardware has to do with anything here.

 

[Mussels]

bricks have been shat, the 280x has died a horrible death. perhaps these odd benchmark results were the warning signs it was on its way out.
This card was a warranty return from gigabyte, the original failed the same way :/

Thanks heaven, nice resu...fuck.

good thing giga reset the warranty on the effing card, now i wonder if i'll get a crossfire compatible card back from them or not :/

 

[AsRock]

Well i believe you may of find the issue ^^, well at least it's under warranty.

 

[m0nt3]

Sorry to hear about that. How is Gigabyte's RMA process?

 

[Mussels]

Sorry to hear about that. How is Gigabyte's RMA process?

well the first time it took approx 3 weeks :/

Good thing i've got a working 7970 in the mean time, this would be a nightmare if i didn't.

 

[m0nt3]

Maybe its just me, but I have always had better luck with reference designs. 3 weeks is a terrible wait. Do they have an option to hold a charge on a credit card until the bad unit arrives? I just ask because I have bought a lot of motherboards from them, but never had to RMA.