Haswell CPU Ring Ratio Tests and Results

[ProKoN]

I was curious to see how the cpu ring ratio (cache ratio) affected performance. everyone suggests for best performance run at a 1:1 ratio. I was not able to, perhaps limitations of my board or cpu. so i decided to test how it really affected performance. i decided to use cinebench because its universal and its quick to bench mark. at 4.4GHZ i could no longer maintain a 1:1 ratio.

 

please keep something in mind. all results from 3.9GHz to 4.5GHz, I have NOT validated at stable settings. I was only able to benchmark these settings but did not run any stability tests from 3.9GHz-4.5GHz.

 

4.6GHZ validated for 20 hours stable

4.7GHz validated for 12 hours stable

 

what i found was cpu frequency is still king!

 

 

stock 3.9GHZ 3900MHz ring bus 1:1 ratio

 

 

 

4.0GHz 4000MHz ring bus 1:1 ratio

 

 

4.1GHz 4100MHZ ring bus 1:1 ratio

 

 

4.2GHz 4200MHz ring bus 1:1 ratio

 

 

4.3GHz 4300MHz ring bus 1:1 ratio

 

 

4.4GHz 4300MHZ ring bus

 

 

4.5Ghz 4400MHz ring bus

 

 

4.6GHz 4000MHz ring bus

 

 

4.7GHZ 3000MHz ring bus

 

 

 

 

 

 

 

[Alex780]

I don't understand why they tell us to bring the cache ratio up in oc guides. I did tests myself and it doesn't affect performance at all. I just got 4.4ghz stable @1.29V. Should I even bother overclocking the cache ratio?

[Prathamesh Patil]

not all the applications benifit from cpu cache speed but very few do and that too its very minimal. I would rather clock my core higher. Moreover the cpu cache has double the bandwidth of the core unlike how it used to be in ivy and sandy days. So the cpu cache need not be in 1:1 ratio with the core speed.
A nice guide from Sin0822 for gigabyte motherboards but the haswell oc basics remain the same.
Here: http://www.overclock.net/t/1401976/the-gigabyte-z87-haswell-overclocking-oc-guide

[Alex780]

 

 

Btw @ProKoN said "do not allow your ring bus voltage to exceed your vcore voltage"  at 3:41

 

I have mine set at 4.3ghz 1.29V while my vcore is set at 4.4ghz 1.275V. Is it bad if I leave it like this? Can it harm my cpu?

[Jumper118]

It. Does seem that cinebench r15 doesn't get effected that much. What programs do benefit for it?

[deathjester]

It. Does seem that cinebench r15 doesn't get effected that much. What programs do benefit for it?

 

If anything cinebench would benefit the most from it. It certainly isn't games.

 

Glad the op posted results. I have been saying this for a long time, I just never bothered to prove it. 

 

That is why I say just set cache to 3500 (you can try 3900 first, but not all Haswells play ball). Dial in clock first. Raise cache/uncore after and test with only cache in aida clicked. Saves you HOURS of testing and crashing late in tests. I wasted a ton of time dialing in a friends 4770k. The chip simply wouldn't do even 3900. If I would have set it to 3500 I could have dialed that thing in, in a fraction of the time. Kept crashing hours in from the damn cash. That 4770k drove me nuts.

[ProKoN]

 

I have mine set at 4.3ghz 1.29V while my vcore is set at 4.4ghz 1.275V. Is it bad if I leave it like this? Can it harm my cpu?

 

it is fine if your ring voltage exceeds you vcore voltage.  i would not go past 1.45V on the ring voltage.

 

your settings are fine.

 

I adopted the platform when haswell launched and i was going by MSI recommendation of not exceeding the ring voltage. after my own personal testing and other peoples results it is fine to have a higher ring voltage.

 

It. Does seem that cinebench r15 doesn't get effected that much. What programs do benefit for it?

when i ran these tests i was adjusting the ring multi on the fly through xtu in the OS environment.

 

I discovered by doing it that way the results may not have been as accurate as restarting every time and changing the ring muti in the bios.

 

 

cpu frequency is still king. the only reason i can even get close to  @CthulDude r15 score is simply because i can push the ring bus to a 1:1 (ish) ratio for benchmarking. my chip is nowhere as good. the average clock speed of my i7 is 4.85GHz with a 4.7GHz ring bus. that is almost as good as cthuldude's 4.9GHz with a 3.9GHz ring. I think the ROG HERO is limiting his cpu ring but it is clear his 4.9GHz with a 39x ring across all cores is still faster than 4.85GHz with a close to 1:1 ring ratio.

 

https://docs.google.com/spreadsheet/ccc?key=0AlC81MjwelBgdEZNV3l6aHl1eUNwSUR4Rml0MXMzN1E&usp=sharing#gid=0

 

i really havent done to many real world benchies gaming or rendering. i really doubt you will see significant difference from a high or low ring bus.

 

i still advocate you push for the highest ring frequency you can get. in order to achieve the best performance \ watt 

[deathjester]

it is fine if your ring voltage exceeds you vcore voltage.  i would not go past 1.45V on the ring voltage.

 

your settings are fine.

 

I adopted the platform when haswell launched and i was going by MSI recommendation of not exceeding the ring voltage. after my own personal testing and other peoples results it is fine to have a higher ring voltage.

 

when i ran these tests i was adjusting the ring multi on the fly through xtu in the OS environment.

 

I discovered by doing it that way the results may not have been as accurate as restarting every time and changing the ring muti in the bios.

 

 

cpu frequency is still king. the only reason i can even get close to  @CthulDude r15 score is simply because i can push the ring bus to a 1:1 (ish) ratio for benchmarking. my chip is nowhere as good. the average clock speed of my i7 is 4.75GHz with a 4.7GHz ring bus. that is almost as good as cthuldude's 4.9GHz with a 3.9GHz ring. I think the ROG HERO is limiting his cpu ring but it is clear his 4.9GHz with a 39x ring across all cores is still faster than 4.75GHz with a close to 1:1 ring ratio.

 

https://docs.google.com/spreadsheet/ccc?key=0AlC81MjwelBgdEZNV3l6aHl1eUNwSUR4Rml0MXMzN1E&usp=sharing#gid=0

 

i really havent done to many real world benchies gaming or rendering. i really doubt you will see significant difference from a high or low ring bus.

 

i still advocate you push for the highest ring frequency you can get. in order to achieve the best performance \ watt 

 

Is his memory running the same speed though? Cinebench seems to really like higher clocked memory. Me and Superbee are 40 points higher at almost the same clock speed as another guy, who must be running 1600mhz ram and my uncore is at 4000.

 

When I OC'd my 1600 to 2133 Cinebench went WAY up. The only game I saw an increase in was Guild Wars 2 (maybe 1.5 fps if that), but cinebench skyrocketed.