Sign in with
Sign up | Sign in
Your question

LGA 1156 - optimal uncore/QPI/RAM frequencies?

Last response: in Overclocking
Share
June 18, 2010 11:40:18 AM

Hi,

I have a Core i3-530, overclocked to 3.52 GHz (by setting BCLK to 160 MHz). I do not wish to set the BCLK any higher or touch the voltages. And that's final! :p 

Other frequency settings:

RAM: I have 2x2GB running at its spec frequency/timings: 800 MHz (1600 MT/s), 8-8-8-20.

Uncore: I understand that the uncore multiplier is locked in LGA 1156 CPUs, so it appears that I'm limited to 2.56 GHz here (up from 2.13 GHz).

QPI: currently set to 2.56 GHz as well (resulting in 5.12 GT/s).


My questions:

1. I understand that for optimal performance, uncore needs to be at least 2x the RAM's transfer rate - which means 3.2 GHz in this case. But as mentioned, I'm limited to 2.56 GHz uncore.

So my question is: since uncore is not synced with RAM at all, will it be better to underclock the RAM to 667 MHz (1333 MT/s), and change CL from 8 to 7? Or does it make no difference?

2. Since my uncore is limited to 2.56 GHz, will I gain anything from increasing QPI frequency? If so, what should I set it to?


Thanks a bunch!
a c 100 K Overclocking
June 18, 2010 3:01:20 PM

Hmm well my understanding is matching RAM freq with FSB and all that was for the older set ups. QPI is a bit different as there's more direct connections for everything so it's already rather fast. I would set the QPI multiplier to it's highest but there's only 2 options, 16x and 18x right? Can't remember specifically... but anyway you should be able to get away with the highest multiplier without issues or voltage changes. I wouldn't reduce the RAM speed if you're keeping the 160 base clock as your only option would be 1280mhz which isn't a good choice. My 1600mhz CL8 RAM is running 1400mhz CL7... I would hate to waste a bunch of speed at CL7 going with 1333.
Even with the not synced speeds you're still getting better performance. From my tests a higher CPU speed (and therefore base clock/QPI) resulted in lower memory latency overall. Your best bet tho is to do some benchmarking... try these things out and you can run Everest to bench the transfer rates and latency (note: I ran a bunch and the transfer rates never changed noticeably but the latency did vary).
a b K Overclocking
June 18, 2010 4:26:19 PM

Resh Aleph said:
I do not wish to set the BCLK any higher or touch the voltages. And that's final! :p 

The only problem with that statement is when you upped your Bclk and leave the voltages on "AUTO" the MB will raise voltages for you. So much for leaving the voltages at stock speed, the only way to do that IS to set your voltages in the Bios.
Related resources
June 18, 2010 7:49:53 PM

wolfram23 said:
I would set the QPI multiplier to it's highest but there's only 2 options, 16x and 18x right? Can't remember specifically... but anyway you should be able to get away with the highest multiplier without issues or voltage changes.

QPI goes all the way up to x22, which gives 3.52 GHz (7.04 GT/s). Can I go that high with the stock voltage? Seems a bit excessive when uncore is only 2.56 GHz...

RJR said:
The only problem with that statement is when you upped your Bclk and leave the voltages on "AUTO" the MB will raise voltages for you. So much for leaving the voltages at stock speed, the only way to do that IS to set your voltages in the Bios.

Oh. OH. That's very not quite good!!

I shall restore the original BCLK to find out the stock voltages, then!

Thank you both :) 
June 22, 2010 2:34:02 AM

What did you end up doing?
June 22, 2010 10:45:55 AM

I just set QPI to x20 = 3.2 GHz = 6.4 GT/s. I didn't run any benchmarks - I lost my patience after fighting with voltage and power-saving settings. That included an unexpected no-POST.

Now I sort of got what I wanted, except Vcore is still a bit higher than stock at idle. I'll probably start a new thread about that at some point...
!