• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Haswell uncore overclock

megatron

megatron

megatron

Soldato
Joined
8 Nov 2002
Posts
4,510
Location
Kent
I'm trying to get stable with 4690k + Z97X Gaming 5.

I believe my CPU clock of 4.5Ghz with 1.26v is stable (at least on its own). I tested with 1600ddr and x35 uncore. Also I've tested with 2400ddr. I tried 4.2 on uncore but dropped it back to 4.1Ghz with 1.2v ring voltage.

I've been taking screens of what I found stable, so I can look back as a form of notes. I have mainly tested with media coder h265 encoding with prelimiary testing with intel extreme testing utility, Asus real bench, and I heard windows experience monitor was useful too. I thought I had my system stable finally but I got a another blue screen probably 75% into a h265 encode. I think my main area of instability is centered around the uncore, unless the extra work running uncore overclocked and 2400ddr throws the CPU unstable. I'll try to keep the what ifs to a min and if I think a unit is causing instability then raise its voltage.

Its just that waiting 8 hours for a blue screen makes testing a slow labourious process. Also blue screen, not program crash, even after 8 hours that's a bit extreme?

Note: You can get full size pictures from photobucket, click cog and + sign and then right click picture: view, left click to expand.

Shows stock and I believe a stuck voltage regulator temp sensor:

4.5Ghz uncore x35 xmp 2400 - This was a real world encode of h 265.

4.5Ghz vcore 1.26v 1600ddr uncore x35 (stock)

I need to get a screenshot of an uncore overclock. I've tried 1.2v vRing & a random +0.005 system agent with 2400DDR and my cpu overclock all together but it lead to a BSOD. The limit on air for vRing/uncore v is 1.35v so I still should have room to wiggle. However I was reading the bios description of system agent and it said that it effects L3 Cache. I though that Uncore and cache are both ran at the same speed i.e. the uncore multi. From the description (I will post a screenshot) it looks like uncore (integrated memory controller, pci-e link, system agent link) and L3 cache need to be overvolted in tandem due to linked multi if I'm right.

L3 Cache voltage supplied by system agent, same ratio as uncore?

I'm currently testing (as you can see in last image) 1.2v vRing + 0.020 system agent (1600ddr). Last BSOD was 1.2v vRing + 0.005 system agent (2400 ddr). Trouble is I don't have an screens of uncore stability yet, but it has managed several hours on h265 encoding 100% load. Do you think I should up the vRing or will upping system agent help?

8 hrs Aida64 Stable now for h265 encode (its 6 degrees hotter running).
 
Last edited:
Thanks for the reply. I decided to test battlefield 3 with my 8 hr Aida64 stable uncore clock and got a BSOD within 1 round! So I tested with stock uncore and x36 and its stable. Going to go up slowly and test each multi at a time in BF3. Although haven't got further than x36 cause I've been looking at Ebaying my old mobo. The prices which the P5Q deluxe are going for is quite heart warming.
 
There is very little to be gained from 1:1 clocking,only in benchmarks.

I just set my Uncore to 40x at 1.15v and leave it there.Trying to clock it high just adds more variables to consider if/when you have issues.Not worth it IMO.

At least all you on 4690k's get good clocks at reasonable voltages,it takes 1.3v to get 4.5 on my 4670 :/
 
A high uncore cache and some fast memory with low timings will give an extra few FPS in gaming, worth doing if the cpu can handle it.
 
Bit of a late bump, rather than make my own thread. But I too have been looking into uncore clocking. It was running at 4ghz, with my overclock of 4.7ghz on the main core. But after everything I've read, you'll gain minimal performance if any at the risk of instability.

So I will probably go back to 4ghz tomorrow

http://www.overclock.net/t/1427411/haswell-uncore-same-or-lower-than-cpu-speed/10
http://forums.anandtech.com/showthread.php?t=2396155&page=2
https://pcdiy.asus.com/2014/06/6-49...ns-of-devils-canyon-on-asus-z97-motherboards/


Additionally initial finding show that the cache ratio or what is also refereed to as the ring bus has a higher degree of sensitivity and can impact overclocking more so than previous Haswell K series CPUs. Previously our auto rules would define a cache ratio of 39 while this slightly reduces efficiency as opposed to maintaining a native cache ratio ( 1 to 1 ) with the turbo value defined for the CPU multiplier it helps to place less concurrent stress on the CPU and allows for improved scaling of the CPU frequency. For this generation we have revised our auto rule and defined one of 44. This helps efficiency and improves performance but may reduce some 4.6GHz CPUs from being able to reach this frequency. As such if pursuing manual overclocking it is initially recommend you manually define a cache ratio of 39. Once you have verified your max CPU frequency and ensured it is stable you can increase the cache ratio incrementally attempting to see if you increase or even potentially match it to our CPU ratio / multiplier.
Click to expand...
^
 
Last edited:
If left to auto setting in bios, my stock cache voltage is 1.2700. But I've manually set it down to 1.200 without issue. For benchmarks I've had it at 4400mhz on 1.2500, cpu at 4.7ghz.
 
I've in the meantime added an OcUK AIO GPU cooler, but I believe I'm at stable settings atm. I have 4.5Ghz CPU clock, 4.0Ghz Uncore, 2400DDR XMP profile. Only yesterday I tried setting the RAM to command rate 1T but was unstable, and in belt and braces fashion I set it back to 2T +0.02 vcore (there is some ambiguity in the XMP profile, 1T as read from CPU-z but sets itself 2T and was the same in a review I saw. It gives 26 points in Intel Extreme Tuning Utility benchmark, so it might be worth going higher than 1.65v on RAM.

I have 1.27v vcore, 1.1 uncore (+0.05), System Agent +0.055, 1.65v DDR voltage (as defined in the XMP profile). I'm using Battlefield 3 to check stability as its the best test I've encountered (better than H265 encoding and Asus benchmark).

I realised that setting my RAM to some settings I found in a review of my RAM in a slower speed was unstable with 1.5v, stable with 1.55v DDR voltage, so this was complicating things; I was considering only CPU or GPU for sources of instability.

I'll probably leave it at 4Ghz uncore maybe try to see if I can get 1T command rate on RAM.
 
Last edited:
With your ddr at 2.4 T2 timing you should be able to run your cpu and cache at 4.5Ghz what mother board are you using.
 
shamus21 said:
With your ddr at 2.4 T2 timing you should be able to run your cpu and cache at 4.5Ghz what mother board are you using.
Click to expand...

That is a pretty bold statement. I've been tinkering with my 4690k since May when I bought it, what makes you able to offer such advice?
 
megatron said:
That is a pretty bold statement. I've been tinkering with my 4690k since May when I bought it, what makes you able to offer such advice?
Click to expand...

Just personal experience. You may have a poor CPU, memory etc. many variables when it come to over clocking. :)
 
You must log in or register to reply here.
Back
Top Bottom