Intel Haswell Overclocking Clubhouse.

[radrok]

I gotta see more results on Haswell, if it scales well with cold (single stage cold) I may pick up one for 24/7.

 

[Jstn7477]

My new chip arrived today, batch L312B326, and my system is ready for me to simply drop in the chip and bolt on the TPC-812. Purchased from Newegg on June 10 and it is of course going in my Z87 Extreme6 with roughly 2 year old 4x4GB DDR3-2133 Ripjaws-X sticks.

 

[TheHunter]

Ok i've put all together

This batch L308B202 is actually very good, default voltage in UEFI @ default speed 3.5ghz? was 0.967v on 2 cores and 0.92v on other 2, not bad ^^


But I think i have some issues with XMP profile, not sure though, when i selected XMP for 2133mhz and rebooted it then also OC'ed my cpu to 4.250mhz @ 1.20v and set blck to 125mhz lol

I need to get use to all these bios settings and see what's what, but so far so good.



EDIT: I figured out XMP and it sticks, also command rate @ 1T works ok

But i think i used 39x multi and now its at 3900mhz all the time? also auto voltage climbed to 1.096v.

WMI looks nice too

 

[Womper]

Ok i've put all together

This batch L308B202 is actually very good, default voltage in UEFI @ default speed 3.5ghz? was 0.967v on 2 cores and 0.92v on other 2, not bad ^^


But I think i have some issues with XMP profile, not sure though, when i selected XMP for 2133mhz and rebooted it then also OC'ed my cpu to 4.250mhz @ 1.20v and set blck to 125mhz lol

I need to get use to all these bios settings and see what's what, but so far so good.



EDIT: I figured out XMP and it sticks, also command rate @ 1T works ok

But i think i used 39x multi and now its at 3900mhz all the time? also auto voltage climbed to 1.096v.

WMI looks nice too

Is that Aida64 memory benchmark in the free version? I can compare some cache speeds with that.

 

[springs113]

CPU 4770 L311B411 fr newegg/mpower with some ddr3 2000 ballistix

Well I got mine to 4.4 with little effort but it wasn't stable after loading windows. I was hovering around mid to high 40 degrees c
CPU-Z kept reading my voltages wrong though because in the bios I set it to 1.22(cpu) and it was fine.

Also at stock MSI already oc'd the processor, the multi is already set 39 and while stock I was around 35-40 idle and as high as 45 load...using the h100. With stock cooler I was idling around high 40s and load in the mid to high 50s.

 

[Jstn7477]

Ugh, got the CPU installed, had to fight with my memory to get it to work without debug code 53, and now none of my USB ports work in Windows for some reason. The only thing I can use at the moment is a PS/2 keyboard to even do anything on it right now...

EDIT: USB fixed, turns out it was some old Z77 USB root hub driver that screwed it over. Used extensive keyboarding to get into Device Manager and update it off the support CD. The board however still hates my June 2011 pair of 4GB G.Skills and only works properly with the Jan 2012 set which coincidentally has XMP 1.3 vs. the XMP 1.2 of the other sticks (no, I wasn't using XMP). The sticks worked fine with my 2600K and 3770K so I have no idea what the big fuss is.

 

[Womper]

Manual 1.41v is the minimum required to boot @ 50x core, 46x cache @ 1.325v, DDR 2133. SuperPI looped 11 times before freezing (good thing I saved a screenshot just after starting).

Now would be a good time to ask- how much voltage can Ivy Bridge take on a consistent basis? My usage model will be speed step with adaptive voltage, as opposed to 24/7 manual voltage override.

 

[cadaveca]

Intel says(I personally dunno how real this is, I've just been gathering info the past several weeks and am working on conglomerating it all into one "easily digestible" meal. ):

At the same time, whitepaper says 153 W max. I'd aim for that, but I think that about 125W is more realistic for 24/7 use.


If you buy "Tuning Plan" warranty, I'd say those voltages listed above are OK, but do keep in mind that's maximums, and not all chips can take the maximums. Most should, but Intel offers that Tuning Plan for a reason...since most chips used by most users will never see voltages as high as are listed above. Set 2.0 V to vCore on air, I'm sure you won't have a pleasant experience.


Still working on the OC guide, as well as review right now, will update the member ranking in a bit here.

 

[MxPhenom 216]

Ugh, got the CPU installed, had to fight with my memory to get it to work without debug code 53, and now none of my USB ports work in Windows for some reason. The only thing I can use at the moment is a PS/2 keyboard to even do anything on it right now...

EDIT: USB fixed, turns out it was some old Z77 USB root hub driver that screwed it over. Used extensive keyboarding to get into Device Manager and update it off the support CD. The board however still hates my June 2011 pair of 4GB G.Skills and only works properly with the Jan 2012 set which coincidentally has XMP 1.3 vs. the XMP 1.2 of the other sticks (no, I wasn't using XMP). The sticks worked fine with my 2600K and 3770K so I have no idea what the big fuss is.

Why didn't you formatting your system and reinstall clean windows when you switched boards?

 

[Jstn7477]

44x/41x @ 1.22v for both vcore and cache seems to be completely stable for me, but under the workload I do 24/7 (World Community Grid), the CPU hovers at 100c which is a LOT worse than my IVB chips by like 15-30c. I seriously don't know what I'm doing wrong to deserve these temperatures. Gaming is completely fine in the 70c range, but as soon as my distributed computing fires up, the temperature shoots up to mid-90s within 5 seconds on my TPC-812.

Can I continue running my chip like this or will it fail very soon? I just can't get this temperature under control with something I do nearly 24/7 that loads down the CPU like IntelBurnTest.

Why didn't you formatting your system and reinstall clean windows when you switched boards?

As much as I would love to do that, my Win7 install is only about 5 months old and honestly it works fine after the drivers got updated. I saw a 14% FPS increase over my 3770K at the same clock in a timedemo I ran this morningin Team Fortress 2. You have no idea how happy I was about that.

Temps:

 

[TheHunter]

Is that Aida64 memory benchmark in the free version? I can compare some cache speeds with that.

Its a full version







Ok managed to downclock my cpu, it was high performance power plan fault (balanced fixed it).. Go figure lol, my old Q9450 downclocked no matter what power plan.

 

[Womper]

Why didn't you formatting your system and reinstall clean windows when you switched boards?

Sysprep, brah. I'm far too lazy to reinstall all my programs and/or deal with blue screens due to swapping the motherboard and stuff. Sysprep is a built-in windows tool to generalize all of your drivers and stuff before migrating to a new machine. You have to reactivate Windows still.

44x/41x @ 1.22v for both vcore and cache seems to be completely stable for me, but under the workload I do 24/7 (World Community Grid), the CPU hovers at 100c which is a LOT worse than my IVB chips by like 15-30c. I seriously don't know what I'm doing wrong to deserve these temperatures. Gaming is completely fine in the 70c range, but as soon as my distributed computing fires up, the temperature shoots up to mid-90s within 5 seconds on my TPC-812.

Can I continue running my chip like this or will it fail very soon? I just can't get this temperature under control with something I do nearly 24/7 that loads down the CPU like IntelBurnTest.



As much as I would love to do that, my Win7 install is only about 5 months old and honestly it works fine after the drivers got updated. I saw a 14% FPS increase over my 3770K at the same clock in a timedemo I ran this morningin Team Fortress 2. You have no idea how happy I was about that.

Temps:

I'd keep temps below 90 under long-term loads like WCG. For me, I just want to run games, but otherwise my CPU will sit in idle state for web browsing and email, so I only care about game temps (which, awesomely enough, are super low).


I was going to make a TF2 demo myself and benchmark overclock speeds. I wish I had done it on my i7-930 before upgrading, but all I know is that TF2 got noticeably slow with the 930. Since the 4770k, I haven't even caught a glimpse of anything under 90fps. Mostly around 120fps (I bought that 120hz monitor for a reason) during the heaviest action, otherwise it's at the 299fps cap.

Planetside 2 can still tank under large battles, down into the low 40's even though I'm at 47x core 46x cache. Of course, my 930 would go into the 20's in these situations, so it's still probably a 1.5x speedup. Sony needs to spend some time optimizing their game engine, since I'm only hitting 65C on my hottest core even at 48x.

 

[Womper]

44x/41x @ 1.22v for both vcore and cache seems to be completely stable for me, but under the workload I do 24/7 (World Community Grid), the CPU hovers at 100c which is a LOT worse than my IVB chips by like 15-30c.

Forgot to say, those temps are normal. The only tricks left are to reduce DDR speed to 1600 (2133->1600 saves me 10C under max load), or get better cooling/delid.

I might go get a h100i tomorrow myself to help me play around at dangerous voltages.

 

[sneekypeet]

There is an EDIT button so that you don't have to double quote and double post

 

[cadaveca]

Forgot to say, those temps are normal. The only tricks left are to reduce DDR speed to 1600 (2133->1600 saves me 10C under max load), or get better cooling/delid.

I might go get a h100i tomorrow myself to help me play around at dangerous voltages.

Don't bother. Go custom water(meh as well), or wait for some new blocks to come out. I noticed that die for Haswell is offset from center, and due to this, most AIO water-coolers under-perform. The older H70 is best, since it feeds water from one end to the next. I think many coolers may get slight updates to deal better with Haswell. I'm looking at waterblocks as well, EK Supremacy with de-lidded chip and EK delid-mount is going to work best.

I think THIS is part of why Haswell seems "hotter" than IVB for some. At the same time, given the power consumption numbers I'm seeing, TDP increase was not enough.

 

[radrok]

Don't bother. Go custom water(meh as well), or wait for some new blocks to come out. I noticed that die for Haswell is offset from center, and due to this, most AIO water-coolers under-perform. The older H70 is best, since it feeds water from one end to the next. I think many coolers may get slight updates to deal better with Haswell. I'm looking at waterblocks as well, EK Supremacy with de-lidded chip and EK delid-mount is going to work best.

I think THIS is part of why Haswell seems "hotter" than IVB for some. At the same time, given the power consumption numbers I'm seeing, TDP increase was not enough.

I fully agree.

I think that we all agree that Haswell problem is not "raw" heatload but it is more how to effectively channel that heat onto a cooling system.

Maybe Haswell really needs a fluxless solder IHS.


For all I can see to get good clocks out of this platform we need to step into SS territory.

 

[Jstn7477]

Forgot to say, those temps are normal. The only tricks left are to reduce DDR speed to 1600 (2133->1600 saves me 10C under max load), or get better cooling/delid.

I might go get a h100i tomorrow myself to help me play around at dangerous voltages.

Wow, the memory speed makes that much of a temperature difference? Is there any "real" performance degradation as a result of lowering the memory speed in games and in general? Would my 2x8GB G.Skill Sniper 1866 CL9 sticks be better for this system than 4x4 DDR3-2133 CL11 sticks?

 

[Womper]

Don't bother. Go custom water(meh as well), or wait for some new blocks to come out. I noticed that die for Haswell is offset from center, and due to this, most AIO water-coolers under-perform. The older H70 is best, since it feeds water from one end to the next. I think many coolers may get slight updates to deal better with Haswell. I'm looking at waterblocks as well, EK Supremacy with de-lidded chip and EK delid-mount is going to work best.

I think THIS is part of why Haswell seems "hotter" than IVB for some. At the same time, given the power consumption numbers I'm seeing, TDP increase was not enough.

From some quick research, I wasn't seeing much temp difference between an h100 and custom water for 3770k overclocks. Custom water costs what, like $250 to put together? I'd rather save the $ for now and see if 14nm acts right. The h100i should fit perfectly in my case (and free up a DDR slot), and keep things 10-15C cooler than what I currently have, maybe better. I'm lazy, and any kinda of water cooling system sounds like an inherently poor choice for electronics so I'm even worried about the h100i. At this point, I'm questioning what the h100i would truly get me for my usage- if I'm under 80C at 4.8-5.0GHz during gaming, then it's probably not worth it.

As for Haswell temps...gonna blame the FIVR and memory controller changes.

Wow, the memory speed makes that much of a temperature difference? Is there any "real" performance degradation as a result of lowering the memory speed in games and in general? Would my 2x8GB G.Skill Sniper 1866 CL9 sticks be better for this system than 4x4 DDR3-2133 CL11 sticks?

Yeah, big temperature difference. I haven't tested whether CL affects temps. I bought my memory before Haswell reviews were released... Corsair Dominator Platinum 2133 1.5v CL9. Doh! If I had known, I would've found some 1.5v DDR 1600 at the lowest possible latencies. Or maybe some 1866 and drop it to 1600 with lowered latencies. The Dominator Platinum does look pretty sweet though.

Performance wise, 2133 versus 1600 gets you a very small gain, pretty much single-digit FPS gains. 1333 is below some sort of threshold, and you see a disproportionately large drop in synthetics and games. Intel puts these nice 8MB caches on their chips to hide the limited bandwidth to DDR. In fact, I would expect that overclocking your cache will have the same effect as using faster DDR.

 

[cadaveca]

As for Haswell temps...gonna blame the FIVR and memory controller changes.

That's part of it too, for sure.


To help combat this, default Intel CPU behavior has NB/RING/CACHE/whateveritiscalled adjusting frequency according to workload, and some users are setting it to match or sit just under max CPU multi, 24/7.

To me, hat seems a bit misguided. Minimum should match Intel's minimum, just like CPU multi, and max only should be adjusted, for 24/7 use. At least, I think this might help with temps...something to look into, anyway.

 

[TheHunter]

My OC atm 4.4Ghz @ 1.15v, Cache ring 42x @ 1.21V(auto set) and rest at auto, temps in Aida64 stress test 60-70C @ H90 and lowered pump/fan speed ~ 1200rpm. In any cpu bound game never exceeds 55C, usually up to 50C.

But it failed at 4.5ghz @ 1.15v, got a bsod in Aida64 stress test, backed down to 4.4ghz and all is good.

And from what i saw @ OC'ed 570GTX its not worth 4.4ghz+, im gpu bottlenecked now.. Actually even stock cpu speed I get gpu bottleneck quickly., while before with Q9450@3.6ghz I got big cpu bottlenecks in something like Bf3, Natural Selection2, Hitman Absolution, Serious Sam3, Ghostbusters the game,..


Btw what's with this offset cpu voltage, how can i set it so my cpu voltage down clocks? Well now its at 1.15v no matter 800mhz or 4.4ghz. Any tips?

 

[PolRoger]

That's part of it too, for sure.


To help combat this, default Intel CPU behavior has NB/RING/CACHE/whateveritiscalled adjusting frequency according to workload, and some users are setting it to match or sit just under max CPU multi, 24/7.

To me, hat seems a bit misguided. Minimum should match Intel's minimum, just like CPU multi, and max only should be adjusted, for 24/7 use. At least, I think this might help with temps...something to look into, anyway.

I've been matching the min/max (i.e. 40/40 or 43/43) cpu cache multi so far while testing overclocks. What is Intel's minimum? and if you run them at different speeds does BIOS dictate the cache multi according programmed rules?

 

[cadaveca]

I've been matching the min/max (i.e. 40/40 or 43/43) cpu cache multi so far while testing overclocks. What is Intel's minimum? and if you run them at different speeds does BIOS dictate the cache multi according programmed rules?

I am not sure, haven't had time to investigate, was just a thought. It seems that it's recommended to use Turbo to OC anyway, so...it kinda makes sense???


But who knows if it actually even works...they have those options adjustable just like CPU multi for a reason, I suppose.


I understand, that for benchmarks, you'd want to keep it at the max, but I think that if it can manage itself effectively(and it should), then it might help keep the temps in check. To me, this is Intel's first full customizable chip, so there could be many things overlooked when it comes time to OC.

 

[PolRoger]

4.5Ghz DDR3-2400 full bank (4x4gb) with SA and CPU A/D IO all on auto. (Corsair H110):

 

[Jstn7477]

After some more testing tonight, I think my target frequency is going to be 43x/43x at >= 1.17v. 44x core regardless of the uncore frequency takes about 1.22v which makes my CPU temps increase by 10c to hit 100c regularly in my screenshot from earlier. At 4.3GHz/1.17v currently, my highest core temperature with distributed computing load is 89c. IMO, this particular chip seems to be responding like my main 3770K did, even down to the voltages. Changing DRAM speed had no effect on my temperatures so I'm still running full banks of 4GB sticks at 2133 CL11. My main priority now is probably going to be voltage tuning to find the threshold of my chip at 43x/43x so it runs as cool as possible.

 

[TheHunter]

@cadaveca

Can i ask you something about cpu sys agent offset voltage,

how can i set it so my cpu voltage down clocks? Well now its at 1.15v no matter 800mhz or 4.4ghz. Any tips?