Intel Haswell Overclocking Clubhouse.

[TheHunter]

And i found another sweet spot too

personal best Cinebench11.5 score yet @ 4.5Ghz

cpu 45x @ 0.085v offset
cache 40x @ 1.10v


2nd best so far with 9.92
cpu 45x
cache 41x @1.12v

 

[Womper]

What I have been finding is that memory itself it quite important for stability, and sticks that were great on IVB.. not so much any more. Looking at other forums and such, other people are reporting the same as well, and I noticed that secondary timings for Haswell are really different, so XMP settings for some users are going to pose stability issues.

I've been trying to smoke this BSOD out, no luck so far. Core and cache speed and voltages have no effect.

1. Loaded Optimized Defaults, no BSOD so far.
2. Loaded Optimized Defaults, then changed AI Tuner to XMP mode, and set memory to its 2133 XMP profile. Didn't touch anything else, so I'm running stock voltage and frequency. Occasional BSOD during boot, guaranteed BSOD during Prime95 blend.
3. Reduced DDR to 1600, keeping XMP timings. Three consecutive blue screens during boot. Eventually booted, but didn't have time for stress testing.
4. Reduced DDR to 1333, keeping XMP timings. No BSOD so far, even during P95 blend.

And yet, I completed a 3 hour Aida64 stress test at 4.7GHz core, 4.6GHz cache, and DDR 2133. And then BSOD 124 every 20 minutes in Borderlands 2 no matter how much voltage I shove into the core and cache.

Recommendations? Maybe increase system agent voltage? VCC IN? Suspect bad RAM or incompatible timings?

 

[cadaveca]

VCCSA, maybe VCC SIO/DIO.


I suspect incompatible XMP profile. got a few sets here that work great on 1155, and are crap on Haswell, a few that won't even boot with XMP enabled, too.

Be careful with VCC-in. I suggest no more than 1.9V. Others, too, try to stay under 1.2V for VCCSA, VCC -SIO/-DIO

 

[Womper]

VCCSA, maybe VCC SIO/DIO.


I suspect incompatible XMP profile. got a few sets here that work great on 1155, and are crap on Haswell, a few that won't even boot with XMP enabled, too.

Be careful with VCC-in. I suggest no more than 1.9V. Others, too, try to stay under 1.2V for VCCSA, VCC -SIO/-DIO

Thanks. Assuming the XMP profile is the problem, then can I just manually tweak the timings to be more Haswell friendly? I paid good money for these Dominator Platinums, I'll be annoyed if I can't utilize them at CL 9.

 

[cadaveca]

Thanks. Assuming the XMP profile is the problem, then can I just manually tweak the timings to be more Haswell friendly? I paid good money for these Dominator Platinums, I'll be annoyed if I can't utilize them at CL 9.

I do understand, although, those sticks were meant for IVB, not Haswell. Corsair has great warranty service in my experience, if you got 1.5 V DIMMs, those seem less problematic.


And yes, set timings manually, but there are two secondaries that show no "live settings" in ASUS BIOS, those two are determined by other timings with Haswell, and are very different than past platforms.

First things first though, be sure to clear CMOS properly, with battery removal method. BIOs are currently corruptible, seem common to all brands, I think Intel MEI may need an update from its current version.

 

[Womper]

I do understand, although, those sticks were meant for IVB, not Haswell. Corsair has great warranty service in my experience, if you got 1.5 V DIMMs, those seem less problematic.


And yes, set timings manually, but there are two secondaries that show no "live settings" in ASUS BIOS, those two are determined by other timings with Haswell, and are very different than past platforms.

First things first though, be sure to clear CMOS properly, with battery removal method. BIOs are currently corruptible, seem common to all brands, I think Intel MEI may need an update from its current version.

Thanks again. These are 1.5 V DIMMs. If I understand correctly, if these secondary timings are causing problems then all I can do is use different memory sticks?

The issue has persisted across 4 BIOS updates and a CMOS clear using the jumper method, so I'm not suspecting BIOS corruption. Unless my motherboard is broken.

 

[cadaveca]

All OEMs seem to be having some issues with Samsung-based DIMMs (which yours are).

 

[TheHunter]

Thanks again. These are 1.5 V DIMMs. If I understand correctly, if these secondary timings are causing problems then all I can do is use different memory sticks?

The issue has persisted across 4 BIOS updates and a CMOS clear using the jumper method, so I'm not suspecting BIOS corruption. Unless my motherboard is broken.

Is it XMP v1.3?

My 4x4Gb 1.65v Crucial Balistix Elite is v1.3 and runs fine at default 2133mhz CL9, I even lowered it to 1T.


What is your ring bus cache multi? Did you try lower ie 40x @ 1.11v?

 

[Womper]

Is it XMP v1.3?

My 4x4Gb 1.65v Crucial Balistix Elite is v1.3 and runs fine at default 2133mhz CL9, I even lowered it to 1T.


What is your ring bus cache multi? Did you try lower ie 40x @ 1.11v?

It is XMP 1.3. 1T and 2T make no difference, and neither does 2133 vs 1600. I'm running stock settings, which is 37x core turbo (when 4 cores active) and 39x cache, and still bluescreening during Prime95 blend or on boot.

I bumped up SA, DIO, AIO, and VCC-In by 0.050v, so if I still BSOD's I think it falls on the RAM.

 

[MxPhenom 216]

It is XMP 1.3. 1T and 2T make no difference, and neither does 2133 vs 1600. I'm running stock settings, which is 37x core turbo (when 4 cores active) and 39x cache, and still bluescreening during Prime95 blend or on boot.

I bumped up SA, DIO, AIO, and VCC-In by 0.050v, so if I still BSOD's I think it falls on the RAM.

What board do you have? Sorry if you have already mentioned it previously in the thread. A guy was saying at OCN that he has the Gigabyte OC Force board with no issues with the 1.5v Samsung based Dominator Platinum's on his rig. Can clock them to 2600+ claiming stable.

 

[Grnfinger]

New member




Have the day off so want to learn the new bios and get some overclocking started.

 

[cadaveca]

What board do you have? Sorry if you have already mentioned it previously in the thread. A guy was saying at OCN that he has the Gigabyte OC Force board with no issues with the 1.5v Samsung based Dominator Platinum's on his rig. Can clock them to 2600+ claiming stable.

If you take a look on forums that have benching teams, there are many mentions that Haswell can be funny with ram. Personally, I think board may play a large role in how well ram clocks, but BIOS is an important factor as well. The guy that said he has no issues has Gigabyte's high-end OC board, and it is possible that "lesser" boards may not get such good clocks.

I've noticed my own "oddities" when it comes to ram, too. I am not the only one, either, this review that was posted by hookielumnus on overclockers of this G.SKill kit shows a 2933 MHz kit that is slower than a 2666 MHz kit. In his conclusion, he blames the ram.


http://www.overclockers.com/gskill-tridentx-ddr3-2933-memory-kit-review


He did manage to get 3330 MHz as an OC. But benchmarks for "stock" were very low. My set hit a bit over 3400 MHz, which I posted a couple of pages back.


I have this same set, review is waiting to be published. What I noticed, myself, is that although the kit performs "poorly" @ 2933, if you drop the divider down to 2800 MHz, performance increases. So to me, that's the board giving the low score, or something specific to that 2933 MHz divider. It's not the ram's fault.


Hookielumnus obviously didn't notice that lower-divider performance is better. I also noticed that this kit, with one of my CPUs, won't boot at all @ 2933 MHz, in the ASUS Maximus VI Extreme. Take the memory and that same funky CPU and install into the ASRock Z87 Extreme6/AC, and the kit boots right up.

Therefore, memory clocking is not as "black and white" as it has been with past platforms. I need to do way more testing myself before I can come to any solid conclusions. But what I CAN say, is that from the hardware I have on-hand, clocking with Samsung can be difficult. It might be down to individual CPUs, boards, or BIOSes, or some weird mix of all three.

New member

[url]http://i237.photobucket.com/albums/ff267/Grnfinger/stock_zps5bf0c287.png[/URL]


Have the day off so want to learn the new bios and get some overclocking started.

Welcome to the club, man. I'd love to add you to our chart, take a look ath the 4th post for the info I am looking for there and if you can provide that, it'd be much appreciated!

 

[Womper]

What board do you have? Sorry if you have already mentioned it previously in the thread. A guy was saying at OCN that he has the Gigabyte OC Force board with no issues with the 1.5v Samsung based Dominator Platinum's on his rig. Can clock them to 2600+ claiming stable.

I've the Z87-Plus, and my Dominator Platinums are listed in the Qualified Vendors list for it. Once I rule out VCC-SA, VAIO, VDIO, and VCC-In, I'll contact Asus tech support. I think I have a pretty good CPU, so I'd rather exhaust other options before blaming it.

@cadaveca- In your experience, can RAM issues/oddities cause 0x124's? Is the 0x124 BSOD kind of an umbrella for all things unstable?

 

[cadaveca]

@cadaveca- In your experience, can RAM issues/oddities cause 0x124's? Is the 0x124 BSOD kind of an umbrella for all things unstable?

Typically, it's just a general catch-all, yes. It can be CPU, cache, or sometimes ram...but ram is the least-likely scenario. However, when clocking up both CPU and ram, when both speed are tested stable separately, this error can appear, for sure.

What I have found, is that if you get 0x124, and then increase vcore, and it changes error to 0x101, then CPU is the likely culprit. I also watch POST code display for slow-downs in memory training at POST, you sometimes see hesitation @ 40 or there-abouts when memory might be a bit funky.

I nearly always boot into memtestx86+ when doing such testing. Where errors appear, and how they appear can indicate where the problem is. I really need to get myself an RSTPro for memory testing.

 

[dumo]

I've noticed my own "oddities" when it comes to ram, too. I am not the only one, either, this review that was posted by hookielumnus on overclockers of this G.SKill kit shows a 2933 MHz kit that is slower than a 2666 MHz kit. In his conclusion, he blames the ram.


http://www.overclockers.com/gskill-tridentx-ddr3-2933-memory-kit-review


He did manage to get 3330 MHz as an OC. But benchmarks for "stock" were very low. My set hit a bit over 3400 MHz, which I posted a couple of pages back.


It's not the ram's fault.

QFT

There still settings on this particular single sided MFR chip which need to be explore.
Actually, with its high MHz advantages this type of ram can be competitive too.

 

[MxPhenom 216]

This whole issue with Samsung memory seems very much based on the board and BIOS. But it doesn't add up when there are a ton of BIOS released for multiple vendors that you would think would improve Samsung DIMM memory compadibility. I am assuming vendors know about the issues???

 

[cadaveca]

This whole issue with Samsung memory seems very much based on the board and BIOS. But it doesn't add up when there are a ton of BIOS released for multiple vendors that you would think would improve Samsung DIMM memory compadibility. I am assuming vendors know about the issues???

I have told ASUS, Gigabyte, and MSI about the issues I encountered. So yes, they know, in some aspect.

Because it is still early in the release, and others are reporting issues on forums, that I have seen, identifying the actual problem's cause might not be so easy.

Could simply be a CPU-IMC-quality thing. Could be the individual sets. Could just be BIOS.

Anyway, A little bit of research will provide you with lots of info about memory clocking ,already pointed you to a thread on XS about ASRock boards and Samsung DIMMs, and that thread did indicate that settings used on past platforms may be sub-optimal for performance.


http://www.xtremesystems.org/forums/showthread.php?286755-How-to-OC-Samsung-with-ASRock-on-z87-OCF

Resist the urge to tighten the settings "You know" will be faster for now.

 

[PolRoger]

And i found another sweet spot too

personal best Cinebench11.5 score yet @ 4.5Ghz
http://i.imgur.com/KfcKyPH.png

cpu 45x @ 0.085v offset
cache 40x @ 1.10v


2nd best so far with 9.92
cpu 45x
cache 41x @1.12v

It took some fiddling but I was able to match your Cinebench score using a 4x4GB Samsung rev.D kit. I tweaked voltages a tad from your settings and also tried cache at 40x/41x/42x

Here is my run @ 45x/42x:

If you take a look on forums that have benching teams, there are many mentions that Haswell can be funny with ram. Personally, I think board may play a large role in how well ram clocks, but BIOS is an important factor as well. The guy that said he has no issues has Gigabyte's high-end OC board, and it is possible that "lesser" boards may not get such good clocks.

I've noticed my own "oddities" when it comes to ram, too.

Therefore, memory clocking is not as "black and white" as it has been with past platforms. I need to do way more testing myself before I can come to any solid conclusions. But what I CAN say, is that from the hardware I have on-hand, clocking with Samsung can be difficult. It might be down to individual CPUs, boards, or BIOSes, or some weird mix of all three.

I'm finding higher overclocks at higher dram speeds (especially full bank) with Samsung to sometimes be rather challenging.

I don't have any 4x4GB high speed Hynix kits to test at this time just 2x4GB kits using CFR and MFR. I also have an older 4x4GB 2133 BFR kit.

I've also noticed that the 2133 divider seems to get some pretty nice bandwidth numbers in AIDA even when compared to AIDA results running at higher dram speeds? Here is a Samsung rev.D 4x4GB kit with the same Bios settings... @ 2133 vs. 2200:

 

[cadaveca]

Yep, for sure some dividers are funky.

I like to avoid those 2000, 2200, 2600, 2933 dividers myself, seems better to use lower divider and use BCLK to get the ram speed up. hookielumnus's review I linked above greatly illustrates why. Not sure why he blames the DIMMs though, maybe he's seen something I haven't...I am not sure.

I feel like I got to try more chips before I can suss out more tweaks currently, had some hardware die that is preventing me from doing reviews right now, have to wait for replacements, so I'll get some work done on the clocking guide and other stuff the next few days.

 

[Jstn7477]

I enjoy BCLK overclocking as well. Makes more sense to me to have lower multipliers based on a higher base clock than vice versa.

My current OC:
4400 core/4275 cache 35x/34x* 125.8MHz BCLK at 1.22v (bad chip, even though starting vcore was 1.007v.)
Memory: 2*8GB G.Skill Sniper 1866 CL10-11-10-30/2T @ 2012MHz (1:6 ratio) @ 1.55v

I think I have plans to get some Tridents or other high speed DDR3 soon, but can't shell out $150 just for RAM at the moment. This chip will do 166 BCLK but the RAM dividers are laughable at that point, and I'm either stuck with crappy sub-1800MHz clockspeed or 2200 which my RAM cannot do period.

 

[Grnfinger]

Grnfinger|4770K/L312B341|1.024/1.021|4500MHz|1.25|45|1.021|1866|Corsair H100|Asus MVI Hero

Currently at 4500 @ 1.25vcore ( 30 min prime so far) heat might be an issue she spikes for a few seconds to 85c from time to time.

 

[cadaveca]

Grnfinger|4770K/L312B341|1.024/1.021|4500MHz|1.25|45|1.021|1866|Corsair H100|Asus MVI Hero

Currently at 4500 @ 1.25vcore ( 30 min prime so far) heat might be an issue she spikes for a few seconds to 85c from time to time.

Nice chip. ADDED!!!

I notice that you vCPU and vCache are very similar, and you can run CPU multi and cache multi at same value. Most other chips do not have such similar voltages, and most report issues trying to run 1:1. I wonder...

 

[Grnfinger]

I haven't touched the cache volts.
Asus set it at 1.021 and it seems to work with manual vcore set to 1.25 and off set 0.856

Just ran 3D11 http://www.3dmark.com/3dm11/6848547

 

[springs113]

I haven't touched the cache volts.
Asus set it at 1.021 and it seems to work with manual vcore set to 1.25 and off set 0.856

Just ran 3D11 http://www.3dmark.com/3dm11/6848547

I see you finally have things up and running, nice specs.

 

[TheHunter]

It took some fiddling but I was able to match your Cinebench score using a 4x4GB Samsung rev.D kit. I tweaked voltages a tad from your settings and also tried cache at 40x/41x/42x

Here is my run @ 45x/42x:



I'm finding higher overclocks at higher dram speeds (especially full bank) with Samsung to sometimes be rather challenging.

I don't have any 4x4GB high speed Hynix kits to test at this time just 2x4GB kits using CFR and MFR. I also have an older 4x4GB 2133 BFR kit.

I've also noticed that the 2133 divider seems to get some pretty nice bandwidth numbers in AIDA even when compared to AIDA results running at higher dram speeds? Here is a Samsung rev.D 4x4GB kit with the same Bios settings... @ 2133 vs. 2200:

Cool, so you're fine at 42x cache @1.075v? I never tried that low yet


Here are my specific default timings

And I noticed the same read/copy drop at 2200mhz.. Weird stuff, thanks Cadaveca for that extra info.




@dumo

Nice, only 1.248v at 5Ghz? Is that lower voltage possible at lower multi + higher Bclk 125mhz?