Intel Haswell Overclocking Clubhouse.

[The Von Matrices]

I just picked up a new i7-4770K. It's to replace my Xeon E5649 (six core) system that died after about a year of usage. I'm pretty sure I killed the CPU with too high a VTT/QPI (1.365V), but I did buy that processor used so I have no clue if it was abused by its previous owner. That said, I'm hoping to have my new CPU last longer than my previous one and I'm going to be a bit more conservative with the overclock this time around. Below is the picture of the box of the CPU I got. I haven't seen this lot documented anywhere on the internet so I was curious as to what I could do with it. I'm planning to keep all the internal voltages below 1.3V for safety's sake. You can see my exact system to the left.

I started by doing the "litmus test" of Haswell overclocking and set the processor to 4.6GHz and 1.2V. It booted to UEFI, and I went no further because I was unsure if it would boot to Windows. So I started low and worked my way up.

After some testing with AIDA64, I couldn't get the CPU past 4.2GHz with 2400MHz memory. I did a little more reading of various Haswell overclocking guides and decided to drop the memory clock to 1600MHz. Now I can get 4.4GHz core & 3.9 GHz cache at 1.225V. AIDA64 has been stable at this for an hour and I have temperatures in the mid 70's. I want to do more testing and also tweak the memory before I call this a completely stable overclock.

What's really interesting is that the 4.4GHz overclock is stable after an hour of testing, but I can't get to 4.5GHz (at least without going outside my margin of safety). With the core at 1.3V and the cache at 3.9GHz I can't get the system to remain stable at 4.5GHz in AIDA64 for more than 10 seconds. I'm not willing to try higher voltage not just to avoid damaging the CPU but especially because I get 95C temperatures at this setting. This CPU must need 1.35V or more to reach 4.5GHz. I've overclocked quite a few CPUs and I've never encountered such a steep change in voltage for such a small increase in clock speed. Does anyone else have this experience with such a wall?

I'm going to keep the 4.4GHz and 1.225V for a few days to confirm stability then i will go on to tweaking the memory. I'll also provide the obligatory CPU-Z screenshots then.

 

[TheHunter]

Try 2133MHz @ CL9-10-10-28 maybe you will have more luck, btw only 2133MHz uses full memory bandwidth ~34gb/s.


And Sorry to say this but batch# L307 is kinda bad, that your voltage speaks for itself.


I have some odd batch too L308B202 although its 100% stable 4.6Ghz @ adaptive 1.234v, cache min 35x; max 42x @ adaptive 1.122v


It can go up to 60-70C per core in x264 video encoder (uses avx) at 100% usage, this is the max i saw.. Usually below 55C in games, ok linx, Prime95, etc can make it 90C (because of adaptive cpuv offset) but i know i will never see such temps real world apps.

 

[EarthDog]

Batches mean nothing these days... Look at all the information on hwbot for example. People will tell you all over the place that batch really doesnt matter much with SB/IB/Haswell.

Does anyone else have this experience with such a wall?

Every chip has this. It is just where the wall is that is the difference. Yours, from this limited preliminary testing, seems like a pretty early wall.

 

[TheHunter]

But so far almost any L307 seems to need more voltage.

 

[EarthDog]

Almost.

Batch does not matter nearly as much as it did prior to SB. To look at a batch only and say Yay/Nay is premature.

 

[Womper]

I just picked up a new i7-4770K. It's to replace my Xeon E5649 (six core) system that died after about a year of usage. I'm pretty sure I killed the CPU with too high a VTT/QPI (1.365V), but I did buy that processor used so I have no clue if it was abused by its previous owner. That said, I'm hoping to have my new CPU last longer than my previous one and I'm going to be a bit more conservative with the overclock this time around. Below is the picture of the box of the CPU I got. I haven't seen this lot documented anywhere on the internet so I was curious as to what I could do with it. I'm planning to keep all the internal voltages below 1.3V for safety's sake. You can see my exact system to the left.


I started by doing the "litmus test" of Haswell overclocking and set the processor to 4.6GHz and 1.2V. It booted to UEFI, and I went no further because I was unsure if it would boot to Windows. So I started low and worked my way up.

After some testing with AIDA64, I couldn't get the CPU past 4.2GHz with 2400MHz memory. I did a little more reading of various Haswell overclocking guides and decided to drop the memory clock to 1600MHz. Now I can get 4.4GHz core & 3.9 GHz cache at 1.225V. AIDA64 has been stable at this for an hour and I have temperatures in the mid 70's. I want to do more testing and also tweak the memory before I call this a completely stable overclock.

What's really interesting is that the 4.4GHz overclock is stable after an hour of testing, but I can't get to 4.5GHz (at least without going outside my margin of safety). With the core at 1.3V and the cache at 3.9GHz I can't get the system to remain stable at 4.5GHz in AIDA64 for more than 10 seconds. I'm not willing to try higher voltage not just to avoid damaging the CPU but especially because I get 95C temperatures at this setting. This CPU must need 1.35V or more to reach 4.5GHz. I've overclocked quite a few CPUs and I've never encountered such a steep change in voltage for such a small increase in clock speed. Does anyone else have this experience with such a wall?

I'm going to keep the 4.4GHz and 1.225V for a few days to confirm stability then i will go on to tweaking the memory. I'll also provide the obligatory CPU-Z screenshots then.

Thanks for sharing your data points! And yeah, maybe we should call it Haswall instead of Haswell.

Try 2133MHz @ CL9-10-10-28 maybe you will have more luck, btw only 2133MHz uses full memory bandwidth ~34gb/s.

Ya, 2133MHz seemed to be the sweet spot. In some benchmarks I remember it beat out the next notch up. Turns out Haswell overlocking likes slower memory, so even more reason.

 

[freeboy]

"Haswall instead of Haswell."

lmao

I found an issue for my numbers.. as I was getting stuck with wierdness in mid fours.. I had an error in TOO low uncore ...
goodluck.

Im posting here.. hopefully the correct spot..
After lots of frustration stuck in the mid 4's... 4.5 4.6 with low temps reading in the 50's C under high load... I realized my UNCORE at 34 was auto correcting and screwing with the mem....
Changed the uncore to 4.1 to 4.2 and now am running a stable and very hot 4.9.. ouch, it needs some tweeking but Im changing fans to see if I can lower the temps as 4.8 stable and under high load tested at about 85C...

I noted Sin 0822 changed the C3 state from disabled to Auto and got some better results, so that and the Uncore boosted to over 4 has unclogged my "HASWALL" and returned to it Has "WELL" !..
I see and feel the frustrations people have and after a bad mobo and many hours of "walls", to actually have a breakthrough is great and hopefully will encourage those here with decent temps and hitting walls...

My 4.9 settings

100 BCLK setting at MANUAL
49 ratio
vcvore 1.427
vrin 2.020
ring 1.17
slight boost under .2 to the cpu agent
vdram 1.7
ram at an underclocked 1333


Good luck all..

 

[Kursah]

Just got my 4770k back from RMA. The new one is still a Malay and has a Batch of L315B355. Here's to hoping I have a decent clocker and performer this time! At least average! Gonna slap it in...might not have time to clock it till this weekend due to my homework load but we'll see what I can accomplish!

Alright Dave...so far this is a MUCH MUCH MUCH better chip...at least on first impressions. Booted to a freshly cleared BIOS. Stock CPU Voltage is 1.024v, which is quite a bit lower than my first chip's 1.072v. Cache stock is at 1.062v...I don't recall if that was any different on my other chip or not. Did the Asus recommendation, set the voltage to 1.25, and started with 46X CPU multi, left cache on Auto. Booted into windows but BSOD'd shortly after getting to desktop. Hell my last chip wouldn't even POST at that speed iirc! Then I dropped it to 45X, left voltage at 1.25v and so far it's in windows, idling in the low 30's. I haven't stressed it much...and might go back to stock while I get my homework done...but I'm happy I RMA'd so far!

 

[cadaveca]

I doubt 1.06V is stock cache, is probably the board doing that due to settings used. I've seen the same for "stock" cache on all the ASUS boards I've tested, but change cache multi settings, or turn off fully manual mode, and it changes.

But yeah, sounds like you got a better chip, that 1.070V was enough to say it was a dog for sure, and I did say so earlier....

Tuning Plan paid off for you.

You might pull 46 or 47 if lucky, but I think 45 @ 1.25V should be good to go! By my estimations, 1.050V stock = 4.5 GHz @ 1.265V or so.

 

[Kursah]

Well I ran AIDA for a while...but then started up a bunch of stuff and hit a BSOD x124 code at 45X 1.25v...but I also upped memory to 2133. Well I loaded XMP...which now that I think about it you said was a bad idea...so I will redo that whole deal here shortly. But ya so far a much better runner!

I had hope with the 1.070 chip earlier because the first guy on your chart has one and was able to get 4.5 out of 1.2x v. If I can do 4.4 or 4.5 and find easy and close to 1.25v stability I think I'll be good to go.

 

[The Von Matrices]

How do you check the stock voltage on the chip? Is there a standardized way to do this?

 

[freeboy]

Set the system to stock speed and look at the listed in bios vcore volts reading?

 

[Kursah]

How do you check the stock voltage on the chip? Is there a standardized way to do this?

Set everything back to AUTO or clear CMOS and go in to BIOS, check CPU voltage before modifying it.

 

[Womper]

Did the Asus recommendation, set the voltage to 1.25, and started with 46X CPU multi, left cache on Auto. Booted into windows but BSOD'd shortly after getting to desktop. Hell my last chip wouldn't even POST at that speed iirc! Then I dropped it to 45X, left voltage at 1.25v and so far it's in windows, idling in the low 30's. I haven't stressed it much...and might go back to stock while I get my homework done...but I'm happy I RMA'd so far!

Glad to hear this chip is more promising! That first one was definitely sad. Looking forward to more data points as you get time to test!

 

[The Von Matrices]

Set everything back to AUTO or clear CMOS and go in to BIOS, check CPU voltage before modifying it.

Set the system to stock speed and look at the listed in bios vcore volts reading?

Yes, but the CPU goes to many frequencies. Is the "stock" CPU voltage for the 4770K at 3.5GHz with turbo off or at what CPU speed with turbo on? Or is it something else inbetween?

 

[freeboy]

I would say at stock with turbo off

 

[cadaveca]

Yes, but the CPU goes to many frequencies. Is the "stock" CPU voltage for the 4770K at 3.5GHz with turbo off or at what CPU speed with turbo on? Or is it something else inbetween?

Turbo on, in BIOS, after loading "optimized defaults". CPU cache to 35, if board reports cache higher than vCPU.

Some boards use custom Turbo Profiles, for both CPU and cache, which can give false readings, however.

 

[The Von Matrices]

Turbo on, in BIOS, after loading "optimized defaults". CPU cache to 35, if board reports cache higher than vCPU.

Some boards use custom Turbo Profiles, for both CPU and cache, which can give false readings, however.

I know that my board sets the cache multiplier to 39, even with multi-core enhancement disabled. Does it change the ring voltage because of this?

 

[cadaveca]

I know that my board sets the cache multiplier to 39, even with multi-core enhancement disabled. Does it change the ring voltage because of this?

Some do. I'm sick, can't recall how your board works at the moment.

 

[Mull1s]

Hello guys, i need some advice on how to get my 4670k stable
I got the batch L312B529 (no idea if this is a good one or not)

I have been trying to get stable on the following settings but im getting 124 bsods or deep freezes:

Cpu multi 46
Cpu vcore: adaptive 1.332v
Cpu cache multi: 43
Cpu cache voltage: adaptive 1.18V + 0.020 offset = 1.2v total
VCCIN: 1.82
VCCSA offset: +0.050v
IO-Digtal offset: +0.10v
IO-Analog offset: stock
RAM OC: 1600Mhz 1.35V (8-8-8-24 2T) to 2000Mhz 1.5V (9-9-9-24 2T)

Edit: Same settings but Cpu multi at 44 and Vcore on 1.25V has been stable on longer stress testings.

Temps doesnt seem to be an issue for now (im using Corsair H90 Pull+Push).. any ideas on how i can get this OC stable?
Thank you

 

[vega22]

what is considered safe 24/7 volts for these chips?

 

[freeboy]

Hello guys, i need some advice on how to get my 4670k stable
I got the batch L312B529 (no idea if this is a good one or not)

I have been trying to get stable on the following settings but im getting 124 bsods or deep freezes:

Cpu multi 46
Cpu vcore: adaptive 1.332v
Cpu cache multi: 43
Cpu cache voltage: adaptive 1.18V + 0.020 offset = 1.2v total
VCCIN: 1.82
VCCSA offset: +0.050v
IO-Digtal offset: +0.10v
IO-Analog offset: stock
RAM OC: 1600Mhz 1.35V (8-8-8-24 2T) to 2000Mhz 1.5V (9-9-9-24 2T)

Edit: Same settings but Cpu multi at 44 and Vcore on 1.25V has been stable on longer stress testings.

Temps doesnt seem to be an issue for now (im using Corsair H90 Pull+Push).. any ideas on how i can get this OC stable?
Thank you

There is a good guide by sin0822 at
http://www.overclock.net/t/1397693/...scussion-issue-report-club/1020#post_20701051


I recommend trying 45 before 46 and gradually increase vcore
I would suggest clicking CPU and moving memory lower until CPU stable.
For me, I lower uncore to 4.0 at those ranges

 

[Womper]

Hello guys, i need some advice on how to get my 4670k stable
I got the batch L312B529 (no idea if this is a good one or not)

I have been trying to get stable on the following settings but im getting 124 bsods or deep freezes:

Cpu multi 46
Cpu vcore: adaptive 1.332v
Cpu cache multi: 43
Cpu cache voltage: adaptive 1.18V + 0.020 offset = 1.2v total
VCCIN: 1.82
VCCSA offset: +0.050v
IO-Digtal offset: +0.10v
IO-Analog offset: stock
RAM OC: 1600Mhz 1.35V (8-8-8-24 2T) to 2000Mhz 1.5V (9-9-9-24 2T)

Edit: Same settings but Cpu multi at 44 and Vcore on 1.25V has been stable on longer stress testings.

Temps doesnt seem to be an issue for now (im using Corsair H90 Pull+Push).. any ideas on how i can get this OC stable?
Thank you

Increase cache voltage, try 1.25v adaptive or add more offset.
Increase VCCin to 1.85-1.89.

And remember, a 1.332v adaptive voltage means your chip will be at 1.45v during AVX workloads. If you plan on encountering AVX stuff, then you will probably need to increase VCCin to the 1.9v range, maybe even 2.0v.

For 24/7 I decided to stay under 1.4v even after the +AVX voltage. Also, I have only had success with adaptive voltage in the 1.2x range; when I go after higher frequencies that require static voltages in the 1.3x range, I cannot get a stable adaptive setting, even if it's much higher than the static voltage.

-----> NEWSFLASH <-----

Z87-PLUS/PRO/A/C BIOS 1405
-Improve system stability.

Nom nom nom. Time to revisit my little 4.8GHz project. And see if this has new microcode too.

 

[Kursah]

Where is this BIOS you speak of? I see 1205 which has been out since July 5th but nada on 1405. Maybe it's not for my PRO?

 

[Anusha]

Anyone using Intel XTU for stress testing? It doesn't spike up the Vcore when using Adaptive Vcore. Does that mean ti is not using AVX instructions in its stress test?

My 4.4GHz/1.285V (static)/RAM at 1333/9/9/9/24/2t/1.5V and everything else AUTO overclock is not stable at all in Prime95 27.9. I'm wondering if it is not stable only with AVX loads. So I'm running Prime 26.6 (the last one without AVX) to see if it passes. If it does, that probably means it is the AVX loads that is causing the issues. Intel XTU's stress test runs for 12+hrs without an issue.

If Prime95 26.6 passes and 27.9 doesn't, it could also mean that Intel XTU is not using AVX instructions in its stress test. Someone found out that the benchmarking portion of it is using Prime95, but don't know which version. (i.e. whether that supports AVX or not) no idea what the stress test is using. Funnily, the temps are almost identical between H.264 encoding and XTU's stress test.

Then there is another issue. Yesterday I ran a long H.264 encode queue and the system rebooted (no BSOD, no crash dump, no bugcheck entry in eventlog nothing! This is not the first time either.) after 10hrs. So far I haven't found the cause of it. First I thought it was caused by lack of VCCIN (input voltage) as I was at 1.75V which is the AUTO setting for my M6H board. But then again, everyone says you only need to be 0.4V above Vcore, so I definitely am there with only 1.285V static (1.296V is the highest I've seen and 1.75V VCCIN is actually 1.776V after LLC at extreme. That's almost +0.5v).

It could even be a driver problem.

Any thoughts on these?