Intel Core i7-4770K Haswell

[cadaveca]

Thanks.

I actually had a few hours and went to test things.

Max multi was x39 (sync all cores)
Max BCLK freq was 105, and anything higher made the system hang right after the post (interesting enough this option was not locked in the bios).

Tested two chips, same results. 4.3Ghz would be enough for me, but I'm not sure about 4.1 tbh. I know it's not much of a difference, but you know how it is, sometimes you only need that extra 2-3 fps for happyness

BCLK clocking is not the way it should be done. This is still IVB in it's essence, and although there are "straps" available to allow for higher BCLK, there are still HUGE dead-zones. Multi-clocking FTW.

 

[Ikaruga]

BCLK clocking is not the way it should be done. This is still IVB in it's essence, and although there are "straps" available to allow for higher BCLK, there are still HUGE dead-zones. Multi-clocking FTW.

Well, I know but that was the max I could achieve in such a short time, but I doubt it that any of those chips could do 4.3Ghz tbh (with any kind of witchery whatsoever), while 4.3 with any of the 3770 I met was a piece of cake.

 

[cadaveca]

Well, I know but that was the max I could achieve in such a short time, but I doubt it that any of those chips could do 4.3Ghz tbh (with any kind of witchery whatsoever), while 4.3 with the any of the 3770 I met was a piece of cake.

If you take a look at the Haswell clubhouse I started last night before going to bed, 4.2 is the lowest clock reported there, @ 1.175V.


So yeah, MOST CPUs will hit 4.3 GHz. SOME won't, but those will be rare. And cooling used is very important.


And I cannot stress this enough.. although there are BCLK options available, they should not be used directly. Using the BCLK dividers and BCLK tweaking is mainly for those playing with memory, and to allow them to get higher memory clocks.


All those screenshots with 3000 MHz+ memory so far...all done with BCLK.


That's why I started the clubhouse. I am working on an OC guide that will help everyone get that 4.3 GHz or more, but there ARE specific ways in which you need to set things in order to truly make that realistic. You'll note that my OCs are far different than what other reviewers are posting, and it's not just because I got retail chips. Personally, I expect 4.6 GHz out of every chip, not 4.3 GHz, but if you follow traditional clocking methods and current guides, yeah, 4.3(or directly, 43x100) should be the average.

 

[Ikaruga]

If you take a look at the Haswell clubhouse I started last night before going to bed, 4.2 is the lowest clock reported there, @ 1.175V.


So yeah, MOST CPUs will hit 4.3 GHz. SOME won't, but those will be rare. And cooling used is very important.


And I cannot stress this enough.. although there are BCLK options available, they should not be used directly. Using the BCLK dividers and BCLK tweaking is mainly for those playing with memory, and to allow them to get higher memory clocks.


All those screenshots with 3000 MHz+ memory so far...all done with BCLK.


That's why I started the clubhouse. I am working on an OC guide that will help everyone get that 4.3 GHz or more, but there ARE specific ways in which you need to set things in order to truly make that realistic. You'll note that my OCs are far different than what other reviewers are posting, and it's not just because I got retail chips. Personally, I expect 4.6 GHz out of every chip, not 4.3 GHz, but if you follow traditional clocking methods and current guides, yeah, 4.3(or directly, 43x100) should be the average.

- But all those chip in the clubhouse thread are from the K series, it's almost paramount to hit 4.3 with those chip, and you really need to be unlucky to get a K chip which won't reach that clock. I was and I'm still talking about the 4770, not the 4770k. I really like TSX tbh, and I want it :B

- Yes, seem like Haswell OC is not really "sensitive to"/"bond with" memory speed.

 

[cadaveca]

AH, my bad. yeah, getting 4.3 GHz out of a normal 4770 is likely impossible. Intel stated that all CPUs other than "K" parts are 100% frequency-locked.


Sheesh. I totally missed what you were saying there, I apologize.

 

[Ikaruga]

AH, my bad. yeah, getting 4.3 GHz out of a normal 4770 is likely impossible. Intel stated that all CPUs other than "K" parts are 100% frequency-locked.


Sheesh. I totally missed what you were saying there, I apologize.

No worries, only those who do nothing make no mistakes

 

[Vlada011]

I don't see better temps than Ivy. I see 10C higher temps than on IB.
Intel go in very very bad way and I hope, but now I'm not sure any more at least to leave us Extreme Platform capable to OC.
I mean 3970X 6 Cores OC on 4.8-5.0GHz this martyr can't turn LinX AVX Edition on 4.2GHz.
Some user say I enter in BIOS on 4.5GHz but on 4.6GHz I can't, and AIDA64 Stability.
Before 1-3 years people laugh when think on AIDA64 Stability test, we are encouraged to push 12-24h Prime95 and LinX AVX Edition with last linpack 100 Run.

Than I had stable 5.0GHz+ because I enter in BIOS on less than 1.3500V on 5.1GHz,
LinX AVX Edition can work days oin 4.8-4.9GHz summer-winter but without thermal throttling and little under 95C. AIDA64 Stability I didn't try that is probably 5.0-5.1 GHz full stable on same temps. I notice throttling chasing once stable 4.9GHz during summer with 1.300V on 102C not with lower clock, but I stop test. HW Info show good Min and Max Clock and Throttling.

And mine temps are 80C for highest core on 4.5GHz with last Prime95 and LinX AVX Edition.
That is not higher temp than Haswell and Haswell don't work on 65-70C like IB with Turbo Enable (3.9) in LinX AVX Edition and Prime95.
I saw some people even problems on only Turbo with 80-90C, that is not better temps than IB.
Temps and overclocking is far worse from IB than SB>IB, far worse. Only that AVX instruction he work very good. But for that better waiting AVX 3.2.

I'm sorry of people who save money to buy i7 Instead of i5 and wait one really good and promised platform. Communities talk about Haswell after SB show up.
Than Intel start to work on this Project.
They want at least 30% OC for K models. Because they invest in 100$ cooler and in 250-350$ motherboard.
For at least 30% completely stable.
Without that it's pointless everything if someone have 4-5C better temps on best cooler than on stock because heat transfer.
And that is not end for 30% OC (about 4.5GHz) on long time you need at least 4.7GHz stable because last stable clock is never good option except for people with lot of money who will sell for 12 months and buy new.

After November 2011 3930K still best options for gamers.
For him worth paying and for H100 and H100i and H220 and Rampage and Sniper and Classified.
This in HTPC with one little Noctua NH-L9i, Turbo Enable and nothing more.

 

[Ikaruga]

I don't see better temps than Ivy. I see 10C higher temps than on IB.
Intel go in very very bad way and I hope, but now I'm not sure any more at least to leave us Extreme Platform capable to OC.
I mean 3970X 6 Cores OC on 4.8-5.0GHz this martyr can't turn LinX AVX Edition on 4.2GHz.
Some user say I enter in BIOS on 4.5GHz but on 4.6GHz I can't, and AIDA64 Stability.
Before 1-3 years people laugh when think on AIDA64 Stability test, we are encouraged to push 12-24h Prime95 and LinX AVX Edition with last linpack 100 Run.

Than I had stable 5.0GHz+ because I enter in BIOS on less than 1.3500V on 5.1GHz,
LinX AVX Edition can work days oin 4.8-4.9GHz summer-winter but without thermal throttling and little under 95C. AIDA64 Stability I didn't try that is probably 5.0-5.1 GHz full stable on same temps. I notice throttling chasing once stable 4.9GHz during summer with 1.300V on 102C not with lower clock, but I stop test. HW Info show good Min and Max Clock and Throttling.

And mine temps are 80C for highest core on 4.5GHz with last Prime95 and LinX AVX Edition.
That is not higher temp than Haswell and Haswell don't work on 65-70C like IB with Turbo Enable (3.9) in LinX AVX Edition and Prime95.
I saw some people even problems on only Turbo with 80-90C, that is not better temps than IB.
Temps and overclocking is far worse from IB than SB>IB, far worse. Only that AVX instruction he work very good. But for that better waiting AVX 3.2.

I'm sorry of people who save money to buy i7 Instead of i5 and wait one really good and promised platform. Communities talk about Haswell after SB show up.
Than Intel start to work on this Project.
They want at least 30% OC for K models. Because they invest in 100$ cooler and in 250-350$ motherboard.
For at least 30% completely stable.
Without that it's pointless everything if someone have 4-5C better temps on best cooler than on stock because heat transfer.
And that is not end for 30% OC (about 4.5GHz) on long time you need at least 4.7GHz stable because last stable clock is never good option except for people with lot of money who will sell for 12 months and buy new.

After November 2011 3930K still best options for gamers.
For him worth paying and for H100 and H100i and H220 and Rampage and Sniper and Classified.
This in HTPC with one little Noctua NH-L9i, Turbo Enable and nothing more.

It's worth mentioning that Sandy Bridge is from Intel Haifa/Israel, while Ivy and Haswell are from the US (Oregon iirc). Different team, different people, different management, different goals, etc... Let's see what the Haifa team can do with Skylake, and only draw a conclusion about Intel after we have it...................... but I have to agree that Intel doesn't seem to care about enthusiast anymore. I guess that the corresponding market segment is so small, they decided it's just not worth to interfere with their business strategy for such a low number of costumers.

About the high temps: Don't forget how incredibly small CPUs Ivy/Haswell are, so the high temps are understandable. You just need adequate (read: very good) cooling if you want to push things beyond their stock settings (which I have absolutely no problem with, since it's just more heat and nothing more in the case of Haswell).

Generally, I think the main "problem" here is that Sandy Bridge was so awesome, it's really hard to make something better in such a short time

 

[Vlada011]

That is same, decision came from same place, nobody can't decide without top and special to make some mistakes in design.
No one can put cheap grace or decide no we go on flux solder except example first circle of 30 people.
This is architecture design problem and limitation like decision.
And they didn't explain us after two year why cheap paste is inside.
I think lose on good paste can't be more than 100.000$ for all CPU.
I mean on so big quantity price drop rapidly and that is nothing for Intel.
Saving money on paste is not only reason.
In all part of life customer deserve and get answer only enthusiasts and overclocker and gamers don't deserve answer what is behind flux solder and what is behind "cheap" or not cheap paste.
I think Sandy Bridge is more rear occurrence not always guaranteed performance.
But he is paste and he is PCIe 2.0 and that story is finished, I don't like explanation how I don't get much with PCIe 3.0.
But Haswell is bottom.

If you say Skylake will make in Israel, where they decide to make Haswell E ???
Skylake is first something what worth investment for people with SB and IB. Haswell is downgrade for me.
Not on stock but OC Profil 100% downgrade.
Even what I saw example CINEBENCH result are absolutely same, difference can make better sample of one or other(IB or Haswell).
You can't say he is clearly if you buy 100% 10% stronger, not even 5%. CPU sample could decide not architecture.

 

[Ikaruga]

That is same, decision came from same place, nobody can't decide without top and special to make some mistakes in design.
No one can put cheap grace or decide no we go on flux solder except example first circle of 30 people.
This is architecture design problem and limitation like decision.
And they didn't explain us after two year why cheap paste is inside.
I think lose on good paste can't be more than 100.000$ for all CPU.
I mean on so big quantity price drop rapidly and that is nothing for Intel.
Saving money on paste is not only reason.
In all part of life customer deserve and get answer only enthusiasts and overclocker and gamers don't deserve answer what is behind flux solder and what is behind "cheap" or not cheap paste.
I think Sandy Bridge is more rear occurrence not always guaranteed performance.
But he is paste and he is PCIe 2.0 and that story is finished, I don't like explanation how I don't get much with PCIe 3.0.
But Haswell is bottom.

If you say Skylake will make in Israel, where they decide to make Haswell E ???
Skylake is first something what worth investment for people with SB and IB. Haswell is downgrade for me.
Not on stock but OC Profil 100% downgrade.
Even what I saw example CINEBENCH result are absolutely same, difference can make better sample of one or other(IB or Haswell).
You can't say he is clearly if you buy 100% 10% stronger, not even 5%. CPU sample could decide not architecture.

I think you greatly misunderstood me. What I was trying to say is that different people have different priorities, and that's true inside Intel as well. Certain goals can be achieved several ways, and different engineers take different path to reach it, this doesn't mean that the end product will be different. Haswells using paste are doing their work perfectly on stock speeds, they are cheaper to make and there is no soldiering needed from the management/leadship point of view. The engineers on the project said it's not needed at all for safe , and normal operation, and that's exactly why it's not used. No sane management will spend extra money in mass production if it's not needed. (please don't get me wrong again, I do agree that we enthusiast would need it badly tho, but that's irrelevant apparently )

I also have no idea about Skylake, but that wasn't my point. I was just trying to say that perhaps it will be closer to your enthusiast desire than how Ivy or Haswell are, or maybe it will be even further away. Only time will tell.

 

[Deleted member 24505]

Maybe they should sell all Haswell k's with a non glued IHS, so you can use it naked, or use paste on the cpu and sit the IHS on it in the socket if you want.

 

[Vlada011]

I didn't saw how Haswell work perfect and on stock speed.
Maybe 3.5GHz but even that is hot for our experience and what we see last years.
What I saw is some samples are on 3.9GHz where close to point of throttling.
Net is full of people who have 90C on 3.9GHz if they start stability test and now suddenly easy people start to change way of testing. Last linpack and AVX instructions heat Haswell as crazy.
I'm sorry but last 5-6 years when I came home with CPU first week I start Linx, Prime95, OCCT, than go 3DMark Vantage, Fritz Chess, CINEBENCH and comparing and than if one week everything work OK start OC, again same way LinX(AVX Edition necessary in last time)Prime95, OCCCT with AVX Enable, wPrime, X264, CINEBENCH, ycruncher, ....and that I declare OC and save OC Profil usually 200-300MHz than last clock.
I'm happy now because I have 4.8GHz stable 15h Prime95 and I can 4.5GHz clock for every day, not more....Prime95 is in my experience most required test for Ivy Bridge, harder even than LinX AVX Edition and even temps are 2-3higher and I need 0.0025V more than for LinX only Prime95 Blend test show real temps after
2-3h example maybe even little more.
Everything on side but when you find volttage and clock and Prime95 past 12h 2-3 times and every worker without error you know everything is OK.
I think even 6h is enough. Without drop clock, throttling, temps in border and everything pass, you can confirm with other test and save profil.
OK everything can go wrong after some time special with abnormal high clocks, example over 4.5GHz for 24/7 for my taste is to much. That is 30% enough.
But like I sad you must 200MHz more in stability tests because 100% stability. I do like that.
AIDA64, than my CPU is better than SB 2500K if we look that test as stability or SuperPI, or CINEBENCH.
AIDA64 heat my CPU to 72C on 4.5GHz.
Prime95 is same temp on stock clock.
I mean how big is difference between that two test.