Haswell i5 4670K/i7 4770K batch and o/c results

[cstkl1]

i've only seen that 71 dead cpu phenomenon on M6E, what boards u using

its 78. and normally its that imc dram clock thingy 1-14. setting it very tight with low trfc... get rainbows before dying. ( literally top part of the screen)..
sandy did get that too but never to a point of dying after that.

but so far disable the ram fast boot, enable mch recheck, enable dram extra training..
seems to be much safer.

[Alex-Ro]

78 i got that when damaged cache(blasted some 1.8V on air to see how much does it take),setting dram clock period 1-14 does not damage the cpu,altough gains are minimal,usually a value around 5-8 is good,but lower you go in this you have to increase vcore.low trfc does not damage also,you get rainbows from graphic corruption resulted from memory errors.

I had tested so far about 5 cpu's,one on dice.

Endurance test consisted seeting vcore 1.6,cache voltage 1.6 and vccsa 1.6(all on air,28 degrees ambient).Also disable vrm protections.Ran like these for hours,basically 32M some 2001,stuff.Got back to previous know settings,all good to go.0 degradation.Did this with 2 cpu's i was going to rma anyway.

On dice i went max with 1.8Vcore and 2.4VCCIN.Only for 2d tests and some 2001,temperature about -70 degrees.Ran for hours straights,0 degradation,no deaths.Truth is i did not test 4c8t intensive benches,this could be harmfull to them.

The way i see it,haswell cpu's are not that sensitive or it's either me i got some tough bastards over here

[cstkl1]

78 i got that when damaged cache(blasted some 1.8V on air to see how much does it take),setting dram clock period 1-14 does not damage the cpu,altough gains are minimal,usually a value around 5-8 is good,but lower you go in this you have to increase vcore.low trfc does not damage also,you get rainbows from graphic corruption resulted from memory errors.

I had tested so far about 5 cpu's,one on dice.

Endurance test consisted seeting vcore 1.6,cache voltage 1.6 and vccsa 1.6(all on air,28 degrees ambient).Also disable vrm protections.Ran like these for hours,basically 32M some 2001,stuff.Got back to previous know settings,all good to go.0 degradation.Did this with 2 cpu's i was going to rma anyway.

On dice i went max with 1.8Vcore and 2.4VCCIN.Only for 2d tests and some 2001,temperature about -70 degrees.Ran for hours straights,0 degradation,no deaths.Truth is i did not test 4c8t intensive benches,this could be harmfull to them.

The way i see it,haswell cpu's are not that sensitive or it's either me i got some tough bastards over here

interesting . but thats how i killed two procs. btw it was on prime and on air and not sp32m. Internal gpu disabled. Total prime time around 24-48 hours of stressing.

another sign was when it was closed to death the vcore on core 1-3 will be flux a lot from the set voltage. example 1.2v set on m6e norm is 1.19-1.22. but this one will flux up to 1.3.

but so far on the current proc been more careful and only increase voltages when necessary with small increments its lasting after nearly 72 hours worth of priming.

[owikh84]

I think we should create a "Haswell Death Report" thread.

[cstkl1]

Ueah n voltages of the two death were vcore 1.2-1.25. Temps below 95c. Vccin max 1.95. Cache voltage at auto. Digital/analogue i/o max 1.2v.

Both deaths were from ram timings or messing with vttddr.

All happened at dram clock 6 2250c7. New proc at 10 no issue so far.

[ChanceCoats123]

interesting . but thats how i killed two procs. btw it was on prime and on air and not sp32m. Internal gpu disabled. Total prime time around 24-48 hours of stressing.

another sign was when it was closed to death the vcore on core 1-3 will be flux a lot from the set voltage. example 1.2v set on m6e norm is 1.19-1.22. but this one will flux up to 1.3.

but so far on the current proc been more careful and only increase voltages when necessary with small increments its lasting after nearly 72 hours worth of priming.

These voltage fluctuations that you mention make me wonder about the durability of the internal VRM parts. I mean we have already been using 22nm tech from Intel without any serious, or common deaths at low voltages. The only real changes to this Silicon are related to the VRM, and since you mention a large fluctuation of core voltage, I see the internal regulation being the issue. Just my $.02.

[cstkl1]

These voltage fluctuations that you mention make me wonder about the durability of the internal VRM parts. I mean we have already been using 22nm tech from Intel without any serious, or common deaths at low voltages. The only real changes to this Silicon are related to the VRM, and since you mention a large fluctuation of core voltage, I see the internal regulation being the issue. Just my $.02.

me think-est the same. hence i even down the loadline from the recommended 8 to 6 on my latest proc as i find that is the most stable and close to what is set according to dmm read on load on the vccin

[stasio]

I think we should create a "Haswell Death Report" thread.

Not bad idea......

[RPGWiZaRD]

All this Haswell CPU dying talk has me worried. Makes you wonder if one shouldn't go Ivy instead xD BTW is there any airclockers around here, I see mostly WC setups... I don't like WC so will go air. Phanteks bigass cooler works equally good as the Corsair WC kits anyway it seems.

[Hazzan]

seem like L312B prety good bacth now..today I tested 5 CPU all can boot from bios 1.2 volt 4.7ghz 4C/8T looking CBB soon under Ln2

[chispy]

seem like L312B prety good bacth now..today I tested 5 CPU all can boot from bios 1.2 volt 4.7ghz 4C/8T looking CBB soon under Ln2

Good Luck Hazzan , I hope is a good one with at least -120c cb or better

[l0ud_sil3nc3]

seem like L312B prety good bacth now..today I tested 5 CPU all can boot from bios 1.2 volt 4.7ghz 4C/8T looking CBB soon under Ln2

Nice to hear, be careful though I have been killing everything that starts with L312B lol

The first good one (B332) was a 1.28v 5Ghz 4c 8t 32m chip and it died in like 4 hours.

The 4 pieces I picked up yesterday are L312B516 and L312B332.

[C-N]

New Bios on UD5H runs lower vcore

4.5GHz L308B202

[Hazzan]

Nice to hear, be careful though I have been killing everything that starts with L312B lol

The first good one (B332) was a 1.28v 5Ghz 4c 8t 32m chip and it died in like 4 hours.

The 4 pieces I picked up yesterday are L312B516 and L312B332.

I have already kill one George....

3 Chip already 5 Ghz 4C/8T Vcore 1.25 Volt

[ChanceCoats123]

Hazzan, do you have the full batch number? I've got a spreadsheet going on and you've not given enough information! :p

[cstkl1]

New Bios on UD5H runs lower vcore

4.5GHz L308B202

nice and at uncore 1:1 at 4.5ghz.

whats ure vccin , digital/analogue , vssa voltages??

btw ures delid?? ure min to max is pretty good if its antec kuhler 920.

[C-N]

nice and at uncore 1:1 at 4.5ghz.

whats ure vccin , digital/analogue , vssa voltages??

btw ures delid?? ure min to max is pretty good if its antec kuhler 920.

Thanks mate, its a good enough 24/7 clock for me.

vccin/vrin = 1.9 ioa & iod = +.050v vssa = stock

Yes its had a roof chop & has liquid ultra on the core & yes its on the 920. Normal silent idle is 35 ~ 40 then the fans kick up after 40oC

[chispy]

Well my only good Haswell just got wounded tonight and is dying slowly ,IMC is hurt badly and cpu wont clock to 6Ghz stable for 3d benchies anymore , now max for 3D is 5924Mhz , @ -115c , that's 3 completely dead cpus and one dying = 4 cpus in 3 weeks out of 6 cpus . I will stopped right now bining cpus , is not worth it , they are way too fragile and die too easily + low clocks + require massive voltage + Heat Monsters + very low cb + cbb = no no .....
I can safely say it , Intel Haswell it is a Big Flop. the only good thing are the awesome motherboads out there , amazing new features. but Intel drop the ball on us.

[FlanK3r]

so, last one is decrease now? Do you think, its CPU voltage or CPUIN voltage? I want not use more than 1.725V and I have a bit fear from VCCIN

[Vivi]

so, last one is decrease now? Do you think, its CPU voltage or CPUIN voltage? I want not use more than 1.725V and I have a bit fear from VCCIN

vccin is not the problem, whats been killing cpu's in my experience is too high uncore / cache volts. 1.65 on ln2 was working nice, above that started killing and degrading some chips. I might be totally wrong, but i didnt have any issues running 1.74vcore 2.85vrin/vccin on ln2.

[aussie-revhead]

but Intel drop the ball on us.

Thats to make us want Ivy-E when it comes out ...

[l0ud_sil3nc3]

vccin is not the problem, whats been killing cpu's in my experience is too high uncore / cache volts. 1.65 on ln2 was working nice, above that started killing and degrading some chips. I might be totally wrong, but i didnt have any issues running 1.74vcore 2.85vrin/vccin on ln2.

See I don't think it is cache voltage for me, as I have not used more than 1.6v under LN2 and input on the last dead one was 2.5, then only thing I have in common is high vdimm/vttdr i.e. >2.00v, so maybe thing a connection there?

[chispy]

See I don't think it is cache voltage for me, as I have not used more than 1.6v under LN2 and input on the last dead one was 2.5, then only thing I have in common is high vdimm/vttdr i.e. >2.00v, so maybe thing a connection there?

+1. Bingo , George I think you are right on that statement , my 4 chips that died were all running high v.dimm/vttddr , there is gotta be a conection somewhere along that line

@Flanker , maybe that also , but more accurate would be the above statement.

[ChanceCoats123]

I guess it's a good thing I don't have a kit of ram worth taking subzero then! This seems like a pretty strong correlation.

[cstkl1]

+1. Bingo , George I think you are right on that statement , my 4 chips that died were all running high v.dimm/vttddr , there is gotta be a conection somewhere along that line

@Flanker , maybe that also , but more accurate would be the above statement.

Think ure on to something here. One of the proc that died.. Found lowering vttdr n running high vccin can pass sp32m....
So the vttdr is linked to vccin??

Max vdimm i was doing was 1.7v only.