Intel Haswell Overclocking Clubhouse.

[Vego]

btw, Ram has a little perf impact, i noticed it in cinebench , any 3dmark physics cpu score, cpu bound games when not gpu limited..
http://www.xbitlabs.com/articles/memory/display/haswell-ddr3_7.html#sect0

to be 100% honest i play only LoL now so i dont need much more fps
always had a better pc than i needed, meybe exept TERA - there no matter what pc u have the game doesnt allways give you smoth performance, especialy when u like max details at 2560x1440 resolution like i do

 

[v12dock]

Just delided two 4770k's hammer and vice with 100% success. Replaced shit TIM with decent antec Formula 7. I am able to run 1.385v@4.5ghz <85c IntelBurnTest 100% stable so far. AIDA64 stable test 6h avg temp 65c.

 

[MxPhenom 216]

Just delided two 4770k's hammer and vice with 100% success. Replaced shit TIM with decent antec Formula 7. I am able to run 1.385v@4.5ghz <85c IntelBurnTest 100% stable so far. AIDA64 stable test 6h avg temp 65c.

Then the chip will die in a month..

 

[v12dock]

Then the chip will die in a month..

Just in time for 4790K

 

[petedread]

Just in time for 4790K

I'm sure I have read somewhere that Devils Canyon has been put back to a mid summer launch. Next month we get a 4790s and a 4790 but no K version. And Devils Canyon won't work in Z87 boards. Please some one tell me this is old news and no longer true. I'm dying to get me a Devils Canyon chip.

 

[TheHunter]

I'm sure I have read somewhere that Devils Canyon has been put back to a mid summer launch. Next month we get a 4790s and a 4790 but no K version. And Devils Canyon won't work in Z87 boards. Please some one tell me this is old news and no longer true. I'm dying to get me a Devils Canyon chip.

4970K is set for June 2nd 2014.. Just saw this @ guru3d




Z87 is compatible, all big mobo manufacturers released IME Firmware uefi update.
My mobo Asus Z87-deluxe (c1) is ready for it since February, first was MSI, 2nd Asus, then Gigabyte and the rest Asrock, Evga.



I think I'll keep this 4770k @ 4.7ghz 1.274v for now, looks plenty. Atm downclocked to 4.5ghz @ 1.17v and still overkill in everything Maybe I'll upgrade to Broadwell in 2years


Speaking of Broadwell, it should be compatible the same way, firmware update..

Haswell IME 9.0.2.xxxx
Haswell refresh IME 9.0.30.xxxx
Broadwell IME 9.1.xxxx or maybe 10.x.xxxx (both uefi firmwares are already @ stationdrivers)

what is IME, etc
http://intelmanagementengine.com/intel-technologies/all-about-intel-management-engine-interface.html


Intel Ceo/rep said it himself back in mid September 2013, it will be compatible with existing systems aka Z87 and brand new systems aka Z97,..
http://www.youtube.com/watch?feature=player_detailpage&v=KeDtXucTwRI#t=22

 

[petedread]

Thank you TheHunter. It is nice to have that finally cleared up. I can plan to put a 4970k in my Z87 board that I love. If My haswell was as good as yours I would probably keep it too. The only reason I'm looking forward to devils canyon is because 4.5/4.5 takes 1.270v/1.225v (cache). 1.170v for 4.5ghz is a dream chip, probably one in a million.
Thanks for the links.

 

[vega22]

Then the chip will die in a month..

why?

 

[cadaveca]

why?

Been a few reports of dead IMC after de-lid, some from people that I trust 1000000%. Dunno if it's luck of the draw, a cooling issue, or what... but there is a chance that de-lid can make a chip die quickly. I'd say 10%-15%, given the numbers I have heard.

 

[vega22]

i don't doubt your sources dude, it was more the certainty of the post i quoted which i was questioning.

the question that springs to mind with what you're saying is, was it the delid which killed them, or the insane volts people will give them once they see how cold they run?

only i know more than a few who have delid their's too and yet none of them are dead either. i know a few who have get them cold and they killed them almost instantly without a delid.

if i was a betting man i would say luck of the draw more than delidding kills.

 

[cadaveca]

i don't doubt your sources dude, it was more the certainty of the post i quoted which i was questioning.

the question that springs to mind with what you're saying is, was it the delid which killed them, or the insane volts people will give them once they see how cold they run?

only i know more than a few who have delid their's too and yet none of them are dead either. i know a few who have get them cold and they killed them almost instantly without a delid.

if i was a betting man i would say luck of the draw more than delidding kills.

Of course.

 

[vega22]

well. depending on the ape who is delidding

i can't imagine the ic would work for too long if it was hit with a hammer you know.

 

[MxPhenom 216]

I think ive seen people run delid chips naked, and then they put cooler on, and too much pressure cracks the die.

 

[TheHunter]

Was playing around with Cinebench15 @ 4.6ghz, cache 4.2ghz, ram 2400mhz oc

best score @ 4.6ghz so far

 

[MxPhenom 216]

@TheHunter Do you find that increasing cache multi improves performance a bit in benchmarks like these? I have mine at 39 when I have it overclocked (I am going to update to the newest BIOS and start over when I get back to school/my apartment) and was thinking about putting the cache to 42 as well. What kind of cache voltage did you need for 42?

 

[TheHunter]

Yes it scales up to ~ 41x - 43x (tested in Cinebench15) after that I saw no performance gains,. 42x seems like the golden middle


for 41x - 1.123v
for 42x - 1.137v
for 43x - 1.149v
for 44x - 1.165v




by all there is a 0.002v + offset for some reason. Ie 42x @1.137v can be max 1.158v by full load.

 

[MxPhenom 216]

Yes it scales up to ~ 41x - 43x (tested in Cinebench15) after that I saw no performance gains,. 42x seems like the golden middle


for 41x - 1.123v
for 42x - 1.137v
for 43x - 1.149v
for 44x - 1.165v




by all there is a 0.002v + offset for some reason. Ie 42x @1.137v can be max 1.158v by full load.

Okay, I dont think my chip is nearly as good as yours since i need 1.125v for 39x multi. Ill probably take the cache voltage to like 1.2v to start for 42, and lower till it crashes in my testing.

 

[cadaveca]

Yes it scales up to ~ 41x - 43x (tested in Cinebench15) after that I saw no performance gains,. 42x seems like the golden middle


for 41x - 1.123v
for 42x - 1.137v
for 43x - 1.149v
for 44x - 1.165v




by all there is a 0.002v + offset for some reason. Ie 42x @1.137v can be max 1.158v by full load.

Shamino suggested that cache @ 3-4 multis below CPU speed is most efficient. So @ 4.6 GHz, 42-43 makes sense. Increase CPU multi, and then the gains should scale upwards again. Haswell cache is nearly 25% faster than IVB, and is where most of the performance gains offered by Haswell are contained. But the question remains on how much it affects temps and power consumption...if lowering cache allows you to increase CPU multi, you should lower cache.

 

[TheHunter]

Shamino suggested that cache @ 3-4 multis below CPU speed is most efficient. So @ 4.6 GHz, 42-43 makes sense. Increase CPU multi, and then the gains should scale upwards again. Haswell cache is nearly 25% faster than IVB, and is where most of the performance gains offered by Haswell are contained. But the question remains on how much it affects temps and power consumption...if lowering cache allows you to increase CPU multi, you should lower cache.

I used the same 4.2ghz by 4.7ghz and it looks to be at its full potential too (same as 41x),

Tried 4.4ghz but it scaled negative in Aida64 memory benchmark, it showed lower bandwidth by L2 & L3, mostly ready and copy. Same by cinebench15 I got 2-4points less. Idk maybe it didnt give it enough volts..

also yeah anything higher heats up more., 4.2ghz vs 4.4ghz can save me ~3-4C on hottest core0.


Speaking of this hottest core, why is core0 always hotter then core3, is it because its so close to System agent controller or poor IHS contact?

ie Full load core0 68C, core3 59C.

 

[cadaveca]

Speaking of this hottest core, why is core0 always hotter then core3, is it because its so close to System agent controller or poor IHS contact?

System Agent doesn't produce that much heat, really. It's cache and iVR that produce the most heat.

It could be has IHS TIM, sure. It could just be the iVR, or a hot-spot in cache. The inclusion of the iVR, and it's lack of a temp sensor, is why I recommend NOT delidding Haswell CPUs.

 

[LAN_deRf_HA]

Has anyone else found these chips hard to cool at the top end? My friend just went to a H110 on his 4770k, runs cooler than his H80i and about the same temp as my 3770K in normal gaming, like say Crysis 3's first level. Then there's level 2 which has a lot of grass, and with a custom config to speed up the grass rendering and increase the amount of grass cpu usage goes to 100%, hitting 100C. Was the same with the H80i, totally different mounting systems. Just seems like the top end of the thermals are nuts on this chip. Reminds me of my 920. No matter what cooler I used nothing could bring it under 90c.

 

[MxPhenom 216]

System Agent doesn't produce that much heat, really. It's cache and iVR that produce the most heat.

It could be has IHS TIM, sure. It could just be the iVR, or a hot-spot in cache. The inclusion of the iVR, and it's lack of a temp sensor, is why I recommend NOT delidding Haswell CPUs.

I have been contemplating the idea of using the Ivy Bridge/haswell Naked kit from EK for my chip, but I would really only do it if I had enough funds to buy a back up chip or 2 if things go south either from the delid (hammer vice method) or from running it delidded.

Has anyone else found these chips hard to cool at the top end? My friend just went to a H110 on his 4770k, runs cooler than his H80i and about the same temp as my 3770K in normal gaming, like say Crysis 3's first level. Then there's level 2 which has a lot of grass, and with a custom config to speed up the grass rendering and increase the amount of grass cpu usage goes to 100%, hitting 100C. Was the same with the H80i, totally different mounting systems. Just seems like the top end of the thermals are nuts on this chip. Reminds me of my 920. No matter what cooler I used nothing could bring it under 90c.

Yeah they get hot. I found that the temps on my chip dropped a bit after a BIOS update weird enough.

Depends a lot on the settings you use for overclocks too.

Cadaveca can give you much more knowledgeable info on this stuff. The complexity of Haswell is beyond me.

 

[cadaveca]

Cadaveca can give you much more knowledgeable info on this stuff. The complexity of Haswell is beyond me.

Ha! Thanks. You can handle it, but I think school is first in your head and you are focused.

And yeah, they are hard to cool. The iVR makes it so much worse than Ivy, and the fact we can get similar or better performance within the same power envelope is kinda nice.


Working on Haswell OC Guide v 2.0...

 

[Kursah]

I have found it odd that my settings with the previous BIOS are no longer OCCT PSU stable. It hard locks every time...I'm actually thinking about going back as this update didn't seem to give me anything really. Everything else I do is totally stable...buuuuut it's just kind of an annoyance when it was stable prior. My default voltages in BIOS are different..where they read 1.15 before, both read 1.024 for core and cache.

Looking forward to reading and sharing Dave's Haswell 2.0 OC guide!

 

[TheHunter]

Well latest V1802 has "bugged" c-states again if enabled. Cpuv drops to 0.7v though, but not c-states.

v1707 worked properly and turned off c-states once cpu hit light load and then returned to c3 (if I enabled all c-states instead of auto), but now with v 1802 it doesnt enable any c-states anymore, kinda like c-states @ auto. 1602 kept using c3 even by load and that was a bit unpredictable.

If im at balanced windows plan it starts to use even c7 but then after idk 10sec it turns them all off completely (checked with RealTemp TI).


No big deal though, cpu stays cool @ c-states -auto or enabled, but I have to make sure I use balanced plan ( c-states - auto) when in idle or it runs at fixed voltage/max freq. all the time then.