Intel Haswell Overclocking Clubhouse.

[cadaveca]

Prime95 pushes the load voltage up to 1.44v. But here's Spi32m, and again, it bounced around on the voltage and frequency just like at stock. It's not hitting throttling by a long shot, am I missing something?

http://img829.imageshack.us/img829/4359/48ddr2133.png

Nope, but you're using an offset voltage instead of a consistent one. For initial testing, it's better to test with a set voltage, find stability, and then take that knowledge to apply the offset.


When the multi changes, the voltage changes, as each multi has it's own VID associated with it. As well, there are other conditions that can adjust that VID, so rather than dealing with that for testing, simply avoiding that is best.

That's why I didn't include your OC voltage in the chart. It really could be anything, and as you can see, as load changes, so does the voltage, so who's to say what you are really running?

 

[Delta6326]

Dude, yeah, you'd think it would be one of the larger factors, for sure, but since I don't really have any inside line with Intel to ask these questions and wasn't part of the launch media, I couldn't tell you what's really the story. As you can see by the diagram I posted way up above, there's apparently only VCCIN and DRAM-in for power supplied to the CPU, but looking at these ROG boards, there's a whole huge whack of things to set, and a lot of other boards don't have these options.

That's why I made this thread in the first place. I need to know more, and if I do, so does everyone else. So let's see what we can find out.

I think I got the CPU multi binning part down. Roughly going as I expected(I've talked to a couple of PC builders and what they got for chips, too). But we still have iGPU, VCCSA/AIO/DIO, VCCIN, and others to figure out yet. None of those videos talk about any of THAT, really. What I did hear was a bunch of "We'll do that for you". As if that's good enough for me.

plus this whole "set 1.2V, try 4.6 GHz" seems..well...like something I told W1zz to do before the launch with his chip, before any of this "info" was out. Really. That's exactly what I told W1zz to do. I had tested three chips. ASUS et al apparantly tested hundreds of chip, and they cannot contribute more than I did with just three chips? WTF....

Those overclock guides contain ZERO info, IMHO. Anyone here could have figured that out. I did...

That's what I don't understand they should know everything about how haswell works. Which either they do, but won't tell or they wasted a lot of time... If I had a 4770K and a board I would help you out, but that won't be for a long time unless you can tell me when the Maximus VI Gene will be out and for how much

 

[cadaveca]

That's what I don't understand they should know everything about how haswell works. Which either they do, but won't tell or they wasted a lot of time... If I had a 4770K and a board I would help you out, but that won't be for a long time unless you can tell me when the Maximus VI Gene will be out and for how much

..Or simply..it's just too early.

I prefer to think it's just to early, so we have just as much a chance of compiling enough info to figure it out as they do. Maybe they need to crowd-source like we do.

 

[Delta6326]

It's never to early I want to know everything now

But yes you are probably very much correct, I'm just impatient.

 

[DOM]

idk whats wrong but now im getting vcore 1.048v ring 0.976v not its just changed to 1.056v 0.984v when the ss put it under 0c

any clue where I can get a beta bios seen 1.2B7 but cant find to dl it anywhere

 

[cadaveca]

idk whats wrong but now im getting vcore 1.048v ring 0.976v not its just changed to 1.056v 0.984v when the ss put it under 0c

any clue where I can get a beta bios seen 1.2B7 but cant find to dl it anywhere

I think the iVR changes it's behavior according to temps with some sort of throttling mechanism. I wonder then if how it changes might indicate how good a chip is for sub-zero.

I found Shamino's guide for M6E late last night. He's still working on it too, it seems.


CL3P0 on overclock.net says he gets the best results matching iVR to double the vCore. There is a mode for LN2 that can force vCore as high as iVR, and keep that and cache in sync, so it'd then kind of act like IVB I think. Balancing the load on the iVR seems pretty important, hadn't considered that really until reading Shamino's guide.

 

[DOM]

What's iVR haha ring volt ?

Most I tryed was around 1.380v and was able to bench nb at 5GHz

Not sure how high on the ring volt is safe so haven't pushed it more don't wanna kill it yet and leaving out of town today and won't be back till Fri Sat death in the family

 

[Womper]

Nope, but you're using an offset voltage instead of a consistent one. For initial testing, it's better to test with a set voltage, find stability, and then take that knowledge to apply the offset.


When the multi changes, the voltage changes, as each multi has it's own VID associated with it. As well, there are other conditions that can adjust that VID, so rather than dealing with that for testing, simply avoiding that is best.

That's why I didn't include your OC voltage in the chart. It really could be anything, and as you can see, as load changes, so does the voltage, so who's to say what you are really running?

I initially used manual settings, but since it still uses a higher voltage than what I set, it was just the same to use offsets.

Anyways, for what it's worth, here's 1.325v manual (BIOS and CPU-Z measure 1.344v). I could probably push that voltage lower, but it'll bluescreen under heavier loads even at 1.325v (which is why I'm questioning the utility of SuperPI results here). At the other end of the spectrum, there's P95, but that's not helpful either since no real program I know of can kick power up that high. Is Aida64 a better all-around choice? Load voltage is the most interesting to me, otherwise you have to say a) what your manual voltage settings was, b) what the BIOS measures the manual voltage as, and c) what your load voltage is. I'm not sure what a) and b) tell you that c) doesn't.

Also, I'll start manually setting my ring voltage to find out where it's at for 4.5GHz. Good thing I found the clubhouse here so I'm not shooting in the dark like before.

 

[cadaveca]

What's iVR haha ring volt ?

Most I tryed was around 1.380v and was able to bench nb at 5GHz

Not sure how high on the ring volt is safe so haven't pushed it more don't wanna kill it yet and leaving out of town today and won't be back till Fri Sat death in the family

iVR is the onboard voltage regulator, that splits the VCCin from board VRM into the different parts of the chip. It's right in the chip itself. I haven't investigated playing with that too much, but I've noticed a couple of times that when increasing volts on some boards VCCin automatically jumped up to 2.0 V from 1.8 V. Has to be useful to tweak that.

I initially used manual settings, but since it still uses a higher voltage than what I set, it was just the same to use offsets.

Anyways, for what it's worth, here's 1.325v manual (BIOS and CPU-Z measure 1.344v). I could probably push that voltage lower, but it'll bluescreen under heavier loads even at 1.325v (which is why I'm questioning the utility of SuperPI results here). At the other end of the spectrum, there's P95, but that's not helpful either since no real program I know of can kick power up that high. Is Aida64 a better all-around choice? Load voltage is the most interesting to me, otherwise you have to say a) what your manual voltage settings was, b) what the BIOS measures the manual voltage as, and c) what your load voltage is. I'm not sure what a) and b) tell you that c) doesn't.

Also, I'll start manually setting my ring voltage to find out where it's at for 4.5GHz. Good thing I found the clubhouse here so I'm not shooting in the dark like before.

http://img839.imageshack.us/img839/3606/48ddr21331344v.png

I'll update your entry on the chart.

I know, Spi32m is hardly a worthy stability test. Reading provided clocking guides and watching videos form the guys that do them, it seems that p95 is a special case and other stress tests may not get that higher VID to enable. Using Spi32m ensures that this vid doesn't enable, from my testing. I'll be investigating that too as I work through testing these boards. I noticed the jump on P95 myself with the first few days of testing, and had already been using wPrime for testing performance a bit already, so swapping over to wPrime's more consistent workload made a lot of sense, and I actually shelved P95 from my testing routine. I had been testing boards before, with at least 24-hour runs of P95...and if it failed, then I'd start over. In the past 8 months or so, I've had about 30 new tests into my testing routine that overall allow me to get a better idea of true stability, faster, than what P95 offers me. If that's just because my PC usage changed or apps are different, or these new instruction sets playing a role...lol... overclocking can be pretty easy...or pretty hard.


What cooling you got, and where'd you pick your chip up from?

As to AIDA64 being a better choice, ASUS is promoting it a fair bit, so they obviously find good value in it. AIDA64 is one app I trust implicitly, the guys/gals over there are very good at keeping things up to date the best they can, and are really good at giving support to paid subscribers. I have used/trusted AIDA64 and other software Fiery has worked on for many many years now.

 

[Womper]

I'll update your entry on the chart.

I know, Spi32m is hardly a worthy stability test. Reading provided clocking guides and watching videos form the guys that do them, it seems that p95 is a special case and other stress tests may not get that higher VID to enable. Using Spi32m ensures that this vid doesn't enable, from my testing. I'll be investigating that too as I work through testing these boards. I noticed the jump on P95 myself with the first few days of testing, and had already been using wPrime for testing performance a bit already, so swapping over to wPrime's more consistent workload made a lot of sense, and I actually shelved P95 from my testing routine. I had been testing boards before, with at least 24-hour runs of P95...and if it failed, then I'd start over. In the past 8 months or so, I've had about 30 new tests into my testing routine that overall allow me to get a better idea of true stability, faster, than what P95 offers me. If that's just because my PC usage changed or apps are different, or these new instruction sets playing a role...lol... overclocking can be pretty easy...or pretty hard.


What cooling you got, and where'd you pick your chip up from?

As to AIDA64 being a better choice, ASUS is promoting it a fair bit, so they obviously find good value in it. AIDA64 is one app I trust implicitly, the guys/gals over there are very good at keeping things up to date the best they can, and are really good at giving support to paid subscribers. I have used/trusted AIDA64 and other software Fiery has worked on for many many years now.

I have a Zalman CNPS10x Quiet cooler, and I bought the chip through Intel's employee purchase program 8^)

Maybe I'll give Aida64 a shot. P95 isn't helpful for stability testing at the clock speeds I want to use due to the heat. Gaming at 4.8GHz is easy, with temps hitting only 64C. But it would be nice to run into bluescreens sooner than 1 hr.

 

[cadaveca]

I have a Zalman CNPS10x Quiet cooler, and I bought the chip through Intel's employee purchase program 8^)

Maybe I'll give Aida64 a shot. P95 isn't helpful for stability testing at the clock speeds I want to use due to the heat. Gaming at 4.8GHz is easy, with temps hitting only 64C. But it would be nice to run into bluescreens sooner than 1 hr.

Really? that's kind of a poopy cooler, so you getting those clock is pretty nice, I think.

As to P95, I guess the problem in the end is that if it is making VID go higher than everything else, and everything else is actually getting a lower voltage, then truly P95 is useless.

So you test P95 @ 1.3V, but other apps are only getting 1.225V, and aren't stable, but P95 is. So, you set 1.3V, other apps are now stable, but P95 now gets 1.375V. So these guides recommend 1.25V max to make sure that you'll only give 1.325 V max...so maybe that should be the cap we aim for.


I think that other apps aren't fully using the chip like P95 does, or one of the new extensions changes how the iVR gives voltage. I think that's what I heard JJ say, I've gone over way too much info the last little bit.

I'm working through my apps and clocking and playing with stuff to see what I get. Then I try the other chip. I could probably go through this faster if I had more chips.

 

[allen337]

 

[Womper]

Really? that's kind of a poopy cooler, so you getting those clock is pretty nice, I think.

As to P95, I guess the problem in the end is that if it is making VID go higher than everything else, and everything else is actually getting a lower voltage, then truly P95 is useless.

So you test P95 @ 1.3V, but other apps are only getting 1.225V, and aren't stable, but P95 is. So, you set 1.3V, other apps are now stable, but P95 now gets 1.375V. So these guides recommend 1.25V max to make sure that you'll only give 1.325 V max...so maybe that should be the cap we aim for.


I think that other apps aren't fully using the chip like P95 does, or one of the new extensions changes how the iVR gives voltage. I think that's what I heard JJ say, I've gone over way too much info the last little bit.

I'm working through my apps and clocking and playing with stuff to see what I get. Then I try the other chip. I could probably go through this faster if I had more chips.

You disrespectin' my cooler?! But seriously, I'll need to see temps in the 90's on a meaningful workload before I consider an h100i or something. For 4.8GHz, I've been using adaptive Vcore with 0.200 offset and 0.050v additional turbo, which allows plenty of voltage headroom for those P95'ish cases (which will ultimately throttle). A game load only requires 1.3v, so that works.

P95 loads up all CPU execution lanes using those AVX instructions, so you really hit peak instruction throughput. I think it's safe to say that most real programs do something besides fast fourier transforms.

 

[v12dock]

I just realized that my BIOS is from april now I am eager to update

WOOT! Able to achieve 4.5Ghz now

4.5Ghz @ 1.285V | OC Cache x39 @ 1.15V | 75c load on Noctua NH-D14

 

[Womper]

For 45x cache, 1.25v is the ticket.

The Intel tool shows "Default" when I'm at default voltages rather than a value. So I don't know what my default cache voltage is yet.

 

[cadaveca]

For 45x cache, 1.25v is the ticket.

The Intel tool shows "Default" when I'm at default voltages rather than a value. So I don't know what my default cache voltage is yet.

http://img198.imageshack.us/img198/2258/45core45cache126c125u.png

Temps are getting up there a bit. Did you track the increase?

What about the performance difference?

 

[Womper]

Temps are getting up there a bit. Did you track the increase?

What about the performance difference?

Maybe I'll run P95 large FFT with 45x cache @ 1.25v and default cache ratio/voltage and compare temps.

More data:
45x cache @ 1.23v will reboot under load.
46x cache @ 1.30v will reboot under load.
46x cache @ 1.325v seems stable. I ran P95 blend with a single core at 46x, the other cores at 43x for a while without issues (didn't throttle). All cores at 46x works, but throttling occurs.
47x cache @ 1.350v and 1.375v both reboot under load. I'm giving up on 47x, that's just too much voltage.

Also, for fun I used the Intel tool to test 45x cores at 1.25v, 1.24v, 1.23v, 1.22v, and 1.21v core voltages. I ran P95 large-fft for 6 passes on each, and only 1.21v caused a reboot. I ran P95 blend on 1.23v, and it ran for 30 minutes or so but crashed as soon as I tried to come back from screen saver. I saw 98C as my max core temp just before it shut down. I still need 1.25v to 1.26v on the core to safely boot into windows, but it was still interesting.

Here's a a picture that includes the Intel tool. The Intel tool is just reporting the BIOS voltage settings.

 

[TheHunter]

Just received mobo+cpu today., finally


Hmm my cpu is a black sheep lol

i haven't seen this batch before >>> L308B202<<<,

hopefully it will OC ok without to much volts, I will see tomorrow when i'll put it together

 

[Jstn7477]

24 hours until my 4770K gets here and is installed.

 

[Womper]

Maybe I'll run P95 large FFT with 45x cache @ 1.25v and default cache ratio/voltage and compare temps.
http://img13.imageshack.us/img13/4884/48ddr21331344c125u.png

Temps showed no difference in P95 large FFT. Planetside 2 at 46x cache @ 1.325v was perfectly stable and stayed under 64C just like 45x cache @ 1.25v.

Meanwhile, simply dropping DDR from 2133 to 1600 will reduce max temps by a good 7-10C, and dropping core voltage sees immediate gains as well. Looks like the cache is either not a big heat producer, or tough to really stress.

Anyone have a good test that should benefit from cache speed?

 

[HammerON]

This is at stock. Temps are a bit high for me
However I do like the low voltage

 

[FireKillerGR]

Just got my new chip:
Batch # L312B508
Stock Voltage: 1.02V
Stock VRing/Cache Voltage: 1V
Stock V Input: 1.754V
4.5 GHz -> 1.22v
4.6 GHz -> 1.28v
4.7 GHz -> 1.30v boot need 1.35 for most benchmark. Didnt try more
5.0 GHz -> 1.45v under bad SS
5.1 GHz -> 1.50v under bad SS

Uncore -> I tried x41 without problems. Didnt need for more.

 

[razaron]

Replacing the stock el cheapo die TIM with Coollaboratory Liquid Pro seems to drop temps by 20C, sometimes more.
Unfortunately, chips that struggle with 4.5GHz with the stock TIM probably won't benefit much from the better temps, from what I've seen.

 

[radrok]

Just got my new chip:
Batch # L312B508
Stock Voltage: 1.02V
Stock VRing/Cache Voltage: 1V
Stock V Input: 1.754V
4.5 GHz -> 1.22v
4.6 GHz -> 1.28v
4.7 GHz -> 1.30v boot need 1.35 for most benchmark. Didnt try more
5.0 GHz -> 1.45v under bad SS
5.1 GHz -> 1.50v under bad SS

Uncore -> I tried x41 without problems. Didnt need for more.

for bad SS you mean 200w load capable unit?

 

[FireKillerGR]

Its a good ss, goes @-42~ C but need to refill it cause under load it goes @-10C :/