980X 'safe maximums'?

[Zithras]

First time overclocking one of the (newish) i7 chips, and I've got lots of questions (thanks in advance to the incredibly helpful and patient people on this forum!)

I've already read the stickied 3-step guide to iX overclocking, and its answered most of my questions as to how i7 overclocking is different than the last few chips I tried to fry

Sooo...most of my questions are related to what people have been seeing performance-wise for this particular CPU (I tried searching for 980x in thread titles, and I'm sure this information is out there, but I couldn't find it in the two pages of results that came up).

First, some partial system specs:
980X on watercooling (AK Supreme HF), hyperthreading enabled (because, well, why not)
Rampage III Extreme MB
2x G.SKILL Trident 6GB (3 x 2GB) DDR3 2000 RAM F3-16000CL9T-6GBTD (12 GB total RAM, registers at 1066 with MB settings on auto but is designed to be overclockable to 2000)
3x5870 on watercooling, other assorted parts that shouldn't affect a CPU overclock (I'll overclock the GPUs next, but one stap at a time)


As far as use goes, I'll keep the computer on 24/7, use it daily, and it will probably be folding (whenever I actually remember to start the threads) when I'm not using it.

Now, questions!

Overclocking questions:

1. The guide mentioned on i7 chips I should try and find the max bclock first, and then mess around with the multiplier. I know that on other unlocked multiplier chips in the past, some people have gotten better performance with very low blocks and high multipliers - is this a better option, or should I stick to the testing order in the guide? (max block, then increase multiplier until temps and voltages get to the limits)


980X specific questions (i.e. what have others found while testing?):

1. What's the maximum 'safe' temperature for 24/7 use on the 980X? Qxxxx chips can be little heat factories, and are safe at 90C. AFAIK, doing this with the i7s is a bad idea. What should I stop at here? 70C (although that seems extremely conservative - maybe 75-80C would be better)?

2. Voltages: What are the maximum safe voltages on this chip (going by the intel spec sheet is one option, as always, but going by what forum members have found through testing is always better ):
Vcore: (not sure what's a good maximum here, although the watercooling should help poke it up to a tiny bit more than an air setup)
QPI/VTT: (I've heard that the i7 chips have not-so-great IMCs, so no clue what to use as a max here):


Memory questions:
I'm guessing I probably want to follow the recommended manufacturer specs: 9-9-9-24, 1.65V, as close to 2000 as possible.
Should I bump memory voltage up a bit to maybe 1.7 or even 1.8, or will that likely result in frying? The Gskills are designed to OC quite high, so there may be a bit of wiggle room here (on the other hand, ram is easy to destroy)?


Video cards questions:
I'll do the CPU and memory first, then post video card overclocking questions in another forum if needed - OCing video cards is generally much simpler than CPU/memory anyways)

If I can get the CPU to 4.5Ghz, I will be absolutely estatic (and shocked). More realistically, 4.0 is my 'minimum acceptable standard', and I'll be quite happy with a final OC of 4.2-4.3; we shall see what I get...(I'm also hoping for a 33% video card increase given the temps I'm getting on water and the OC tests I've read, but that's a different step)

Summary:
a) what's the max temp I should keep the CPU at under load?
b) What's the max settings I can use for: VTT, VCore, Memory
c) Do I want a high block and then increase the multiplier, or do very high multipliers work better?

Thanks for the help!
Zithras

 

[JenBell]

ras...let us begin:
1. Stick to the guide. Now remember Intel change some of the voltages for 32nm Gulftown's. 24/7 most people will say 1.35vcore. I'd actually be inclined to go to maybe 1.36 if I the OC was worth it. It is worth noting though that the r3e mobo (have 980, r3e and 12GB of 2000C8 RAM) does have the auto or crazy OC to 4.00 and 4.2Ghz. Now when u enable these settings the BIOS automatically sets the vcore to 1.4 (gave me a heart attack the first time I saw it)...still I booted into windows and because I had the thermal controls still enabled the CPU vcore would only rise to 1.4 when I was benching, otherwise it stayed around 0.9 most of the time.
2. I dont know off the top of my head but honestly. If u are running it 24/7 like I do then u want to remain as close to Intel spec as possible. This does not mean u dont adjust the voltages, just that u remain cautious as when u set u will probably leave for a long time. Google 980x overclocking...I also found some useful tips via youtube.

Few comments. U have 12GB of RAM. Like my corsair its rated to run at 2000mhz. Realistically, be very happy if u can get to 1800-1900 and stable. OCing to 2000mhz is hard enough on single set of DDR3s. U also have multiple GPUs so I am pretty sure will need to ensure u use the both molex connections on the mobo for additional power and ensure that as u OC that enough vcore it provided.

Temps. Subjective but there is always common ground. Personally, I'd keep it around 70-75 range. People might be happy with more...for 24/7 use then anything over this on full load is not something I am happy with.

Multi or blck...combination rudi. Easy 4Ghz is via multi usually but as u have the r3e I'd go down the blck route...its way more fun and seeing blck over 200 always makes u smile like a little kid;-).

I am sure the people posting after me will be able to give more direct technical and informative responses to ur questions.

Just a note of temps. If I OC my CPU to 4.4Ghz via 235blck x 19multi, 1.36vcore on my 720 xt setup I hit around 74c under load. System is on 35% power so got 65% in reserve...saving the rest for when I get bored and really want to go kamikaze on the whole thing;-). No OC on stock its 44c under load (still on 35% WCing power).

 

[Zithras]

Thanks JenBell - The system has a 1200W PSU and all the MB molexes are in, so there should be plenty of power available. Temperatures I'll keep an eye on, but am not too worried about - I doubt it'll end up with temps being the limiting factor on the OC of any of the components. 70-75C sounds good for the CPU, and I know the GPUs can get to about 90C. I use the computer occasionally for gaming, and video editing, but the most intensive task it's going to be doing most of the time is folding. That being said, I'm planning to start and stop the folding tasks manually, so if it's folding, or if I forget to start it folding and am not using it, it's not going to be doing anything else. Torrenting, chatting, and suchlike will get outsourced to the 'old' computer this one's replacing

According to the Intel spec sheet, the VTT and Vcore both have a max of 1.4. Yipes!

http://download.intel.com/design/processor/datashts/323252.pdf

Is it really okay to run the processor with that much juice? The spec sheets usually are conservative, not high, but this seems wrong. Maybe I'm looking at the wrong thing?

Also, any advice on memory voltage?

Edit: Reading more pages, it looks like setting VTT and VCore at 1.4 ea. might indeed be safe, but it should go no higher. Still not sure on the memory voltages, but it seems ocing 6 dimms may be problematic Ofc, Gskill does guarantee 2000, so if it's not hitting anywhere near there (i.e. not even hitting 1600 at 9-9-9-24), I can always return it and get new RAM...Thoughts?

Update: Current planned 'maximums': (PLEASE PLEASE PLEASE comment, especially on the voltages!)
Vcore: 1.4V
VTT/QPI: 1.4V
VDimm(RAM voltage): specs say 1.65V, so maybe 1.7V (not sure here)
CPU Temp: 75C(ish - I'm not gonna be too worried if 1 core it hits 76 occasionally)
GPU Temp: 90C (from what I've heard, I'll hit the upper limits of stability loooong before this)

 

[JenBell]

Vcore = cool. I'd aim for lower pesonally but if ur comfortable then its all good. I go according to 24/7 use. If I was benching specific then I'd happily go 1.45v if my temps are good. This is only for benching though. Vdimm. 1.66 would my personal limit but if I knew that 1.67 ot 1.68 would give me 2000 solid with 12GB then I would seriously consider it but my common sense would bring me back down I think in the long term.
GPUs dont run freekin hot I cannot comment of the best temps - I am sure someone can answer that for us? VTT/QPI at 1.4 also seems to be a setting done by the mobo when doing the crazy OC to 4.0/4.2Ghz. U got a PM.

 

[Bobnova]

If your vcore is inside the Intel recommended maximum, temps don't matter that much.
That said, running it right next to shutdown temps (~100*c) will shorten it's life. In theory it has a 20 year life to start with, how much it shortens it is unknown.

Personally, i would like to see <90*c for 12 thread P95/IBT, and <80*c for daily stuff. That's just me though.

 

[Zithras]

My guess is that I'll want to use this computer for 4-6 years before it becomes woefully obsolete. I think I'll try it with both 1.375/1.375/1.66/75C and 1.4/1.4/1.67/80C VCore/VTT/VDimm/CPUTemp ; If there's a big difference, or if the latter can hit 4.5 stable and the former can't then I'll go with it, else use the more conservative figures.

Again, anyone with any insight into this is more than welcome to comment

 

[Falkentyne]

I'm going to say something right now.

If anyone noticed, the "absolute maximum" voltages for the core i7 non gulftowns (45nm) were 1.55v, and the absolute maximums for the core i7 980X is 1.4v. Please note that there was a time when Intel listed the "functional voltages" and the absolute max in the same pdf. it seems that they stopped doing the "functional limits" when the core 2 series came out, instead just giving a VID range possible.

The "functional limits" were the voltage ranges where long term reliability of the processor were guaranteed. Now, there was a range OUTSIDE the functional limits but BEFORE the absolute maximums, where long term reliability and functionality of the processor were NOT guaranteed, but if the processor was returned back to its functional limits, it would still function, but *might* have its long term reliability compromised. Then it said that if a processor were operated outside the absolute maximum voltages, then its long term reliability and functionality *WOULD* be compromised, even if the processor were later returned back to functional limits.

The absolute maximum, in this explanation on the old pdf's, were supposed to be the absolute highest range that the processor could be run at, for SHORT periods of time. But Intel completely removed the range BETWEEN functional limits and absolute limits, for some reason, on all newer pdf's starting with the core2, I think.

Putting 1.40v into the core i7 980X, is basically, according to the "Absolute Maximums", the same as putting 1.55v through a core i7 920. If we are going "by the book". So if you're on air, and you feel COMFORTABLE with putting 1.55v into a core i7 920, on air or on water, then that should be the "same effect" as putting 1.4v into the 32nm gulftown, with the only difference being lower temps from the die shrink.

We won't know if 1.4v on air or water is safe for these processors for awhile; when someone reports serious overclock degradation from 24/7 1.4v @ 4.5 ghz on air feedback over months, then we will know.

 

[Zithras]

Okay I set the VTT to 1.375, the memory to its minimum settings, and the multiplier to 12X.

Bclk 220 will pass Prime95, but 222 won't post at all.

I then raised VTT to 1.4, expecting to get a bit more bclk out of the chip.

Again, 220 passes all tests, 222 won't post.

I think I'm running into a BIOS limitation - is there anything that might help? (If I need to post bios settings, please tell me what to post - pretty much everything else it on auto, HT is enabled, all the C state stuff is disabled).

(Good point Falkentyne - I'll mostly likely stick with the 1.375V overclock on this CPU unless I have a good reason not to - I wanted to test out both profiles and see the difference though.)

(oops forgot to mention: temperatures definitely aren't the issue here - they're well below limits - maybe if I play with the RAM voltages a bit...)

 

[Mickster]

Some "deja vu" reading there Zithras - i too have had the same results with my 980x and R3E. I can get the bclk at 220 with vtt at 1.35v, anymore than that is irrelevant to my bclk -it wont budge!.
If i try running with a cpu multiplier of 20, the vcore has to be in excess of 1.4v for primes95 to run stable. For now ive dropped the multi to 19 for 24/7 but will continue to play about with different combos. Im running a custom WC cpu loop so reckon ive still plenty of headroom as temps are only high 40's - low 50s @100% load. Bit frustrating as i was hoping to hit at least 4.4Ghz without resulting to a vcore over the intel spec.
- let us know how you get on

 

[MIAHALLEN]

Okay I set the VTT to 1.375, the memory to its minimum settings, and the multiplier to 12X.

Bclk 220 will pass Prime95, but 222 won't post at all.

I then raised VTT to 1.4, expecting to get a bit more bclk out of the chip.

Again, 220 passes all tests, 222 won't post.

I think I'm running into a BIOS limitation - is there anything that might help? (If I need to post bios settings, please tell me what to post - pretty much everything else it on auto, HT is enabled, all the C state stuff is disabled).

(Good point Falkentyne - I'll mostly likely stick with the 1.375V overclock on this CPU unless I have a good reason not to - I wanted to test out both profiles and see the difference though.)

(oops forgot to mention: temperatures definitely aren't the issue here - they're well below limits - maybe if I play with the RAM voltages a bit...)

Some "deja vu" reading there Zithras - i too have had the same results with my 980x and R3E. I can get the bclk at 220 with vtt at 1.35v, anymore than that is irrelevant to my bclk -it wont budge!.
If i try running with a cpu multiplier of 20, the vcore has to be in excess of 1.4v for primes95 to run stable. For now ive dropped the multi to 19 for 24/7 but will continue to play about with different combos. Im running a custom WC cpu loop so reckon ive still plenty of headroom as temps are only high 40's - low 50s @100% load. Bit frustrating as i was hoping to hit at least 4.4Ghz without resulting to a vcore over the intel spec.
- let us know how you get on

That's plenty

You guys are off to a great start and Zithras, YOU are asking all the right questions

I'll try to follow this thread more closely....you're inspiring me to make another revision to my OCing guide Thanks, it's much needed.

Back to my first comment....for 24/7 use, 200MHz bclock is plenty The old 220MHz bclock limit is very well known, and there is no for-sure cure for it. There are many things to try, but nothing guarenteed to fix it But, that's OK....go back to 200 and work on the memory/CPU (steps 2 & 3 in the guide) OCing now

As far as my opinion on voltages??? With Bloomfield/Lynnfield, I was very liberal with voltages, and only ever killed one chip (a really nice i5 750 ). But with Gulftown, I'm a little more careful. Probably because they're so expensive, but also because they have had a higher failure rate (I've killed three 980x CPUs already). Personnaly, I would limit myself to 1.4/1.4/1.7V for 24/7 use as long as the temps are below 90C for LinX stress testing.

I've got an i7 970 on the way that I'll be doing a bit of stress testing on, so I'll be playing along with you

BTW - the 980X CPUs I killed all died with around 1.8-2.0V with extreme cooling while I was benching at over 5.6GHz....so don't be scared by my comments earlier.

 

[Zithras]

nice to know I'm not the only one who's running into this wall Thanks for the comments Mihallen - I'm definitely not running it above 1.4 - here's to hoping I can get good performance out of 1.375

Not much to post; just a quick update:

Memtest is currently running - I prefer 6 passes for stability checking, but that'll be on my final settings - 2 is good for basic testing, since they take forever with this much RAM

Anyway, VTT is at 1.375 for now, since 1.4 is rather pointless.
Bclk 220 gives me RAM stable at 9-9-9-24 @ 1763
bclk 200 gives me RAM stable at 9-9-9-24 @ 2005
(the ram's rated for 2000, and at 220 bclk the the options are 1763, and over 2100(which doesn't post)

now to start on the multipiers:
so far I want to test (for highest multipliers):
vcore at 1.4 and 1.375, with both 220 and 200 bclks.
will also experiment with raising vtt once vcore bclk and mult are set, and see if that has any effect, but it seems like it shouldn't.

 

[MIAHALLEN]

Extra VTT may help to get your memory running faster/tighter, but it won't help CPU clocks

Hey man, if you have a camera, while your doing all this stuff could you take a few minutes and help me out?
http://www.overclockers.com/forums/showthread.php?t=653470

 

[Zithras]

Well. I must say that I am rather disappointed. The 980X performs amazingly well at stock settings, compared to other chips on the market right now, and has an impressive 6 cores. Still, Intel Extreme Edition chips are MADE for overclocking, and it is a sad, sad day when Intel releases a modern chip that can't even hit a 1 GHz overclock under whats pretty much the very best non-extreme cooling.

After spending the night tinkering with settings, here are my final results.

This chip and RAM is stable at 4200 Mhz, bclk 200, mult 21.0, 12GB DDR3 RAM (6 sticks) 9-9-9-24@2005 Mhz
Vcore and VTT are both 1.375, RAM voltage is 1.66950 (as close as I can get to the manufacturer's recommended 1.66) (Ram was 6+ hours memtest-3 passes stable before I stopped it, CPU was smallfft Prime stable for 2.5 hrs before I stopped it) It didn't want to hit 2100+, but seems to have no problems with 2005.

Trying to get to that 1 Ghz overclock of 4.33, I pushed the vcore slowly up. Eventually, at 196 bclk, multiplier 22, 4.312 Ghz, Vcore 1.45 (!) and VTT at 1.4, I got the computer to run Prime stable for all of 7 minutes and 12 seconds before a thread crashed due to a rounding error. At no time did the temp on any core go over 66C, so that's not an issue here. I searched around quite a bit more for any other islands of stability but no dice.

Seeing as this was going nowhere fast, and having quite exceeded the Intel limits (1.45 vcore is probably not good for daily use anyway ), this gives the final processor stats for the overclock:

VTT: 1.375V
Vcore: 1.375V
VDimm: 1.66950V
bclk: 200
mult:21
CPU final speed: 4200Mhz with 6 cores (12 virtual)
RAM timings: 9-9-9-24@2005 Mhz
Now for a 24 hour Prime95 test and 6+ pass memtest86+ to guarantee rock solid stability, then on to the graphics cards (final step).

So. It seems this processor seems to have a pretty solid wall around 4.2. Admittedly, LLC is disabled, and HT is enabled, but that's really no excuse for failing a 1 GHz OC on an extreme chip even when voltages were pushed fairly high.

In any case, I'd like to thank everyone who posted offering suggesions, advice, and ESPECIALLY voltage information, and Miahallen specifically for the wonderful guide that saved me a TON of time trying to figure out this new chipset; it turned out to be not all too different after all->I think the ix series is actually easier to overclock with the new architecture! I went ahead and posted R3E BIOS screenies in the thread you mentioned - tell me if you have any other questions/comments/requests and I'll do by best to help out

If there's anything obvious I'm missing, please feel free to chime in, but I think I've gotten all I can out of this CPU wthout resorting to LN

On the other hand, I'd like to recommend GSkill's Trident RAM to any interested parties - it seems to be one of the few 2000 MhZ modules out there that can reliably reach its advertised speed and timings under some pretty demanding conditions! (I haven't run the extra long memtest yet to doublecheck, but after almost 7 hours and over 3 passes, it was going great!)

Zithras

Edit: Is there any way to let the processor transition to a lower multiplier when it doesn't need the extra power, but still have no impact on stability (and keep 'turbo mode' off)? It'd be nice if it didn't run up my electricity bills unnecessarily What if I disable (keep disabled) 'CPU Turbo Power Limit' but re-enable C1E support, SpeedStep, and C-STATE Tech? I think this would allow it to adjust its multiplier lower (but not over the 21 I set in the BIOS), but I'm not sure if this could make the system more unstable? (The TM function is also currently enabled, but ofc will never come into effect unless the pumps break - I think the load temps are barely 50C right now)

Edit 2: What's the current wisdom on prime settings? I've always used smallfft for maximum CPU stress, and relied on memtest for RAM testing. I've heard that Memtest86+ is no longer a completely reliable indicator of RAM stability on Windows anymore though, and so some people are using Prime95 blend instead. Thoughts?

Update: Got bored. 17 hours 8+ pass Memtest86+ stable and 13 hours Prime95 Small FFT stable before I turned them off is plenty stable. Now for the GPU overclocking!