X79 and/or rampage IV OC'ing thread for those of us still left....

[agent_x007]

@dalekdukesboy What's you VTT/VCCSA ?

 

[dalekdukesboy]

Surprisingly low, part of how I've gotten better with OC'ing is let the board auto set a few voltages for starters and I get an idea of what memory/vtt/vccsa actually NEED versus me just pumping in what I think they need. I wish I could see settings IN windows but unless any of you know of a program that plays nice with windows 10 and this board I can't see all those settings except in bios itself. The vtt is 1.25 and the vccsa is 1.17.

 

[agent_x007]

Thanks.
Hwinfo64 or AIDA64 ?

 

[dalekdukesboy]

I have a cpuidhw monitor and no on that, AIDA64 I don't use for that but obviously have for benchmarks I'll look around and see if it lists it there, that's a good idea. Yeah I used to have both vtt/vccsa 1.28'ish but it appears this chip really doesn't need it, least not on the RIVE black, and with a Power supply that actually supplies power adequately.

Good show on recommending AIDA64 for it indeed does have those listed once you go deep enough into the program to find where they are listed under simply "sensor".

 

[mouacyk]

So after digging through some old ASrock X79 threads on OCN, I decided to give 4.7GHz another try on my 1680 V2. Enabled Internal PLL Overvoltage, because it seems to be needed past a certain frequency for Sandy Bridge-E (predecessor). Increased PLL Voltage to 1.903v (will lower later, but this is what got 4.7GHz on my previous 1660 V2.) Increased VCore to 1.415v fixed. Set multiplier to 38, strap to 125, and FSB to 123.7MHz.

Prime95 is maxing around 87C, AIDA FPU maxes in low 90's. Realbench reaches 80C, as well as my standard 10 compiles of gcc. Not too bad, +0.1v for +2GHz. Passmark 9.0 scores 22,000. 1680 in CBR15.

 

[dalekdukesboy]

So after digging through some old ASrock X79 threads on OCN, I decided to give 4.7GHz another try on my 1680 V2. Enabled Internal PLL Overvoltage, because it seems to be needed past a certain frequency for Sandy Bridge-E (predecessor). Increased PLL Voltage to 1.903v (will lower later, but this is what got 4.7GHz on my previous 1660 V2.) Increased VCore to 1.415v fixed. Set multiplier to 38, strap to 125, and FSB to 123.7MHz.

Prime95 is maxing around 87C, AIDA FPU maxes in low 90's. Realbench reaches 80C, as well as my standard 10 compiles of gcc. Not too bad, +0.1v for +2GHz. Passmark 9.0 scores 22,000. 1680 in CBR15.

I have that pll voltage and I believe it's set to off at the moment so glad you mentioned that haven't looked at it in some time, great job on OC'ing though, it definitely takes work with a decent water cooler to even try to get the 1680's 8 cores cooled at 4.7 ghz, no easy feat.

Speaking of what is the cooler/fans you got on that rig? I thought you may have mentioned it but yeah I admit I don't want to go through all these pages to find it lol.

 

[mouacyk]

I have that pll voltage and I believe it's set to off at the moment so glad you mentioned that haven't looked at it in some time, great job on OC'ing though, it definitely takes work with a decent water cooler to even try to get the 1680's 8 cores cooled at 4.7 ghz, no easy feat.

Speaking of what is the cooler/fans you got on that rig? I thought you may have mentioned it but yeah I admit I don't want to go through all these pages to find it lol.

EK Supremacy Universal on a 240mm + 120mm custom loop. Even though the CPU overclock seemed stable at 4.7GHz with 5x compilations of GCC, the memory is now unstable at 2307MHz. Lowering to 1600MHz is better, but still has intermittent instability. It's probably the 123.7MHz FSB that the RAM doesn't like, or isn't a good combination with high overclock.

 

[dalekdukesboy]

EK Supremacy Universal on a 240mm + 120mm custom loop. Even though the CPU overclock seemed stable at 4.7GHz with 5x compilations of GCC, the memory is now unstable at 2307MHz. Lowering to 1600MHz is better, but still has intermittent instability. It's probably the 123.7MHz FSB that the RAM doesn't like, or isn't a good combination with high overclock.

How many fans or what kind/specs on your rad? I have just a (relatively) crappy 280 mm AIO Corsair h110 extreme or something like that but I put 4 of the 140 mm 175 cfm silverstone fans and I seem to be in the ballpark of your temps as well as speed of processor with relatively the same voltage.

Partially I ask because I want to upgrade what I've got but I want an idea of how much a custom loop and/or better pump will help. I have a bigger rad here it's the 480 mm AIO from Alphacool but pump is fairly lame....however the pump on this Corsair has NO specs listed anywhere that I've seen so I can't even really compare the two. I think the corsair has a stronger pump but no way to know. The occool I want to add a stronger pump to because it's an expandable model but not real confident on how to do so properly.

 

[mouacyk]

240mm and fans are from the original Swiftech H2O (not X) kit. 120mm rad is EK PE with a gentle typhoon fan. All fans are in push configuration, as exhaust. The pump is a D5 in PWM mode, running at about 65% max speed (3000/4700).

 

[dalekdukesboy]

240mm and fans are from the original Swiftech H2O (not X) kit. 120mm rad is EK PE with a gentle typhoon fan. All fans are in push configuration, as exhaust. The pump is a D5 in PWM mode, running at about 65% max speed (3000/4700).

Wow I actually have a variant of the d5 pump but hesitant to hook it up because it's by OCCOOL and the directions are in chinese or something so which 2 holes I need to use without a reservoir is a bit of a mystery to me. At any rate, I do have a slightly bigger rad and definitely my fans are pushing a ton of air.

Speaking of rads I tried my occool again and temps were terrible, just like last time I put it on, not sure what the deal is but either the fittings are somehow not locking in and fluid isn't flowing right, I got hoses on wrong (yes I've moved them around and directions to it suck) or the pump has some issues. Regardless, was kinda sad I have this huge all copper rad and can't use it but for a paperweight till I figure out what the deal is, it used to work well.

However that forced me to obviously re-apply my corsair and in doing so I cleaned it all up and was very careful getting tight fit and got good tim and got it spread well after using alcohol to clean the cpu and AIO copper surface...now I'm running 10 gig memory LINX with 4.6 ghz same as before and highest temp core is 79! So, I obviously didn't have as good contact or tim spreading as I should have before. So with that added buffer I'm hoping now I can get 4.7 or maybe some higher memory #'s with cpu being cooler and happier.

However, better yet I realized one fan of the 3 on the rad wasn't running wasn't plugged back in! So....Shaved a degree or two or more off depending on which core! Wow!

However....once again I'm trying for 4.7 ghz and now at least temps are fine lower 80's at highest but still getting errors if I try above 2 gigs of memory. I actually got it to pass 3 runs with 1 gig and I think 2 gigs of memory. But that means it's not exactly stable but I may now have better chance for 4.9 ghz benchmark and keep it benchmark stable well over 4.8 ghz....more to come.

 

[mouacyk]

Nice progress. A straight 47x100 or FSB overclock like 127x125 wasn't easy to stabilize and required lots of voltage 1.45v+. Again, my strange 4.7GHz result was with under-clocking the 125MHz strap to 123.7 and use multiplier 38. Be warned though, this seems to have de-stabilized my memory at 2300MHz.

 

[dalekdukesboy]

Nice progress. A straight 47x100 or FSB overclock like 127x125 wasn't easy to stabilize and required lots of voltage 1.45v+. Again, my strange 4.7GHz result was with under-clocking the 125MHz strap to 123.7 and use multiplier 38. Be warned though, this seems to have de-stabilized my memory at 2300MHz.

Thanks! Yes when my bigger rad/pump was giving me 50 celcius idle temps at 3.0ghz stock speed I was pretty bummed. I grumbled and put old corsair rad back on but instead of half ass slapping it on there in a "rage quit" mode I was very careful in every way and obviously it made a difference! Speaking of....I managed 4.7 ghz @ 130.58 FSB with memory I believe at 2440 or so...a 4 gig run so that's enough so it's at least stable enough I should be able to game etc and I'd probably not have any issues with it.

Yeah, I had to back off of the 131.5 fsb and thus also the memory to get it to pass, I tried 4.725 or whatever that came out to and really didn't pass many runs unless it was less than a gig used which is nothing. So yeah as you said next after I dial this in bit better I'll try with 100 fsb and see if that allows for any more raw clock speed and either try 2333 mhz memory or 2666 mhz which I believe are the options at 100 fsb. Regardless, as you said this is major progress I thought I was going to slink in the corner with pure failure tonight with OC'ing this thing.

Dammit.. I want the 1680 v1... I know it will do 4.7 stable....

Behold....you were right.

 

[dalekdukesboy]

Best result thus far with 10 gig memory tested, 4.62 ghz with 132 fsb but due to some weird error the bus speed was left blank in the screenshot but every other value is correctly filled in CPUZ. I had the voltage high for 4.7 ghz runs so will lower and see what is max it needs and see if I can indeed get over 132 fsb or that is just it, so far it seems about far as it wants to go. Like I said for some reason the 100 fsb strap really gave fits when I tried 47 multiplier for 4.7 ghz and have no idea why but it really didn't like it and restarted rather quickly under stress test.

 

[mouacyk]

That 4.62GHz is still with 2400+ memory? The core voltage is a bit high, but if that's what it takes, it's still under 1.45v and 85C. There is something fundamentally limiting the 1680v2's max 24/7 clocks when compared to lower chips like 1660v2 and under, which can achieve 4.7-4.8GHz on the high-end. Significantly more cache, +2 cores, and originally a 10-core SKU with disabled parts might be significant factors.

For 4.7GHz and 4.625GHz, I've tried memory at 1600 and 1833 to no avail. While the CPU is stable (Linx, compiling, realbench), the memory cannot be stabilized at all -- sounds like a cache instability, since it is tied to the core clock on Ivy-Bridge. This could be why in Haswell, the cache clock was separated out and still rarely any can run above 4.6GHz on the cache (ring) clocks, while core sores to 4.8 and 4.9GHz on good bins.

It probably doesn't help that I run a mix of 2x8GB 2666MHz + 2x8GB 2993MHz for 32GB of RAM, but it was stable at 2333MHz 10-12-12-31-2T and 1.65v with 4.7GHz on my previous 1660 V2. Running 4x4GB is definitely helping you get stability, compared to my results at 4.6+. I've even tried swapping the memory modules around, and while it helped at first, longer memory tests always ended in errors again.

I'm convinced my 24/7 is stuck 4.5GHz at 1.312v and 2333MHz. By the way, the PLL voltages turned out to not have any effects.

 

[dalekdukesboy]

That 4.62GHz is still with 2400+ memory? The core voltage is a bit high, but if that's what it takes, it's still under 1.45v and 85C. There is something fundamentally limiting the 1680v2's max 24/7 clocks when compared to lower chips like 1660v2 and under, which can achieve 4.7-4.8GHz on the high-end. Significantly more cache, +2 cores, and originally a 10-core SKU with disabled parts might be significant factors.

For 4.7GHz and 4.625GHz, I've tried memory at 1600 and 1833 to no avail. While the CPU is stable (Linx, compiling, realbench), the memory cannot be stabilized at all -- sounds like a cache instability, since it is tied to the core clock on Ivy-Bridge. This could be why in Haswell, the cache clock was separated out and still rarely any can run above 4.6GHz on the cache (ring) clocks, while core sores to 4.8 and 4.9GHz on good bins.

It probably doesn't help that I run a mix of 2x8GB 2666MHz + 2x8GB 2993MHz for 32GB of RAM, but it was stable at 2333MHz 10-12-12-31-2T and 1.65v with 4.7GHz on my previous 1660 V2. Running 4x4GB is definitely helping you get stability, compared to my results at 4.6+. I've even tried swapping the memory modules around, and while it helped at first, longer memory tests always ended in errors again.

I'm convinced my 24/7 is stuck 4.5GHz at 1.312v and 2333MHz. By the way, the PLL voltages turned out to not have any effects.

Everything you said is spot on, 2 more cores and more cache obviously is a drag on these cpu's once you get to a certain mhz ESPECIALLY if you are pounding the memory and trying for good mhz AND timings. Speaking of....

These are exact settings I had when I made that run I just neglected to have memory tab open to show it. Your "mix" of memory, has everything to do with your limitations not only somewhat with overall clocks but definitely will hold back your memory overclocks unless you got lucky and got two types of similar memories with same chips and similar timings etc etc. So yes memory on this platform even "matched" sets are picky, but you mix sticks and it's purely just luck of the draw if they will cooperate when you try to push them.

Also I mentioned it in post but this voltage was what I had already set to get 4.7 ghz meh/reasonably sort of stable....I'll try again just to see but if you go back I have a few runs on here I posted recently at up to 4.6 ghz requiring 1.375 volts with occasional flickers to 1.39 volts and I ran a few runs at 4.62 similar to this run I just posted and it passed fine. So yes voltage I had there was just for higher mhz and I didn't remember to change it back when I changed other settings in bios.

I also changed the vdroop voltage setting on CPU from high to ultra high so going at it a different way and having to adjust voltage accordingly to how it now doesn't droop. This run was at 1.408 volts under load. So it definitely didn't need the 1.440 or so it was getting in the run the other day, not sure what minimum is but I'll play around with that. Problem is the 132 fsb you see here is about as far as I have pushed it successfully so far so not sure I can go any more with FSB on this multiplier.

 

[dalekdukesboy]

Already edited and added the last message like hell so new run new post and the load voltage is exactly what shows in this screenshot minus a couple moments it bumped to 1.408 so I'll have to drop several levels down in voltage to see what the least voltage it needs for this OC is. However regardless of what that is I'm quite happy with temps and voltages and really glad I did a much better job of applying tim/watercooler AIO block to the cpu, it shows in even temps with every core minus a couple which are obviously just under-reporting the temps and always have so those are just outliers.

 

[dalekdukesboy]

Haven't tried video/games/GPU but at least for LINX purposes 4.62 ghz with 2462 mhz memory and 1.376 volts under load is stable for 10 runs anyway.

Best stable result I can get thus far 4.641 ghz with 2475 mhz Cas 10 memory with 1.392-1.408 volts under load. I'm going to slip down the voltage and see if it can do 1.392 or less under stress. I tried 1.360 volts under load for 4.62 ghz and it passed 6 runs so it was pretty close but just not quite enough.

 

[dalekdukesboy]

That was the best I could get with hynix or whatever the high speed RAM was put my Micron back in and had to fight with it but realized it was only "seeing" 2-3 of my Ram sticks instead of 4 because of bad settings. I thought this black edition RIVE just didn't like the micron before but just have to adjust couple settings in timings or it won't work properly or at all. Still working on stability but as expected without even getting to ideal settings they already are getting better bandwidth #'s.

Micron does significantly better write/copy times both near 70k mbs and primary timings are "looser" than the Klevv memory, but being double sided and having much tighter secondary/tertiary timings makes all the difference in actual performance. I had to boost VTT/VCCSA voltages a bit though as a tradeoff for stability compared to Klevv memory. Micron FTW.

 

[dalekdukesboy]

Crickets....lol.

 

[dalekdukesboy]

That was the best I could get with hynix or whatever the high speed RAM was put my Micron back in and had to fight with it but realized it was only "seeing" 2-3 of my Ram sticks instead of 4 because of bad settings. I thought this black edition RIVE just didn't like the micron before but just have to adjust couple settings in timings or it won't work properly or at all. Still working on stability but as expected without even getting to ideal settings they already are getting better bandwidth #'s.View attachment 112248

Micron does significantly better write/copy times both near 70k mbs and primary timings are "looser" than the Klevv memory, but being double sided and having much tighter secondary/tertiary timings makes all the difference in actual performance. I had to boost VTT/VCCSA voltages a bit though as a tradeoff for stability compared to Klevv memory. Micron FTW.

I only reply to myself because....this result was me just plugging my microns in with settings mostly the same as the Klevv's with couple small changes to get it stable....I have tinkered, toiled, and tried every setting from primary to tertiary to even RTL/IO's which are tough to figure out; point is the memory benchmarks were practically identical or unstable lol. I just want 70kmb/s across the board...the write time as you can see is in mid 69's and I barely saw 70k a couple times no matter what I changed. With read/copy ditto 76k mb/s for read and 70k mb/s for copy at best no matter what I did.

Not sure what the technical "maximum bandwidth" for this platform/cpu is but my cpu is listed at 59,000 mb/s on Intel's site so obviously that isn't truly the max. I've seen threads about that # that Intel lists on their Ark cpu info webpage but not sure what real answer is.

 

[mouacyk]

Your max theoretical peak bandwidth is: 79,232MB/s

One of my old goals on Z97 was maximizing this bandwidth, and I came so close: At 2666MHz in dual channel, I reached about 94% efficiency at best, which is 40.1GB/s out of 42.7GB/s max.

For now, I've settled for 4.5GHz (36x125) at 1.320v on my 1680v2 with 32GB 2333MHz memory for 24/7. The memory is tightened as much as possible to pass 2hours of GSAT. Timings:
10-12-12-31-2T 1.65v, VTT/SA both at 1.1v
tRFC =220
tRRD = 4
tFAW = 16
tRTP = 4
tCWL = 7
AIDA64 was 69GB/s read 67GB/s write and 65GB/s copy out of max 74.6GB/s, so about 90% efficient.

I ended up going back one release in BIOS, because the latest is unpredictable when it comes to cold booting 125MHz strap overclocks. The prior BIOS also doesn't cold boot at 36x125, but it's predictable that if a successful 45x100 cold boots, then 36x125 warm boot will be stable afterward. 4.625GHz was semi-stable but required 1.4+ voltage and operating temps were high 80's, so decided against that for daily setting.

 

[dalekdukesboy]

Your max theoretical peak bandwidth is: 79,232MB/s

One of my old goals on Z97 was maximizing this bandwidth, and I came so close: At 2666MHz in dual channel, I reached about 94% efficiency at best, which is 40.1GB/s out of 42.7GB/s max.

For now, I've settled for 4.5GHz (36x125) at 1.320v on my 1680v2 with 32GB 2333MHz memory for 24/7. The memory is tightened as much as possible to pass 2hours of GSAT. Timings:
10-12-12-31-2T 1.65v, VTT/SA both at 1.1v
tRFC =220
tRRD = 4
tFAW = 16
tRTP = 4
tCWL = 7
AIDA64 was 69GB/s read 67GB/s write and 65GB/s copy out of max 74.6GB/s, so about 90% efficient.

I ended up going back one release in BIOS, because the latest is unpredictable when it comes to cold booting 125MHz strap overclocks. The prior BIOS also doesn't cold boot at 36x125, but it's predictable that if a successful 45x100 cold boots, then 36x125 warm boot will be stable afterward. 4.625GHz was semi-stable but required 1.4+ voltage and operating temps were high 80's, so decided against that for daily setting.

Thanks for that good information! I did't realize I was so close to the max I could possibly do but I had a feeling that was the case because #'s I'm getting are as high as I've seen in all the X79 threads old or new. How did you figure out my theoretical max bandwidth by the way? Also I would love to see your screenshots of your E1680v2 setup and nice results on the 4790k.

 

[mouacyk]

The bandwidth formula is: (speed * 64 * channels) / 8 = MB/s

Former 1660v2 at 4.7GHz with 4x8GB, which was actually stable:

1680v2 at 4.7GHz for benchmarks -- way too hot and not 100% stable:

 

[dalekdukesboy]

The bandwidth formula is: (speed * 64 * channels) / 8 = MB/s

Former 1660v2 at 4.7GHz with 4x8GB, which was actually stable:
View attachment 112503View attachment 112504View attachment 112505

1680v2 at 4.7GHz for benchmarks -- way too hot and not 100% stable:
View attachment 112506

Been a while since I've done all those tests especially at once and never have I done them at the settings I have now which are 100 % stable and I'll have to play around with highest benchmark stable settings see what I can do, I'd be pretty happy to get cinebench etc at 4.8 ghz + and see what it does. I know it can do 4.7 already did that only difference is I have windows 10 64 bit but at 4.64 ghz I'll run all those and I think you'll be surprised at how high I get.

Quickly upped the multiplier by 1 and upped voltage to what I guess'd would be benchmark stable, and there's an almost 4.8 ghz cinebench lol.

 

[dalekdukesboy]

Justy FYI on those last benchmarks technically NOT stable only because the 130 Refresh cycle time I had set on my memory was way too tight thought I had tested it but somehow I didn't, however the #'s in the benchmarks are close to what are stable and I'll rerun them and see what I get.

Speaking of, despite past failings I decided to try to inch up my OC by just upping the Bus speed slightly and so far managed 2 "bumps" up with a slight bump of voltage with each to get them totally stable. So definitely near the point of diminishing returns or just no returns, but there is a little left beyond the 4.640 I got stable. So now I'm at 4.65 lol. A whole 10 megahertz but, I guess it's something and how you have to do it when you're near the limit of what your setup will be able to do.

Above 4.645 @ 1.392 load

and above 4.65 @ 1.408 load

4.663 ghz passes with 1.424/1.440 under load.