1366 x58 Xeon Enthusiast overclocks club

[Rock&Roll]

Guess I spoke too soon Finally broke the barrier and got above 220 BCLK be setting QPI compensation to "less" and then hard lock at 226MHz. Currently testing for min vCore at 4.4ghz.

Something is a tad odd about those temperature readings though. It's like a few of the cores are reading fake values. I'd love to have sub-ambient temps of 4C My basement is probably 20-24C ambient....

 

[GushpinBob]

Thread subbed. Replaced my i7 920 with an X5650 for 70 bucks just after I found this discussion. I've got a ton of home movies and videos sitting on my hard drive still waiting to be reencoded. So far it seems like this CPU runs x264 like a hot knife through butter compared to my old i7, and I haven't even touched the clock settings yet.

 

[MongGrel]

Still want a SR2 with dual X5690's myself like someone I know elsewhere, but have been happy with just popping a X5650 in here awhile now myself.

Even have a L5639 I've been too lazy to pop into a I-7 920 awhile now.

 

[novadaemon]

That cooler on the x79 setup is huge.

Is that really the memory modules hiding below both fans lol.

When I first looked I was like "Where do hell are the ram slots"

haha yeah. that heatsink is quite hefty. I'll move it over to the X99 setup when I'm ready to upgrade but for now, it would be too much for the X58 setup.

 

[Zoson]

*snip*

Guess I spoke too soon Finally broke the barrier and got above 220 BCLK be setting QPI compensation to "less" and then hard lock at 226MHz. Currently testing for min vCore at 4.4ghz.

Something is a tad odd about those temperature readings though. It's like a few of the cores are reading fake values. I'd love to have sub-ambient temps of 4C My basement is probably 20-24C ambient....

Remember, QPI is a DDR link, so when it shows 4000+ MHz in CPUz, you're actually running 8000+ MHz.

Most x58 motherboards will not be stable above ~7800MHz.
You won't see a large performance loss until below about 5800MHz, which is the original speed that QPI operated at upon release. It was later upped to 6400MHz.

So, in short, drop that QPI multiplier even more.

 

[Zoson]

What speed are you guys running your ram? Have any tips/guides for ram overclocking? I've been wanting to overclock my l5639 that has the Samsung magic ram for the longest time but there's just so much values to the point of intimidation.

6GB kit: Limit will be the kit itself.
12GB kit: Limit will be around 2000MHz CAS 9.
>12GB: Your limit will be your IMC, which will be between 1800-1900MHz, with most falling closer to 1800MHz.

The only way to learn about memory overclocking, though, is to just google all the various settings and then trial/error with your own personal kit. Just like CPUs, all memory is different.

Just remember, for maximum performance UCLK speed should be 2x memory speed. Just don't exceed 1.35v QPI/Uncore/Vtt voltage trying to get there.

 

[GonzoP]

6GB kit: Limit will be the kit itself.
12GB kit: Limit will be around 2000MHz CAS 9.
>12GB: Your limit will be your IMC, which will be between 1800-1900MHz, with most falling closer to 1800MHz.

The only way to learn about memory overclocking, though, is to just google all the various settings and then trial/error with your own personal kit. Just like CPUs, all memory is different.

Just remember, for maximum performance UCLK speed should be 2x memory speed. Just don't exceed 1.35v QPI/Uncore/Vtt voltage trying to get there.

Good Stuff Zoson. Fair to say 1.65v is still the ram voltage ceiling or is anyone going past that? [And I'm not talking about the guys who run 1.66v since their mobo might not have 1.65v itself]

 

[Zoson]

Good Stuff Zoson. Fair to say 1.65v is still the ram voltage ceiling or is anyone going past that? [And I'm not talking about the guys who run 1.66v since their mobo might not have 1.65v itself]

I've never had the courage to exceed 1.656v. Also, I'm running my kits at 1.575v, since that's all they need to do 1866MHz CAS 8-8-8-27. Since I'm running 24GB, I'm limited by my IMC long before my memory.

I will say that in the XS R3E thread there was talk about how the maximum memory voltage was not actually a hard and fast 'do not exceed 1.65v' but rather was 'vDDR should not exceed .5v over QPI/UCLK/Vtt'. This was also investigated by Anandtech, and Anand agreed, but I don't think I ever saw them actually practice this themselves. So, put into numbers, if you're running 1.35v QPI/UCLK/Vtt, you should be able to safely hit 1.85v DDR without causing problems according to that line of thought. Again, I did not try this, and do not intend to.
Reference Link: http://i4memory.com/wp/article/327

 

[GonzoP]

I've never had the courage to exceed 1.656v. Also, I'm running my kits at 1.575v, since that's all they need to do 1866MHz CAS 8-8-8-27. Since I'm running 24GB, I'm limited by my IMC long before my memory.

I will say that in the XS R3E thread there was talk about how the maximum memory voltage was not actually a hard and fast 'do not exceed 1.65v' but rather was 'vDDR should not exceed .5v over QPI/UCLK/Vtt'. This was also investigated by Anandtech, and Anand agreed, but I don't think I ever saw them actually practice this themselves. So, put into numbers, if you're running 1.35v QPI/UCLK/Vtt, you should be able to safely hit 1.85v DDR without causing problems according to that line of thought. Again, I did not try this, and do not intend to.
Reference Link: http://i4memory.com/wp/article/327

Yeah I had read the .5v thing before and then I read a .3v thing as well. With your QPI 1.35v max recommendation the .3v on the ddr would be the mystical 1.65v everyone seems to hold to. Yeah I'm in no rush to exceed 1.65v just curious

 

[ssnyder28]

6GB kit: Limit will be the kit itself.
12GB kit: Limit will be around 2000MHz CAS 9.
>12GB: Your limit will be your IMC, which will be between 1800-1900MHz, with most falling closer to 1800MHz.

The only way to learn about memory overclocking, though, is to just google all the various settings and then trial/error with your own personal kit. Just like CPUs, all memory is different.

Just remember, for maximum performance UCLK speed should be 2x memory speed. Just don't exceed 1.35v QPI/Uncore/Vtt voltage trying to get there.

Thats interesting, I had no idea that was the case. Thanks for the good info!

Does a 12gb kit mean specifically a 3x4gb kit or does a 6x2gb kit count?

 

[Zoson]

Thats interesting, I had no idea that was the case. Thanks for the good info!

Does a 12gb kit mean specifically a 3x4gb kit or does a 6x2gb kit count?

It really depends on the kit itself. In this situation specific sticks being in specific sockets can make a difference also. Basically though, if you're going to be overclocking you don't want to populate all your memory sockets. Whenever I'm benching and going for top scores, I switch to my 6GB kit, which does 2200MHz CAS9 easily at 1.65v.


So yeah, I have 6x 4GB sticks for my daily use, and 3x 2GB sticks for benching.

 

[Rock&Roll]

Remember, QPI is a DDR link, so when it shows 4000+ MHz in CPUz, you're actually running 8000+ MHz.

Most x58 motherboards will not be stable above ~7800MHz.
You won't see a large performance loss until below about 5800MHz, which is the original speed that QPI operated at upon release. It was later upped to 6400MHz.

So, in short, drop that QPI multiplier even more.

Thanks! I remember reading a long spiel on XS years ago about Uncore/QPI limits....just couldn't reliably recall the details. I'll look into doing that.

EDIT:

Seems the only way to reduce QPI multi anymore is the disable "QPI control" in evga's bios. Everytime I did, it was a no-boot situation. Closest I got was to see first line of POST and then hard lock. Not sure I understand what would be made more unstable by slowing QPI...

If what I've read is true, it may not matter. I've read that it's likely at 225 and above I would need to start substantially raising PCI-E freq.

 

[Deimos]

Some interesting results running RAID 0 on the built in controllers. The ICH10R actually bests the LSI controller in small file transfers!

Unfortunately I can't find the Atto screens from the LSI RAID 0 tests but max file transfers were around 1GB/s with a 2 disk array. I can only assume that the ICH10R will match it with a 4 disk array but at this stage I'm not willing to spend another $1k on SSDs to try it.

I've updated my 1st page post with this info.

 

[Deimos]

What speed are you guys running your ram? Have any tips/guides for ram overclocking? I've been wanting to overclock my l5639 that has the Samsung magic ram for the longest time but there's just so much values to the point of intimidation.

Didn't see anyone respond to this. I would say don't worry about overclocking ram, triple channel controller means that memory bottlenecks are minimal. Have seen benchmarks that show no tangible benefit going above 1600MHZ and in some cases the result was actually worse performance because of the looser memory timings.

Sure, synthetic benchmarks will show crazy increases but this is only useful if you want to show off your e-penis.

 

[Deimos]

My X5670s arrived today (finally).

Not going to delid this time. Running IBT is still under 60C at 4.2Ghz so I doubt there is any point.

Testing at the following settings before I try 4.4Ghz:

200BCLK
21 Multi
1.29 VCore
1.3 QPI

 

[Deimos]

Low PPL voltage doesn't work very well for me, infact, even default PLL is worse than bumping it up a little.

pumped over 1.4V core to get 4.4 through IBT with 1.8PLL but increasing PLL to 1.88V and dropping core to 1.33V is stable and around 6-7C cooler.

Might be able to go lower still on the vcore, don't have time to test ATM, I'm also testing lower memory timings, currently running at 8 8 8 24 with a 1 CPC 1.6V (memory is rated for 9 9 9 24 2 CPC at 1.5V)

Interestingly, going higher still on the PLL introduces instability, 1.92 just outright failed. Apparently 1.8 to 1.88 is within spec.

 

[combat fighter]

Managed to snag a deal for a good price on a hardly used set of 12GB Corsair Dominator GT tri-channel DDR3 RAM CMT12GX3M3A2000C9 2000, 9-10-9-27

Should go nicely with my X5675 when it arrives.

Never thought I would get so excited over my X58 platform again! haha

I know it's not worth much but I will be selling my old i920 DO (complete with box and unused cooler) along with my old set of Corsair CMT6GX3M3A2000C8 Dominator GT 6GB Memory (complete with box)

So if they sell that will make this upgrade pretty cheap and a right bargain in my eyes.

Gotta love the X58 platform. Still kicking ass all these years on.

 

[kniaugaudiskis]

Ordered an X5670 today. Hopefully it plays nicely with my P6T Deluxe.

 

[Zoson]

Some interesting results running RAID 0 on the built in controllers. The ICH10R actually bests the LSI controller in small file transfers!

Unfortunately I can't find the Atto screens from the LSI RAID 0 tests but max file transfers were around 1GB/s with a 2 disk array. I can only assume that the ICH10R will match it with a 4 disk array but at this stage I'm not willing to spend another $1k on SSDs to try it.

I've updated my 1st page post with this info.

Our Marvell controller is connected by a single PCIE x1 2.0 link.
That means the controller cannot exceed 500MB/sec, no matter what the drives or configuration. You're seeing less than that due to command overhead/controller inefficiency.

 

[GonzoP]

Some interesting results running RAID 0 on the built in controllers. The ICH10R actually bests the LSI controller in small file transfers!

Unfortunately I can't find the Atto screens from the LSI RAID 0 tests but max file transfers were around 1GB/s with a 2 disk array. I can only assume that the ICH10R will match it with a 4 disk array but at this stage I'm not willing to spend another $1k on SSDs to try it.

I've updated my 1st page post with this info.

Yeah I found out a long time ago [i7] the genuine SATA2 was better on my P6X58D-E than the so called SATA3 via the Marvel controller. I did re-test both when I got my Xeon but the results were mostly the same. I've actually disabled the marvel controller in my BIOS.

Didn't see anyone respond to this. I would say don't worry about overclocking ram, triple channel controller means that memory bottlenecks are minimal. Have seen benchmarks that show no tangible benefit going above 1600MHZ and in some cases the result was actually worse performance because of the looser memory timings.

Sure, synthetic benchmarks will show crazy increases but this is only useful if you want to show off your e-penis.

Yeah I'm running a 200BCLK and thus a 1600 speed at 9-9-9-24 which is decent considering my ram is just 1333. I said I would one day test to see if I could run 2000Mhz on the ram but surely with my 1333 I'd have to drop the timings to probably 11-11-11-30+. I have googled some of the comparisons between 1600cas9 and 2000cas11 and overwhelming it shows the greater speed despite loose timings is superior. Granted most of that was in the aforementioned synthetics that no one cares about and don't really matter. In the real world stuff and games the differences appeared to be small to non existent. But it wasn't a bad idea if you could pull them off.
I'm kinda out of the testing mode right now but maybe one day I'll give this a research project.

Low PPL voltage doesn't work very well for me, infact, even default PLL is worse than bumping it up a little.

pumped over 1.4V core to get 4.4 through IBT with 1.8PLL but increasing PLL to 1.88V and dropping core to 1.33V is stable and around 6-7C cooler.

Might be able to go lower still on the vcore, don't have time to test ATM, I'm also testing lower memory timings, currently running at 8 8 8 24 with a 1 CPC 1.6V (memory is rated for 9 9 9 24 2 CPC at 1.5V)

Interestingly, going higher still on the PLL introduces instability, 1.92 just outright failed. Apparently 1.8 to 1.88 is within spec.

Yeah I've hit most of my speeds with 1.8v PLL which is the lowest my mobo can supply I too found going to high made things worse from a stability point. I'm still at just 1.8v and if memory serves I was on 1.8v when I got to 4.8 which granted wasn't completely stable. Again were I in the testing mode which I'm not I'd play around with this again. Because I'd surely love some lower cpu temps. But at the end of the day given everything I've read, especially from people like Zoson in the choice between vcore and pll it seems pll was the one more likely to fry your cpu. So I'm up at 1.375vcore with the 1.8v pll at 4.4 myself. If memory serves when I did set 1.88v pll I was a few notches lower on vcore maybe 1.34375v. If I recall correctly though I'm more stable with the higher vcore and lower PLL. At the end of the day I just kinda have set it and forget it. I suppose with more hours of fiddling I could find nirvana but I have to assume I'm at 99% right now.
Bottom line my system does everything I want it do no stability issues. I've given up chasing one degree or one fraction of a volt less here or there.

Our Marvell controller is connected by a single PCIE x1 2.0 link.
That means the controller cannot exceed 500MB/sec, no matter what the drives or configuration. You're seeing less than that due to command overhead/controller inefficiency.

This is pretty much one of those aforementioned synthetics versus real world issues. I too got the same numbers as Deimos on the Marvel controller but it wasn't as stable friendly as the SATA2. In reality how much difference does SATA2 vs SATA3 make to what I actually do with my PC in terms of load times, start up and shut down? Almost none.


Well, like said above I was out of the mode of testing but after this series of replies maybe I'll make one short stint at it again. Just to specfically address the two issues brought up here. VCore vs PLL and the ram memory.....eh maybe not. If I do I'll of course report back my findings

 

[GonzoP]

Well I did do the PLL experiment.
Raising my PLL from 1.8v to 1.88v I dropped my vcore from 1.38125v bios [1.368 cpuz] to 1.3375 bios [1.328cpuz] just to give me a starting point and it seemed OK until I stressed it. 1.35v seems fine and stable across my games. 1.35v in my bios is 1.336 in cpuz and hardware monitor.
I don't know raising PLL .08v to save .02-.03v vcore. I might switch back. Leaving it for now.

On to the ram testing

 

[GonzoP]

So this is interesting.
My old ass DDR3 1333 will run windows at 2000Mhz with timings of 11-11-11-31. It did fail the burn test though with the classic 0x1a BSOD. Since I was getting bored from all the PLL stuff earlier I didn't mess around with either A] relaxing the timings even more or B] actually trying a higher vdimm like 1.7v.
What I did instead was drop back down to 1600 and tighten the timings and I'm happy to report I'm IBT passing with 8-8-8-22 timings. Pretty good for some 9-9-9-24 1333 ram.
Gotta hit the door but I might try some 7-7-7-20 lovin' later. Or since I could navigate windows at 2000 try that again perhaps with either the loose timings or maybe exceeding the 1.65v barrier.

It might have been in my mind but stuff did seem to open faster with 2000 vs 1600

 

[Zoson]

So this is interesting.
My old ass DDR3 1333 will run windows at 2000Mhz with timings of 11-11-11-31. It did fail the burn test though with the classic 0x1a BSOD. Since I was getting bored from all the PLL stuff earlier I didn't mess around with either A] relaxing the timings even more or B] actually trying a higher vdimm like 1.7v.
What I did instead was drop back down to 1600 and tighten the timings and I'm happy to report I'm IBT passing with 8-8-8-22 timings. Pretty good for some 9-9-9-24 1333 ram.
Gotta hit the door but I might try some 7-7-7-20 lovin' later. Or since I could navigate windows at 2000 try that again perhaps with either the loose timings or maybe exceeding the 1.65v barrier.

It might have been in my mind but stuff did seem to open faster with 2000 vs 1600

You might want to see if you can get your IMC up to 3733MHz and then try to hit 1866MHz ram with tighter timings.

You also might want to try like 7-7-7-24, 7-8-7-24, or 7-9-7-24 etc. A lot of ram from x58 generation did really well with a relaxed Trcd and Tras.

 

[GonzoP]

You might want to see if you can get your IMC up to 3733MHz and then try to hit 1866MHz ram with tighter timings.

You also might want to try like 7-7-7-24, 7-8-7-24, or 7-9-7-24 etc. A lot of ram from x58 generation did really well with a relaxed Trcd and Tras.

Well with my BCLK at 200 I only have 8x or 10x available to me. I'm sure I could run something in between 1600 and 2000. I mean I did 1750 at 9-9-9-24 when I dialed my BLCK up to 219 when I was trying to max it out. But again I'm looking for the best 24/7/365 set up and 200x22 4.4Ghz works too well for me to reduce the speed of that. 220BCLK x 20 is not as stable for me as 200x22. Like I said I'll fiddle a little more when I get back but I kinda think I've hit the ceiling. I'm sure I could run 215x20 but why? Faster memory at the expense of 100Mhz clockspeed?

Again I was quite content where I was at. Now sure I did learn I can tight the timings and that was a nice gift. I'll tinker just a bit more.

If anything seeing how I can save a little vcore with 1.88v on my PLL I might try for a little 205x22 with the same vcore I ran at 200x22 on the lower PLL or even 209x22 for a nice little 4.6Ghz. To get there before I needed over 1.4v but that was with 1.8vPLL with 1.88v I might need less than 1.4v.

 

[Deimos]

Our Marvell controller is connected by a single PCIE x1 2.0 link.
That means the controller cannot exceed 500MB/sec, no matter what the drives or configuration. You're seeing less than that due to command overhead/controller inefficiency.

The only reason I benched the Marvell with RAID 0 was to show everyone here how shit it is, I wasn't expecting any improvement over single disk performance, it delivered on that expectation.

What I was surprised about was the ICH10R besting the LSI controller in transfers under 32MB in size. For some reason the LSI controller's small file transfer speed suffers when you set it up in RAID 0. I did some benchmarks on the LSI in RAID 0 and took screenshots but for some reason I can't find them

 

[Deimos]

This is the original configuration that I abandoned a while ago, its a 6 disk RAID 0 array with OCZ Vertex 3 120GB drives on the LSI controller. As you can see, small file transfer performance is awful. The array is also pretty slow, the hardware is capable of around 3GB/s but going from 2 disks to 6 only increases max speed by 300MB/s

I still have the 6 Vertex 3 drives, I might see if I can get them all back in my system and see how the ICH10R does with a 6 disk stripe.

 

[GonzoP]

I still have the 6 Vertex 3 drives, I might see if I can get them all back in my system and see how the ICH10R does with a 6 disk stripe.

Yeah I'd like to see that if you can. Kinda debating whether grab a second 840EVO myself

 

[Zoson]

Well with my BCLK at 200 I only have 8x or 10x available to me. I'm sure I could run something in between 1600 and 2000. I mean I did 1750 at 9-9-9-24 when I dialed my BLCK up to 219 when I was trying to max it out. But again I'm looking for the best 24/7/365 set up and 200x22 4.4Ghz works too well for me to reduce the speed of that. 220BCLK x 20 is not as stable for me as 200x22. Like I said I'll fiddle a little more when I get back but I kinda think I've hit the ceiling. I'm sure I could run 215x20 but why? Faster memory at the expense of 100Mhz clockspeed?

Again I was quite content where I was at. Now sure I did learn I can tight the timings and that was a nice gift. I'll tinker just a bit more.

If anything seeing how I can save a little vcore with 1.88v on my PLL I might try for a little 205x22 with the same vcore I ran at 200x22 on the lower PLL or even 209x22 for a nice little 4.6Ghz. To get there before I needed over 1.4v but that was with 1.8vPLL with 1.88v I might need less than 1.4v.

You can certainly gain performance with uncore/ram speed that's greater than the performance gained by 100MHz core clock.

I don't have any good comparisons, but I can run some tonight for you. I'll even do them at 4.4GHz 200BCLK on a 22 Multi for you.

 

[Zoson]

Spent all night doing this.

4.4GHz CPU 1600MHz 6-8-6-24 1T RAM Results:
3.2GHz UCLK (81.4GFlops) | 3.6GHz UCLK (83.5GFlops) | 4GHz UCLK (85GFlops)
. . . .

4.4GHz CPU 2000MHz 7-9-7-24 1T RAM Results:
3.2GHz UCLK (83.5GFlops) | 3.6GHz UCLK (85GFlops) | 4GHz UCLK (86.4Gflops)
. . . .

If you're feeling brave, 4.6GHz CPU 2000MHz RAM 4GHz UCLK (90.3GFlops):


So, what do we learn from this?
UCLK benefits more than I thought, get it as high as you can within 1.35v.
For every 100MHz CPU above 4.4GHz you get about 2GFlops.
From 3.2GHz UCLK and 1600MHz RAM to 4GHz UCLK and 2000MHz RAM you gain 5GFlops.
That would sit a 4.5GHz CPU 3.2GHz UCLK 1600MHz RAM system at about 83GFlops.
You can match that with a 4.4GHz CPU 3.6GHz UCLK 1600MHz RAM or a 4.4GHz CPU 3.2GHz UCLK 2000MHz RAM system, and a 4.4GHz CPU 3.6GHz UCLK 2GHz RAM system would be faster.
A 4.4GHz CPU 4GHz UCLK 2GHz RAM (86.4GFlops) system would beat a 4.6GHz CPU 3.2GHz UCLK 1600MHz RAM system (estimated 85GFlops).

DELIVERING THE GOODS.

 

[CrimsonKnight13]

So... I have 1600MHz RAM & 3200Mhz UCLK. What should it really be?

 

[Deimos]

One of the benefits to running a low PLL is it also often reduces your core temperatures by a good 2-4C.

After fiddling again with PLL voltage, I'm going to say that it makes little to no difference to CPU temps. going from 1.88v down to 1.35v changed the CPU temp by maybe 1C if at all.

It might make a difference on a 990X but it certainly didn't do anything for my X5670

Spent all night doing this. (snip)

Good effort.

One point though, Linpack is not a benchmark and the "GFlops" reported by Linpack is worth shit, just change the size of memory tested and you will notice a huge variance in reported speed. Test 24GB and the GFlops for me drops to 50.

Secondly, this is purely synthetic. If you want the best scores, great, but how about usable performance? I haven't tested temps with the higher UnCore but I did notice a decent temp increase when going above 2x memory speed so people should consider if the trade off is worth it.

 

[GonzoP]

Appreciate the efforts Zoson. If anything your cpu tweaker settings will prove most valuable.
I know there is no best benchmark to test with. Even for gaming performance how would one test across various settings and have it be accurate, reliable and consistent?
If nothing else this gives me and us a base point of reference to work off of.
I was not at all surprised to see higher ram speed with looser timings beating out tigher timings at slower speeds as I eluded to earlier.

 

[Zoson]

After fiddling again with PLL voltage, I'm going to say that it makes little to no difference to CPU temps. going from 1.88v down to 1.35v changed the CPU temp by maybe 1C if at all.

It might make a difference on a 990X but it certainly didn't do anything for my X5670

Way to quote me out of context. How about you also include the part where I mention that every cpu is different, and that the PLL voltlage I'm running won't necessarily do anything for you? Did you just do the stock test to find where you would be stable? Or did you find your optimal PLL for your own CPU and then overclock and test with the lower PLL vs higher? That's where you see the temp difference, at overclocked loads.

Based on your comments sounds like you just plugged in my PLL value and expected a drop in temps.

One point though, Linpack is not a benchmark and the "GFlops" reported by Linpack is worth shit, just change the size of memory tested and you will notice a huge variance in reported speed. Test 24GB and the GFlops for me drops to 50.

This is a known bug in *hyperthreading* not linpack where your cores start thrashing due to race conditions. Either set LinX/IBT to use only 6 threads, or disable HT in bios and the problem will disappear.

OR, you could do what we did on the XS forum and test with 9216MB for large size memory pools with HT on (which I mentioned IN the post you quoted).

Secondly, this is purely synthetic. If you want the best scores, great, but how about usable performance? I haven't tested temps with the higher UnCore but I did notice a decent temp increase when going above 2x memory speed so people should consider if the trade off is worth it.

This is to test that the specified settings are stable and to provide a ballpark as to what's reasonable and what kind of performance gains you'll see associated with raising/lowering specific things. If it can pass 20 loops of updated Linpack, it can pretty much handle anything.

This is my streaming PC. I encode x264 realtime WHILE playing demanding games. I also use it for virtualization purposes for work. I've yet to find anything that can simulate a load like that OTHER than Linpack.

 

[Rock&Roll]

Zoson,

What's your thoughts on this thread from a couple years ago?

When I was running my i7 920, I admit I went a little lax on my stringency for stability in favor of being greedy for clocks. It was never unstable But really wasn't stable enough for mission critical stuff.

This time, my aim is 100% unwavering stability. Think there's any merit to doing linpack via linux as instructed in that thread?

 

[Zoson]

First, you need to know that there are two kinds of Machine Check Exceptions: Warnings and Fatal.

In both Windows and Linux, Warning events do not cause the OS to stop functioning. In both Windows and Linux, a Fatal event causes a hardlock(BSOD in windows with code MACHINE_CHECK_EXCEPTION).

Windows Vista and up implements a similar feature to dmesg logging. MCE warnings are thrown to the system event log(MCE events that do not cause a crash).

It's also worth noting that MCE's can be caused by things other than overclocking.

 

[GonzoP]

So somehow along the way I guess I forgot/never realized/didn't care that my ram was running 2T instead of 1T. Thanks to Zoson et al I'm back at 1T. I also, with the slight adjustments I learned about, have bumped up to 4.5 [205x22] as my new daily driver. I'm still under all the QPI/PLL/VCore warning levels so I'm fine here. I can run 4.6 stable but then I am maxed on all the recommended voltage levels QPI 1.35v/ PLL 1.88v and Vcore is at 1.45v and my temps are up at that speed as well. I'll call it good here at 4.5 and likely be done. I'm sure when winter hits I'll make some nonsense attempts at 5Ghz with my office window wide open but really this is good.
I mean until yesterday I had spent a few weeks running and was content to run 4.4Ghz, 200x22 with ram at 9-9-9-24-2T and now with just as much if not more stability I'm at 4.5Ghz, 205x22 with ram at 8-8-8-22-1T. Plus to gain stability at 4.6 I had to set the ram back to 2T or change the CAS timing to 9 from 8. So yeah I'll sacrifice 100Mhz of clockspeed for less vcore/PLL/QPI and tighter ram timings. But yeah I really do think this is it for me here.
At some point this fall I might replace my 6x2GB with a 3x8GB. I'll probably get at least ddr3 1600 if not 1866 or whatever the best deal is. So this is not the end all of my tweaking but assuredly this represents 99% good for me. I'd drive myself crazy trying to turn 4.6 into my daily driver with 100% stability. So I'm done for now.

 

[Zoson]

So somehow along the way I guess I forgot/never realized/didn't care that my ram was running 2T instead of 1T. Thanks to Zoson et al I'm back at 1T. I also, with the slight adjustments I learned about, have bumped up to 4.5 [205x22] as my new daily driver. I'm still under all the QPI/PLL/VCore warning levels so I'm fine here. I can run 4.6 stable but then I am maxed on all the recommended voltage levels QPI 1.35v/ PLL 1.88v and Vcore is at 1.45v and my temps are up at that speed as well. I'll call it good here at 4.5 and likely be done. I'm sure when winter hits I'll make some nonsense attempts at 5Ghz with my office window wide open but really this is good.
I mean until yesterday I had spent a few weeks running and was content to run 4.4Ghz, 200x22 with ram at 9-9-9-24-2T and now with just as much if not more stability I'm at 4.5Ghz, 205x22 with ram at 8-8-8-22-1T. Plus to gain stability at 4.6 I had to set the ram back to 2T or change the CAS timing to 9 from 8. So yeah I'll sacrifice 100Mhz of clockspeed for less vcore/PLL/QPI and tighter ram timings. But yeah I really do think this is it for me here.
At some point this fall I might replace my 6x2GB with a 3x8GB. I'll probably get at least ddr3 1600 if not 1866 or whatever the best deal is. So this is not the end all of my tweaking but assuredly this represents 99% good for me. I'd drive myself crazy trying to turn 4.6 into my daily driver with 100% stability. So I'm done for now.

Lets see the CPUz, CPUt and HWMonitor shots of your fine tuned beast!

 

[primetime]

Lets see the CPUz, CPUt and HWMonitor shots of your fine tuned beast!

What is cput and what does it show that the others apps don't? was just wondering since i don't recall using that program

 

[Zoson]

What is cput and what does it show that the others apps don't? was just wondering since i don't recall using that program

CPU-Tweaker, it's in all my screenshots, just take a look to see!

 

[Rock&Roll]

Tried some more tonight to work on my overclock. Currently got a few things I'm not sure I'm gonna be able to overcome. These may just be common issues with the E760 Classified......


1) I can lower the QPI mufti to x16 alleviating the 3900MHz limit by turning off "QPI Fast Mode" in my bios. However, doing so has some odd side effects. For one, when the screen redraws/clears after some commands, it will sometimes do so very slowly. It will literally do a slow vertical scan down the screen. When QPI Fast Mode is enabled, still happens, but so fast you won't notice it.

Once in windows, the systems just feels noticably slower and theres more hitches and catches in programs loading and drawing on the screen. I've tried adjusting RAM timings, MCH strap, and voltages everywhere I could think of....still hard lock the system at 225-226 BCLK. Interestingly enough, when I was running a QPI of over 4GHZ (224 BCLK x18) over the weekend, the system was remarkably stable for short 20-30minute runs of LinX.


Voltages shown are my baseline for running 211 BCLK. I left QPI Signal Compensation at less. I figure if it's good for getting me to 224BCLK, it can't be bad for 211. I have yet to have any short-run stress test fail at these settings. System is snappy and responsive.

2) My turbo multi doesn't work (BIOS 83)... My CPU seems to be more than good to run stable at 4400+MHz....just can't get the turbo multi to kick in. My BIOS even reports target frequency of 4400MHz


3) Cold boot issues. I've read it's common. Just never found anyone with a solution. My Cold boot issues (Code 25) were temporarily resolved by raising PCIE frequency to 101 (reports 99mHz in windows when set to 100 in bios). Today, cold boot was back with a vengance. Sometimes had to reset 3 or 4 different times to get to POST. Adjusting voltages didn't seem to make a difference here either. (QPI PLL seems to be very important on this board since upgrading to this x5650)

4) Amazingly, this CPU seems quite happy to run 4200mHz with HT turned on with little-to-no voltage increase. I really don't know what to do with that, since I've always just turned off HT. Since all the stress tests I know cause contention issues by running 12 threads (unless set to 6), how to I 100% load the CPU to test for stability? Or is 6 threads effective for a proper 100% stable overclock with HT enabled?

Anyway, here's where I stand tonight, in terms of looking for a good 24/7 hard-as-nails overclock. Only thing that really bugs me about it is I just wanted to know how far I could push this chip. Looks like the motherboard is going to get in the way first.