Stress Testing - but NOT all cores...

[Scoob]

Hi all

I'm new here today, but wanted to post a new topic based on my overclocking experiences last night.

This is based on my system, which at it's heart is a 2500k on an Asus P8Z68-V Pro motherboard.

Now, I've been running at 4.5 since the day I got this chip - very, very hapy with Sandy Bridge I must say.

At these speeds I can run wPrime without issue, leave Prime 95 blend running for hours. I can even run the massively demanding IBT high stress test without issue - though my temps get into the low to mid-70's - a good 10c higher than either Prime pushes them.

Anyway, these multi-core processors are great. We all know this. More and more games too are starting to utlise this. Fantastic

However, some games still won't fully utilise all cores, generally only pushing a couple of them. After some random game crashes with games who exhibited this exact behaviour - just pushing two cores - I decided to do some additional stability testing on my rig.

I ran Prime 95 for a while, blend, four threads, just to prove nothing had gone screwy with my OC. All worked perfectly. Next I chose to run only TWO threads and I set processor affinity to cores 2 and 3. The test failed within a few seconds! After a few more tests I could see that if only two (or even three, though it took longer) cores were pushed my machine was less than stable. I wondered if I was just having a problem with Prime 95 so I fired up the MUCH more stressful IBT - the same IBT I been running without issue for ages previously.

So, IBT, two threads, affinity to just a pair of cores...go. Within 10 seconds I saw the blue screen - a sight I'd near forgotten!

Well this was vexing...why would my PC be sooo unstable when only a pair of cores were pushed when the vCore was more than adequate for all FOUR cores to be working hard? Were my "idle" cores somehow unstable because there was too much vCore for the work they were (or weren't really) doing?

I tried giving my machine a bit more vCore...IBT = CTD. I tried tweaking things like SysIO etc. but that didn't help.

So, I had a machine that was perfecly stable in Prime 95 for hours and survived IBT without a hitch when pushing four cores, but when just a pair of cores were stressed - a more realisic "test" for many games - it failed badly.

This did now explain certain CTD's I'd been getting in games that generally just push two cores. I'd hear the fans ramp up, then the game would drop...even though I'd say I was rarely hitting more than 80/90% on two cores...troubling.

Ok, to cut my evening of tweaking short, I DID manage to sort things. Basically I'd been running an LLC of "High" - which is a middle-ish setting on my motherboard. However, with two cores being push I needed to set this to "Very High" to get any sort of stability. Just that one change made be 100% stable in all the test I'd previously failed - I changed nothing else, having reverted to a prior profile after the earlier changes to be sure.

So, during testing, I now see slightly higher vCore reported when all FOUR cores are stressed and somewhat lower when just a pair of cores are.

I do wonder how many other peoples "stable" overclocks might fail such a test as I've performed? A test that, I think I've proven, is likely more realistic in terms of actual gaming system stress.

I'd be interested to hear peoples thoughts on this - it did vex me for quite some time yesterday evening

Apologies for the long post, hopefully it wasnt' too dull for all my fellow enthusiasts out there

Cheers,

Scoob.

 

[SmoulderingCircuits]

Thanks for sharing your venture with us

I was thinking LLC as I was reading down. The trouble I had with my 1090 was if you try using turbo (3 cores) at a 0.5 higher multiplier you wind up with insane frequency and voltage you never asked for. Might be the same even with Sandy Bridge that once you get to larger frequencies they aren't as consistant when not using every single thread as they should be.

Infact (some may remember the post) I got 4.8Ghz (according to AOD) by setting my turbo to 18 from the base 17.5 multi - which, with a 232Mhz BUS never should have happened and with 1.65v propping it up couldn't be kept

 

[Scoob]

Hi JADOOM,

I think I may have been initially overly cautious on my LLC settings. I did, for a short while, have an MSI GD45 P67 motherboard which had TERRIBLE voltage control. I remember trying the higher LLC setting (it had two) on that and it added 0.15v under load! Scary! Vdrop was quite high, vDroop was minimal but it waaay over-compensated for even that small amount - making getting the desired vCore under changing load conditions impossible.

The Asus by contrast offer MUCH finer and more accurate control, and also gives much faster switching during sudden load changes. Ended up getting a refund on the MSI as it started doing weird things, the NIC failed to connect at more that 100mbps after day 1 and I lost all mouse control in the new UEFI BIOS...

Also my old MSI would never properly lower the vCore when dropping into idle state if you'd set anthing other than auto - sorta spoilt my initial Sandy B fun. The Asus on the other hand bases it vCore on the VID at the given multiplier, you then provide an adjustment to this.

Ultimately I'd love to see a motherboard that lets you set individual vCore for EACH multipler and different settings for high and low load at said multi...

Cheers,

Scoob.

 

[SmoulderingCircuits]

Hi JADOOM,

I think I may have been initially overly cautious on my LLC settings. I did, for a short while, have an MSI GD45 P67 motherboard which had TERRIBLE voltage control. I remember trying the higher LLC setting (it had two) on that and it added 0.15v under load! Scary! Vdrop was quite high, vDroop was minimal but it waaay over-compensated for even that small amount - making getting the desired vCore under changing load conditions impossible.

The Asus by contrast offer MUCH finer and more accurate control, and also gives much faster switching during sudden load changes. Ended up getting a refund on the MSI as it started doing weird things, the NIC failed to connect at more that 100mbps after day 1 and I lost all mouse control in the new UEFI BIOS...

Also my old MSI would never properly lower the vCore when dropping into idle state if you'd set anthing other than auto - sorta spoilt my initial Sandy B fun. The Asus on the other hand bases it vCore on the VID at the given multiplier, you then provide an adjustment to this.

Ultimately I'd love to see a motherboard that lets you set individual vCore for EACH multipler and different settings for high and low load at said multi...

Cheers,

Scoob.

Maybe they'll bring vcore and other variable to variable tethered profiles in future bios updates but it's more likely they'll come with new board releases (more money for them). I don't really know why they currently class 100% LLC on S/B as 'extreme' it's not really extreme when you consider 890FX has all or nothing as regards LLC (don't know about 990FX).

As for MSIs products I find if you intend to leave them at stock then they will perform exceptionally but soon as you tweak them they become unstable or outright keel over beyond repair. Definately buying from Asus in future. MSI may have better aesthetics and cooling but the individual chips underneath the heatsinks are not that great IMO

 

[SPS]

Thanks for sharing your venture with us

I was thinking LLC as I was reading down. The trouble I had with my 1090 was if you try using turbo (3 cores) at a 0.5 higher multiplier you wind up with insane frequency and voltage you never asked for. Might be the same even with Sandy Bridge that once you get to larger frequencies they aren't as consistant when not using every single thread as they should be.

Infact (some may remember the post) I got 4.8Ghz (according to AOD) by setting my turbo to 18 from the base 17.5 multi - which, with a 232Mhz BUS never should have happened and with 1.65v propping it up couldn't be kept

I truly believe that was AOD's fault and not the correct reading because I did the same test and got that on AOD but on other monitoring software like CPUID and HWMonitor the readings were as expected. Like you said prevoiusly your AOD had wrong readings.

 

[SmoulderingCircuits]

I truly believe that was AOD's fault and not the correct reading because I did the same test and got that on AOD but on other monitoring software like CPUID and HWMonitor the readings were as expected. Like you said prevoiusly your AOD had wrong readings.

I am more inclined to believe that as I still have a working CPU today

Nonetheless it never did that when I had turbo on at stock

 

[SPS]

I am more inclined to believe that as I still have a working CPU today

Nonetheless it never did that when I had turbo on at stock

Yeah

Might be a bug with AOD reading turbo above a certain speed - who knows

 

[SPS]

Hi all

I'm new here today, but wanted to post a new topic based on my overclocking experiences last night.

This is based on my system, which at it's heart is a 2500k on an Asus P8Z68-V Pro motherboard.

Now, I've been running at 4.5 since the day I got this chip - very, very hapy with Sandy Bridge I must say.

At these speeds I can run wPrime without issue, leave Prime 95 blend running for hours. I can even run the massively demanding IBT high stress test without issue - though my temps get into the low to mid-70's - a good 10c higher than either Prime pushes them.

Anyway, these multi-core processors are great. We all know this. More and more games too are starting to utlise this. Fantastic

However, some games still won't fully utilise all cores, generally only pushing a couple of them. After some random game crashes with games who exhibited this exact behaviour - just pushing two cores - I decided to do some additional stability testing on my rig.

I ran Prime 95 for a while, blend, four threads, just to prove nothing had gone screwy with my OC. All worked perfectly. Next I chose to run only TWO threads and I set processor affinity to cores 2 and 3. The test failed within a few seconds! After a few more tests I could see that if only two (or even three, though it took longer) cores were pushed my machine was less than stable. I wondered if I was just having a problem with Prime 95 so I fired up the MUCH more stressful IBT - the same IBT I been running without issue for ages previously.

So, IBT, two threads, affinity to just a pair of cores...go. Within 10 seconds I saw the blue screen - a sight I'd near forgotten!

Well this was vexing...why would my PC be sooo unstable when only a pair of cores were pushed when the vCore was more than adequate for all FOUR cores to be working hard? Were my "idle" cores somehow unstable because there was too much vCore for the work they were (or weren't really) doing?

I tried giving my machine a bit more vCore...IBT = CTD. I tried tweaking things like SysIO etc. but that didn't help.

So, I had a machine that was perfecly stable in Prime 95 for hours and survived IBT without a hitch when pushing four cores, but when just a pair of cores were stressed - a more realisic "test" for many games - it failed badly.

This did now explain certain CTD's I'd been getting in games that generally just push two cores. I'd hear the fans ramp up, then the game would drop...even though I'd say I was rarely hitting more than 80/90% on two cores...troubling.

Ok, to cut my evening of tweaking short, I DID manage to sort things. Basically I'd been running an LLC of "High" - which is a middle-ish setting on my motherboard. However, with two cores being push I needed to set this to "Very High" to get any sort of stability. Just that one change made be 100% stable in all the test I'd previously failed - I changed nothing else, having reverted to a prior profile after the earlier changes to be sure.

So, during testing, I now see slightly higher vCore reported when all FOUR cores are stressed and somewhat lower when just a pair of cores are.

I do wonder how many other peoples "stable" overclocks might fail such a test as I've performed? A test that, I think I've proven, is likely more realistic in terms of actual gaming system stress.

I'd be interested to hear peoples thoughts on this - it did vex me for quite some time yesterday evening

Apologies for the long post, hopefully it wasnt' too dull for all my fellow enthusiasts out there

Cheers,

Scoob.

This maybe why my OC failed on games but could run Prime all day long. I'm running a 1090T though and I don't have an option to change the LLC state I don't think. I got tired and gave up so I'm running stock at the moment

 

[SmoulderingCircuits]

This maybe why my OC failed on games but could run Prime all day long. I'm running a 1090T though and I don't have an option to change the LLC state I don't think. I got tired and gave up so I'm running stock at the moment

Not even in the 1304 update? I thought you could...

 

[SPS]

Not even in the 1304 update? I thought you could...

Possibly only had that for a little while and haven't had chance to look through it

 

[Scoob]

Hi,

It may be that SB chips just work a little differently to those that went before.

When I OC'd my Q6600 (2.4) to 3.6 I changed little other than the vCore yet it ran trouble-free for years. I did however lock it at 3.6, unlike the 2500k which I allow to idle. Allowing chips to idle like that used to be a no-no to get a stable OC - not true with SB.

Afraid I cannot really comment on AMD overclocking, my last AMD was the 4200x2 and, while it's still going, really doesn't like being overclocked any more. Back then getting it to 2.8 from stock 2.2 was considered rather good - all on air of course.

Still, back to stability, I've had friends that cannot comprehend that their machine can be unstable because it runs wPrime for a few minutes ok. Though they do like to comment on how all modern games seem to crash all the time...

I'm pleased my machine is now behaving its self properly, but was surpised by the results of my testing.

JADOOM: I agree, 100% should be "normal" or "default" but I guess they don't sound as sexy as "Extreme" or "Ultimate" or whatever..

On the MSI board... it was one of their cheaper ones, but seemed to be well put together using decent components. Being a "P" chipset it was in my mind an overclockers board, but it simply could not offer the stability I expected...it was still a £100 board at the time after all. I would possibly buy MSI again, but not from their low-end range. Thing that annoyed me was how they fixed a bug in a BIOS release, but that the same time removed a feature/voltage control option to separate the cheaper board from it's more expensive siblings...grumble...

Cheers,

Scoob.

 

[SmoulderingCircuits]

Hi,

It may be that SB chips just work a little differently to those that went before.

When I OC'd my Q6600 (2.4) to 3.6 I changed little other than the vCore yet it ran trouble-free for years. I did however lock it at 3.6, unlike the 2500k which I allow to idle. Allowing chips to idle like that used to be a no-no to get a stable OC - not true with SB.

Afraid I cannot really comment on AMD overclocking, my last AMD was the 4200x2 and, while it's still going, really doesn't like being overclocked any more. Back then getting it to 2.8 from stock 2.2 was considered rather good - all on air of course.

Still, back to stability, I've had friends that cannot comprehend that their machine can be unstable because it runs wPrime for a few minutes ok. Though they do like to comment on how all modern games seem to crash all the time...

I'm pleased my machine is now behaving its self properly, but was surpised by the results of my testing.

JADOOM: I agree, 100% should be "normal" or "default" but I guess they don't sound as sexy as "Extreme" or "Ultimate" or whatever..

On the MSI board... it was one of their cheaper ones, but seemed to be well put together using decent components. Being a "P" chipset it was in my mind an overclockers board, but it simply could not offer the stability I expected...it was still a £100 board at the time after all. I would possibly buy MSI again, but not from their low-end range. Thing that annoyed me was how they fixed a bug in a BIOS release, but that the same time removed a feature/voltage control option to separate the cheaper board from it's more expensive siblings...grumble...

Cheers,

Scoob.

Maybe MSI are one of those companies with a guy in one department who organises a really good product them some annoying troll further up tells him to remove some of it's best features.

Old AMD chips had horrific OCing

I can get 4.4Ghz bootable but get blue screens and 4.3Ghz was tempremental but it could well have been the linked HT with the CPU/NB @2.7 and 2.9Ghz that were causing the problems

 

[Scoob]

Maybe MSI are one fo those companies with a guy in one department who organises a really good product them some annoying troll further up tells him to remove some of it's best features.

Old AMD chips had horrific OCing

I can get 4.4Ghz bootable but get blue screens and 4.3Ghz was tempremental but it could well have been the linked HT with the CPU/NB @2.7 and 2.9Ghz that were causing the problems

Heh, that could well be the case

Yet we thought they were the mutts back in the day...and indeed they were a respectable bit of kit but Core2 came along...

I don't know your chip but that doesn't sound too shabby. I get the urge to "have another go" at overclocking every few weeks (as I did last night, but a slightly different agenda) yet even when successful it yields no real gain in gaming...but it's fun.

Cheers,

Scoob.

 

[SmoulderingCircuits]

Heh, that could well be the case

Yet we thought they were the mutts back in the day...and indeed they were a respectable bit of kit but Core2 came along...

I don't know your chip but that doesn't sound too shabby. I get the urge to "have another go" at overclocking every few weeks (as I did last night, but a slightly different agenda) yet even when successful it yields no real gain in gaming...but it's fun.

Cheers,

Scoob.

Run games with lots of tesselation/ physics/ objects and you'll see a difference

 

[Scoob]

Run games with lots of tesselation/ physics/ objects and you'll see a difference

True enough, however the jump from my stock of 3.3ghz to the 4.6ghz I'm running now has given the lions share of any improvement. Another 1 or 200 mhz on that is not really going to be too noticable, except in benchmarks.

I imagine your CPU, like mine, is actually more than capable at stock for the most part.

Oddly enough I'm mostly playing older games at the moment...though one is particularly CPU heavy, hence my investigation.

Cheers,

Scoob.

 

[SmoulderingCircuits]

True enough, however the jump from my stock of 3.3ghz to the 4.6ghz I'm running now has given the lions share of any improvement. Another 1 or 200 mhz on that is not really going to be too noticable, except in benchmarks.

I imagine your CPU, like mine, is actually more than capable at stock for the most part.

Oddly enough I'm mostly playing older games at the moment...though one is particularly CPU heavy, hence my investigation.

Cheers,

Scoob.

Stock 3.2Ghz is actually quite sluggish. I'd say you need 3.5 - 3.7 minimum to feel it smooth

 

[Scoob]

Stock 3.2Ghz is actually quite sluggish. I'd say you need 3.5 - 3.7 minimum to feel it smooth

Ah, I didn't know that..good that you can get things smooth for free though.

I must admit, I'm really pleased with this 2500k, I mean I've given it a 1.3ghz OC without too much effort...that should see this CPU doing the business for some time. Though saying that I'd thought of this initially as an interim system, however it was better than I expected.

My old Q6600 lasted as long as it did because of a 1.2ghz, or 50% OC - if I'd been stuck at 2.4 it would have been left behind sooner.

Cheers,

Scoob.

 

[SmoulderingCircuits]

Ah, I didn't know that..good that you can get things smooth for free though.

I must admit, I'm really pleased with this 2500k, I mean I've given it a 1.3ghz OC without too much effort...that should see this CPU doing the business for some time. Though saying that I'd thought of this initially as an interim system, however it was better than I expected.

My old Q6600 lasted as long as it did because of a 1.2ghz, or 50% OC - if I'd been stuck at 2.4 it would have been left behind sooner.

Cheers,

Scoob.

That's in my opinion at least. 3.2Ghz is certainly not terrible though

 

[Scoob]

That's in my opinion at least. 3.2Ghz is certainly not terrible though

As long as it does the job mate, it's all good.

Scoob.