Overclocking Question

[Sartanius]

In an effort to squeeze all that I can out of my i7-5930 (spec in my signature) I have been digging more intentionally into overclocking. While I have used the builtin ASUS UEFI overclock presets for a long time, I decided to start trying my own settings.  It's never too late to teach and old dog new tricks! 

I eventually arrived at a 3 core ratio of 4.5 GHz with the remaining 3 cores at 4.3 GHz with a voltage of 1.350v.  Stressing with ASUS' RealBench for 2 hours and all is stable.  Used Prime95 for about 20 minutes, and it was stable, but CPU package temps were pretty high (~82C).

Thinking I had a pretty stable O/C, I did a flight and about half way through the computer froze with the typical blue screen crash, then restarted.  It is disappointing to see that P3D stresses the O/C more than Prtime95 and RealBench!  Anyway, is the best thing to do at this stage to leave the core speed settings alone and try adjusting voltage incrementally first?  Just looking for general guidance on which parameter is more likely to improve stability.

Thanks in advance for any suggestions.

[Benjamin J]

So let me just say upfront that I'm a total novice when it comes to overclocking, but from what I've read a 1hr Prime95 test seems to be the recommended length for that test. You might want to redo it and see how it goes if you stress the computer for more than 20 minutes? I wouldn't be surprised if the length of your flight was quite a bit longer than the 20 minute Prime95 test.

[martin-w]

On 08/02/2018 at 3:42 PM, dmiannay said:

In an effort to squeeze all that I can out of my i7-5930 (spec in my signature) I have been digging more intentionally into overclocking. While I have used the builtin ASUS UEFI overclock presets for a long time, I decided to start trying my own settings.  It's never too late to teach and old dog new tricks! 

I eventually arrived at a 3 core ratio of 4.5 GHz with the remaining 3 cores at 4.3 GHz with a voltage of 1.350v.  Stressing with ASUS' RealBench for 2 hours and all is stable.  Used Prime95 for about 20 minutes, and it was stable, but CPU package temps were pretty high (~82C).

Thinking I had a pretty stable O/C, I did a flight and about half way through the computer froze with the typical blue screen crash, then restarted.  It is disappointing to see that P3D stresses the O/C more than Prtime95 and RealBench!  Anyway, is the best thing to do at this stage to leave the core speed settings alone and try adjusting voltage incrementally first?  Just looking for general guidance on which parameter is more likely to improve stability.

Thanks in advance for any suggestions.

 

Which version of Prime95? Newer versions run AVX 2 and thus generate a great deal of heat. It stresses aspects of the CPU that you probably wont in everyday use. 

It is a 6 core CPU, so it wont overclock as high as 4 core CPU's and indeed, there will be more heat generated.

82C in Prime95,  if it's a newer version of Prime is to be expected. What was the temp when running RealBench, and what's your ambient temp? Also how are you cooling your CPU?

OCCT you could try, which often picks up instability other stress tests don't, but yes, the real way to determine stability is running the application you use the PC for. 

If it were me, I would not use anything newer than Prime 95 version 26.6 which is without AVX. Normally I use RealBench for 4 hours.  And sometimes OCCT. You can also try XTU, Intel's Extreme Tuning Utility. 

I would also sync all cores, running them at the same frequency rather than per core. 

As I said though, before advising you increase voltage, I would be interested in the Prime version, and temp in RealBench. 

[Sartanius]

14 hours ago, Benjamin J said:

So let me just say upfront that I'm a total novice when it comes to overclocking, but from what I've read a 1hr Prime95 test seems to be the recommended length for that test. You might want to redo it and see how it goes if you stress the computer for more than 20 minutes? I wouldn't be surprised if the length of your flight was quite a bit longer than the 20 minute Prime95 test.

That's a good point, and lines up with what said regarding how long to run the stress tests.  Thanks for weighing-in.

 

45 minutes ago, martin-w said:

 

Which version of Prime95? Newer versions run AVX 2 and thus generate a great deal of heat. It stresses aspects of the CPU that you probably wont in everyday use. 

It is a 6 core CPU, so it wont overclock as high as 4 core CPU's and indeed, there will be more heat generated.

82C in Prime95,  if it's a newer version of Prime is to be expected. What was the temp when running RealBench, and what's your ambient temp? Also how are you cooling your CPU?

OCCT you could try, which often picks up instability other stress tests don't, but yes, the real way to determine stability is running the application you use the PC for. 

If it were me, I would not use anything newer than Prime 95 version 26.6 which is without AVX. Normally I use RealBench for 4 hours.  And sometimes OCCT. You can also try XTU, Intel's Extreme Tuning Utility. 

I would also sync all cores, running them at the same frequency rather than per core. 

As I said though, before advising you increase voltage, I would be interested in the Prime version, and temp in RealBench. 

Very detailed and considered response... thanks, Martin.  

Using Prime95 v29.3.  The RealBench temps were 78C for the 2 hour run.  My ambient here in the room is 72*F (22.2*C).

When I tried running all cores at 4.5 with 1.350v I couldn't get through a 15 minute RealBench run.  I dropped down to 4.4 at 1.350v, and still no joy.  Probably didn't do well in the silicone lottery with this CPU.  I then changed the settings to a 3-core ratio of 4.5, the remaining 3 cores at 4.3, and the voltage at 1.350.  This was what I used for the testing that started this thread. 

After my flight crashed, I posted this thread to get suggestions on the best way to proceed.  When I didn't hear anything right away, I decided to change to 2 cores at 4.5, and the remaining 4 cores at 4.3, which is where I am now. I did the same 3 hour flight with that new setting with no crashes.  But, still not sure this setting will be stable for a longer flight.

I'll run RealBench for 4 hours later today to see how that goes.  I'll also look for the earlier version of Prime95 you mentioned and run that for at least 1 hour.

Again, thank you for helping me with this.  I have much to learn.

[MDFlier]

I'm not a "pro" overclocker, but I've been doing it since the 486 days. I once got a 486 to run at  133Mhz!. That baby ran FS9 pretty well. I was in tall cotton, as they say. 

It has been my experience that FSX (and I'm certain that P3D is the same) can be a better diagnostic tool than many actual diagnostic tools. I've seen plenty of systems that will pass stress tests and deep level diagnostic routines, but will fail to run an entire flight in FSX. FSX excels in utilizing HW resources (excluding CPU cores, but they only gave cursory consideration to dual core threading as there were only a handful of dual core CPUs arrived in the scene in the early 2000s. The ACES team did not believe that multi core CPUs would be the new normal at the time they built FSX). My point? Prime95 is a boring waste of time. Just test your OC straight away in FSX. As you can see, running Prime95 for X hours doesn't mean it will run FSX for 5 minutes.

I'd be curious to know if the mismatched core speeds might be causing timing problems between threads. It would seem to me that if some threads are running slower, the faster ones might be completed and sitting around waiting for the slower ones to finish. It just doesn't seem that any code waiting for other code to complete would make for a nice, steady overall flow from a code execution standpoint.  

[Sartanius]

10 minutes ago, MDFlier said:

It has been my experience that FSX (and I'm certain that P3D is the same) can be a better diagnostic tool than many actual diagnostic tools. I've seen plenty of systems that will pass stress tests and deep level diagnostic routines, but will fail to run an entire flight in FSX. ... Just test your OC straight away in FSX. As you can see, running Prime95 for X hours doesn't mean it will run FSX for 5 minutes.

I am beginning to understand that.  Thanks for your observation.

 

11 minutes ago, MDFlier said:

I'd be curious to know if the mismatched core speeds might be causing timing problems between threads. It would seem to me that if some threads are running slower, the faster ones might be completed and sitting around waiting for the slower ones to finish. It just doesn't seem that any code waiting for other code to complete would make for a nice, steady overall flow from a code execution standpoint.  

That's a great point, but I hope it's not true.  I have had no success with all cores at 4.5 or 4.4 and 1.350v (using RealBench and P3D).  But there does seem to be some stability with mismatched core speeds.  If I go back to all cores at 4.5, do you believe adjusting voltage up/down would be a way to find stability?

[lownslo]

3 hours ago, MDFlier said:

It has been my experience that FSX (and I'm certain that P3D is the same) can be a better diagnostic tool than many actual diagnostic tools.

This is especially important if P3D is your only "game"... using it to stress test the system will make your overclocking life much easier.

I agree with Martin about Prime95 and syncing the cores.

Couple of questions... is the CPU delidded?  If not then your overclocking will usually be limited (and more difficult).  Also, please tell us more about your memory... is it overclocked, and if so how did you do it.  Did you record/research the BSOD code when the sim crashed (that incident can be very useful in troubleshooting your overclock)?

Greg

[martin-w]

19 hours ago, lownslo said:

 

Couple of questions... is the CPU delidded?  If not then your overclocking will usually be limited (and more difficult).  

Greg

 

I'm pretty sure the i7 5930 is soldered Greg, not TIM. So no, no delidding.

 

On 10/02/2018 at 1:06 PM, dmiannay said:

 

Very detailed and considered response... thanks, Martin.  

Using Prime95 v29.3.  The RealBench temps were 78C for the 2 hour run.  My ambient here in the room is 72*F (22.2*C).

 

 

29.3 runs AVX then, and is notorious for very high temps. precisely why the latest platforms have an AVX offset in the BIOS. That way you can add a value that it will clock down to to avoid the issue.

If it's 78 degrees in RealBench, and that's core temp, measured by something like RealTemp or CoreTemp, or HWMonitor, then I wouldn't exceed that if I were you, that's my preference anyway. TJMax is around 100 degrees, but conventional wisdom has it that anything over 80 is a bit too close to TJMax for long term use.  Especially in the sim, because as you've discovered, it's quite demanding on the CPU and generates higher temps than a less CPU biased application. 

The most important factor is your cooler of course? If you have a cooler that's less than the best, then upgrading would allow you more thermal headroom. 

If you would like to run per core instead of syncing all cores, that's fine, in fact it's the way Intel Turbo works anyway, but many overclockers prefer all cores the same frequency. I doubt there would be any negatives to per core, except perhaps in terms of which cores your P3D is actually utilising and for what. You may find that aspects of the sim you would like to be running on a high clocked core, actually aren't.

At the end of the day, I would say don't get obsessed with just a couple of hundred megahertz, in terms of frame rate increase at 30 FPS it's a very small percentage and would generate 1-2 frames per second extra at best.

 

[MDFlier]

7 minutes ago, martin-w said:

If you would like to run per core instead of syncing all cores, that's fine, in fact it's the way Intel Turbo works anyway, but many overclockers prefer all cores the same frequency. I doubt there would be any negatives to per core, except perhaps in terms of which cores your P3D is actually utilising and for what. You may find that aspects of the sim you would like to be running on a high clocked core, actually aren't.

At the end of the day, I would say don't get obsessed with just a couple of hundred megahertz, in terms of frame rate increase at 30 FPS it's a very small percentage and would generate 1-2 frames per second extra at best.

Good advice. My only concern with per core is as I stated, there was not a lot of thought given into thread safety (preventing threads from stepping on each other's resources) or thread scheduling when FSX was complied. ACES just didn't think it was necessary because they though that multi core CPUs were a novelty at the time, only around until the new single core CPUs exhibited performance well beyond what the dual core CPUs of the day had. History shows that they were wrong. So... Modern SW written and compiled with multi threading in mind should handle unsynced cores just fine, because all of the threads are aware of each other and the the SW has built in provisions for playing "traffic cop" to make sure that everything gets done in the proper order. When we run FSX on a multi core CPU, all of the thread distribution and timing is being handled by the hardware and OS. FSX isn't really managing all of those threads. I'm just not 100% confident that given a lack of thread awareness in FSX that the HW and OS will always get things done in the correct way. It's a bit of a hack.  

Doug, I would look for a BIOS option that will allow the MB to increase the voltage to the CPU as needed. My old ASUS board called it "Turbo boost" or something like that. My ASRock board had a different name for the same feature. My Gigabyte board calls it "CPU Vcore loadline calibration". Changing that setting to "Turbo" was the last thing I had to do in my current setup to get it stabilized at 4.8Ghz, which was my intended target. It simply applies more voltage to the CPU when needed, which allows you to set the actual CPU voltage lower to keep temps down. When the CPU needs more power, the MB can give it some.  

[martin-w]

2 hours ago, MDFlier said:

Good advice. My only concern with per core is as I stated, there was not a lot of thought given into thread safety (preventing threads from stepping on each other's resources) or thread scheduling when FSX was complied. ACES just didn't think it was necessary because they though that multi core CPUs were a novelty at the time, only around until the new single core CPUs exhibited performance well beyond what the dual core CPUs of the day had. History shows that they were wrong. So... Modern SW written and compiled with multi threading in mind should handle unsynced cores just fine, because all of the threads are aware of each other and the the SW has built in provisions for playing "traffic cop" to make sure that everything gets done in the proper order. When we run FSX on a multi core CPU, all of the thread distribution and timing is being handled by the hardware and OS. FSX isn't really managing all of those threads. I'm just not 100% confident that given a lack of thread awareness in FSX that the HW and OS will always get things done in the correct way. It's a bit of a hack.  

Doug, I would look for a BIOS option that will allow the MB to increase the voltage to the CPU as needed. My old ASUS board called it "Turbo boost" or something like that. My ASRock board had a different name for the same feature. My Gigabyte board calls it "CPU Vcore loadline calibration". Changing that setting to "Turbo" was the last thing I had to do in my current setup to get it stabilized at 4.8Ghz, which was my intended target. It simply applies more voltage to the CPU when needed, which allows you to set the actual CPU voltage lower to keep temps down. When the CPU needs more power, the MB can give it some.  

 

I would have to defer to your better knowledge regarding that. I guess as I haven't done any simming in a long time, it's not something I consider these days. 

[Sartanius]

On 2/10/2018 at 1:03 PM, lownslo said:

Couple of questions... is the CPU delidded?  If not then your overclocking will usually be limited (and more difficult).  Also, please tell us more about your memory... is it overclocked, and if so how did you do it.  Did you record/research the BSOD code when the sim crashed (that incident can be very useful in troubleshooting your overclock)?

Greg

Hi Greg.  As Martin correctly noted, no, the 5930 is not delidded.  As for memory, I'm running 32GB of Corsair Vengeance DDR4 2400 (PC4-19200).  I have XMP enabled in the BIOS and no other overclocking.  As for recording the BSOD fault, I am not sure how to check into that.  I do run AppCrashView, but am not sure how that information helps me.

 

4 hours ago, martin-w said:

If it's 78 degrees in RealBench, and that's core temp, measured by something like RealTemp or CoreTemp, or HWMonitor, then I wouldn't exceed that if I were you, that's my preference anyway. TJMax is around 100 degrees, but conventional wisdom has it that anything over 80 is a bit too close to TJMax for long term use.  Especially in the sim, because as you've discovered, it's quite demanding on the CPU and generates higher temps than a less CPU biased application. 

The most important factor is your cooler of course? If you have a cooler that's less than the best, then upgrading would allow you more thermal headroom. 

If you would like to run per core instead of syncing all cores, that's fine, in fact it's the way Intel Turbo works anyway, but many overclockers prefer all cores the same frequency. I doubt there would be any negatives to per core, except perhaps in terms of which cores your P3D is actually utilising and for what. You may find that aspects of the sim you would like to be running on a high clocked core, actually aren't.

At the end of the day, I would say don't get obsessed with just a couple of hundred megahertz, in terms of frame rate increase at 30 FPS it's a very small percentage and would generate 1-2 frames per second extra at best.

 

Martin - in the 4 hour RealBench run I completed yesterday the max temp was 79*C for the Package as displayed in HWMonitor.  I am currently near the end of a long (6-hour) flight with the PMDG 747 with significant weather and haven't have any issues yet.  Highest temp recorded so far is 77*C.  So far, so good.  The cooler is the Corsair Hydro Series H110 Extreme.  I removed and replaced the heat sink on the CPU with a fresh dab of thermal paste, so I believe I have good contact for heat transfer.

I agree with your "obsessed" comment... I'm not getting overwhelmed with splitting hairs here.  Most of the exercise was intended to help me learn more about overclocking.  While still a newbie, I am making some progress... thanks to the help for folks like you.

3 hours ago, MDFlier said:

Doug, I would look for a BIOS option that will allow the MB to increase the voltage to the CPU as needed. My old ASUS board called it "Turbo boost" or something like that. My ASRock board had a different name for the same feature. My Gigabyte board calls it "CPU Vcore loadline calibration". Changing that setting to "Turbo" was the last thing I had to do in my current setup to get it stabilized at 4.8Ghz, which was my intended target. It simply applies more voltage to the CPU when needed, which allows you to set the actual CPU voltage lower to keep temps down. When the CPU needs more power, the MB can give it some.  

That's a great suggestion.  As soon as this flight completes, I'll dive back into the BIOS to see what is offered.  I know I've seen something like that, so I'll make the change and see if things remain stable.

I really do appreciate all of you weighing in on this.  Using P3D to do the actual stress testing is indeed the best way forward, and it already seems to be more stable.  Time will tell!

[martin-w]

I Agree with your "obsessed" comment... I'm not getting overwhelmed with splitting hairs here.  Most of the exercise was intended to help me learn more about overclocking.  While still a newbie, I am making some progress... thanks to the help for folks like you.

Oh yes, I know you aren't, it wasn't a criticism. It's just that many don't know that overclocking in a well balanced system is linear. And thus, a few hundred megahertz is minimal in terms of frame rate. So it's important to point that out so that a decision can be made in regard to whether any negative aspects of overclocking are worth it.

[Sartanius]

5 hours ago, martin-w said:

Oh yes, I know you aren't, it wasn't a criticism.

The thought never even crossed my mind, Martin.  Again, thanks!

[lownslo]

12 hours ago, martin-w said:

I'm pretty sure the i7 5930 is soldered Greg, not TIM. So no, no delidding.

Ahh, thanks Martin.  I recalled that the 5xxx series were the first to use non-soldered TIM... not so though.

Doug... I have a low tech solution for capturing the info of a BSOD.  I just take a picture of it while it's still on my screen.  Then I research the info (particularly the fault code).  Also, Windows Event Viewer can be useful in diagnosing a system crash (although a BSOD is more often than not hardware related).

The reason I asked about your memory has to do with the voltage you're running on it.  Sometimes adding a bit more (.1 volt for example) can lend stability to the memory.  But I wouldn't recommend doing that until you have a better idea what is causing the system to crash.  I agree with the suggestions of pulling the frequency back until the system is stable, then tweaking for more overclocking performance if you wish.  And if you do get more crashes try to collect info as possible.  Plenty of folks here who can diagnose that crash info.

HTH,

Greg

[Sartanius]

48 minutes ago, lownslo said:

Doug... I have a low tech solution for capturing the info of a BSOD.  I just take a picture of it while it's still on my screen.  Then I research the info (particularly the fault code).  Also, Windows Event Viewer can be useful in diagnosing a system crash (although a BSOD is more often than not hardware related).

I would normally do this as well, Greg, but the BSOD is graphically scrambled on the screen with no discernible text. Didn't think to look at WEV, but will do so now to see if there are some clues. 

The good news, though... just finished a 7 hour flight in the PMDG747 from KSDF to PANC with no issues.  CPU package temp peak was 77*C, which is acceptable.  I just might have finally stumbled upon a stable o/c.  Time will tell.

52 minutes ago, lownslo said:

The reason I asked about your memory has to do with the voltage you're running on it.  Sometimes adding a bit more (.1 volt for example) can lend stability to the memory.  But I wouldn't recommend doing that until you have a better idea what is causing the system to crash.  I agree with the suggestions of pulling the frequency back until the system is stable, then tweaking for more overclocking performance if you wish.  And if you do get more crashes try to collect info as possible.  Plenty of folks here who can diagnose that crash info.

As I said above, I think I'm stable now.  If I start to see instability again., I'll give this a look.  Thanks again!