Overclocking i7 6800K + Memory - P95 issues

[Djxinator]

Evening folks,

 

I'm having difficulty identifying a failure in my system. I recently built my PC into a new case and for some reason my Overclock reset. I've re-installed Windows for good measure and begun the gruelling process of overclocking once again.

 

Finally got what I believed to be a semi-stable overclock of 4.2Ghz (3.1 Cache) with memory @ 3000 C15 1T. It passed 1 hour of P95 Blend and 1 hour of P95 Small-FFTs (I usually do this to test stability whilst tuning)

 

I ran a custom P95 and was aiming to run it for 4 hours before declaring the system stable enough for me, setting memory usage to 28000mb for my 32GB system. It failed within 9 minutes.

 

Tried to increase VCCIO+System Agent as I thought it was a memory issue. Same problem.

Increased memory voltage. Same issue

Loosened timings. Same issue

Dropped memory back to 2133, VCCIO+System Agent to auto with stupidly loose timings of C16 - same issue.

 

I'm running Blend right now with Memory @ 2133 and its running 100% fine, on the same tests its was failing at with Custom using 28GB of memory.

 

Im considering calling this P95 custom crap a lost cause but I'd like to ask you guys for advice first.

 

 

 

[done12many2]

9 minutes ago, Djxinator said:

Evening folks,

 

I'm having difficulty identifying a failure in my system. I recently built my PC into a new case and for some reason my Overclock reset. I've re-installed Windows for good measure and begun the gruelling process of overclocking once again.

 

Finally got what I believed to be a semi-stable overclock of 4.2Ghz (3.1 Cache) with memory @ 3000 C15 1T. It passed 1 hour of P95 Blend and 1 hour of P95 Small-FFTs (I usually do this to test stability whilst tuning)

 

I ran a custom P95 and was aiming to run it for 4 hours before declaring the system stable enough for me, setting memory usage to 28000mb for my 32GB system. It failed within 9 minutes.

 

Tried to increase VCCIO+System Agent as I thought it was a memory issue. Same problem.

Increased memory voltage. Same issue

Loosened timings. Same issue

Dropped memory back to 2133, VCCIO+System Agent to auto with stupidly loose timings of C16 - same issue.

 

I'm running Blend right now with Memory @ 2133 and its running 100% fine, on the same tests its was failing at with Custom using 28GB of memory.

 

Im considering calling this P95 custom crap a lost cause but I'd like to ask you guys for advice first.

 

You've mentioned everything, but CPU voltage?  Are you running auto VCCIN?

 

I personally don't use Prime95, but there's no reason you can get it to pass if that's what floats your boat.

[Djxinator]

15 minutes ago, done12many2 said:

 

You've mentioned everything, but CPU voltage?  Are you running auto VCCIN?

 

I personally don't use Prime95, but there's no reason you can get it to pass if that's what floats your boat.

CPU core voltage is unchanged from passing Blend and Small FFTs as the core voltage should not have to be increased just because I'm using more RAM.

 

Input voltage is sat at 1.91 LLC 7 so no vdroop occurs. 

 

I'm running another custom test with cache voltage and cache multi on auto (x28) to see if it's the cache.

[done12many2]

12 minutes ago, Djxinator said:

CPU core voltage is unchanged from passing Blend and Small FFTs as the core voltage should not have to be increased just because I'm using more RAM.

 

Sounds like you're not too open to this suggestion.

 

Running more through the CPU whether that be through a CPU cache speed increase, memory speed increase, or simply the type of load type can call for slight bumps in CPU Voltage.

 

You have a fairly modest overclock for Broadwell-E and I'm sure that you have plenty of voltage overhead unless you are thermally limited.

[Djxinator]

1 minute ago, done12many2 said:

 

Sounds like you're not too open to this suggestion.

 

Running more through the CPU whether that be through a CPU cache speed increase, memory speed increase, or simply the type of load type can call for slight bumps in CPU Voltage.

 

You have a fairly modest overclock for Broadwell-E and I'm sure that you have plenty of voltage overhead unless you are thermally limited.

You got a free best answer. Wrong button.

 

I'm running it with cache at stock now using 28GB memory and it seems rock solid.

 

My chip isn't the best overclocker and I'm using 1.275v just for 4.2

 

Thermal overhead is fine up until around 1.375v but I'm using manual voltage (adaptive voltage nearly killed my CPU with spikes) so im trying to keep it down. Pushing 1.375v into the cpu while idle seems a bit on the stupid side.

[done12many2]

1 minute ago, Djxinator said:

You got a free best answer. Wrong button.

 

I'm running it with cache at stock now using 28GB memory and it seems rock solid.

 

My chip isn't the best overclocker and I'm using 1.275v just for 4.2

 

Thermal overhead is fine up until around 1.375v but I'm using manual voltage (adaptive voltage nearly killed my CPU with spikes) so im trying to keep it down. Pushing 1.375v into the cpu while idle seems a bit on the stupid side.

 

So with cache at stock you've effectively slightly reduced running through the CPU and no longer require additional voltage instead of maintaining/increasing the CPU workload and voltage.

 

You're cache voltage, which you never mentioned may have been a contributing factor as well.  Broadwell-E's cache for the most part seems to do very well from the 3.6 to 3.8.  Of course this requires a cache voltage bump and some VCCSA and VCCIO work with some chips, but I haven't seen reports of any not reaching those numbers.

 

Adaptive is the only way to go in my book.  I don't like running voltage through a chip when it doesn't need it.

 

You may want to read up on Broadwell-E/Haswell-E overclocking guides released by Raja@ASUS.  I think you find a lot of excellent information there.  They literally go through hundreds of BWE and HWE chips before sharing their findings.

[done12many2]

17 minutes ago, Djxinator said:

You got a free best answer. Wrong button.

 

Forgot.  Thanks.  That should help pay the bills around here.  

[Djxinator]

1 hour ago, done12many2 said:

 

Forgot.  Thanks.  That should help pay the bills around here.  

I've read the overclocking guide for BW-E and they don't mention P95 as a stability test

[done12many2]

6 minutes ago, Djxinator said:

I've read the overclocking guide for BW-E and they don't mention P95 as a stability test

 

This could be because they actually recommended that folks stop using Prime95 Small FFTs back when Haswell-E launched.  They cautioned against it's use not because of heat, but due to the fact that it ran entirely too much power through the CPU for no good reason.   They may have assumed that everyone knew this and left it out of the guide.

 

Prime95 in no way proves stability by itself.  It does however do a great job of proving that your cooling solution works.

 

Here's a clip of the Haswell-E (5960x) guide. 

[Djxinator]

Just now, done12many2 said:

 

This could be because they actually recommended that folks stop using Prime95 Small FFTs back when Haswell-E launched.  They cautioned against it's use not because of heat, but due to the fact that it ran entirely too much power through the CPU for no good reason.   They may have assumed that everyone knew this and left it out of the guide.

 

Prime95 in no way proves stability by itself.  It does however do a great job of proving that your cooling solution works.

 

Here's a clip of the Haswell-E (5960x) guide. 

I use 26.6 anyway, it doesn't use AVX.

Small FFTs on 26.6 produce the most accurate real world maximum temps. I use them for thermal testing.

 

I actually play a lot of Mechwarrior Online which is built on the Cryengine. This uses AVX instructions and produces temps lower than Prime95 26.6. Unfortunately this limits me to using manual voltage, as when I play MWO I get massive voltage spikes from adaptive voltage requesting more volts to do the AVX instructions.

[done12many2]

5 minutes ago, Djxinator said:

I use 26.6 anyway, it doesn't use AVX.

Small FFTs on 26.6 produce the most accurate real world maximum temps. I use them for thermal testing.

 

I actually play a lot of Mechwarrior Online which is built on the Cryengine. This uses AVX instructions and produces temps lower than Prime95 26.6. Unfortunately this limits me to using manual voltage, as when I play MWO I get massive voltage spikes from adaptive voltage requesting more volts to do the AVX instructions.

 

Another thing that confuses me is why you have to run your memory slower than necessary.

 

It sounds like you already had most of your own answer so I'll leave you to it.  Good luck with the overclock.  It's a nice build.

[MageTank]

3 hours ago, Djxinator said:

Evening folks,

 

I'm having difficulty identifying a failure in my system. I recently built my PC into a new case and for some reason my Overclock reset. I've re-installed Windows for good measure and begun the gruelling process of overclocking once again.

 

Finally got what I believed to be a semi-stable overclock of 4.2Ghz (3.1 Cache) with memory @ 3000 C15 1T. It passed 1 hour of P95 Blend and 1 hour of P95 Small-FFTs (I usually do this to test stability whilst tuning)

 

I ran a custom P95 and was aiming to run it for 4 hours before declaring the system stable enough for me, setting memory usage to 28000mb for my 32GB system. It failed within 9 minutes.

 

Tried to increase VCCIO+System Agent as I thought it was a memory issue. Same problem.

Increased memory voltage. Same issue

Loosened timings. Same issue

Dropped memory back to 2133, VCCIO+System Agent to auto with stupidly loose timings of C16 - same issue.

 

I'm running Blend right now with Memory @ 2133 and its running 100% fine, on the same tests its was failing at with Custom using 28GB of memory.

 

Im considering calling this P95 custom crap a lost cause but I'd like to ask you guys for advice first.

 

 

 

Let's hope I am not too late. First of all, do not use 28GB of memory out of 32GB. You use 75% (24GB) so that you leave enough for the OS and background tasks to prevent swapping. If it swaps, it will defeat the purpose. Second of all, you use small FFT to test cooling solution, not memory. 48k FFT will bring your cooling solution to its knees. 512k-4096k will test the IMC, IO lanes, and ram itself. Do 512k small, 4096k large, 15 minute intervals and test for 8 hours (or however long you personally see fit, i do 8 hours myself, after passing overnight memtest86 loops)

 

As for "the issue", what exactly is the issue? You mentioned it failing, but what do you mean by that? Is it a rounding error? If so, did it round up? Or down? All of this is important information and will dictate how I can help solve the issue. If it is a BSOD, which one is it? 0x124 (WHEA UN-CORRECTABLE ERROR)? Is it just a hard crash (screen freeze with no input response)? 

[Djxinator]

6 minutes ago, MageTank said:

Let's hope I am not too late. First of all, do not use 28GB of memory out of 32GB. You use 75% (24GB) so that you leave enough for the OS and background tasks to prevent swapping. If it swaps, it will defeat the purpose. Second of all, you use small FFT to test cooling solution, not memory. 48k FFT will bring your cooling solution to its knees. 512k-4096k will test the IMC, IO lanes, and ram itself. Do 512k small, 4096k large, 15 minute intervals and test for 8 hours (or however long you personally see fit, i do 8 hours myself, after passing overnight memtest86 loops)

 

As for "the issue", what exactly is the issue? You mentioned it failing, but what do you mean by that? Is it a rounding error? If so, did it round up? Or down? All of this is important information and will dictate how I can help solve the issue. If it is a BSOD, which one is it? 0x124 (WHEA UN-CORRECTABLE ERROR)? Is it just a hard crash (screen freeze with no input response)? 

You aren't too late buddy.

 

I've used Small FFTs for temp testing and they are fine. Loads of breathing room (66c hottest core) with all my case fans on low and my H110 GTX on quiet. And these BWE chips run super hot.

 

I was getting a rounding error 0.4/0.5 on one core.

 

Current situation - CPU @ 4.2 - Cache @ 3.1 - Mem @ stock (2133 loose timings)

All voltages except for core and cache voltage are on auto. 

 

LLC level 7 and other Digi power settings are on the usual Extreme/Asus optimised.

 

I'm currently bumping up the cache by 1x per 30 minute run of custom. Once I fail I will bump the voltage by .05  and test again, repeating until I hit max stable Cache, after which I will run a longer test. Then I will bump up the memory speed step by step, with voltages (DRAM, VCCIO, VCCSA) set manually at .05 over stock and bump voltages where necessary to reach my highest stable memory clock.

 

Once core, cache and memory are at desirable levels I will run another long test and that should be ok.

[MageTank]

10 minutes ago, Djxinator said:

You aren't too late buddy.

 

I've used Small FFTs for temp testing and they are fine. Loads of breathing room (66c hottest core) with all my case fans on low and my H110 GTX on quiet. And these BWE chips run super hot.

 

I was getting a rounding error 0.4/0.5 on one core.

 

Current situation - CPU @ 4.2 - Cache @ 3.1 - Mem @ stock (2133 loose timings)

All voltages except for core and cache voltage are on auto. 

 

LLC level 7 and other Digi power settings are on the usual Extreme/Asus optimised.

 

I'm currently bumping up the cache by 1x per 30 minute run of custom. Once I fail I will bump the voltage by .05  and test again, repeating until I hit max stable Cache, after which I will run a longer test. Then I will bump up the memory speed step by step, with voltages (DRAM, VCCIO, VCCSA) set manually at .05 over stock and bump voltages where necessary to reach my highest stable memory clock.

 

Once core, cache and memory are at desirable levels I will run another long test and that should be ok.

Was it 0.4 and expecting 0.5? Or 0.5 expecting 0.4? The order is very important. Also, more VCCIO/VCCSA does not mean more stability. You can actually cause instability by having too much of these two voltages. It is all about finding a sweet spot with these two. You will eventually find a value that allows you to not only overclock your memory higher, but does so while maintaining rock solid stability. The IMC can be very picky at times, and can fall outside of stability when excessive VCCIO/SA voltages are used. 

 

My experience with Broadwell-E is non-existent, but if its anything like Haswell-E, then be very wary of its IMC.Haswell-E was known to have terrible IMC's when it came to overclocking and memory stability in general. They tend to hate XMP kits, and require a lot of fine tuning to work properly. 

 

I am curious. If you would not mind, can you run memtest86? Specifically test #6 (block move). Do 5 passes of it. Be sure to select "All cores" before starting the test. Turn off all other tests and just do test 6. I can't help but feel this is a training issue. 

 

Another thing to keep in mind, is DRAM REF value. It should always be one half of VDIMM. DRAM REF is a threshold voltage. Cells below the DRAM REF value are evaluated as logical zero, while cells above DRAM REF value are evaluated as logical one. An incorrect value will cause rounding errors with almost absolute certainty. If you do not have DRAM REF voltage, then ignore this part. The board should (hopefully) be smart enough to auto assign that value correctly. I have yet to see this voltage listed on any Skylake board I own, but it was present on every Haswell board I had. Let me know if its rounding high or low. Once we figure that out, I can tell you how to fix it. 

[Djxinator]

45 minutes ago, MageTank said:

Was it 0.4 and expecting 0.5? Or 0.5 expecting 0.4? The order is very important. Also, more VCCIO/VCCSA does not mean more stability. You can actually cause instability by having too much of these two voltages. It is all about finding a sweet spot with these two. You will eventually find a value that allows you to not only overclock your memory higher, but does so while maintaining rock solid stability. The IMC can be very picky at times, and can fall outside of stability when excessive VCCIO/SA voltages are used. 

 

My experience with Broadwell-E is non-existent, but if its anything like Haswell-E, then be very wary of its IMC.Haswell-E was known to have terrible IMC's when it came to overclocking and memory stability in general. They tend to hate XMP kits, and require a lot of fine tuning to work properly. 

 

I am curious. If you would not mind, can you run memtest86? Specifically test #6 (block move). Do 5 passes of it. Be sure to select "All cores" before starting the test. Turn off all other tests and just do test 6. I can't help but feel this is a training issue. 

 

Another thing to keep in mind, is DRAM REF value. It should always be one half of VDIMM. DRAM REF is a threshold voltage. Cells below the DRAM REF value are evaluated as logical zero, while cells above DRAM REF value are evaluated as logical one. An incorrect value will cause rounding errors with almost absolute certainty. If you do not have DRAM REF voltage, then ignore this part. The board should (hopefully) be smart enough to auto assign that value correctly. I have yet to see this voltage listed on any Skylake board I own, but it was present on every Haswell board I had. Let me know if its rounding high or low. Once we figure that out, I can tell you how to fix it. 

I'll be doing a lot more testing tomorrow night. I'm at 3.3 Cache right now so once I get to 3.6-3.7 stable I'll do some DRAM testing.

 

I'm currently stable with an hour of 512 - 4096, 24GB used - 4.2 Core, 3.3 Cache and stock memory.

 

Tomorrow night after work I'll be testing 3.4, 3.5, 3.6

 

The cache is currently only at 1.1v which is way lower than I had before. I may have been pushing too much voltage into it. And VCCIO/SA I was also liberal with the voltages. Im taking it super slow this time and once I have the cache and core stable I'll move on to the memory. Once I do I'll revive this thread or message you on the forums.

 

Cache on BWE is a bitch and nowhere near as stable at high clocks, so it might be a while.

[done12many2]

21 hours ago, Djxinator said:

I'll be doing a lot more testing tomorrow night. I'm at 3.3 Cache right now so once I get to 3.6-3.7 stable I'll do some DRAM testing.

 

I'm currently stable with an hour of 512 - 4096, 24GB used - 4.2 Core, 3.3 Cache and stock memory.

 

Tomorrow night after work I'll be testing 3.4, 3.5, 3.6

 

The cache is currently only at 1.1v which is way lower than I had before. I may have been pushing too much voltage into it. And VCCIO/SA I was also liberal with the voltages. Im taking it super slow this time and once I have the cache and core stable I'll move on to the memory. Once I do I'll revive this thread or message you on the forums.

 

Cache on BWE is a bitch and nowhere near as stable at high clocks, so it might be a while.

 

Our conversations on the other thread sparked some thought.  Have you set up AVX offset with that chip?  Also, have you messed with Asus' TCT at all.  Pretty slick if you'd like to get your single core performance up while maintaining a somewhat decent multi core clock.  I used both on a 6800k and 6950x that I had and they work great.  I continue to use the Asus TCT on my 5860x.

 

[Djxinator]

2 minutes ago, done12many2 said:

 

Our conversations on the other thread sparked some thought.  Have you set up AVX offset with that chip?  Also, have you messed with Asus' TCT at all.  Pretty slick if you'd like to get you single core performance up while maintaining a somewhat decent multi core clock.  I used both on a 6800k and 6950x that I had and they work great.  I continue to use the Asus TCT on my 5860x.

 

Still overclocking the cache right now. I managed to get x34 stable with 1.12v Cache but jumping to x35 is proving difficult - I'm at 1.2v right now and still unstable - gonna try pushing up the VCCIO and System Agents a notch.

 

Since re-installing Windows and starting the OC from scratch I've shaved 0.04v off my core voltage (It was @ 1.312 now @ 1.275) so I'm gonna try and push for 4.3 soonish.

I'll probably play with those settings once I have my Core, memory and cache all clocked at speeds I'm happy with - I had considered trying to push Core 5* to 100mhz higher also. Not sure how TCT works and AVX offset will definitely be useful in the future - it might enable me to run with Adaptive voltage without fear of too many volts being pushed through the CPU.

[Djxinator]

On 09/08/2016 at 10:49 PM, MageTank said:

Was it 0.4 and expecting 0.5? Or 0.5 expecting 0.4? The order is very important. Also, more VCCIO/VCCSA does not mean more stability. You can actually cause instability by having too much of these two voltages. It is all about finding a sweet spot with these two. You will eventually find a value that allows you to not only overclock your memory higher, but does so while maintaining rock solid stability. The IMC can be very picky at times, and can fall outside of stability when excessive VCCIO/SA voltages are used. 

 

My experience with Broadwell-E is non-existent, but if its anything like Haswell-E, then be very wary of its IMC.Haswell-E was known to have terrible IMC's when it came to overclocking and memory stability in general. They tend to hate XMP kits, and require a lot of fine tuning to work properly. 

 

I am curious. If you would not mind, can you run memtest86? Specifically test #6 (block move). Do 5 passes of it. Be sure to select "All cores" before starting the test. Turn off all other tests and just do test 6. I can't help but feel this is a training issue. 

 

Another thing to keep in mind, is DRAM REF value. It should always be one half of VDIMM. DRAM REF is a threshold voltage. Cells below the DRAM REF value are evaluated as logical zero, while cells above DRAM REF value are evaluated as logical one. An incorrect value will cause rounding errors with almost absolute certainty. If you do not have DRAM REF voltage, then ignore this part. The board should (hopefully) be smart enough to auto assign that value correctly. I have yet to see this voltage listed on any Skylake board I own, but it was present on every Haswell board I had. Let me know if its rounding high or low. Once we figure that out, I can tell you how to fix it. 

Core and Cache are stable, and memory seemed stable up until 2800 (loose timings) using stock volts for VCCSA and VCCIO. I know the IMC for BW-E can handle much higher volts than HW-E and 1.3 VCCSA and 1.25 VCCIO are quite common.

 

- Enabled XMP and the PC turned off 1.5 hours into a test (No BSOD, no errors)

- Dropped my overclock to 4.2 from 4.3 and tried again and I got FATAL ERROR: Rounding was 0.4952392578, expected less than 0.4

- Increased VCCIO, VCCSA and a small notch of Core Volts, now I'm getting different errors entirely. Just seems to be errors everywhere when I enable XMP.

 

Specifically the errors occur on the final test of the 15 minute 1680K FFT, on one specific core/thread (Worker 9) - I've noticed the worker is lagging behind the others too.

Error is as follows.

 

FATAL ERROR: Resulting sum was -3.491616648422552e+150, expected: 293897265119006.8

OR

FATAL ERROR: Resulting sum was 183228083011059.4, expected: 1032412882043832

 

I'm running P95 again right now (passed 1680K and currently working on 512K) and at 1.2 VCCSA and 1.15 VCCIO it has passed where before I received the above errors.

 

I THINK I may have a better understanding of how this all works so I will work with Prime95 until I find I can't get it stable no matter what. Once that happens I'll test for the "training issue" you have suggested. 

 

Do you find that tweaking memory timings with these chips also requires VCCSA and VCCIO tweaking? If that is the case then I might pass on trying to get to C15 1T, its been hard enough with standard XMP settings.

 

[EDIT] Rounding error on a completely different core (Worker 2) on the 512K FFTs - FATAL ERROR: Rounding was 0.4931640625, expected less than 0.4

[Djxinator]

Final update I hope.

 

Overclock is now stable with XMP Profile at 1.225v VCCSA and 1.15v VCCIO

 

It was not stable with 1.25 VCCSA or 1.2 VCCSA - very interesting.

[MageTank]

2 hours ago, Djxinator said:

Final update I hope.

 

Overclock is now stable with XMP Profile at 1.225v VCCSA and 1.15v VCCIO

 

It was not stable with 1.25 VCCSA or 1.2 VCCSA - very interesting.

As I said before, too much voltage can cause just as much instability as too little. It's a balancing act. As for your previous questions involving timings, VCCIO/VCCSA can help with making primary timings stable IF the IMC is the one holding the memory back. If it's the IC's themselves, then that can only be solved with VDIMM. How do you know which one is the bottleneck? Simple. Go until it no longer budges, then add more VDIMM. If it fails to help it budge, try more VCCIO/SA. If that makes it work, its the IMC not liking the configuration. Now, that does not mean you NEED to keep the higher VCCIO/SA voltages in order to obtain your desired primary timings. You can technically change a few tertiary timings to relieve stress off of the IMC. I personally advise against this, because you will see far more performance from optimized tertiary timings than what you will ever see from tight primaries. 

 

Those rounding errors were indeed very off. I highly doubt you would be able to pass block move on memtest86 without errors being thrown with that previous setup. If you have any additional questions, let me know. I'll try to help in any way I can. 

[Djxinator]

11 hours ago, MageTank said:

As I said before, too much voltage can cause just as much instability as too little. It's a balancing act. As for your previous questions involving timings, VCCIO/VCCSA can help with making primary timings stable IF the IMC is the one holding the memory back. If it's the IC's themselves, then that can only be solved with VDIMM. How do you know which one is the bottleneck? Simple. Go until it no longer budges, then add more VDIMM. If it fails to help it budge, try more VCCIO/SA. If that makes it work, its the IMC not liking the configuration. Now, that does not mean you NEED to keep the higher VCCIO/SA voltages in order to obtain your desired primary timings. You can technically change a few tertiary timings to relieve stress off of the IMC. I personally advise against this, because you will see far more performance from optimized tertiary timings than what you will ever see from tight primaries. 

 

Those rounding errors were indeed very off. I highly doubt you would be able to pass block move on memtest86 without errors being thrown with that previous setup. If you have any additional questions, let me know. I'll try to help in any way I can. 

What are your thoughts on HCI Memtest vs Memtest86/86+?

 

I've heard that HCI memtest is more likely to find errors that result from overclocks.

[MageTank]

36 minutes ago, Djxinator said:

What are your thoughts on HCI Memtest vs Memtest86/86+?

 

I've heard that HCI memtest is more likely to find errors that result from overclocks.

I have personally never used it, but memtest86 is perfectly capable of finding errors that result from overclocking. Memtest86 doesn't stress memory. It is mostly used to check to see if the IMC is fine with the ram configuration. If you want to stress the ram, you have to use something like Linpack or Prime95 with high memory workloads. I'll look into HCI though, thanks for bringing that up. 

[done12many2]

2 hours ago, Djxinator said:

What are your thoughts on HCI Memtest vs Memtest86/86+?

 

I've heard that HCI memtest is more likely to find errors that result from overclocks.

 

HCI MemTest without a doubt.  That or Google Stressapp via Linux, which is actually better than HCI MemTest.   Memtest86/86+ is a walk in the park after passing either one of those.

[MageTank]

17 minutes ago, done12many2 said:

 

HCI MemTest without a doubt.  That or Google Stressapp via Linux, which is actually better than HCI MemTest.   Memtest86/86+ is a walk in the park after passing either one of those.

Yeah. I have a hard time explaining to most people that Memtest86 is not a stress test, and will not show you ram instability. It helps with seeing if your IMC likes the timings or not (Specifically Test6 Block Move) but for the most part, it means very little until real stress is applied.

 

I personally use Prime95 512k FFT - 4096k FFT 75% ram capacity (Stresses everything from ram itself, all the way to the IMC, and everything in between). I do want to give HCI a try though, after hearing you two speak about it. I spend far too much time overclocking ram to not know about this, lol. 

[done12many2]

1 minute ago, MageTank said:

Yeah. I have a hard time explaining to most people that Memtest86 is not a stress test, and will not show you ram instability. It helps with seeing if your IMC likes the timings or not (Specifically Test6 Block Move) but for the most part, it means very little until real stress is applied.

 

I personally use Prime95 512k FFT - 4096k FFT 75% ram capacity (Stresses everything from ram itself, all the way to the IMC, and everything in between). I do want to give HCI a try though, after hearing you two speak about it. I spend far too much time overclocking ram to not know about this, lol. 

 

The free version of HCI works great.  I calculate 93-95% of total memory and divide that by total number of CPU threads.  I open one instance of HCI per available thread and input the portioned RAM for that instance.

 

The HCI MemTestPro allows you to type in a simple command to do all of this and a bit more, but I find myself doing it the manual way despite having paid for the $5 pro version.  Save yourself $5. 

 

HCI is great for cache/RAM and Google Stressapp is RAM only.