LTT Forums CPU Overclocking Database!

[MeltyMoon]

I agree with ProKoN points, but i think it will hurt expanding of SB, besides, in most cases, users will not be able to replicate any settings from DB even if all components are the same. But they can use logic behind it, and use DB as reference point.

 

Also 8 hours of Prime95 is impossible for some users, in reality, if you are stable for 20 minutes, you will be stable for 8 hours, also, Prime95 (or any similar software) isn't the best way to test stability to it's finest. System stability also depends from other factors, OS for example, and how is configured.

 

I agree that is wide topic, but i don't agree that making 8 hours Prime as requirement is a good idea.

I currently don't have all of the stability testing suites ironed out yet for my revision, or time lengths for that matter. Give me a little more time, things will start to look a lot more organized.

[leipero]

I currently don't have all of the stability testing suites ironed out yet for my revision, or time lengths for that matter. Give me a little more time, things will start to look a lot more organized.

Ok, i just strongly advice you to not include as required 8 hours of prime stability test. Few reasons for that, one of htem is that will cut DB in half, 2nd it will not gain anything compared to 20-30 min. test in majority of cases. Just something to consider ;).

[ProKoN]

I agree with ProKoN points, but i think it will hurt expanding of SB, besides, in most cases, users will not be able to replicate any settings from DB even if all components are the same. But they can use logic behind it, and use DB as reference point.

 

Also 8 hours of Prime95 is impossible for some users, in reality, if you are stable for 20 minutes, you will be stable for 8 hours, also, Prime95 (or any similar software) isn't the best way to test stability to it's finest. System stability also depends from other factors, OS for example, and how is configured.

 

I agree that is wide topic, but i don't agree that making 8 hours Prime as requirement is a good idea.

why is it impossible for some users to not be able to run and provide a valid screenshot stable for 8 or more hours of stability testing?

 

only reason i can think of is their overclock will crash sometime before it starts and the 8 hour mark.

 

start the stress test before you go to bed and when you wake up, capture a stable screenshot...easy.

 

I agree 100% software variables can cause system instability however these variables are often very unique and cannot be taken into consideration as we try to measure hardware stability as a whole.

 

stability testing is a very wide topic i do agree.

 

i would disagree with 20 minutes of stability testing as an accurate measurement. i have had systems crash anywhere from 1minutes to 30 hours of stability testing, i disagree with 20 minutes of stabilty means you will have 4 or more hours of stability.

 

seems all advanced users have their own stress testing methodology that works for them.

 

the database should provide validation or you may as well just post your unstable, unvalidated results in the 5ghz oc club thread.

 

melty has done a good job with this db, it takes time and energy to maintain such a thing.

 

my biggest point is, it is fundamental to promote stabilty and to provide valadition to have a strong and healthy  Database.

[ProKoN]

First run through at this 

 

https://docs.google.com/file/d/0B9vsJFi2SpbEdXAxRUdiaGN3WGM/edit?usp=sharing

 

Any feed back would be great, still waiting for new Ram to arrive so running my old DDR 3 from old build

i was also able to boot and to stress test  at 1.285V

 

reality is i bluescreened after 37 minutes of stress testing and watching an hd video.

 

your screenshot shows you stress testing for fifteen minutes or less.

 

provide a stable screenshot of 8 hours or more while stability testing.

 

im not saying you cant do 4.6 at 1.28V.. just provide some validation please.

[old reece frost]

My overclock for my old Pentium E5700, couldn't really get it past 4GHz on the stock cooler.

Gunna try and get my i7 somewhere to 4.6GHz when I get some Corsair Hydro Series goodness.

[danthe1man1989]

i was also able to boot and to stress test  at 1.285V

 

reality is i bluescreened after 37 minutes of stress testing and watching an hd video.

 

your screenshot shows you stress testing for fifteen minutes or less.

 

provide a stable screenshot of 8 hours or more while stability testing.

 

im not saying you cant do 4.6 at 1.28V.. just provide some validation please.

Yeah same as me about 30 minutes in blue screened, still got some work to do

[Wasabiz]

In the OP I have a link to the form so you can submit your entry. Do yourself a favor, being the DB is currently going under revision, make sure you submit your screenshot after 8 hours of prime, aida, intel burn test, etc running!

yeah dw i stress test mine minimum of 12 hours before i can call it stable

[leipero]

why is it impossible for some users to not be able to run and provide a valid screenshot stable for 8 or more hours of stability testing?

only reason i can think of is their overclock will crash sometime before it starts and the 8 hour mark.

start the stress test before you go to bed and when you wake up, capture a stable screenshot...easy.

my biggest point is, it is fundamental to promote stabilty and to provide valadition to have a strong and healthy Database.

Good point, hard if they use PC for other work (eg rendering etc.) while sleeping, not impossible you are right, but my point was, very few will actually bother with it.

I would love that also, to promote stability over overclocking for no reason at all (most people do overclock when they actually don't need it, or problem is elsewhere, not in speed of the system). I agree it is wide topic, and even Prime95 (and other stability tests) are unable in some cases to detect obvious stability problem, for example, 10 hours + prime stable, but system is lagging (in games, in sense of control response) due to the lack of power delivery to the CPU, while for work is perfectly stable.

[MeltyMoon]

Please help me out by voting in the poll! Future of how this DB is run is going to be decided by you beautiful lot ;)

[Art Vandelay]

I'd prefer a 4 hour stress test, with each FFT being tested for 2-3 minutes at a time. I know you're supposed to do a full 24 hour stress test, but I find that it's a bit of a waste of time to have my entire computer wasted for 24 hours,

[MeltyMoon]

I'd prefer a 4 hour stress test, with each FFT being tested for 2-3 minutes at a time. I know you're supposed to do a full 24 hour stress test, but I find that it's a bit of a waste of time to have my entire computer wasted for 24 hours,

Exactly why I'm using the communities voice. After all, it's all thanks to you that I own one of the biggest threads on this forum :P. Least I can do is run the thread the way you guys want!

[ProKoN]

I'd prefer a 4 hour stress test, with each FFT being tested for 2-3 minutes at a time. I know you're supposed to do a full 24 hour stress test, but I find that it's a bit of a waste of time to have my entire computer wasted for 24 hours,

the basline i would like to push for is 8 hours minimum. 24 hours is excessive, i agree.

 

set it before you go to bed, wake up and capture your valid screenshot.

 

I just failed a  stress test after 9 hours this morning, would you consider  that overclock stable? (in your opinon)

[Art Vandelay]

the basline i would like to push for is 8 hours minimum. 24 hours is excessive, i agree.

set it before you go to bed, wake up and capture your valid screenshot.

I just failed a stress test after 9 hours this morning, would you consider that overclock stable? (in your opinon)

Well, the reason it takes 24 hours to do a blend test in prime95 is because each FFT takes 15 minutes, so the whole test takes 16 hours or so. At 2 minutes per FFT, it would take significantly less time, and still be a reasonable test of stability.

I personally don't like to run my system overnight on a stress test, due to paranoia and the fact that my room heats up fast.

I find that as long as your can pass a 4 hour stress test with 3 minutes per FFT, your system will be fairly stable, and since degradation occurs, there's no such thing as 100% stable.

My goal in these kinds of things is just to try for as high of a clock speed as possible, not a clock speed I will be using for every day use. I don't really see the point in making sure it is extremely solid, if that's my goal. Maybe there should be a separate section for 24/7 overclocks.

[ProKoN]

Well, the reason it takes 24 hours to do a blend test in prime95 is because each FFT takes 15 minutes, so the whole test takes 16 hours or so. At 2 minutes per FFT, it would take significantly less time, and still be a reasonable test of stability.

I personally don't like to run my system overnight on a stress test, due to paranoia and the fact that my room heats up fast.

I find that as long as your can pass a 4 hour stress test with 3 minutes per FFT, your system will be fairly stable, and since degradation occurs, there's no such thing as 100% stable.

My goal in these kinds of things is just to try for as high of a clock speed as possible, not a clock speed I will be using for every day use. I don't really see the point in making sure it is extremely solid, if that's my goal. Maybe there should be a separate section for 24/7 overclocks.

you raise some valid points about prime95 and bring insight to the discussion.

 

please consider this.

 

Prime 95 is NOT VALIDATED for Haswell Cpus. Even though i think p95 is a good and valadited stability test i think it has absolutley no relavence to stress testing haswell since it has not been updated to stress the New instruction sets.

 

http://www.mersenne.org/freesoft/

 

 

feel free to prove me otherwise.

 

your only validated haswell stress tests are aida64 and the ietu.

 

degradiation/ micron migration, yup that happens over time.

 

how long should your computer be stable for at stock settings if all the power\ thermal envelopes are appropriate? ( from a pure hardware perspective, In theory)

 

the answer is infinity 100% stable.

 

there is no reason for an overclock not to be 100% stable, no reason at all. in theory your overclock should stress test for infinity.

 

We all have personal goals and comfort zones, i respect that.

 

in my opinon,as an advanced user with more than 13 years of overclocking experience, i believe appropriate fundamental guidlines are in the best interest of the ocdb.

[ProKoN]

Perhaps Linus can shed some insight as to Overclocking Valdation

 

please advance to 6:20 in the video. linus recommends 8-24 hours of stability testing

 

" generally speaking you want to run 8 hours of prime before you can consider anything stable"

 

linus is an experienced overclocker.

 

 

Linus recommends 24-48 hours in this video. advance to 5:40 and watch a bit

 

 

linus recommends 8-12 hours stability testing  here when overclockng and testing his memory advance to 10:00

 

 

Linus recommends 12-24-48 hours of stress testing  advance to 12:50

 

[Wasabiz]

If I'm not mistaken, in the latest overclocking video Linus made he said something like prime95 is not really validated for haswell

Sent from my iPhone using Tapatalk 2

[xBlizzDevious]

It was requested that we have a prime95 run for a minimum of 8 hours. My CPU apparently wasn't stable though as at 10 or so hours, I crashed. I've upped the voltage a little and now I've got 16 hours out of it, but I'm wanting to use my PC, so Prime is now off. Haha.

 

 

[leipero]

Perhaps Linus can shed some insight as to Overclocking Valdation

 

I just quoted you in order to see my post.

 

Since you are all for stability, and I am too, i would like to point out at one specific problem in those "validation" methods, and would love to hear your opinion, I want to explain why I'm against those "validations", it's because i think they are pointless.

 

Here is an example, i can bring my CPU at current motherboard at 4000MHz easily, with just 1.2745v set in BIOS (in reality, under 100% stressed CPU, it is 1.240-1.248v). It is at least 4 hours prime95 stable, and most probably 8, 24, 48+ hours stable, so what is the problem with it than? Why i don't use that voltage? It is stable right? NO.

 

Under specific situation, and i will give best example and most obvious, you will recognize that. Say you fire up specific game, go to online server, start recording game at 30 FPS, bam, you get LAG (no stuttering, but controls lag). Other side effect is, when you did recorded, output file doesn't look smooth as your game at time you recorded (there is slight "micro-stutter", not so easy to detect, but if you know how and where to look, it is very easy). Now, by upping voltage, you solve both of those side effects and go up with the heat.

 

Another problem, if you up voltage way too high, you will get same side effect while playing/recording, at the same time, output file would give you normal picture, with exception of lag in most cases (not present).

 

Problem? CPU don't get enough, or do get too much voltage. Even tho in all cases it appears to be stable, it isn't actually. Few things can cause this:

 

1. Unstable PSU (weak PSU).

2. Not sufficient VRM on motherboard.

3. The very basic limitation of silicon in the chip.

 

You will hit barrier one way or another. Problem is, if you don't know how to detect it, you will end up thinking that your system is stable, when in fact, it isn't. And that beats the purpose of validation in my opinion, i can prove to you that my CPU is "stable" with 8+ hours of Prime95 or any other program, while in fact it isn't, and i know that (and that isn't the problem, problem is when someone don't know).

 

Now i know that perception of LAG is highly subjective and hard to measure (it depends mainly on user reflexes and perception), and it is different for every person, same as perception of smoothness. Tho i think there are valid ways to measure those effects, we (regular or professional users) don't have equipment for that (while MAD, Intel, nVidia have, probably ASUS, MSI etc. but not necessarily). From that perspective, i would say that Prime95 etc. stability validation is pointless, while it can be very useful on the other hand (personal validation).

 

I'm interested about your opinion from this perspective.

[ProKoN]

I just quoted you in order to see my post.

 

Since you are all for stability, and I am too, i would like to point out at one specific problem in those "validation" methods, and would love to hear your opinion, I want to explain why I'm against those "validations", it's because i think they are pointless.

 

Here is an example, i can bring my CPU at current motherboard at 4000MHz easily, with just 1.2745v set in BIOS (in reality, under 100% stressed CPU, it is 1.240-1.248v). It is at least 4 hours prime95 stable, and most probably 8, 24, 48+ hours stable, so what is the problem with it than? Why i don't use that voltage? It is stable right? NO.

 

Under specific situation, and i will give best example and most obvious, you will recognize that. Say you fire up specific game, go to online server, start recording game at 30 FPS, bam, you get LAG (no stuttering, but controls lag). Other side effect is, when you did recorded, output file doesn't look smooth as your game at time you recorded (there is slight "micro-stutter", not so easy to detect, but if you know how and where to look, it is very easy). Now, by upping voltage, you solve both of those side effects and go up with the heat.

 

Another problem, if you up voltage way too high, you will get same side effect while playing/recording, at the same time, output file would give you normal picture, with exception of lag in most cases (not present).

 

Problem? CPU don't get enough, or do get too much voltage. Even tho in all cases it appears to be stable, it isn't actually. Few things can cause this:

 

1. Unstable PSU (weak PSU).

2. Not sufficient VRM on motherboard.

3. The very basic limitation of silicon in the chip.

 

You will hit barrier one way or another. Problem is, if you don't know how to detect it, you will end up thinking that your system is stable, when in fact, it isn't. And that beats the purpose of validation in my opinion, i can prove to you that my CPU is "stable" with 8+ hours of Prime95 or any other program, while in fact it isn't, and i know that (and that isn't the problem, problem is when someone don't know).

 

Now i know that perception of LAG is highly subjective and hard to measure (it depends mainly on user reflexes and perception), and it is different for every person, same as perception of smoothness. Tho i think there are valid ways to measure those effects, we (regular or professional users) don't have equipment for that (while MAD, Intel, nVidia have, probably ASUS, MSI etc. but not necessarily). From that perspective, i would say that Prime95 etc. stability validation is pointless, while it can be very useful on the other hand (personal validation).

 

I'm interested about your opinion from this perspective.

 

 

I will try my best to provide you with my opinion \ insight in regards to your comments.

 

First off I do believe you are trying to explain the effects of Vdroop . Setting your voltage at 1.275V but in reality when stress tested ( or cpu is under load) the Voltage is 1.24V?

 

This is Natural, in fact your voltage will scale accordingly as you increase it or decrease it. Vdroop sucks, but I can honestly say on the Haswell Platform we are not experiencing such Voltage differences when under load.

 

All I can Say about Vdroop is report the Voltages you manually input, and use these voltages when keeping a log of your results.Never go by what the actual voltage is when the cpu is under load.

 

So if you input 1.275V, report and log  1.275V not 1.24V, even tho 1.24V is the actually Voltage reading. The Vdroop voltage should scale accordingly so if you set your voltage for 1.285V the real voltage when under load should be 1.25V, the .01 Voltage increase affects both idle\ load scenarios.

 

Because you are stable for four hours Does Not mean you have 8, 12, 24, 48 Hours of stability. It is a indication the system is fairly stable but certainly does not mean you have more than 4 hours of stability. If you are stable for four hours, you are stable for four hours, not eight, not twelve, not 24 or 48, being stable for 4 hours means your stable for four hours. My current Overclock fails after 9 hours of stress testing, I do not consider this overclock or stable or Validated, for my personal validation My systems require no less than 18-24 hours of stress testing before I will consider Validating my system as being stable.

 

 

 

As far as these Variables are concerned

 

1. Unstable PSU (weak PSU).

2. Not sufficient VRM on motherboard.

3. The very basic limitation of silicon in the chip.

 

All of these mentioned variables are legitimate from a hardware perspective and yes can effect your overclock results and stability as a whole. If any of these variables are making a negative impact on your overclock you may reach lower OC results and\ or Not find stability for 8 or more hours.  All I can say is turn down the frequencies and voltages to find stability if your PSU, VRM, or CPU are unstable.

 

 

The rest of your post basically went into describing "software" as a variable. in your case recording your gameplay and having different results with different voltages.

 

Please do not confuse hardware variables and software variables when overclocking. The ONLY software  variables that should be considered when overclocking is the health of your operating system, and the stress testing software you are using, thats it. You must understand keeping hardware + software variables separate   is very very important.  If you have to tweak your hardware to make your software applications happy, thats fine, however, it does not offer any insight as to overclocking from a pure hardware perspective.

 

we need to measure are results as best we can from a pure hardware standpoint, we have to eliminate software as being any type of variable, as most individual use cases will vary.

 

There are alot of software variables that may cause instability.

-corrupt OS

-Outdated\ bad drivers

-Program conflicts.

-windows updates...

 

many many more, but just to name a few.

 

To sum this up.

 

stress test for more than 8 hours for personal validation

 

tune your system for your individual use case, also for personal validation

 

 

Science is all about data measurement. We have a better chance of measuring data accurately if we can isolate as many variables as possibly while still focusing on the important information we are trying to record.

 

please comment if you found any of this helpful, or if I possibly mis-interpreted your points.

 

Thank you for adding to the discussion as I would encourage anyone to do so.

[leipero]

I will try my best to provide you with my opinion \ insight in regards to your comments.

Vdroop aside, event tho i did explained effect of vdroop as not relevant one (only to explaining voltage CPU actually run at in the full stress), that wasn't the point, and ofc. i always report voltage set in BIOS, not the actual one, but it is good for clarify for others potentially.

 

My point was that what i explained in latter sentence isn't software variable, but hardware, and only way to measure those effects is by measuring resistance of individual components on basic level witch is impossible for anyone of us users to do. Only measurement that is valid for that is the scale of our own perception. In that sense, i consider stability tests invalid way to indicate system stability. That's what i was interested in to hear opinion, because it is not logical to require stability test (error free test), and not the actual stability of system with all those variables, if our goal is to create database that would be valid for users to replicate, otherwise, if that isn't the goal, i don't see the point doing so at all.

 

Hope i was not way too confusing.

[MeltyMoon]

Bump, final decision on where I'm taking this DB has been decided. Please look at the OP

[leipero]

I think everyone would be satisfied with that solution, very good idea in my opinion.

[Jumper118]

My overclock for my old Pentium E5700, couldn't really get it past 4GHz on the stock cooler.

Gunna try and get my i7 somewhere to 4.6GHz when I get some Corsair Hydro Series goodness.

you should put the hydro on the p4 and go for 4.6ghz  also just out of interest, what board did you use to overclock the p4 with? 

[old reece frost]

you should put the hydro on the p4 and go for 4.6ghz  also just out of interest, what board did you use to overclock the p4 with? 

GIGABYTE G41MT S2PT

[ProKoN]

Bump, final decision on where I'm taking this DB has been decided. Please look at the OP

 

 

I think everyone would be satisfied with that solution, very good idea in my opinion.

 

Yes I agree 100%. good job Melty.

 

You basically figured out a way to keep existing entries without deletion or upsetting anyone, and left it up to the user to provide there own form of validation.

 

Very well done I commend you on a good job.

 

I guess it would be nice to see the validated Overclocks in Green Vs Purple.

 

Green means go bro, dude wheres my car?

 

5 out of 82 entries  validated ( i still feel 8 hours is minimum absolute minimum). I hope users see the importance of providing validation for future entries.