Haswell-E impossible to get it prime95 stable with a reasonable OC?

[joeyjojo]

So to clarify, you think Prime 95 has the same AVX 2.0 routines as the two latter tests? This place has really gone down hill.

Maybe this will help:

http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club/2390#post_22900116

Well of course it does, there's only about a dozen AVX extensions. Linpack is even used as an example benchmark in this Intel AVX white paper. I get the impression more and more that you're blagging TBQH.

http://www.intel.com/content/dam/ww...on-e5-v3-advanced-vector-extensions-paper.pdf

Regarding that forum post, an Asus marketing rep spreading rumours about competing software to the benefit of their own Realbench software isn't exactly a convincing source.

I can overclock about 2-300mhz more in general if I ignore the Elephant in the room that AVX 2 prime can't pass but I don't.

This is a good way to put it.

It's like ignoring 50-70% of a CPUs performance (from AVX extensions) just to get a few hundred MHz more overclock.

The real "problem" here (from an overclocking point of view) is these chips get hot when used to their full potential. I don't know if it's the transistor density or the die-IHS interface or something else, but there's no way around it. The solution isn't to stick your fingers in your ears and say you don't like this or that program, it's to back off a few hundred MHz and test again. Saying it's "unrealistic" is a cop out.

 

[Justintime]

Yeah I read that Asus reps post awhile back, seems a lot of people quote it. I've used that so called realbench and passed with no issues, same with aida etc.. but had instability in gaming or general use. When I prime pass it generally means everything is stable because if the voltage supplied can pass avx 2 then anything else is a cakewalk.

 

[Silent_Scone]

Yeah I read that Asus reps post awhile back, seems a lot of people quote it. I've used that so called realbench and passed with no issues, same with aida etc.. but had instability in gaming or general use. When I prime pass it generally means everything is stable because if the voltage supplied can pass avx 2 then anything else is a cakewalk.

I take it you've seen the current ramps when running the test too?

That's rhetorical, because clearly you wouldn't have that insight or equipment.

You're clearly experienced enough to know better...Much like Joey, who doesn't realise the white paper is in relation to Xeons which throttles with the loads incurred with Prime.



I wish it was hilarious trying to get through to you but actually it's quite frustrating seeing fellow enthusiasts are so utterly clueless.

 

[joeyjojo]

Look, we agree AVX codes are more intensive. You seem to think that means they're a danger to the CPU, that they will "degrade both cache and core" as you wrote earlier.

What you're really saying is a massive overclock + Prime95 is dangerous. Well duh! Overclocking itself is a danger to the CPU and will shorten its life, we all know that. That's not Prime's fault, it's the 35-40% overclock on a CPU that was never designed for anywhere near that.

Where the bad advice really starts is then telling people to keep their max overclocks and ignore the fact that some codes will push their chips to dangerous levels. Er, what happens when they do try and run some intensive programs? Suddenly the overclock they thought was stable isn't so stable, because they didn't test with the worst case scenario. This guy a few weeks ago comes to mind... https://forums.overclockers.co.uk/showthread.php?t=18727575

Instead we need more Justintimes on the forum to say "that's a nice overclock but do anything really intensive on it and it'll melt! dial it back a few hundred MHz".

Unless of course people are doing 4.7 GHz overclocks just to surf facebook, in which case...

 

[Justintime]

I take it you've seen the current ramps when running the test too?

That's rhetorical, because clearly you wouldn't have that insight or equipment.

You're clearly experienced enough to know better...Much like Joey, who doesn't realise the white paper is in relation to Xeons which throttles with the loads incurred with Prime.



I wish it was hilarious trying to get through to you but actually it's quite frustrating seeing fellow enthusiasts are so utterly clueless.

All I know is I've been using the latest versions of prime with all of my cpus and all of the overclocks I do for friends/clients and have come to the conclusion that in many scenarios if its not fully tested theres a chance something might come along and cause instability. Passes with earlier versions of prime and Asus realbench, Aida etc.. with a lower vcore and or a higher clock never seem to be quite rock solid as a pass with prime avx 2 etc... I've not experienced any degradation and sometimes even with tweaking other parameters later on have even managed lower vcore to pass avx 2, certainly not the behaviour of a cpu with symptoms of degradation due to electro migration.

I strongly doubt any damage is occurring as all of my avx 2 passes are done at a capped reasonable voltage for the process used and more than adequate cooling. Gives me the peace of mind that nothing will take the machine down and so far for the last 17yrs or so of using p95 its never failed me and passed an unstable machine, quite likely as I usually only clock the cpu, usually leave memory speed and timings etc.. up to the spd and rarely even oc my gpus.

So call me clueless, un-insightful and lacking equipment the average user or even technician will never have but I'm happy to continue what I'm doing.

 

[tyler_jrb]

While prime MAY be more stable. I've ran prime for hours on end to then compress a video with handbrake and have it crash due to instability.

Like 8 pack said the best thing is to set your stability for what you actually need it for. No point in making it prime stable and run your CPU underclocked and overvolted when you only use it for gaming.

I use real bench or aida64. Both have been perfect for what I use my CPU for. Gaming, handbrake etc.

 

[Justintime]

While prime MAY be more stable. I've ran prime for hours on end to then compress a video with handbrake and have it crash due to instability.

Like 8 pack said the best thing is to set your stability for what you actually need it for. No point in making it prime stable and run your CPU underclocked and overvolted when you only use it for gaming.

I use real bench or aida64. Both have been perfect for what I use my CPU for. Gaming, handbrake etc.

What version of prime? Anyhow i build pcs to be as stable as if it were stock, not just for a specific purpose, not to knock 8 pack but i prefer a stable more conservative clock rather than a maxed out potential bsod in the waiting.

 

[tyler_jrb]

Haven't used it in ages. Will be an old version I had with my 3770k.

Just to add my 5930k runs perfect 1.25v. Real bench, AIDA 64. Even XTU. Run prime and it BSOD's in about 5 seconds. Requires 1.35v to be stable. Same voltage I can get 4.7ghz with.

Never had it crash. Never had it BSOD in handbrake or games. Don't need prime so simply do not use it.

 

[Justintime]

Well to be fair everyone has their own definition of stability and more power to you, but that would never leave my door.

 

[tyler_jrb]

True I wouldn't sell it to someone like that. But for my personal usage I'd rather it stable with what I use it for. If there was issues with programs I use, sure I may ensure stable with prime. But runs everything I need perfectly fine so I don't see the point really.

 

[Silent_Scone]

True I wouldn't sell it to someone like that. But for my personal usage I'd rather it stable with what I use it for. If there was issues with programs I use, sure I may ensure stable with prime. But runs everything I need perfectly fine so I don't see the point really.

What's funny is this is also Intel's stand on these things. They were on a recent podcast with ASUS and shared that mentality.

Unfortunately people won't listen as they think they know better, the reality is they do not have the experience or knowledge to understand why Haswell-E and Prime 28+ is a bad combination. This will become more evident with Broadwell-E as these testing conditions become more of a limitation.

I guess seeing things for what they are, and accepting one may have not realised what they were doing is misinformed separates the men from the boys.

Look, we agree AVX codes are more intensive. You seem to think that means they're a danger to the CPU, that they will "degrade both cache and core" as you wrote earlier.

What you're really saying is a massive overclock + Prime95 is dangerous. Well duh! Overclocking itself is a danger to the CPU and will shorten its life, we all know that. That's not Prime's fault, it's the 35-40% overclock on a CPU that was never designed for anywhere near that.

Where the bad advice really starts is then telling people to keep their max overclocks and ignore the fact that some codes will push their chips to dangerous levels. Er, what happens when they do try and run some intensive programs? Suddenly the overclock they thought was stable isn't so stable, because they didn't test with the worst case scenario. This guy a few weeks ago comes to mind... https://forums.overclockers.co.uk/showthread.php?t=18727575

Instead we need more Justintimes on the forum to say "that's a nice overclock but do anything really intensive on it and it'll melt! dial it back a few hundred MHz".

Unless of course people are doing 4.7 GHz overclocks just to surf facebook, in which case...

Everything in your post is bad advice. You are effectively saying that it's better to tell people to push Prime and run dangerous amount of current (that you still seem to be misunderstanding) purely for peace of mind.

I'm more interested at this stage to know who/what it would take for you to realise the error of your ways.

 

[lamboman]

I guess seeing things for what they are, and accepting one may have not realised what they were doing is misinformed separates the men from the boys.

This is just childish.

Everything in your post is bad advice. You are effectively saying that it's better to tell people to push Prime and run dangerous amount of current (that you still seem to be misunderstanding) purely for peace of mind.

At no point did he say anything to that effect. He is stating that if you're willing to run conservative clocks, Prime95 is safe to run on an overclocked system. Due to the non-soldered IHS on "consumer" chips (i.e. Haswell), this is almost impossible as thermal dissipation is poor to start with.

You seem to be taking and running far with Asus' claim that Prime95 can damage processors. I have no idea why you're focusing on Prime95. Running AVX 2.0 capable versions of Linpack, or frankly any synthetic benchmark or any similar workload takes advantage of AVX 2.0 will yield the same result. If you're watching your temperatures and voltages, it's absolutely fine to run heavy AVX 2.0 workloads on overclocked processors. You just need to be conservative and understand what you're running.

Quite frankly, I can't be the only one who is sick and tired of hearing that "an hour of Realbench is plenty", "if it doesn't crash it's stable", "Prime95 will instantly kill your processor", etc. The bar set for what people are calling "stable" is shocking, and the fact that some people are then using these systems for distributed computing projects (even going as far as to use the projects for stability testing) is concerning to say the least.

You should be running stability tests that incorporate error checking, for a significant length of time (24 hours is reasonable). If an overclocked component can be subjected to such a workload free of errors, as you'd expect from a stock clocked component, for a reasonable amount of time, it is stable. An 8 hour run in Realbench without crashing or an hour in a game without visual artifacts doesn't cut it. If that's good enough for your use case and you don't mind the implications, fine - it's your system after all - but don't call it stable.

People seem to have this idea that overclocking is almost a case of set-and-forget. The reality is that for a truly stable, verifiable overclock is a staged process that places your system under load for days. Most don't have time for that.

 

[Justintime]

This is just childish.


At no point did he say anything to that effect. He is stating that if you're willing to run conservative clocks, Prime95 is safe to run on an overclocked system. Due to the non-soldered IHS on "consumer" chips (i.e. Haswell), this is almost impossible as thermal dissipation is poor to start with.

You seem to be taking and running far with Asus' claim that Prime95 can damage processors. I have no idea why you're focusing on Prime95. Running AVX 2.0 capable versions of Linpack, or frankly any synthetic benchmark or any similar workload takes advantage of AVX 2.0 will yield the same result. If you're watching your temperatures and voltages, it's absolutely fine to run heavy AVX 2.0 workloads on overclocked processors. You just need to be conservative and understand what you're running.

Quite frankly, I can't be the only one who is sick and tired of hearing that "an hour of Realbench is plenty", "if it doesn't crash it's stable", "Prime95 will instantly kill your processor", etc. The bar set for what people are calling "stable" is shocking, and the fact that some people are then using these systems for distributed computing projects (even going as far as to use the projects for stability testing) is concerning to say the least.

You should be running stability tests that incorporate error checking, for a significant length of time (24 hours is reasonable). If an overclocked component can be subjected to such a workload free of errors, as you'd expect from a stock clocked component, for a reasonable amount of time, it is stable. An 8 hour run in Realbench without crashing or an hour in a game without visual artifacts doesn't cut it. If that's good enough for your use case and you don't mind the implications, fine - it's your system after all - but don't call it stable.

People seem to have this idea that overclocking is almost a case of set-and-forget. The reality is that for a truly stable, verifiable overclock is a staged process that places your system under load for days. Most don't have time for that.

Great post, and I agree that it seems the bar has been lowered and that a lot of people are claiming much higher clocks with a much lower quality of testing. Its like building a car and claiming it can do 200mph all day but testing it at said 200mph downhill for 10 minutes, i.e lightly loaded and quite short. And while I do tend to refrain on comments about peoples personalities or online personas in general I must say that it does come across as a bit childish, even if said person turned out to be 100% right the point was not made in a particularly great way.

 

[mmj_uk]

The way I see it CPU stability is like a light bulb, it's either on or off, stable or unstable, saying that if it's stable for what you use it for then its stable is being quite ignorant really. All it takes is one tiny error in calculation and the physics of a game might go funny, or an application might do something unintended, okay it won't crash and you probably won't notice it but just because a CPU is unstable doesn't mean that it will only crash, crashing is the most obvious and severe side of processor instability.

It's true there's no way to guarantee stability but tests like Prime95 are the closest thing consumers have to the million dollar equipment Intel will use to test their processors. A processor is a bunch of on off gates, it either switches those gates on off at any given speed or not and the whole argument that processors aren't supposed to be fully loaded is merely an excuse for poor thermal design and/or too aggressive clocking by the manufacturer, or end users who crank up the voltage too high and would sooner sacrifice that peace of mind of knowing it's stable for the sake of an extra 200mhz which in most cases is less than 5% extra performance.