27.7 vs 27.9 for Stress Testing?

[Batalov]

Originally Posted by Belial88

I just have the faintest clue that prime95 was originally designed a long time ago for finding prime numbers. I really havent an idea what you are talking about.

Your whole post pretty much lost me. I'm just trying to stress test...

There's no argument that there's value in stress testing. However, you must ask yourself the main question - stress testing for what? Just protection from regular BSoDs (for gamers who will boast to each other "supa cul, my man, 4900Mhz on air and 120 fps in Crysis 3" with all the rainbow blinking lights and a pink custom case)? That's too easy. Do your stress testing as you did now, never mind what the program was written for. (You probably never asked yourself what was... other than that silly SuperPi test that stubbornly fails, that is...)

If you are just building boxes, then sell them to gamers - this is not the test you should be running anyway. If you thought that "yeah, 12 hour test is not enough, I've seen boxes that fail after 12 hours but before 24 hours. If I could only squeeze 24 hour test into 12 hours", you are not prepared for the idea that there exist systems where 1 bit error in 30 days is unacceptable (and then there are systems that people have the audacity to ran 500-day tests on, and then act all surprised that their test were full of errors), and guess what - all of you "24-hour-rock-stable-systems" will fail that test.

Truth is too harsh?

Take it easy. Or take the hard road and actually try to understand what is going on in the tests that you are running.

[Belial88]

The best way to find out (which does not involve advanced knowledge) is to connect a power-meter to the electrical plug of your system (like a kill-a-watt or so) and see which testing mode takes more power. THAT will be the one which is stressing your system more. Well, this makes me think that a long discussion can be started from here too, regarding of what parts of the system you want tested, because this will arguably stress your power supply too, and if that is what you want tested also, then turning on/off different peripherals can do a better job, or not.

No need to test PSU. I want to test the stability of my system, primarily my CPU and RAM and their interactions. So I want to make sure my overclocks are stable, on both my CPU and RAM. I'm also making sure my hardware isn't faulty but I've already passed 24+ hours on stock settings as well as 24 hours on varying overclocks.

Don't have a kill-a-watt, but there's power readings in hwinfo on how much power my cpu is consuming i believe. I'm sure it's not accurate all the way, but comparing +/- based on test thread counts...

On the other side, P95 was never a good toy to test the memory or the memory interface, the VGA cards, the PCI-e slots, etc. What P95 is "genial to", is testing your CPU, and CPU-cooling assembly, under really hard conditions. It is the simplest and the best program if you want to find issues with your CPU, with your thermal dissipation system, or if you want to identify "on the edge" cooling or power supply problems. No other software beats it at this.

I have found prime95 to be a much more reliable test of RAM stability, and for faulty RAM, than any other program for the same amount of time. I basically test my ram speeds and timing overclocks by using 24 hours prime95.

Currently stress testing 2400mhz 8-12-8-28 1T with some extremely tight secondary and tertiary timings (i got 8/12/8/28 with default loose secondary/tertiary passed 24 hours, and some of them tightened too at 24 hours, but having just a bit of trouble on the last stretch, i think i figured it out that as my last failure was at 21 hours).

Prime95, moreso than any other program ime, seems that it is best for stability for memory - so your memory might work at whatever timing, but the IMC might not like it, which is where p95 becomes important. Full system interaction, see not just how the ram overclock affects the ram, but affects everything.

Small fft is also indeed the best temp testing program I've found, and Ive done a few heatsink reviews.

but prime95 custom blend 24 hours seems to be the best and easiest and quickest way to test for full stability in my experience. I'd love it if there was another program or something that tested better or quicker...

P.S.: by "DC" I was referring to "Double Checks", a kind of "test" we do here around, it is enough stressful for your hardware (beaten only by the LL test, but for LL test you don't know if your hardware was good or bad, unless someone else "Double Checks" your result, which can take months or years). One LL test can take around 7-20 days on your system (depends on exponent, number of threads, etc, even longer for exponents outside of our current testing range). One DC test can take about 2-7 days for your system.

I think 24 hours of prime95 is pretty close to 'rock stable'. If you can assure me that this is necessary for a truly stable system, I'm all ears though.

The reason I find 24 hours necessary, is because most of my unstable overclocks fail at around 12-24 hours. I've even had faulty hardware fail, consistently, at 12-24 hours of prime95. In my experience, 12 hours of stable means nothing and can still crash instantly in games, streaming, etc. But 24 hours stable has never crashed on me in anything, and I've never had faulty hardware pass 24 hours of prime95.

I suggest GNFS (factoring a 130 digits number with all 4 threads is a good workout for your PC, especially the Lanczos phase).

I have no idea what you are talking about.

You can manually set what FFT lengths a torture test runs via one of the drop down menus; fiddle with it a bit.

Yea I know about that. The problem is that p95 runs what, 80+ fft lengths in ~23, 24 hours of prime95. You can set an fft length to any of those 80+ fft lengths, but then you'd have to manually set it each and every time...

I mean I've had overclock consistently fail at certain ranges of FFT length and it's obvious that the issue is the imc, the on-die cache, the ram, etc, from that, but there are so many fft lenths that seem to stress different parts of the system. It'd be really great if you could pause prime95, or start blend from any point at the test, or something like that.

[Belial88]

There's no argument that there's value in stress testing. However, you must ask yourself the main question - stress testing for what? Just protection from regular BSoDs (for gamers who will boast to each other "supa cul, my man, 4900Mhz on air and 120 fps in Crysis 3" with all the rainbow blinking lights and a pink custom case)? That's too easy. Do your stress testing as you did now, never mind what the program was written for. (You probably never asked yourself what was... other than that silly SuperPi test that stubbornly fails, that is...)

Streaming. And if I'm going to run an unstable overclock, that's fine, but I'd like to know my system. I have no problem running an unstable overclock, but I think you should still know your 24 hour prime95 stable overclock still.

So I might know, okay I can do 4.8ghz@1.4v. Since I only game, I'll make that 5ghz@1.4v and call it a day. Now, if I upgrade new hardware and get crashes, I can make sure it's the new hardware and not my overclock causing the problem, or something like that.

I mean I pulled my hair out for a month when a cpu I bought with 6 broken pins, that I had replaced with copper wire in the mobo socket, was unstable at every overclock and seemed 'more' stable at 4ghz@1.3v than 3.6ghz@1.48v. I was consistently failing p95 at the 14 hour mark in prime95. Eventually I realized severely downlocking the ram helped, then I realized it was actually the motherboard in 2 particulary dimm slots.

If I wasnt an idiot and let that first prime95 test finish instead of cut it short at 12 hours, it woudl have definitely failed before 24 hours, and I wouldve realized the board had gone bad in the 2 years I had used it, even though 2 years ago I had p95 tested it for 24 hours.

Thanks for the link. I think 24 hours is much, much more demanding that 12 hours, which is equal or more demanding than any other stress test, and that 24 hours will pass any game or streaming with no issues for years so that's what I settle on. I'm aware stability is relative but i think 24 hours is accepted by most people.

[kevindd992002]

Originally Posted by Batalov

There's no argument that there's value in stress testing. However, you must ask yourself the main question - stress testing for what? Just protection from regular BSoDs (for gamers who will boast to each other "supa cul, my man, 4900Mhz on air and 120 fps in Crysis 3" with all the rainbow blinking lights and a pink custom case)? That's too easy. Do your stress testing as you did now, never mind what the program was written for. (You probably never asked yourself what was... other than that silly SuperPi test that stubbornly fails, that is...)

If you are just building boxes, then sell them to gamers - this is not the test you should be running anyway. If you thought that "yeah, 12 hour test is not enough, I've seen boxes that fail after 12 hours but before 24 hours. If I could only squeeze 24 hour test into 12 hours", you are not prepared for the idea that there exist systems where 1 bit error in 30 days is unacceptable (and then there are systems that people have the audacity to ran 500-day tests on, and then act all surprised that their test were full of errors), and guess what - all of you "24-hour-rock-stable-systems" will fail that test.

Truth is too harsh?

Take it easy. Or take the hard road and actually try to understand what is going on in the tests that you are running.

So you're saying SuperPI is not a good RAM stability torture test program?

Originally Posted by Belial88

Don't have a kill-a-watt, but there's power readings in hwinfo on how much power my cpu is consuming i believe. I'm sure it's not accurate all the way, but comparing +/- based on test thread counts...

Software readings of power are very inaccurate compared to Kill-a-Watt readings. A real power meter is a good investment, really.

[Dubslow]

Originally Posted by kevindd992002

So you're saying SuperPI is not a good RAM stability torture test program?

Doing only cursory research, I think it's not quite as good as Prime95, and for the computers Batalov is referring to, even Prime95 is not a great stress testing utility. So really, it depends on what sort of stability one is looking for. For gaming stability, SuperPI and Prime95 both work fine.



As for altervatives to Prime95, the LINPACK library in the form on the IntelBurnTest is on par with Prime95, at least in terms of CPU testing; I've seen my CPU hit higher temps with that than with the small FFT torture test. I'm inclined to think it doesn't test RAM like P95 does, given the quite-small size of the problems it runs.

[Belial88]

So you're saying SuperPI is not a good RAM stability torture test program?

Why would anyone use SuperPi over HyperPi?

And no, it isn't. I can do 2400mhz 8/12/8/27 or 2600 9/13/9/28 on hyperpi 32m but I can't even do 2 hours of prime95 on that. 24 hours of prime95 for me required 8/12/8/28, although i got some very tight secondaries and tertiaries at least. I was failing a lot of runs between 12-24 hours though.

Software readings of power are very inaccurate compared to Kill-a-Watt readings. A real power meter is a good investment, really.

I'm aware of that but surely I could look at relative increases/decreases in power consumption by the cpu.

As for altervatives to Prime95, the LINPACK library in the form on the IntelBurnTest is on par with Prime95, at least in terms of CPU testing; I've seen my CPU hit higher temps with that than with the small FFT torture test. I'm inclined to think it doesn't test RAM like P95 does, given the quite-small size of the problems it runs.

I've passed 50 runs of max ibt no problem on faulty hardware and very, very unstable overclocks that couldn't do more than 12 hours of p95. doesnt get nearly as hot as small fft either.

[kevindd992002]

Originally Posted by Dubslow

Doing only cursory research, I think it's not quite as good as Prime95, and for the computers Batalov is referring to, even Prime95 is not a great stress testing utility. So really, it depends on what sort of stability one is looking for. For gaming stability, SuperPI and Prime95 both work fine.



As for altervatives to Prime95, the LINPACK library in the form on the IntelBurnTest is on par with Prime95, at least in terms of CPU testing; I've seen my CPU hit higher temps with that than with the small FFT torture test. I'm inclined to think it doesn't test RAM like P95 does, given the quite-small size of the problems it runs.

Thanks.

Try setting up the worker priority in Prime95 to 10 and you will roughly get the same max temps comparing it with IBT.

Originally Posted by Belial88

Why would anyone use SuperPi over HyperPi?

And no, it isn't. I can do 2400mhz 8/12/8/27 or 2600 9/13/9/28 on hyperpi 32m but I can't even do 2 hours of prime95 on that. 24 hours of prime95 for me required 8/12/8/28, although i got some very tight secondaries and tertiaries at least. I was failing a lot of runs between 12-24 hours though.



I'm aware of that but surely I could look at relative increases/decreases in power consumption by the cpu.



I've passed 50 runs of max ibt no problem on faulty hardware and very, very unstable overclocks that couldn't do more than 12 hours of p95. doesnt get nearly as hot as small fft either.

Yes, I know that. SuperPI/HyperPI are only used as a quick test when looking for your potentially stable RAM overclock range. I usually use both programs even though I know that HyperPI is optimized for using all cores. There were a couple of times I passed HyperPI but didn't pass SuperPI, so go figure. The ultimate test would still be Prime95.

For CPU OC, I could pass 100 IBT runs with 90% of available RAM, but would fail 24hours of Prime95 unless I bump vcore two notches.

You do reply here faster than you do in OCN.