Finding DAILY stability on FX chips

[Daveburt714]

This seems to be a matter of contention from alot of folks (and it's somewhat justified)...
There are so many stress tests to consider, LinX, WPrime, SuperPi, HyperPi, CineBench, OCCT and too many more to mention!

The one obviously missing from this list is Prime95. Even folks who use this as a stress test will argue about how long you need to run it to get real stability...
When I bought my 1st FX chip, I actually thought something was wrong with P95 because I could pass everything else and still fail Blend in minutes!

P95 Blend is by far the best stability test for FX chips, it heats up and stresses them way more than anything else.
For that reason, I really don't think you need to run it for hours on end...

If you can get it to blend stable (or Custom (MaxMem)) for 30 minutes you'll be good for everyday use!
Even Video Encoding, which is the most stressful thing I do, will be good if you can MaxMem Blend for 30 minutes...

Depending on cooling, all my chips crap out ~70c, but nothing will heat them up up like P95!
If your going to run BOINC/WCG you may need to lower the HTRef a couple clicks, but the above settings will still get you close.

Prime95 Blend is the best stability test IMHO, but don't try to run it for hours on end!
It's VERY stressfull and you may kill something else in the proccess...

That being said, I swapped out my best chip (the old 8120) for the 8150 and got these results:

[FlanK3r]

And wehre si yor 8120? On the way to me :-P? This 8150 looks good too Dave.

[BeepBeep2]

It is extremely easy to fail blend within the 1-2 hour mark with FX...and not because of heat.
If anyone needs to be 100% stable, 4-8 hours + I would recommend Dave

[Andrea deluxe]

hyperPI 32M 16core

prime 95 (all the test)

prime95+unigine at the same time

[FlanK3r]

no need 4+ h PRIME95, its bull, sorry guys, but really, your PC is overheating in 100% all day?

[tbone8ty]

8150 price has been reduced finally

[FlanK3r]

maybe tomorow coming my next 8150, after wattercooling set....I hope for impressive batch

[BeepBeep2]

no need 4+ h PRIME95, its bull, sorry guys, but really, your PC is overheating in 100% all day?

You're going to call BS?

chew* said anything less than 4 hours was BS on Thuban, rules don't just change with BD because you can get away "most" of the time without being stable...
I had Prime95 fail on BD @ 4.5 GHz with too low vcore after three hours one night with my old chip...and no overheating, I was in mid 60's c...

Also, why complain about the temperature? That's like running Thuban at 4.3 and 1.625v+ then complaining about Prime95 because the CPU heats up too much...

Dave titled this thread "Finding REAL stability on FX chips"...to do that, you need a lot longer than 30 minutes IMHO.

[tbone8ty]

so what should CPU/NB or HT link speed be?

am i going to lose much performance by keeping them both at 2250mhz?

having trouble keeping it stable both cpu/nb and ht link at 2500mhz....cpu/nb volt are at 1.25v and ht link at a little bit above stock at 1.2125v

right cpu is at 4.5ghz @ 1.425v

figure id keep them both at 2250mhz since the cpu/nb volts is making it get super hot

[BeepBeep2]

Try 21 x 214 for DDR3-2000 X-X-X-XX, 4.5 GHz CPU and ~2360 CPU_NB (11x). You can also lax your blackline to 9-11-9 and gain some more headroom, maybe dropping CPU multi to 20.5 x 219, (4.5 GHz CPU) DDR3-2050 X-X-X-XX, and 2.42 GHz CPU_NB.

HT Link isn't quite as important.

[tbone8ty]

cool ill try this tonight...figure ill leave HT link on auto for now

thanks beep

[Gener_AL (UK)]

You're going to call BS?

chew* said anything less than 4 hours was BS on Thuban, rules don't just change with BD because you can get away "most" of the time without being stable...
I had Prime95 fail on BD @ 4.5 GHz with too low vcore after three hours one night with my old chip...and no overheating, I was in mid 60's c...

Also, why complain about the temperature? That's like running Thuban at 4.3 and 1.625v+ then complaining about Prime95 because the CPU heats up too much...

Dave titled this thread "Finding REAL stability on FX chips"...to do that, you need a lot longer than 30 minutes IMHO.

if an AMDFX81X0 can process the same amount of tests in less time then thats fine imo, 30 min seems not enough time chew enough iterations of prime numbers (of course ram also is a factor too)
personally I run vantage and 3dm11 CPU tests looped / memtest for ram.
Bulldozer so far for me has been a great experience overclocking and for power use/and efficiently (yes I know!!!).
5ghz on cool air running full AIDA suite with 2.6ghzNB (back in feb) 1.475CPU NB 1.33V was required just not worth using 24/7 too much W.
Voltage on the CHV is nuts and like the cpu needs attention to get it to behave/ bulldozer also needs very very careful voltage application i have noticed so far.
Some programs/games depending on what FPU/Int load is required can use less voltages. I can happily game at 4.4ghz with stock voltages (using 1.41v which is CPU VID for turbo). 1.365v for 4ghz.
This cpu takes stock NB1.2250 stock Vid is 1.3125 / NB seems stable at 1.25v 2400 but 2.6 is just not worth the voltage wall/loss of stability/heat.
power states and BIOS seem to be the key to unlocking performance / stability can be quite easy if you know your workload.

[NEOAethyr]

Used to be 48hrs+, highest fft's I could use with the amount of ram I have, ram limit set to 1 or 4, I forget which doesn't matter (more then the min tends to swap lots more, no need to hammer the drive using swap in prime).
I would probably use a custom fft, 40960k (40960, not 4096), 6 instances of the prog, one on each core.

I don't see the point in running with lower fft's then 4096k, which is what I used for my nf2 with 512m+ of ram...
Some say it stresses the cpu more because it fit's into cache blah blah, etc.
But in my experience, the higher the better, just 'cause it takes a little longer to find a cpu error with a higher fft doesn't mean it won't find it soon enough .

I won't prime unless I am going for a 48hr run, 48hr's of cpu time, which tends to be more then 48hrs of running prime... (depending on what you're running at the same time).
Also you wanna give yourself enough ram free'd up to use to game with and browse the net with.
4, 6 or 8 instance of prime with a 4096k fft should be no issue.
But I think it's better to go even higher if there is the ram for it.


But anyways...
Prime is in my opinion a waste of electricity and effort checking for stability.
It's hard to say if there's any sort of combination of things that outright beat prime in terms of stability though, except for certain things like 3d.
It's not easy to check stability in any case, with or without it.
But I think there are better ways.

Though, prime's built in benchmark, if you unlock all the fft's (unlock, read only ini, etc) it does take a little bit to finish and tends to be a halfway realistic stability test.
Hard to say, never had a habit of using that much.

I would like to say that intelburntest is crap ^^.
It will pass, always, almost... :\
I mean it's a way to easy of a test to pass.

Y Cruncher is interesting though.
Seems to take 8hrs to really check for stability sometimes, at the min I guess.
That's to long too...

PCSX2, gaming, s&m, browsing and watching movies/music is what I tend to do to test.
When gaming I tend to run it for about an hour or so, and then I move on to the next game.
It's tedious, it would need to be done anyways even you primed ^^.

Edit:
Oh, a call of duty engine baised game works better then s&m's mem test for testing cpu voltage...
Like block ops for example is what I had been using to check for vcore a while back, before I got new mem.

Block ops needed more then s&m, just a tiny bit, at way below stock vcore though the gap was larger (was messing with the msr pstate tweaker prog).
I'll still use s&m for primary testing fo mem timings/speeds though.

Y cruncher was good for checking cpu-nb voltage back when.
Pcsx2 might be much faster and better at that...

PCIE speed, black ops again, just starting a map correctly seems to be a very decent test.
To finish pcie/chipset-nb voltages I take it to mafia2, for now...
Oh and I do make sure I run 5 rounds of nazi zombies on codwaw, it's just something I do for testing, not sure if black ops is better or the same yet lol (I run the same test for vcore that I do for the pcie speed to put it simply I guess, for the most part).
Testing a cod without a mg42 is boring, so I still run waw so the consent testing isn't as boring :|...

Ht voltage, just get it to boot windows seems good enough lol...
For those high fsb speeds, around 300mhz+.

Of course I check every game I have installed... just saying I've got at least half a dozen more games I go through...
It's not fun unless you're breaking limits or playing with new toys I think .

Sorry for the huge post.

[stangracin3]

I may be out of the loop, but im with beep on this one.
You don;t just change the rules because of one chip.

[AlleyViper]

I would like to say that intelburntest is crap ^^.
It will pass, always, almost... :\
I mean it's a way to easy of a test to pass..

Even on my Deneb the only good thing IBT/Linx would do after passing >8h Prime Blend was to find the need for an extra single bump (+0.0125V) of cpu_nb voltage. It would never need more cpu voltage than prime, being happy with up to -0.0250V than a long Prime Blend required, despite slightly higher temps being reached.

[Daveburt714]

OK.... I changed the title of the thread. Real stability may have been a little much.

What I was trying to say is when I get my machine to blend for 30 minutes, it runs daily tasks without problems.
As for running it for hours on end, I personally wouldn't do it! Even after an hour @ reasonable v's/clocks the power
section of the board gets extremely hot to the touch (even with a fan). I don't see the point in pushing your hardware to
those extremes when 30 minutes results in clocks stable enough for everyday use...

Granted, you can't change the rules for new chips, but power requirements for FX are more demanding than previous AMD CPU's.

I was just offering my experiences, if you want to push your machines for hours on end, it's your call.
I was just telling folks what works for me...

[wez]

OK.... I changed the title of the thread. Real stability may have been a little much.

What I was trying to say is when I get my machine to blend for 30 minutes, it runs daily tasks without problems.
As for running it for hours on end, I personally wouldn't do it! Even after an hour @ reasonable v's/clocks the power
section of the board gets extremely hot to the touch (even with a fan). I don't see the point in pushing your hardware to
those extremes when 30 minutes results in clocks stable enough for everyday use...

Granted, you can't change the rules for new chips, but power requirements for FX are more demanding than previous AMD CPU's.

I was just offering my experiences, if you want to push your machines for hours on end, it's your call.
I was just telling folks what works for me...

For daily gaming stability running 4-6 threads in prime can be somewhat successful, just make sure to switch them around. Or if you run shorter prime sessions, like 30min, make sure to reduce the run time on each FFT so you still get to run several different sizes. But there is still no shortcut for real stability, although BD takes it one step further when it comes to power draw during prime, and those values are for many very unrealistic in day to day usage. Encode something with 8 threads and you still wont as high load as with prime.

The main problem I have atm is that I dont have a separate water loop for the CPU anymore. Which makes stress testing so much harder when I have to load the gpu's at the same time to keep things realistic.

[demonkevy666]

I'm a mad man then
I've ran A.O.D, prim95, and Coredamaga0.8h all at once. :]
So far Coredamage seems to heat my cpu's the hottest. Even that does not find stability all the time. After several hours I've still had it crash on light gaming.
I've used OCCT too with furmark, IBT too.
To be honest prime95 and IBT seem suited for the ram A.O.D sometimes finds error earlier then others but not always.
furmarks is good gpu but can make them abnormally hot compared to normal gaming.

use everything at your disposal to find stability imo.

[Miwo]

Battlefield 3 is a great stability tester and will crash both CPU and GPU unstable OCs
Mind as well have some fun too while stability testing right?

[BeepBeep2]

I've measured 88c on VRM on the back side of the board under the backplate with CPU @ 4.7 GHz and too much temperature/voltage.

Perfectly safe

Around 100c, the board will throttle CPU as long as you have VRM over temp protection enabled in bios.

[Plywood99]

If it never crashes with what you do, then its stable. Peeps can argue true stability all they want, it's B.S. If your overclocked cpu does what you need it to do with no crashes or problems, then the rest is just B.S.
I don't run P95 for hours on end. You know why? It's pointless to me.
Take it all with a big helping of salt folks, cause stable to you is not same as stable to me.
Quick rule of thumb: If it does what you need it to do and is stable, then great.
Tell everyone else to go piss in a bucket...

[FlanK3r]

You're going to call BS?

chew* said anything less than 4 hours was BS on Thuban, rules don't just change with BD because you can get away "most" of the time without being stable...
I had Prime95 fail on BD @ 4.5 GHz with too low vcore after three hours one night with my old chip...and no overheating, I was in mid 60's c...

Also, why complain about the temperature? That's like running Thuban at 4.3 and 1.625v+ then complaining about Prime95 because the CPU heats up too much...

Dave titled this thread "Finding REAL stability on FX chips"...to do that, you need a lot longer than 30 minutes IMHO.

I never tested PRIME more than 1h, in real use I have with my FX never crash. Of course, my system is not 24/7 in hard load, but how many people has this? My PC is running about 10h-12h per day, maybe 8h working (microsoft office, sometimes websites etc.) and 2-4h playing games or encoding videos to DVD. And sometimes benchmarking . But if you have not PC in hard time (encoding in handbrake all day, all day rendering scenes, FOlding home 24/7) dont need 4h+ PRIME 95.

[AlleyViper]

OT: prime95 versions - p64266 is way harder than p64259 on my Deneb.

When I got my Dragon on sig 3 years ago, I did a lot of blend tests under prime95 25.9, mostly 12h stuff to check cpu vs cpu_nb voltage at different speeds, so I have quite some tables notated for my leaky C2 under a TRUE P+P (2x~1.5K rpm).
Last year, after a clean OS and downloading the newer 26.6 (which optimizes testing per CPU cache size), I couldn't repeat those results anymore and I thought it was CPU or VRM degradation, due to immediate BSOD failure (less than 5m) on my previous stable settings at 3.7 and 3.8GHz under good temps (<52ºC). The BSOD is always a 0124 (or a reboot), which usually indicates low vcore, and can only be solved by raising it. IBT/Linx/OCCT don't seem to suffer, but they've always been more forgiving on vcore.

Only today I remembered to take a shot back at v25.9, and I'm a bit surprised that it still does OK for the same FFT sizes on older settings. My 3.7/2.6GHz profile just kept going for >2h instead of failing instantly. Under v25.9 I "only" need 1.375V@3.6, 1.4V@3.7 and 1.475V@3.8; while currently on v26.6 it's 1.4V@3.6, 1.45V@3.7 for an 8h test; dropping the cpu_nb OC makes no difference and this is still under 50ºC! Didn't bother to test 3.8GHz, because it might need over 1.525V now.

Well, I guess prime just got harder since the previous 09' stable version we've been using on Deneb/Thuban. And this is why it might be catching errors so well on newer FXs over IBT/OCCT, and requiring way more vcore than everyday apps.

[largon]

For me it's not stable if it can't do same things as it does at stock.
Anything I throw at it crashes -> I judge it unstable.

[Oese]

i test linx 5-10 tests for rough stability, then like 30 min prime blend - If that is stable, I begin to use it. If anything crashes, ofc, its unstable, but it does not occur often that i have stability issues after I tested like this.

I used to run p95 for hours back in the days, and sometimes i do it now too, but only if i am in the mood. Ofc, for OC lists, there need to be standards, and either its validation, max cinebench, p95 blend xx h stable or anything else. Ofc, there's no debate in that case. What you do besides OC lists is your own call and i guess, many people will have their likings