Stability testing on a Linux machine

[knoober]

I put this here instead of the linux section because I think this is more of a stability problem than anything to do with the OS. But Ill check in there if need be. Until then let e know what you guys think

I build a home NAS machine that consists of the following parts:

Gigabyte 970a-d3p
330W Seasonic PSU
2x2GB Elipeda
Im kinda sketchy on the GPU - AMD 5450 maybe? its a low powered AMD card with a passive heatsink. I will dig up the actual part name if it will help
CPU AMD Athlon II x2 270
.... and a bunch of drives Soon to be less drives with a higher capacity.

Here is the issue. I cant stability test my OC accurately in Ubuntu! I pushed the chip from 3.6Ghz up to 4.0 Ghz with no troubles at all. The cooling is more than sufficient as I used the setup from the Phenom II I had in the case before. So Im basically running a 65W chip in cooling meant for twice as much

Linux tools that I can find will stress the cpu and list at 100% stress. Temps approach 45C after hours of testing but are stable right there (also there is a heater vent blowing right into the case when tepms are this high - not ideal I know but I did the best I could for placement and the temps arent out of line ). The problem comes in when I try to encode any of my Mpeg files to h.264 . I cant do it. System shuts down after about 15 min.

I know you guys usually need some looks at CPU-z and such but I cant get them without installing Windows and that is a last resort. I already went a little pushed the CPU voltage up some because I have headroom , the HT Link and NB frequencies are as high as they can go for this chip (2000) , and Im not sure if messing around with the NB voltages is going to gain anything if I cant raise the NB frequency.

As I said before Im really just looking for some guru action from OC'ers that might have an educated guess. I know you might need more info and Ill get it if its called for ... but Id hate to install windows and have looks at CPUz and HwMonitor just to find out that my chip is already maxed out!

 

[DNRDustin]

What do you mean you can't accurately test stability? MPrime is what you need, AFAIK prime 95 was a Linux utility first. Imo you want an nVidia GPU if you're running Linux BTW.

Take a peak at this thread, there might be some little tid bits of info that you find useful. It's an fx6200 OC in a ga-970a-d3 on ubuntu 14.04
http://www.overclockers.com/forums/...houghts-on-my-Overclock-System?highlight=6200

 

[knoober]

I only found something called "stress-cpu" (thats from memory dont quote me on it, but it allows to choose how many cores and other options from the command line) for streassing and lm-sensors with conky for reading my temps. Ill have a read through that thread and hope I dnt feel too embarrased afterwords

Edit: smack your head and sat D'oh ! The repos have got no prime95 but its right there in the download section. I still didnt pick up from that thread what was being used for monitoring. is that conky + lm-sensors? I ahvent had a great deal of luck with that but Ill admit that Im not a real genius with Linux either

Edit2: Haha, prime killed me in less than 5 min. Looks like Ive got some work to do. Or I could just drop it back down to the stock.... NAH! Id really love an AI Suite or something similar for this board because its a headless station thzt I connect to remotely. I guess Ill have to pull the monitor out for a day or so

 

[DNRDustin]

The gui is psensor, but that was just for the screenshot. Really I'd just watch lm-sensors in terminal.

- - - Updated - - -

The gui is sensor+lm-sensors, but that was just for the screenshot. Really I'd just watch lm-sensors in terminal. "sudo watch sensors" if you're wondering.

 

[trents]

4.0 ghz stable on an Athlon II X2 270 is not a given. That one is pretty much maxed out from the factory. Some of them will do it and others just won't. I would lower it to 3.8 ghz and test the stability again at the same voltage. By the way, what is the core voltage set to in bios? Did you ever raise it to compensate for the overclocked frequency? You never said. And exactly what cooler are you using? I would not count on it doing an adequate cooling job of the X2 overclocked just because it handled the X4 Phenom at stock. Overclocking jacks up watts pretty quick.
And with that heater vent and all . . . you are taking a lot for granted. The dual core CPUs are usually failed quad cores and they may not scale well temp wise when overclocking.

If I were you I would do a temporary install of Windows so you can do some adequate testing and monitoring of your overclock and get it stable. You wouldn't even have to activate the Windows. Then put Linux on it.

 

[DNRDustin]

I agree with most of what Trent's is saying, except for installing windows. It will be more of a challenge for you on Ubuntu, it's not what people are used to seeing, and you yourself said youre not all that comfortable with it. So, it will be a little challenge. You can install Wine and run windows programs. I don't know if I'd trust hwmonitor readings in Wine though, I'd watch lm-sensors for that info, but you could run CPU-z in it. Screenshots man! I'd start by setting bios to factory default and running a prime blend.

 

[knoober]

4.0 ghz stable on an Athlon II X2 270 is not a given. That one is pretty much max out from the factory. Some of them will do it and others just won't. I would lower it to 3.8 ghz and test the stability again at the same voltage. By the way, what is the core voltage set to in bios? Did you ever raise it to compensate for the overclocked frequency? You never said. And exactly what cooler are you using? I would not count on it doing an adequate cooling job of the X2 overclocked just because it handled the X4 Phenom at stock. Overclocking jacks up watts pretty quick.
And with that heater vent and all . . . you are taking a lot for granted. The dual core CPUs are usually failed quad cores and they may not scale well temp wise when overclocking.

If I were you I would do a temporary install of Windows so you can do some adequate testing and monitoring of your overclock and get it stable. You wouldn't even have to activate the Windows. Then put Linux on it.

First let me put your mind at ease about the cooling. I glossed over the matter of cooling because I didnt forsee the temps being any part of the equation with results so favorable, buuut... Im not using a cooler meant for a Phenom II, Im using the whole setup! Which means: all the fans + a slightly above average cooler. I would have to dig for the exact name of the cooler but I was previously told it was little better than stock... but it IS rated fr a 95-125W chip. I have intake on the drives (in front and below) 120mm intake under the optical drive. 120mm on the ceiling and at the usual exhaust. This thing blows cold air under load. Honestly I am amazed. And the heater is not piping directly into it either. It is actually a sweet spot for thermal matters. The case sits under a corner bench, open front and open back panel so plenty of flow. When the heater is on though the temp does raise about 10F just from blowback around the rear opening. Air from the front is still cool. As an added bonus, in the summer time air conditioning is pumped through that same register so it will be cooled in the warm months. I was worried too so I tested it. Thermal problems are not even in it... I think Ive got a solid 45C under load (atleast until I got prime95 running. Obviously what I was using wasnt stressing enough)

I did the OC in a real slipshod manner because of how easily I achieved good results. I jumped the FSB as high as I could and added volts until it was stable -> then tested. Temps were still fine so I gave the cpu voltage one more tick in the BIOS and dropped the FSB 2 points and called it good. I realize it was a lazy OC but it was working and I wanted to be done! Now I have to go back in there as a reward for my haste. phooey!

I agree with most of what Trent's is saying, except for installing windows. It will be more of a challenge for you on Ubuntu, it's not what people are used to seeing, and you yourself said youre not all that comfortable with it. So, it will be a little challenge. You can install Wine and run windows programs. I don't know if I'd trust hwmonitor readings in Wine though, I'd watch lm-sensors for that info, but you could run CPU-z in it. Screenshots man! I'd start by setting bios to factory default and running a prime blend.

I agree about Wine. I havent found anything that its really good for. I read that you should use the Linux packages when you can and I do. Ive got to drag out the monitor to be able to run clonezilla anyway (doing an OS drive swap so I can add more storage) so I might as well fix the OC while Im at it. You'll get your screenies And thanks for the heads up on psensor. I recently found "xsensor" and it is pretty okay, but your shots looks nicer.

 

[trents]

When you use the FSB to overclock a CPU the HT Link and CPU/NB frequencies also go up. You may need to lower them, especially the HT Link, or give the CPU/NB some extra voltage.

 

[knoober]

When you use the FSB to overclock a CPU the HT Link and CPU/NB frequencies also go up. You may need to lower them, especially the HT Link, or give the CPU/NB some extra voltage.

You know i kind of glossed over that fact when I did the original OC. THanks for reminding me. However When I raise the FSB (and it has to be the FSB, the multiplier is maxed out) the HT Link and the CPU NB max out at 2000. Are they capped there or just not showing the increase, do you think?

Edit: You were right to caution double checking. I had been reading the thermistor (which I understand is the mobo temp : please confirm if you know) as the thermal diode (which I read to be an on die temp sensor: also please confirm if you know). Basically I was reading the wrong temp. Wouldnt have been a problem in hwmonitor, but Im glad it was caught. Load temps are on the border. Ill be back with screenshots after I tinker around a bit.

Edit again: I gotta hand it to you Trents, you were spot on. This thing is more work than I thought. Ive been at it all day and am having a tough time keeping the temps below 60 until I drop the voltage a bunch. I keep losing cores because of the lack though. If I cant boost it up to 3.9Ghz then I will have to just run it at stock. 200 mhz isnt worth the hassle/risk of instability.

 

[trents]

No can't confirm the thermistor thing. I've fooled a very little bit with overclocking in linux but it was so frustrating to try to monitor stuff and know what you were monitoring I gave up.

The HT Link and CPU/NB must increase with the main bus frequency since they are tied to it. My guess would be the increase is just not being reported by the bios which may not have the microcode to interpret beyond 2000 mhz. The only way that's not going to be the case is if somehow the bios is automatically lowering the multiplier for those two components. I know on budget boards the components have ceilings but I think it's a reporting issue but whether or not that applies with your board I can't say for sure but that would be my guess.

 

[trents]

By the way, do you have the latest bios installed on the motherboard?

 

[knoober]

By the way, do you have the latest bios installed on the motherboard?

The short answer is that Im not sure and Ive already gotten the monitor unhooked and the machine back in its resting place. When I initially set up this board with a Phenom II, I had been havng trouble locking the cores together and needed to switch the BIOS version for testing. I honestly forget what version I ended up staying with. I think I stayed on the last stable version (as opposed to the BETA that is the newest).

Ive got the screenies ready! I had to take a minute and remove anything personally identifiable (home machine, but I do the best I can for security) or I would have had these sooner.



This pic is stock settings (hitting F7 or whatever key is labeled Set Optimized Defaults in BIOS) and no load. There are a couple things to notice here. The first is the tabs on the far left. When I first got this little gem it only had one tab and on box on the tab. The tab was labeled k10temp and the box was labeled "temp1". Put that together with the fact that my cpu stress program was a bit weak and the airflow from the case was cool on the skin and you can see where the problems started A little bit of putzing around with the configurations and now there are more tabs with more info!


Since these are just stock pics and just for show I only ran prime95 for 10min. I am always surprised by how much voltage is poured on "out of the box". 1.4+ V is far to much for this chip to run at a decent temp. Almost enough for full load at 4Ghz (this chip will run for nearly 20min at this voltage/4Ghz before dropping a core). I understand that they are trying to guarantee stability, but its still surprising


OC with no load on the cpu. Plain jane. Ive only raised the FSB at this point and haven't done anything with any other settings. Voltage is still way up there.


Here is that same OC under load for 20min. Man I wish I had been able to get a copy of CPU-G (apparently that is linux cpu-z, or works/looks much the same) but I got a 404 at what seemed to be the official page. If you notice "temp1" in the pics - thats what I was looking at before I got all the other boxes. The middle box -temp2- is the core temp/maybe and temp3 is anybody's guess


I dont have pics of the final result because I got busy and just didnt have time to document like that. I ended up pushing the voltage down to 1.37V and was still hitting load temps of 55-57/58C if I dropped the voltage any more I would lost a core too soon. I just kind of accepted that where it was because I can accomplish the tasks I need. I realize that it is against best practices, but I was born a lazy lazy man I like the results of OC , but I do not care much for the process! The machine can run its encodes at night when the ambient is lowest and this machine isnt my daily driver so I can be patient if it drops a core and slows down. Most of the encodes Ive tested lasted less than 20min anyway (the mark where my core dropped). The next time I have occasion to pull the monitor out I might see if I can improve things, but for now its good-ish. Also with respect to the HT-Link and Cpu NB I believe that they are simply maxing out at 2000mhz on a locked chip. I did end up dropping the HT/NB frequency down to 1800 and saw nice results with no need for increased voltage on the NB.

Edit: After looking at this post I think I wont do the screenshots at full screen next time Sorry if its hard on the eyes.

 

[DNRDustin]

k10 is your package/core temp. temp2 is most likely socket, and temp3 is either your north bridge or vrm(more likely) temps.

 

[trents]

1.5 vcore is not excessive for that line of processors as long as temps are well-controlled. 55-60c is about the stability limit for core temps and 65c is about the limit for the socket temp. I'm speaking of load temps now. Intels will handle much higher core temps without becoming unstable.

 

[knoober]

1.5 vcore is not excessive for that line of processors as long as temps are well-controlled.

I will keep that in mind for my next tussle with this cpu. I guess I just instinctively shy away from anything close to 1.5V (not that I havent set it there just to see if I can pull some more stability). I had assumed that being k10 cores (same as the phenom II right?) they would have the same thermal limitations. I have read that 70 is okay for this chip, but Ive been trying to adhere to 55C as my limit (and that has been a challenge that I can pretend to have met ). I usually try to dial back a step or two from where the stated thermal limits are just because I fear for my chips. Thanks for the help guys. Ive got to go see if I can get that tab with all the extra information back on xsensors. I had to do a fresh install and now its back to just the k10 tab. Ill be sure to update the thread if anything amazing happens

 

[trents]

70c is the max "safe" limit for those CPUs. But safe and stable are two different things.