Overclock.net banner

1 - 16 of 16 Posts

·
Registered
Joined
·
10 Posts
Discussion Starter #1
Howdy overclockers,

I have an overclocked A10-7850 on an asrock FM2A88M pro3+ with 2x8 G.SKILL Ripjaws X Series @ 2400.

It went ok, I'm prime95 stable (mprime) on debian linux (jessie) at these settings:41x mult, 102 fsb, 1900nb, 900mhz gfx. Ram is correctly identified via XMP and the setting is 2400.

temps are 51-53.5 under full load using cpuburn or mprime plus surfing.

I have a few questions.

Does changing the fsb on this chip overclock the ram/pcie as well? The mobo doesn't show actual mhz on the ram, just the 2400 setting which i take to be a divider. Can someone calc this out and explain it if so?

Also if anyone has any easy to run / build gaming benchmarks for debian...I've been hitting webgl bench pages and some x windows bench stuff but nothing great. Can't get anything else to build easily. glxgears, lol.

Thanks
 

·
Registered
Joined
·
972 Posts

·
Registered
Joined
·
10 Posts
Discussion Starter #4
Thanks, I was thinking about getting a haswell with the price drops since v5 came out. That'll come in handy for sure.

I have several rigs, but I'll see what I can do in terms of the sig.
 

·
Banned
Joined
·
6,565 Posts
I'd urge you to try prime95 with "in place fft" and "16K" settings selected - more heat and reliable long term in my experience.
Just follow this guideline to have a brief overview of how much voltage you need for any given cpu multiplier.
Post #45.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #7
Quote:
Originally Posted by 7850K View Post

yes the FSB on kaveri will affect the RAM, pcie, AHCI, NB and possibly USB if I remember right. Generally its better to just use the multiplier.
wrong forum
Yeah I thought so but I was hoping not. Crap.

Thanks.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #9
Quote:
Originally Posted by mtcn77 View Post

I'd urge you to try prime95 with "in place fft" and "16K" settings selected - more heat and reliable long term in my experience.
Just follow this guideline to have a brief overview of how much voltage you need for any given cpu multiplier.
Post #45.
What I do is run torture test with 4 threads and blend (8k to 4096 the defaults) and then pump the ram up 6400 megs and take 4 minutes each. That usually puts a load on it, lol. These are options under mprime on the command line.
 

·
Banned
Joined
·
6,565 Posts
Quote:
Originally Posted by gamer1mug View Post

What I do is run torture test with 4 threads and blend (8k to 4096 the defaults) and then pump the ram up 6400 megs and take 4 minutes each. That usually puts a load on it, lol. These are options under mprime on the command line.
The rest of the system can run fine. You can test out "16-16" for a direct L1 denial of service. It doesn't cause much heat, only local to test if the cores can take the heat. The cold plate thickness and conductance plays a major role.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #11
Quote:
Originally Posted by mtcn77 View Post

It is easier for you this way, the memory controller won't trip at reboot, though unlikely. Memory timings get stretched.
Post #338.
Actually it is easier but I failed a at 43 after around 5 hours running primes with an incorrect result. So I backed off to 42 and nudged the fsb. Stable as a table.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #12
Quote:
Originally Posted by mtcn77 View Post

The rest of the system can run fine. You can test out "16-16" for a direct L1 denial of service. It doesn't cause much heat, only local to test if the cores can take the heat. The cold plate thickness and conductance plays a major role.
And thanks for that voltage table. I suspect I have either a dud chip or dud mobo...kinda think its the mobo. Although i'm running an aging antec 520 gamer which might need replaced too. I'll just push the whole smear till it drops.
 

·
Banned
Joined
·
6,565 Posts
Of course, I was just considering heat as a parameter for the cpu. I didn't intend to overclock the fsb or any other part. Just that. I was actually deliberating to find the inflection point of the cpu cooler where it would always get cooler over time whatever the load. As it turns out, it is at [email protected] I have the stock fan, you see. Works well though. Previously what I thought was a system stability turned out to be gpu related, so two birds with one stone.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #14
Quote:
Originally Posted by mtcn77 View Post

Of course, I was just considering heat as a parameter for the cpu. I didn't intend to overclock the fsb or any other part. Just that. I was actually deliberating to find the inflection point of the cpu cooler where it would always get cooler over time whatever the load. As it turns out, it is at [email protected] I have the stock fan, you see. Works well though. Previously what I thought was a system stability turned out to be gpu related, so two birds with one stone.
If I understand you correctly, your more or using large fft sizes to test the northbridge/southbridge because of cache mismatches on the cpu and additional forced ram reads. Another way to do it "seat of the pants" is run prime at a low priority (I just nice it up on linux) and also run something continuous to stress the gpu. I use glxgears...and then surf around.

BTW, I'm at 1.45ish on the vcore for the proc and thats about all I want to give it. Temps are 53-54ish under massive load. I sprung for a hyper 212. Delivered voltage while running under load is 1.28ish to 1.41ish if lmsensors is accurate.
 

·
Banned
Joined
·
6,565 Posts
Quote:
Originally Posted by gamer1mug View Post

If I understand you correctly, your more or using large fft sizes to test the northbridge/southbridge because of cache mismatches on the cpu and additional forced ram reads. Another way to do it "seat of the pants" is run prime at a low priority (I just nice it up on linux) and also run something continuous to stress the gpu. I use glxgears...and then surf around.

BTW, I'm at 1.45ish on the vcore for the proc and thats about all I want to give it. Temps are 53-54ish under massive load. I sprung for a hyper 212. Delivered voltage while running under load is 1.28ish to 1.41ish if lmsensors is accurate.
I got frustrated when it wouldn't stabilize at 4.5GHz. Direct heatpipes are even more fickle.
I think setting it as 16K is the hardest test for "the cpu cooler" because the cpu can only work as fast as the cooler can displace the heat properly. Overclocking above this point is just messing around with its derivatives. Copper has limited heat conduction(or heat carrying capacity/heat carrier potential) and all the variables relating to lateral conductivity like the cold plate thickness, direct contact heatpipes, IHS & coldplate surface evenness and the thermal paste application will be taken into consideration.
If you have a direct contact heatpipe cooler, like the Hyper 212, you will have better transient response. Water is a good heat carrier in close proximity and cores won't heat up much during lateral heat conduction. However if you test the whole cpu die itself using all sorts of its parts, or in other words just go full blast at total heat, you will be settling for less because the heatpipes don't necessarily determine how much heat you can dump - those take place at radiator fins via forced convection. Besides, water will not condense fast enough near the fins and the cold plate will start running 'dry'(be sure to orient the cpu heatpipe plane perpendicular to gravity to ease capillary action back to the heat source). Your cooler will start falling behind its initial performance and this isn't what direct contact heatpipe coolers do best, either. Big tower coolers with mega fins are more resilient.
Also, disable AI tweaker(OC Tweaker) by selecting "manual" if you will undervolt because it contrarily overvolts the cpu by itself.
 

·
Registered
Joined
·
10 Posts
Discussion Starter #16
Update. Well, since I had to keep the fsb at 100, I changed things around a little. Seem to be stable at 4.3 on the cpu but I had to downclock the northbridge back to 1800 and reduce the overclock on the gfx. This is probably a good compromise setting.

I think my motherboard just doesn't want to deliver good power at any higher clocks. By itself the cpu has no problem with 4.4 although it took too much voltage to get to 4.5, fwiw. I might make one more try at a 1900 nb but I doubt its worth it.
 
1 - 16 of 16 Posts
Top