• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

***official amd 83x0 overclocking thread***

o3SriIM.jpg
http://[/IMG]

8.21 after closing all my background programs
 
This is Cinebench for the 8320/50/9590 at 5GHz:

XlFcqy4.jpg

Some other multithreaded apps do better on AMD, e.g. rendering and encoding, but Cinebench is a typical one to test multithreading.

That 5GHZ FX9 result is fishy, or the stock one is.
The FX95 is 4.7GHZ, at 5GHZ we're talking sub 10% improvement in clock, yet the 5GHZ result is scaling wayyyy higher.

Ignoring the other results, because they're irrelevant to the point.
 
There are 8 threads working together going 10% higher... since farmerboy has 8.21 at 4.8GHz, I don't know why 9.02 at 5GHz is hard to believe.
 
Such a ray of sunshine :)

The higher figure looks fine according to farmerboy's results, maybe the lower figure should be a little bit higher but it doesn't look ridiculous.
 
Meh, I'll just leave it, either way, the figures don't add up.
That 5GHZ FX9 result is fishy, or the stock one is.
The FX95 is 4.7GHZ, at 5GHZ we're talking sub 10% improvement in clock, yet the 5GHZ result is scaling wayyyy higher.

Ignoring the other results, because they're irrelevant to the point.

Would say it depends on how it is overclocked. If its only multiplier OC then i agree with you it seems fishy but if its not done solely by multiplier it becomes more likely.

Look at the picture below.
The 5.02ghz score was archived at 25x200 and 2500 HT Link Speed.
The 4.82ghz score was archived at 24x200 and 2300 HT Link Speed.
The 4.85ghz score was archived at 21x230 and 2500 HT Link Speed.

qsry1f.png


It seems Cinebench response well to a higher Bus Speed.
 
The older generation AMD CPU's responded really well to increased FSB and NB speeds. PD also benefits just not to the same degree.
It is still worth doing though.

How far you can go will depend on your MB, Sabertooths and Crosshairs which I have used can get to 300Mhz quite easily.

You have to balance the FSB clock and available multi's for the CPU, NB and RAM.

Some things you will need to establish if you want to get the most benfit from increased FSB speeds.

1: what is the MAX clock for your RAM.
Set a low CPU speed so there is no chance of it causing any errors and push your RAM clock up.
Adjust the FSB and RAM multi and keep the CPU and NB multi low whilst you do this as the test is only for the RAM.
When testing RAM clocks it is advisable to use memtest before booting into an OS as really unstable RAM can corrupt your OS.
Download memtest86+ bootable USB key version ideally.
When booted into the OS you can run a few large FFT tests in P95 to test the RAM.
I would try a Super PI 32m test first. if that fails don't bother with the P95 tests as the RAM is unstable.
You will need to increase your RAM voltage as you approach the limits of your RAM, going over 1.65v is OK but how far over depends on what you are after long term settings or bench able settings.
As you up the RAM speed you will need to slacken off the primary timings a bit.
Your standard settings at say 1600 may be 9.9.9.27 at 1866 you may have to use 9.10.10.28 for example.
Let the MB sort out the sub timings until you become more proficient, tuning the sub timings is of great benefit but will waste days of your life.

Note: If your NB speed is set too high memtest can throw up RAM errors which are actually the fault of unstable L3 cache in the NB.

2:Find your NB limits.
Keep the CPU and RAM multi's low and up the NB speed.
Test with AIDA, and P95 blend.
P95 will error very quickly with an unstable NB so no need to do any extended testing.
You will need to increase the NB voltage as you go over 2400Mhz.
Above 2700Mhz you will need a lot of NB volts, if you find you are setting greater than 1.4v to try and keep things stable I would stop.
OK for quick benches but no good for any long term stability

3: Find your MAX CPU clock.
Same drill as before keep NB and RAM multi's low to rule them out as possible causes of instability.
Using the 200Mhz base clock adjust the CPU multi up and test for stability with AIDA or similar.
Use the half multi's as well and then increase the FSB a little to find the absolute maximum CPU speed.
You will need to increase vcore as the CPU speed increases.
If you fail a test up the vcore a little until it passes.
There will come a point where increasing vcore just isn't helping any more with the CPU clocks.
You know your getting close to this wall when a minor increase in CPU speed requires a large increase in vcore.

4:Find MAX FSB
Set low multi's for CPU,NB and RAM.
You will be wanting to keep them close to stock when testing the FSB speeds so as not to have them as a possible cause of instability.
As before this is to ensure you are only Working with one variable at a time.
Up the FSB and test.
Keep upping the FSB and asjusting the other multi's as necessary to keep them within limits until you hit instability.

Now you should have found your max RAM clock, NB clock, CPU clock and FSB.
Time to play with different combinations of FSB and multi's on the CPU, NB and RAM.

You will be looking for a combination that runs your CPU, NB and RAM at close to there maximums with as high an FSB speed as you can manage.



Note: keep your HT speed down whilst testing the limits of the other clocks 2 - 2.2Ghz should be plenty.
You might want to increase the HT clock as potentially this can speed up the transfer of data over the bus to the Video Cards.
This is only really of benefit with multi card setups though as the standard HT speed is normally sufficient for single cards.
Some folks sync up the HT and NB clocks as this can increase 3d performance.
 
So fsb overclocking would be better in this case? Will try tomorrow and post results

I think a mix of both. In my case with my hardware it seems like i need to UP the voltage a bit more compared to if i just OC using multiplier. I have a feeling that there is a bit of untapped potential with the piledriver chip if you OC it right but how that is done im not sure yet. This is why i hope 8 Pack does come out with the piledriver OC guide at some point, would be lovely to have some material to gain some more knowledge from.
 
You don't get the same boost on BD/PD though.
High NB speeds are harder to achieve than on Thuban and the performance of the L3 cache is so bad it doesn't make any difference.

When running the memory tests in AIDA you can see that often the RAM has higher read/write speeds than the L3 cache. :(
With the new AIDA bench dll PD appears somewhat improved but is still quite low.

Upping the RAM speed helps a lot, not surprising considering the caches are so bad.
I see big performance improvements going from 1600 - 1866 - 2133.
More so that I do on my Intels.

Here is a radical thought
Instead of crying for daddy (8pack) why don't you do some testing yourselves

Also there are most likely good OC guides over on the ROG forums
 
Last edited:
I have tried upping ram before with epic fail my patriot viper doesnt like 1866mhz I loosened timings upped voltage and still no success I think Samsung green worth investment if ram helps pd will test a mixture of fsb and multi tonight anf let you know how I fair.

What would be good for ht mhz 2600 as standard or lower
And cpu/nb 2200 or up if poss?
 
Back
Top Bottom