FX Overclocking Multi vs FSB

[Priller]

I had heard various things about overclocking the FX series of processors. Now that depends on who you ask but it is often said to OC with the FSB instead of the multi.

 

So I set out to find out for myself if the FSB was truly the best way to OC FX processors.

 

 

The first step was trying to find a frequency that would allow me to balance the multi, FSB, and ram speed without going too far from each other to truly see what the difference was. I quite accidentally stumbled upon that core speed at or around 4.8Ghz.

 

 

 

 

Every test was carried out the same way and many of them ran 4 times.

 

Let's take a look at the first line for 4.8Ghz. You will notice that it is using just the multi and scores 754.

When you compare that to the line underneath it you will notice that only the HT Link & NB Freq. Multi's have been changes to x12. This yields a score of 760.

If you compare that to the next line you will notice that the ram speed has gone up slightly but all of the overclocking is done via the FSB at 240. Even with the massive increase in the FSB the Single Core score for Cinebench remains the same 

 

 

If we look at all of the lines that have the CPU freq. at 4.80-4.83 you can see something curious.

-Increasing the HT Link & NB Freq improves the score at the same core speed.

-Using just the FSB to OC does not increase the single core performance versus the multi.

-With a lower HT Link, NB Freq, ram speed, and FSB the run done at 4.83Ghz spits out both a higher Multi and Single Core performance than the run using just FSB to OC while using a lower FSB speed.

 

 

 

 

I included some runs done just to see the scaling at different core speeds. 

 

 

 

[samuel]

Based on your graph, there is no reason to have more than 2400mhz on the NB. Actually even with 2200 we will see almost no difference in performance. I'd just undervolt the Nb after finding the best CPU clocks for lower thermal output.

TBH, I expected some example of difference in voltage required in FSB vs multi at the same clocks,.

[Priller]

Based on your graph, there is no reason to have more than 2400mhz on the NB. Actually even with 2200 we will see almost no difference in performance. I'd just undervolt the Nb after finding the best CPU clocks for lower thermal output.

TBH, I expected some example of difference in voltage required in FSB vs multi at the same clocks,.

Actually there is a good step forward when the NB freq. & HT link multi. is set to 12. That seems to work up until around 2700-2800 on the NB on fairly normal voltages and to exceed that requires a hefty core speed. 

 

This set of tests was just based on shear performance so all voltages were set at the beginning and remained the same throughout as to not add any additional variables. 

 

Voltage testing will be carried out as well but that takes a while to perform and I have to find a good tipping point that would work one way and not another. That will be coming as well.

[Jumper118]

I had heard various things about overclocking the FX series of processors. Now that depends on who you ask but it is often said to OC with the FSB instead of the multi.

 

So I set out to find out for myself if the FSB was truly the best way to OC FX processors.

 

 

The first step was trying to find a frequency that would allow me to balance the multi, FSB, and ram speed without going too far from each other to truly see what the difference was. I quite accidentally stumbled upon that core speed at or around 4.8Ghz.

 

 

 

 

Every test was carried out the same way and many of them ran 4 times.

 

Let's take a look at the first line for 4.8Ghz. You will notice that it is using just the multi and scores 754.

When you compare that to the line underneath it you will notice that only the HT Link & NB Freq. Multi's have been changes to x12. This yields a score of 760.

If you compare that to the next line you will notice that the ram speed has gone up slightly but all of the overclocking is done via the FSB at 240. Even with the massive increase in the FSB the Single Core score for Cinebench remains the same 

 

 

If we look at all of the lines that have the CPU freq. at 4.80-4.83 you can see something curious.

-Increasing the HT Link & NB Freq improves the score at the same core speed.

-Using just the FSB to OC does not increase the single core performance versus the multi.

-With a lower HT Link, NB Freq, ram speed, and FSB the run done at 4.83Ghz spits out both a higher Multi and Single Core performance than the run using just FSB to OC while using a lower FSB speed.

 

 

 

 

I included some runs done just to see the scaling at different core speeds. 

this is a pretty good set of results.it seems the bulk of the extra point come from the clock speed but the ram and northbridge do give a decent boost even when the core is at the same speed.overclocking on the FSB is clearly better though. you could do a test of the ram speed vs the northbridge to find out at what speeds the ram gets bottle necked by the northbridge and the other way around. try start at really low ram speed and increase the northbridge speed from around 2000-2800 then try at a a few ram speed up to your highest with all the northbridge frequencies.

[Priller]

this is a pretty good set of results.it seems the bulk of the extra point come from the clock speed but the ram and northbridge do give a decent boost even when the core is at the same speed.overclocking on the FSB is clearly better though. you could do a test of the ram speed vs the northbridge to find out at what speeds the ram gets bottle necked by the northbridge and the other way around. try start at really low ram speed and increase the northbridge speed from around 2000-2800 then try at a a few ram speed up to your highest with all the northbridge frequencies.

 

I guess Ill add that to my list...

 

I think I hear my 8350 crying now. haha

[Jumper118]

I guess Ill add that to my list...

 

I think I hear my 8350 crying now. haha

i think the motherboard will be crying as well with those northbridge frequencies. if you have lots of spare time you could run super pi as well as cinebench so that your testing 2 benchmarks. super pi is very depended on ram speed and timings as well so the results will be interesting. one is more computing numbers and the other test is rendering so you'll be covering 2 key areas of cpu useage.

[Priller]

i think the motherboard will be crying as well with those northbridge frequencies. if you have lots of spare time you could run super pi as well as cinebench so that your testing 2 benchmarks. super pi is very depended on ram speed and timings as well so the results will be interesting. one is more computing numbers and the other test is rendering so you'll be covering 2 key areas of cpu useage.

 

I guess we will see. I will try some baselines and see what I come up with.

[Faa]

This is a better example:



Overclocking with the FSB gives a massive performance difference, same goes with Intel but theyre quickly unstable after 2-3MHz.

[Priller]

This is a better example:

Overclocking with the FSB gives a massive performance difference, same goes with Intel but theyre quickly unstable after 2-3MHz.

 

You can tell that the graph is older because it uses Cinebench R11.5 scoring systems (X.XX). Cinebench R15 spits out 3 digit numbers (XXX)

 

Also I would write that test off as the variables are too great. The ram speed it self varies by 235mhz, when compared to my test which only vary by 82mhz.

 

I will continue to do more tests to really see the difference.

[Faa]

You can tell that the graph is older because it uses Cinebench R11.5 scoring systems (X.XX). Cinebench R15 spits out 3 digit numbers (XXX)

 

Also I would write that test off as the variables are too great. The ram speed it self varies by 235mhz, when compared to my test which only vary by 82mhz.

 

I will continue to do more tests to really see the difference.

Yes but the scoring systems make much more sense in Cinebench 11.5 in my opinion. 

[Priller]

Yes but the scoring systems make much more sense in Cinebench 11.5 in my opinion. 

 

In Cinebench R11.5 an FX prossor has 77% the performance of haswell while in R15 it has 88%. Meaning the instruction set has been updated showing more of the true performance of the FXs.