Well for one, the Radeon HD 5850 that you have is actually slightly faster than my 6850.
The numbering by AMD makes things confusing, but the goal of the 6850 was to have performance on par with the 5xxx series, while using less power and making less heat.
Look @ this http://www.hwcompare.com/5916/radeon-hd-5850-vs-radeon-hd-6850/
"Texel Rate: Texel rate is the maximum number of texture map elements (texels) that are processed in one second. This figure is calculated by multiplying the total texture units by the core clock speed of the chip. The better the texel rate, the better the card will be at texture filtering (anisotropic filtering - AF). It is measured in millions of texels applied in a second."
Your card beats mine quite a bit in Texel Rate - Which makes it better at AF than mine. I'm assuming 3d Mark weighs HEAVILY on your GPU - which would likely explain the difference in our scores (Your card is considered 6% "better" than mine as far as benchmarks go. I bought my card with the intention of running it in crossfire, due to its low power consumption, as soon as I pick up a beefier power supply shortly after Christmas.
As far as overclocking your CPU goes - all of them are different, and therefore will overclock differently... I would suspect, however, that your chip should be able to run at similar numbers to mine.
It seems to me, based on your post, that you are kind of willy-nilly trying different numbers in your BIOS to see what your processor will run at, but correct me if I'm wrong.
If you go back a page, to post #25 by "PioneerIsLoud" - You will see a very nice, short and sweet post on how to OC a phenom - and in what order....
The most important thing is to make sure that you have the patience to sit down for probably a good 6 hours while you OC, and record important numbers. Also make sure Cool n' Quiet is disabled in the BIOS.
I began by overclocking my FSB, and my FSB alone, to see how high I could get it. This is with everything else being kept at, or below stock frequency while at stock voltages. I started at 200, and worked my way up by 5's. I just so happened to know that my board was fine with anything up to 240 - so I cheated and started here - you might not want to cheat here if you don't know this for a fact but I would say your board should hit 240FSB on stock NB without any trouble. Each time I raised the FSB, I booted into Windows and checked all of my frequencies with CPU-Z to be sure I was running everything at stock or below, and only the component I was overclocking was above stock. If this was true, I went on to run Prime 95 for about 5 minutes. If prime ran with no errors for five minutes, I rebooted and raised the FSB by another 5. Be sure that each time you change something, you record it on paper so you can go back and see trends.
Raising the NB volts may increase FSB stability at high clocks:
My FSB testing looked like so:
240 = OK @ 1.15V NB (NOT CPU/NB)
250 = OK @1.15V NB
And so on until I got to...
290 = Errors in P95 / No Boot @ 1.2V NB
290 = Errors in P95 / No Boot @ 1.225V NB
290 = Errors in P95 @ 1.2375, 1.24375, and 1.25
So I knew that my MAX FSB came out to 285MHz at 1.15V and anything under that would be just fine.
Moving onto the CPU OC.
Start the BIOS, but the FSB back at 200, and run everything at stock speeds or below.
Your multiplier will be 15 at stock settings.
Try raising the multi to whatever brings you to 3.4 GHz - which I believe is the Turbo Core speed for the 960T's.
Soo 200FSB*17xMulti=3400MHz CPU Frequency. Try booting at around 1.3V here - mine booted at 4000MHz using 1.3 V but we want to start at lower frequency, and work our way up to see at what point does the CPU require move voltage to be stable. When you boot, run Prime95 for ten minutes, if it's okay, reboot and raise the Multi by .5 until you get errors in P95.
By doing the above ^^ - we can get our max possible CPU frequency using the lowest possible voltages available to us in our BIOS. Mine allows me to go up in very tiny amounts, which most boards will. If you find that around 4+ GHz you need more voltage to run P95 successfully - you will not want to just jump up the voltage to some random, higher number... just try giving it the smallest amount of voltage bump necessary to get it stable!
It's important to remember that Phenom's prefer lower temps to higher voltages - which lead to higher temps.
is an excellent little read surrounding the ideas I just talked about.
My CPU testing went like this:
CPU-NB (Shouldn't need to adjust for CPU OC, only for NB/RAM OC) 1.25V
*Black screen on login!
*Worker 4 Error @ 5 MINS
*Worker 4 error @ 2 MINS
*Errors in P95
*Rebooted during testing, no good.
I stopped here BUT - I have plenty of voltage to play with here, and some headroom temps wise to raise this higher. I'm sure I can run the processor faster, with more voltage. The vCore max on this processor is 1.55 - But I wouldn't go over 1.5 on AIR more than likely. I will try going higher but I stopped here while overclocking because my goal was 4 GHz for a 24/7 OC.
Keeping close track of all these numbers makes finding your final OC very simple and almost stress free. Keep track of the lowest possible voltages you can work with by working your way up from a relatively low starting point - remember to make the SMALLEST bump in voltage possible by the BIOS to get the CPU stable.
Moving onto the NB.
Nice little guide:
Raising the NB frequency of the MOBO will sometimes require more CPU/NB but NOT more Mobo NB voltage. Higher RAM clocks also benefit from this, as well as higher RAM voltages at times.
Do not raise the CPU/NB above 1.45 on this processor unless you want to fry your IMC (Integrated Memory Controller)
Again, as with the other tests, set everything at or below stock speeds, and at normal voltages.
Each time we boot, we will run Prime95 for about 5-10 minutes (longer is better) to make sure we are somewhat close to stable with each OC.
Set your NB to 2200 (not the HT, keep HT between 1800 and 2100 or you can see a decrease in performance at anything above 2100!)
With the NB at 2200, boot, run p95 and reboot if no errors.
Try 2400, and so on. Aim for 2800+ I'm at 3000 and going to aim for 3200 but anything over 3000 is a fairly lofty goal I think I esp on AIR.
When you begin experiencing issues, raise the CPU-NB some.
My testing looked like so:
Blue screen @ 1 min in P95
Blue screen @ 5 min in P95
As you can see, the higher you OC, the longer you might want to run P95 - I ran for 5 mins at lower clocks after each restart, and 10 mins when I got higher, just to be sure I was ok.
I could raise my CPU/NB higher to see if I could get 3200 NB stable, but I stopped at 3000 with 1.25V CPU/NB for now.
*NOTE* - When I got to putting my final OC together - after a few failed runs of P95 with frequencies and voltages that should have, in theory, been fine according to earlier testing - I was not stable. I was getting blue screens which usually means memory/IMC related issues. Since I was able to narrow it down to IMC due to blue screen, I tried giving the processor a CPU-NB bump from the 1.25 volts that she ran fine at before, to 1.28125V - The reason I had to do this even though it ran fine previously at lower voltage, is because once I was OCing the NB, as well as the memory at the same time for the final overall overclock, the IMC needed a tiny bit more juice to make everything work together.
It's important to keep the above in mind because yes the NB ran fine at 3000MHz with 1.25 V but when you but load on it with the memory OC too, it may need some more voltage to pass 4 hours of final stress / blend testing in P95.
You can choose to OC your RAM, but if it's a pain you could just run it as close to stock clocks as you can get - even running your RAM slightly below stock, with an OC to the CPU and NB is an acceptable tradeoff.
My ram kit runs 1600MHz stock 1.5V with timings of 9-9-9-24.
I don't know much about overclocking RAM and changing timings.
What I did:
Set everything to stock speeds or below.
Adjusted the FSB to 205 (keeping everything else below stock) and ran the RAM just slightly above its stock 1600 MHz.
If it passed 5 mins of P95 blend tests, I rebooted and checked the memory with the windows memory diagnostics.
I kept going up until I was unstable at 1666 MHz. I gave it a voltage bump to 1.55 volts, and was able to get it stable up to 1730MHz or so - and hit a wall here at stock timings.
I knew anything below 1730MHz at 1.55V should be ok.
Now I made a cheat sheet.
I wrote my high OC's down on one sheet, and recorded at what voltages everything "worked" at.
I could now see:
FSB = 285 @ 1.15 V
So anything under 285 is ok.
CPU 3900 @ 1.3 V
CPU 4000 @ 1.34375 V
CPU 4100 @ 1.3625 V
So anything under 4100 is ok.
NB 2800 @ 1.25 V
NB 3000 @ 1.25 V
So anything under 3000 is ok.
RAM 1600 @ 1.5 V @ 9-9-9-24
RAM 1666 @ 1.55V @ 9-9-9-24
RAM 1733 @ 1.55V @ 9-9-9-24
So anything under 1733 is ok.
So from here I can pick numbers that look good to me.
For a 24/7 OC, and to be sure we don't go totally INSANE trying to get her stable, I recommend selecting a step DOWN from your max OC numbers, but this is totally up to you.
I picked FSB @ 250 because that allowed me to put my memory at 1666, while allowing me to set my NB at 3000 on the dot, HT at 2000 on the dot, and CPU at 4000 on the dot with a multi bump to 16x up from 15x.
You can fine tune your numbers using the FSB, in increments as small as 1 MHz
Also, remember this when assembling your final OC:
1. Some components, esp the CPU/NB may need slightly higher voltages with your final OC than they did when testing each component individually due to a higher load being placed on components once everything is OC'd at once.
2. This is from one of the links above, and is helpful during OCing to assess what could be the cause of any instability you experience along the way while OCing:
“Im sure some of you may have experienced a crash with cinebench………sometime you will blue screen, somtimes you will just black screen and sometimes the bench will just crash ( dissapear, etc just shut down ) and windows will still be up………..the blue screen is NB vid/IMC memory related, the black screen is core clocks/cpu voltage related and the just crash/dissapear from desktop is temp related………….”
The above quote mentions cinebench, but it applies the same way to just about any program you may be using, such as P95. Read the above quote and come back to it as necessary!
Hope this helps / is what you needed. I'm sure it will help somebody either way!
Let me know if you need more help / info / etc.