Crosshair V + Bulldozer Overclockers guide
In this guide I'm just going to give you the basics and some steps to follow in order to achieve a solid all round OC on your new bulldozer setup. I'm not going to delve into drive strengths, secondary subs or anything too complicated as the gains are so minimal for a daily setup but this should give you a head start and help you get off on the right foot.
CPU = AMD FX8150
RAM = GEIL EVO TWO 2000 6-9-6
GPU = ATI HD5870
PSU = Corsair TX950
Motherboard = Crosshair V Formula
Cooling = Venamous X and rather large fans.
Things you will need:
Your preferred stress test tool is obviously necessary but first a word of warning, as previously on AMD platforms Prime95 is the most stressful tool, it seems to stress the northbridge and memory more than linpack but the gap has narrowed, the difference between my max stable linx and prime 95 stable overclocks is under 100mhz with the same voltages etc and I haven't experienced any unexpected crashes on the linx clock so it comes down to just how stable YOU want it to be. After all a stability tests kind of looses it's relevance if you never crash at settings the test fails at.
Prime 95 download
A windows overclocking tool will be helpful just to save you time, ROG Connect is excellent if you have a spare rig. The only problem with it is the temps seem to be a little off, Almost 10C on the cpu in my case.
AMD OverDrive download
ROG connect download The only bad thing I have to say about this tool is that it reads cpu temps about 10°C Lower than they actually are and that you don't have access to multipliers. Other than that it's fantastic and saves a TON of time. Just use k10stat ( update any day now ) should you need access to multi's in windows.
K10 Stat ( when updated )
Monitoring software, cpu-z and hardware monitor are a must.
CPU-Z BD reviewer pack download.
Cinebench and hyper pi, I use these 2 just to find the upper limits of the system and get a ballpark figure for voltages needed and an idea on how to approach the higher clocks. It's always handy to run benchmarks like this throughout testing to make sure your performance is going in the right direction.
The first step is to familiarize yourself with the bios, while you're in there you can shut off a few things, mainly power saving and turbo but if there's anything else you aren't using (audio and lan etc) then it's not helping you so you can go ahead and disable that too. While you're at it switch Ai tuner to manual to unlock the FSB.
Another thing to take note of is the profile settings and even the GO button function.
You're likely be trying a lot of different settings, some will work, some wont. Save them all so when you clear CMOS you aren't back to square one.
Initial testing and the system limits: I like to start things off by testing the ram and northbridge whilst leaving the cpu at safe clocks so it's out of the equation and so if any problems occur, we know the culprit.
CPU-NB frequency: it doesn't scale in the same way as on Deneb and Thuban, temps and silicon quality are going to play a bigger role than voltage, on my sample I could boot the 2600 strap with stock cpu-nb voltage and couldn't move up a strap regardless of voltage. There is also a minimum ratio of 1:1 with your ram so on air this may be the limiting factor in you ram OC ( if running some godly 2400mhz+ ram ).
I usually use hyper pi to give the NB and Memory a quick workout rather than prime 95 or memtest as I'm just looking for ballpark figures at this point. I used the 1866 divider as the 2133 divider wasn't reading ram clocks in windows which I thought was important for this exercise, obviously you can do it differently.
NB voltage: Not cpu-nb voltage, helps with ram stability at higher frequencies ( 2200+ ) If you have a lower speed kit then you probably wont have to worry about it. I found 1.175v to be the sweet spot and helped me gain a few clicks, much over 1.2 and it started to go backwards so a little finesse is required here.
CPU-NB voltage: Although it didn't help cpu-nb clocks as much as I would of liked it made a noticeable difference to FSB/HTT clocks, at 267 base clock I was using up to 1.3v on the CPU-NB. Again too much voltage and your going to see backward scaling or crashes, that goes for everything, you have to test a range of voltages and not just wack everything to the max. Go up step by step.
Ram Voltage: Every kit is different and it's different on every platform, i know my kit scales up to 1.76 on Llano but it seemed to do more on a less with bulldozer and got a bit twitchy with higher dram voltages. 1.69v was used for the final round of testing which I would be happy to use 24/7. Vdimm overvolts quite a lot on my board, 1.7125 set in bios, 1.733 read via software in windows and 1.745 on the read point which is a bit nasty, just bear that in mind and keep en eye on the readings rather than what you think you set.
It is worth trying to factor in a a nice FSB OC, even tho you will probably be limited to around 270 due to the lower NB caps on these chips under ambient cooling along with the unsupported straps in the bios if using faster ram. It's still worth doing as when I got to the cpu overclocking it was difficult to pass 4.8ghz on multipliers alone, FSB was needed to get up into the 5ghz range. A good time to use memtest would be now if you want to be extra thorougher.
Memory OC: If you are a bit new to this then the ram profiles in the timings menu are a great place to start, they are tuned more for frequency and should be good for around 2400 on most half decent PSC modules with reasonable voltages, it may be worth tightening up the trfc a little tho as it runs loosey goosey at 300ns by default.
2400+ without much fuss and without going near the drive strengths is quite impressive, really liking what AMD have achieved on the memory front lately. Anyway, moving on to something a bit more realistic as that voltage was a tad too high for daily use.
CPU OC: A good starting point would be to fire up cinebench to give you an idea of your chips capabilities. Bear in mind this OC will likely be much higher than your stable OC but to give you an idea of voltages etc needed it's a good starting point.
I have no idea of safe voltages as I'm writing this before release and have been given literally no help for this but I presume the temp limit is still in the low 60's and that's where I'm trying to keep it. As with the previous generation temps play a huge role in stability so try and keep it under 60C. You will likely fail at anything close to 70°C anyway.
As I mentioned earlier it was difficult to achieve higher overclocks without touching the FSB so you have to move it off 200.
ALL TOGETHER NOW: Now we have a rough idea of what the system is capable of it's time to put it all together. This is the most challenging part, it's worth toning down memory and NB at first as it's hard to put together both a high memory OC and CPU OC. Can never have your cake and eat it eh!
Near 4.9ghz I had gone well past what I would consider safe temp wise so i went for a more aggressive fan setup on the venX and opened the windows which made all the difference.
Around 4.8 could be had just above the default turbo VID of 1.4 or just under 1.5 whilst maintaining a high OC on the ram with temps in check just hitting the 50°'s under load, that would be fine for me as a daily rig but for the purpose of this guide... We need more speed!
Dropping down the memory and NB OC a touch allows for tighter timings and also a higher cpu OC, similar to how sandy bridge requires higher vcore when you really push the memory. The lower the memory speed the less stress on the chip. For example: 4.95 with 2424mhz @ CL7 takes 1.525 volts whilst 4.96 with slack/default settings takes just 1.5. Obviously performance will suffer.
Bearing all that in mind and finding a good balance between the two I ended up with this OC.
Unfortunately I couldn't crack 5ghz on air with a balanced OC, the temps were simply too high but that's good news for water coolers out there, I believe the gains we'll see from dropping those few degrees is going to be much higher than on phenom II.
Crashes: Crashes are a little less clear cut than on phenom II where you knew exactly what to blame. I haven't had a single hard crash down where you are required to cut power to get a boot out of it so down to OCP so DO NOT change the current protection settings in the bios, they're dangerous and there's no need to on ambient cooling.
I have been getting a lot of machine check errors and the likes, it seems most every bsod without a memory dump or error code seems to trace back to the cpu ( too little or too much voltage ) where as memory dumps trace back to the NB or memory. Lock ups are usually down to NB/MEM again.
After a lock up your reset button wont work a lot of the time (9905 bios) so just hold down the start button and wait for it to shut off then you're good to go again. Sometimes Vcore won't apply after a crash causing another crash under load ( usually failing to get to windows ) so best to either check vcore or power down completely if you come across that.
Tweaking: Now you have a solid OC you can start to tweak it to get those few extra frames or just for the fun of it, raising the FSB one notch at a time may be a good way to go, tightening timings, whatever, just have fun with it. This board really is a dream to use, it doesn't get much easier. If you can use rog connect then go for it, if you have a rounding error during your stability testing it allows you to fine tune voltages on the fly and try again straight away without having some heavy software running. That also mean you can drop clocks down and then back up during benchmarks to give you an edge, say you can pass vantage gpu tests with 5.3 on the cpu but need to dial it down to 5ghz for the cpu tests, you can do it easily without macro keys and awkward programs hogging resources.
After I was confident it was stable at the above settings with the memory OC tied in I tried pushing it a little further. I really should try and get the trcd down.
A tad move cpu voltage was needed and this OC really was riding on the 60°C limit.
Of course there is no set way to go about this, if you want to do it backwards then go ahead, this is just a guide and the way I like to go about things. After all this is YOUR OC. The screenshots I've posted and voltages I've mentioned may be irrelevant when it come to your hardware, ceilings will be different for better or for worse but it's the theory behind it I want you to take note of, add a little methodology to the testing and take it in steps or you will be there for days. *Has spent days messing around and getting nowhere on this platform* It all comes together when you take it in steps.
Thanks for reading and I hope it helps.
Edited by el gappo - 1/18/12 at 8:55am