Can anyone explain how a motherboard determines the default settings for a CPU in it's Bios? I was having a poke about recently and noticed my multiplier was jumping about randomly in CPU-Z. So had a look in Bios and discovered I'd overclocked my old FX8320 at some point but some settings I'd left on auto. Temps were fine and CPU was stable but I decided to turn off the auto settings and start an overclock again. After resetting bios to defaults I noticed my NB voltage was up at 1.34 which seemed kinda high as most people are saying 1.2 or there abouts. Just wondering why so high by default.
It knows your CPU settings due to the microcode that’s on the CPU. The board reads it and applies the correct profile according to spec. Auto is not suggested as there’s once again predefined parameters that they try to stick with. My memory controller voltage tries to set itself at 1.2 volts for the XMP of my memory... but it’s almost bombproof at 0.95v which makes it extremely unstable at 1.2v. It’s all about tuning things correctly.
Cheers mate, I just couldn't figure out why, but didn't know if it was something preset in the motherboard or not. Seems fine at 1.2v though so we'll see how it goes. Got the CPU multiplier up to 21 giving me a 4.2mhz overclock from 3.5mhz and despite some weird temp readings in HW monitor (which aren't replicated in AMD overdrive logging) temps look good (to me), stable and no failures yet in prime 95 or throttling. Still got some thermal margin to go, but need to replace my PSU as it's not the best.
You’ve got to keep those FX CPUs below around 62c or you’ll have issues. Use something like HWINFO as that’s better for AMD.
Yeah buddy, still well below that. HWiNFO doesn't read the temps right on Vishera, possibly down to the temperature scale AMD uses to work out temps on those chips, or maybe just a quirk with my system . I had been using CPU-Z HWMonitor to monitor temps but again it looks like it has some issues as the max temp is way high (129°C though current temp is 41°C, 40mins into prime 95 stress test) but it's not replicated in AMD overdrive (min thermal margin of 26.75°C), HWiNFO doesn't show any throttling of the core speed. I'm logging it just now and stress testing, though I do notice a min core speed of 4052mhz in HWmonitor so will look into that after this testing is finished.
Update: Prime 95 stress test was running for 2 hours or there abouts. AMD overdrive results showed a minimum thermal margin of 26.75°C (so roughly a max temp of 53.25°C assuming the limit is 70°C) HW Monitor showed a pretty steady temp of 41-42°C though min was 35°C and max at 129°C HWiNFO logged a steady overclock not showing any drop in speed on any core though HW Monitor showed a dip to 4GHz Started another stress test now with HWiNFO logging every 200ms. I have now found the temps I should be looking at, must have looking at the wrong values before so apologies for getting that wrong before.
Ok, so after another run for 1hr 40mins CPU temps in HWiNFO was a max of 45.6°C and 63°C on the socket. Speed was a solid 4214Mhz across all 8 cores logged every 200ms. Great program BTW, thanks for the suggestion! HW Monitor was showing a 0°C min and a max of 225°C so I'm calling it an error with that.
Be careful. What motherboard do you have? I'm a bit of an expert on those boards, given I fried two of them lmao. There are not many that allow you to push an 8 core hard. However, those 8320s do 4.9-5ghz quite easily, if you have the right board. I had three, and "crash out" temps depended on the CPU. One of them took a crap at 62c, whilst one was happy to clock to whatever you wanted it to, providing you kept it under 72c. Then, as if by magic, the rig would lock up no matter what the clocks. Providing you don't do anything stupid (IE keep it under 1.45v on a good cooler) you should not damage anything (providing the board is up to it... There are only about four that are). Edit again. DO NOT use Prime 95 on those CPUs. It will cook the VRM. Use something like LinX or many of the other CPU benchmarks. I used to use Asus Realbench personally speaking. The "heavy multitasking" test pretty much assured me a clock speed, if it got through it.
It's an Asus A5M97 Evo R2 I've got. Not raised the voltages at all, in fact I lowered them slightly to 1.35 and 1.22 on the northbridge from what the defaults were in Bios. There's quite a large heatsink on the VRMs so I thought it should probably be ok, and the fans are in direct alignment to blow right over that section of the board. Cooler is a silverstone Argon AR01, not the best cooler but not half bad conssidering the cost and it does seem to be doing the job, as I think I will be nearer the thermal limit on the socket rather than the CPU.
You will only get 4.3 out of that board. It's only 4+2. For the big clock on that socket you need 8 phases. YMMV though. Good thing is it can only supply about 1.35v max, no matter what you set it to, so it's safe for overclocking
Good to know, thanks mate. I'm pretty happy with getting it to 4.2, I think that's been a fair improvement in my fps in game which is good. I think that'd what I'm going to leave it at, just now anyway. Too scared my psu will die if I push it much further.
Think I'll do a couple of benchmarks tomorrow to see how it goes. Did a couple tonight but only got one useable result after remembering to start/stop during gameplay (ie after and before loading screens). I'm mostly a CS:GO player so for me that's where it really matters. Initially my benchmark on Cashe was min fps 85, max 249 average 120ish which puts it in the just match playable category, and still some benefit from my 144mhz monitor. I'm more interested in the lows, but this was real matchmaking not some youtube "yeah, 60 fps in CS:GO" but on a local server with bots. It'd be interesting to get some data on performance per map too. I'll maybe do a wee comparison with my previous cpu core speed as well...
I must have gotten really lucky with my FX chip. I didn’t realize that so many of them clocked this poorly. :/ Mine did 5.1 on air all day long and could go as high as 5.7 on chilled water. If you are really wanting more performance, you will have to consider an upgrade into 2018 tech, rather than using basically 2010 tech.
Wow, 5.1 on air, that's pretty good. Sure, I know it's old tech and I do have an upgrade list picked out but in the meantime just thought I'd explore this rabbit hole a little
It's not the chips dude it's the boards. In good boards all three of mine did 4.9-5.2. However, that was on a Crosshair with 8+2+2 phases (last two separate for RAM). 4 phase board gives you 4.1-4.3, 6 phase (like a couple of the Evo) do you 4.4-4.7 and a good 8 phase you will tap the CPU out. As I say, problem was boards. Most were designed with quad core in mind long before Bulldozer and were not updated. Hell, most didn't have any VRM cooling at all and throttled at stock ffs. And the good boards? there were literally about four. Crosshair, Gigabyte UD7, One of the Asrock (Gaming 9 IIRC, it cost over £300) and the Sabretooth.
