|
Anyways for those that are interested, this is currently the settings I am using in Bios, and right now currently just tinkering my FSB, Multiplier, CPU offset, and CPU_VID. I really like testing by lowering multipliers on everything, raising the FSB a huge leap, and then lowering below stock the CPU multiplier so the net result is a CPU clock gain, and to make sure no matter how high I set my FSB, my other stuff won't be an issue (due to setting their multi's so low). Apparently a lot of stuff is locked, so on my board things like the NB Multiplier is actually just me selecting my NB speed, locked up to 2000mhz, in which case if the FSB is anything but stock, it will result in a tinkered speed. Anyways heres what I'm doing now, I know there isn't a single guide that actually tells you how to mess with your BIos just what your 'supposd' to do so it was kind of confusing at first for me. Also nice to write everything down, if there's soemthing you do just on stock settings, and you need to do a CMOS reset, good to remember what you always do as well as your overclock settings.
also fuond my motherboards limit for FSB is somewhere between 260-270. At 270 it will say corrupt CMOS and I need to hard reset.
MAIN: Floppy Disabled (won't boot OS otherwise, huge headache when first built pc!) Advanced-CPU Config: Powernow Disabled (Biostar equivalent of CoolnQuiet, I'm going to turn it back on once done overclocking. just makes things clearer when oc'ing) ACC - Per Core - 0%/0%/0%/0%. 0% on all cores, had no issues with each core in prime95, failures are usually not worker fails but crashes/resets. ACC on because no point in overclocking on 3 cores when I know 4 cores is better and the oc difference, if it exists, isn't worth it. I know my 4th core is perfectly fine though, having same OC results. Hardware Health - Shutdown 60*C BOOT DEVICE PRIORITY: 1st SATA WDC (hd), 2nd disabled (no floppy!) Have to make sure you can boot into windows silly PERFORMANCE: CPU Over Voltage +.2 (this isn't an accurate overvoltage actually, its more like .075. Also, this is a motherboard overvolt, so generally my software like cpu-z may not correctly report my voltages because the V_drop increases - difference between reported bios and software voltage. Since my Athlon II is voltage locked as well as multiplier locked, I need to use the motherboard to overvolt) CPU FID/VID Control: Core VID: 1.4 Maxed. With my offset, I'm actually overvolting quite a bit. CORE FID: 13.5 (below stock because FSB is high. I actually think unlocked multi oc'ing is stupid, you have to raise your FSB to get the most out of an overclock anyways so you'll probably be below stock OC) NB_FID: 1400mhz. (I believe this translate into a 6 or 7 or 8 or something multiplier, because my system reports a NB of 1750 right now. Kept this below stock while I oc my CPU. NB_VID: stock. I know nb may affect stability, but thats also why I underclocked it so much so it demands less voltage
HT Config: 1.2ghz/16bit
DRAM: "Limit" DDR3-1066 (translate to about a little more than 700mhz dram) Timing Mode: "Both" 9/1t/9/9/auto/24/33/auto/4/autox5 done twice for each stick. Used stock timings, and then 'auto' for anything that wasn't displayed in bios settings for the RAM. definately affects stability, when I ran timings on auto I had a crash because of it on a stable OC.
cheers
|
So before I was using Cinebench and if it ran, I would deem stable, and then run more tests, and then see where the cinebench limit was, then use p95 for long time. Then, I realized that just running p95 for a minute (let alone launching the application) was a much better way to do a quick stress test. A lot of results I have to throw out, because I didn't realize how important voltage is to overclocking, which is made more confusing by both CPU_VID and Motherboard CPU Over Voltage offset, which is pretty much the basis of overvolting on a locked CPU. Thus, i usually do a combination of lowering CPU VID and raising offset.
I find that even at 1.4 + .02 (which the offset is a bit inaccurate, is slightly less than .01 per +.01) I can't reach 3.6. I ran 3.51 on about 9 hours stable p95, but didn't have voltage written down correctly. Hopefully when I crashed at 1.35+.01, it was because the voltage was diffferent just now. So I'm running 1.4+.01 voltage, at 3.510, p95 right now. I know you shouldn't use programs during p95 testing, but it's better to run a test and browse the internet than no test at all. Once I'm done browsing I'll let p95 run for a long time. Seems to be the max stability I get.
Looking over some results, I don't think i can even reach 3.35 without overvolting. It also seems about +1, or 1.47, is the best voltage to be at, give or take .01 voltage. It's hard to push even 200 mhz on this processor apparently. Temperatures are more than okay though, the problem is stability, as I can't get any stability past 3.5ghz with more than 1.5v, a very high voltage as is. 1.55v is apparently the max I should go.
Which is fine I suppose, 200 mhz should be a noticeable increase. I wish I could get more, but oh well. Unlocking a 4th core is the real beauty in this chip, although I guess I didn't really need to get an aftermarket HSF and 7 fans after all. I guess it'll help to keep things healthy, and maybe when I start overclocking my CPU-NB tommorow, I may get better results. I'll post about that when I start doing that, here.
|
Ah and so I stumble on to a problem: http://www.overclock.net/amd-cpus/943109-why-vrms-big-issue-why-usually.htmlhttp://www.overclock.net/amd-motherboards/946407-amd-motherboard-power-phase-list.htmlbasically, my motherboard has a basic VRM/mofset, meaning I really shouldnt push my TDP on it. I will probably just stick to 3.51ghz - at 1.35+.01=1.44v it doesnt work but at 1.4+.1=1.47 it does. I'm already pushing my motherboard, and I can feel the VRM is about 40-50*C (very hot but just slightly not hot to make me take finger off back of mobo where mofset is). I only have 125w rated board with 3+1 (not even 4+1) with 4 pin CPU power supply, so I should avoid the higher voltages. Given that 1.55+ can't even get me 3.6ghz, I'll just stick with 3.5 at 1.44 v, and maybe even just going a bit slower depending on the voltage needed. I think I'll probably end up settling with a lower voltage around 3.4. But I have not had any issues whatsoever with heat though. Very low temps with hyper 212 and 7 fans. Liek i said, I can't accurately read them (below ambient idle is impossible), they are below what load was on stock cooling stock speed locked core. The problem with my tower heatsink is that it also isnt cooling my northbridge or mofset like my stock hsf was.
Anyways, I plan to take a hacksaw to my stock CPU cooler, make a half and half mix of elmers glue and the thermal paste that came with my hyper 212, and glue a mofset heatsink onto my mofset chipsets. Smart? No. Better than nothing? Gotta be. Totally awesome? Yep.
|
here's a picture of my northbridge fan, as when I got my tower cooler, motherboard components by the Cpu were no longer getting cooled. At the moment, I have 3.510 ghz at 1.456v. I won't run p95 again (i did 8 hours before, but can't do 5 minutes on lower voltage) until I mod some heatsinks onto my Voltage Regulator module. I've learned alot about VRMs, and basically my motherboard has a 3+1 as well as NIKOS mosfets, so that's not really good for overvolting, especially since unlocking goes from 95w to 125w, and then I'm pushing 140w with my overvolt on top. I definately need heatsinks (really i need a better motherboard with better VRM with more phases or dedicated mosfet drivers at least) to continue, so just waiting for the mail - I asked how to make homemade thermal epoxy at overclock.net and someone was nice enough to mail me (erm, assuming its in the mail lol) some thermal tape, so i dont have to use a mix of elmers glue + coolermaster TIM LOL
Here's pics of my northbridge DIY cooling solution.No copper wire is touching the mobo, although it looks like it. It's pretty impossible for it to, as its threaded through the fan holes, as well as the electric tape I put on just in case.
|
So while I wait for my fan controller and thermal tape in the mail (sent by a nice, generous person at Overclock.net who was horrified that I would use elmers glue and TIM to put makeshift heatsinks on my VRMs lol), I am going to work on overclocking my RAM now, as I'd rather not push the limits of my CPU/CPUNB until I have some proper cooling heatsinks on my VRM (read: no more prime95 until I get heatsinks).
So, since I hear AMD likes tighter timings without a doubt, and from what I've read tighter timings at lower frequency is always way better, I figured I'd keep my RAM at stock speed (at first), tighten my timings by 1 step for all of them each until I get problems, and then start raising the speed.
Well, there's a lot of mystery timings in my Bios, so I just alter the ones that show up in BIOS as a certain value (at the bottom of Bios ram timings page, it shows my CS, Cas to Ras, the basic timings) and lower them by one, leaving the rest as Auto (This covers my Cas Latency, ras precharge, cas to ras, cycle times, and command rate). Well, I tried 6/6/6/21/30/1T/ and when applying changes, screen wouldnt boot at all. After a hard reset, the system would show everything reverted to stock and a bad checksum - I tried it by upping the vdimm voltage from 1.6 to 1.65, no luck.
So it looks like my timings are going to be 7/7/7/22/31/1T, and from here I'll change the speed of my RAM. I would go crazy trying to alter each individual timing on its own (ie a mix of 7 and 6) so I'm just going to up the speed now. I could drop my speed below stock, but well, that would just be way too much to deal with for nothing anyways. Timings are the biggest thing, and having dropped them from my stock 9 is a big enough deal.
For stability testing, I just run HyperPi@32m. I'm pretty confused on how exactly to test RAM stability and benchmarking, so I just hope I'm increasing system performance lol. I haer memtest86 isn't thorough, and HCImemtest, I dont understand quite exactly how to figure out how to allocate my RAM for multiple instances in a thorough manner, and it also isn't good for stability testing.
I hear prime95 blend may be the best stability testing, but right now I'm just doing 'quick' stability tests that can be done in under 25 minutes as I tinker my settings to the exact right spot, and then i will do a thorough test.
|
In my bios there is "Memclock Value" or something, and I can choose "Auto, Limit, Manual". Limit says something like it won't exceed a certain value, and Manual says it will be this value regardless - the value being my choice in an option right below of DDR3-800, 1066, 1333, or 1600 (note my RAM is Kingston HyperX Blu DDR3 1333mhz CL9 2gb x 2). I'm not really sure of the choices, but I do know that I usually pick "Limit" and it seems to have a certain affect with FSB - ie 200FSB plus "Limit" of 1333 means 667x2=1333 RAM speed, 260FSB+ 1066 = 693x2=1386mhz.
So when I choose "Manual" things never work, but whatever, because its all about speed and CPU-Z is showing me my speed changes. So I stick with Limit.
I choose 1333mhz, and lower the multiplier to everything (CPU, NB, HT) to stock or less. Then, I choose 210FSB - I get 700mhzx2=1400mhz RAM speed. Seems to work. When I choose 220FSB, which I have no idea what the equation is, and therefore what speed is equals, the PC won't boot right at all (usually reset takes long and then everything has reverted to defaults in bios and checksum is bad) - it lets me know pretty quickly it doesn't work. So obviously over around 700 mhz is bad. Maybe its a lot over it, maybe a little, I just know 210=700 works and 220=??? doesn't, which is the next step up. I also tried raising voltage from 1.6 to 1.65, no luck.
My point being, is that I won't overclock the speed lol. My current CPU overclock settings are 260x13.5, which, with DDR3-1066 RAM 'Limit', is 693mhz DRAM speed - pretty darn close to 700mhz. The fact I tightened by timings, each one, by 2, is pretty cool, and would be more than I could ever gain than from overclocking speed, as apparently AMD systems love tighter timings.
On another note, I just received in the mail, the thermal tape from the friendly donor at overclock.net, so I will be sawing my stock CPU HSF into tiny pieces, taping them to my VRM, and then be happy and post pics!
|
I'm surprised there aren't more OC SC2ers willing to share some info.
|
So I think I may have a few stability issues? Every once in a while, while logging on to windows with the circle, the system will freeze. Last night I did run a 12 hour prime95, but I have no clue if I did it correctly as the system went to sleep, so it's possible I simply ran like 30 minutes worth (yea im dumb). And today I had the first crash on SC2 (even though crysis 2 demo ran fine) -a BSOD "memory management 0x0000001a" error - although, after downloading BlueScreenView, it's the only BSOD that reads as memory management issue.
My guess is either memory timings, or that GPU overclock (although just to be ridiculous about 10 minutes ago, I tried an OC of 750 to 900 and had no stability test issues on MSI Kombustor utility). Since I've had BSOD's and crashes from memory timings (come to think of it, i think most were boot/checksum issues not BSOD) and none of them had memory management.... I dunno what happened.
Anyways, I think my GPU is limiting my system. I noticed that leaving my GPU at stock settings, and then going for a crazy 900mhz, 1950mhz, insane overclock on the GPU that borders on being stupid and dangerous to do, there isn't a single fps increase in my GPU tests - it's still an average of 25fps. I'm thinking there's no reason to overclock my GPU, as my 3.5ghz X4 Athlon II is 'bottlenecking' it.
:O
|
heres me applying thermal tape sent by a friend at overclock.net, theonedub!
Here's where I sawed an AM2 stock heatsink (i was going to use my stock am3 from my athlon ii but I traded heatsinks with roommate with the pc he's borrowing lol). Awesome job!
Slapped on the heatsink
Finished! Note the stock fan jerryrigged with copper wire (wrapped with electric tape so it won't make a shocking connection anywhere important) on my northbridge that was necessary when I went from radial stock hsf to tower cooling.
i had some really hard times with the temp sensors, in the end, I put the sensors from my new NZXT Sentry 2 fan controller I installed today and put 1 on the side of the heatsink with thermal tape, and one simply stuck inside the fins.
The back of my motherboard still feels hot from them, but I hope this will go a long way in improving the temps on my mosfets!
|
So i will begin long stability tests now! I think my setup is stable but not 100% stable, as my PC has crashed once or twice - once during sc2 during stupid gpu overclock and one time on prime95 test about an hour in. I am thinking it's the RAM as thats the only thing that's really changed - that or my northbridge.
I think my CPU bottlenecks my GPU, so there's no point in overclocking my GPU. I think my next step is 24 hour testing with my current setup, see if its okay, if those were freak crashes, or if I should drop my timings. I seem stable enough, but maybe not 100% stable. Afterwards, I don't think I really need to test because everything seems stable enough now, so my RAM timings loosened by a step should be fine (bringing it to CL8/8/8/23/32 instead of CL7/7/7/22/31). If loosening it required, I will try to see what kind of speed increase I can get by first increasing the multiplier, and if that doesn't work, I'll lower my FSB, raise my ram multiplier, and then achieve my 3.5 overclock with a higher multi/lower fsb as opposed to higher fsb/lower multi as I'm doing now.
Then, I would move on to my northbridge. Just a simple "does x9 work? does x10 work?" and see what I can get, and I'll be done!
|
|
|
|