I just read that the rule of thumb for a cpu Vcore is not too add more than .25V to the normal Vcore, as long as temps are fine, of course. In my case, 1.65 + 0.25 is 1.9 . What do you guys think ? Do you think the temps are what matters here ? Or do you think that no matter what the temps are, too much is just too much ? I tend to think that sometimes, too much is just too much, but I'd like to know, since I could go a bit higher (not up to 1.9 though, but probably 1.825 as my external die temp is@50C after 24 hours of Prime95 or folding. ABSII (internal temp) is at@60C. The max for the 2500+ is 85C (that's the internal temp, of course) 200*10 sucks right now, I'd like to go higher but not at the price of a new cpu in 2 months. I want it to last a good 6 months, then I'll buy another of the same, probably, unless a new overclocking king (that sounds weird...) is born.
I definatly wouldn't wanna increase my Vcore by more than .25v. My temps dont really increase that much with a Vcore increase.... All cpus are different, Ive seen people overclock their 2ghz P4a to over 3ghz with 1.55v! Standard is 1.5v But I need 1.65v just to get to 2.5ghz. I wouldnt just go by the 'rule of thumb' cos if this guy only needs a .05v increase to overclock that much an extra .2v could possibly fry it. I suppose for a first time or inexperianced overclocker running by the 'no more than .25v' thing is an indicator of the extreme. When I first started overclocking I used to ask how far I should increase the fsb?, I get told ......'it depends'. How high should I increase the Vcore?...... 'not much'! People want a quantitive answer but a proper answer cant give that. It does depend on everything. I think if you know computers you should just trust your own judgement. Every pc ive ever touched is so different from the last, you get to know your own a hell of a lot more than a generalised 'rule of thumb' built up over thousands of comps. As for temps..... an increase in V doesnt always mean an increase in temp. A .2v increase will raise my cpu by maybe 5c. Temps should always be watched, but if its cool it doesnt mean you can overclock more or increase Vcore more. If it did a Vapochill would mean infinite overclocks! Theoretically like Just try and keep em as low as possibe.... or untill your comps stabilty drops. You may find that your P4 is unstable at 55c... other may lose stabilty at 80c! If you've ever heard of a computer or mathematical model..... one would be perfect for applying to overclocking. Graphically you might have a few bars on a chart, one for say temps, vcore, fsb, mem voltage, mem speed etc etc Raise the fsb and see which other bars increase... sounds dead simple but the maths and programing which would analyze your system cos every comp, cpu, ram chips etc are so different, would be proper hard stuff..... more than any of these benchmarking tools you get.
That's actually the kind of answer I was expecting I know computers enough to solve my own problems but I'm very new to overclocking so I don't quite know what I'm doing, but I know enough not to expect to oc as good as one with the same mobo/cpu/ram. What I'll do is increase it to 1.825, see if it makes a difference, if it doesn't or if it's not big enough (what I expect) then I'll back off, if it makes a good difference (let's say another 200mhz) then I'll keep it that way. I wanna' keep my ABSII under 65C at all times, I don'T wanna' go beyond that. I'll keep you posted.
If your kinda new to it all, the way id go at it is to try and find the extremes for each aspect of the overclock. Try and work out how far the cpu will go with each Vcore increment, find out what the highest stable temp is for your cpu, how far your ram will go etc. Its alot easier with an AMD cpu cos you have the multiplier with it. Id love to have an unlocked multiplier. Ive no idea how most amd overclockers go about it but I would just get the fastest ram under the sun and overclock the fsb as much as possible. If you have good pci and mem dividers, even better. I wouldn't worry too much about the cpu speed, if the benchmarks in sandra 2003 pro are anything to go by they perform tons better per mhz against the p4. They have some kind of hyper threading though dont they? Overclock the cpu a bit just for good measures though That fsb and mem increase will do your 3d mark 2001 score a healthy increase too.
You need to fill and connect the L1 bridges if you have a Palimino, or a motherboard that doesn't unlock the Thoroughbred/Barton (and probably the upcoming Thorton---AthlonFX?). Only P4's and Xeon's come with Hyperthreading---and this applies to the newer ones. Sandra is not a very precise benchmark, so take the results with a grain of salt. This isn't to say it has no relevence, but it's nothing more than a package to roughly measure your performance. With the older 1.75V Vcores on the Athlon's, it was recommended that you look into something like watercooling much after 1.85V.
I've tyried with 1.825 and there's just nowhere to go. I think I have a bad ocer, it's as simple as that. I didn'T buy it thinking I'd get a great oc so I'm fine with it. I'll remain at where I am !
What are you trying to get out it? A high fsb or clock speed? Id run it at stock Vcore, and increase the multiplier till it goes no further.... increase the vcore keep upping the mulitiplier etc till it goes no more. Then id run the cpu at the highest clocked speed possible but increase the fsb and lower the multilier etc unitill the bus wont go no further. Id just go for the fsb speed. I dont know how your dividers and motherboard will do but your ram should take you far. For pure cpu speed surely you should get 2.1ghz at least. Looking at some overclocked speeds... they rarely get over 2.3 but they take the power. Some over 2v. Not that id do that
My board can't go over 1.85v unless I write a e-mail to soltek saying I want the "Power user" bios which they don't put on their website so newbies like me don't fry their boards All nforce2 boards lock the pci/agp. what I'd like to get is 3200+ speed (200*11) or a high FSB with at least 2.1ghz, between 210 and 220 would be nice. I put the vdimm to 2.8v and the vcore to 1.825v and it posted at 210*10, it wouldn't below that. I tried upping the vdd to 1.7v but to no avail. I tried 200*11 and it wouldn'T do it either. I think I just have a bad ocer. Would Upping the vdimm make possible lower mem timings ? it's at 8-3-3-3 right now, 7-2-2-3 (turbo settings) doesn't work, aggressive and optimal settings are the same. I would have thought the timings would be better for the price I paid, considering I'm running it at their rated speed. EDIT: 3DMark01 would crash at those settings.
Increasing the vdimm will allow the ram to run quicker but I have no idea to what extent. It makes very little difference with normal sd ram, but apparently it can help quite a bit when it comes to ddr. Its quite a new cpu, let it 'burn in' for a while. When I first got this P4 it wouldnt go past 2.2ghz, now its at 2.5ghz!
ok I'll do that, it's been running 24/24 for a week now. I've read on amdmb that every .5 of cas latency is worth 5mhz of FSB. Is it true ? Does cas latency make that much of a difference ? Personnally, I think not.
Performance wise, I do think cas latency can make a difference. If you think about it, 3 down to 2 is a 33% increase in rate. Its not a small number not matter how insignificant the result may be. A hell of a lot of people dont think it does matter so much. It may well be the case with ddr ram. Ive only ever put together ddr systems, Ive only ever hammered single rate sd ram. But when it comes to the latencys I dont really worry so much about the speed, most cas3 sticks will run at cas2 ok. I suppose signalling that there isnt that much of a performance increase in the differance between cas 3 and cas2. Not many chips or components can handle a 33% rate increase, but the ram can when considering cas 3 and cas2. Thats sounds confusing cos the 'rate' can also apply to the clock speed.Im not referring to clock speeds. I only get cas2 cos it means I can drop back to cas3 if I have to. Carrying on from that, it doesnt matter what cas latency I have my ram set at. Ive got cas2 pc100 @125 with 2-2-2-5. Remember my ram only gets about 800mb bandwidth and my cpu is throwing nearly 4.gbs at it, so its heavily overloaded with work but can still handle cas2. I struggle to get it to go past 126-7 regardless of the ram timings. Overall I do beleive the cas latency can make a differance, but out of all the timings associated with ram I think the precharge delay is the most important. Basically this is the time gap between the blocks of info sent to the ram chips. I cant be bothered explaining myself but its probably why the numbers are so high. Like I said in a post about 3d mark 2001, the higher the number the more accurate it will be. If you dont beleive cas and ras latencys are very significant just set them to the slowest setting you can but with the pre charge on the quickest setting, then overclock till your content and then see if ram will run at the faster cas and ras settings. If it does, its a small bonus, if not im sure the extra cpu or fsb mhz you reach will make more of a difference than the quicker cas setting. Ram clock speeds are what counts imo. Dont worry about the cas and ras. If you can keep the precharge delay at a low timing even the better.
I found what the problem was ! There's a jumper on my board that selects the frequency. Default is 100/133 , and pins 2 and 3 closed is for 166/200. So I had put it like that but I had to put it to default and it works great ! I was able to run a loop of 3dmark01 w/ prime95 in the background while folding @230*[email protected] (which uses as few clock cycles as it can when another program is running, but still) so I moved to FSB240 but it didn't work, I moved back to 230 and it still didn'T work, I'm now at 225*10 and running prime95, I hope it'll hold !
It's been 11 hours of prime95 at 225*10 and still going. If it holds, I'll try at 1.7Vcore instead of 1.75 and even the 10.5x multiplier, to push this thing over 2.3ghz 230*10 seemed to work well but it locked-up once in MOHAA, and if it did it once, it can do it again so I backed off. 225*10 is more than I had hoped for so NOW I'm happy
