Originally Posted by Bartouille
Originally Posted by Belial
Whats going on, did I miss something? Can I get a recap?
It seems to me that ivy, sandy, and haswell all overclock exactly the same - they are binned similiarly (on 1.45v or less 50% do 4.6, 20% do 4.8, 10% do 5) and they all hit similiar frequencies at the same voltages for equal binning (ie an average chip does email@example.com stable, most are below avg and need 1.3+). The only differences is heat, and practically speaking all 3 generations are heat limited and scale almost perfectly linearly with voltage (ie as opposed to AMD's 'heat wall').
I dont think itll be long before we see a delidded haswell on water (or high end closed loop), and possibly air, doing 5ghz 24/7. Just like ivy and sandy we saw only like half a dozen OCN members do 5ghz, and they usually showed up later.
Also, stability is always an issue too with these claims. Most people dont do 24+ hours of p95, especially on haswell, so most claims of 'stable', especially the larger overclocks, are nowhere near rock stable. Like my aid64/h264 firstname.lastname@example.org/4.5ghzuncore/2800 RAM turned into email@example.com/4.4uncore/2666Ram and Im still struggling to even make that p95 stable.
Theres always some guy who does firstname.lastname@example.org for 24 hours of p95, but he'll be the only guy, and he'll have spent a ton of time testing and likely have owned and binned multiple chips.
To be honest I don't see the need to run P95 for 24 hours. Unless you do a lot of rendering or very intensive task with your CPU on a daily basis, a 6 hours run is more than enough, even 3 maybe... Yeah, it might be ROCK stable or whatever, but it still doesn't guarantee you it won't crash, plus running P95 for 24 hours degrades your chip, and if the chip is new (like Haswell) degradation is even worst. Personally, I got 1.35v at 4.7GHz and the way I tested it was so simple, I ran Aida64 at 1.34v, crashed after 2 hours, now I use 1.35v, ran Aida64 a couple mins to see if temps were still okay, nothing else. If you didn't stress your CPU so much and had left it alone it would have been fine at 1.45v 4.8Ghz. If people can't accept the fact overclocking may cause instability (rarely, but you still have to think it can happen), then they shouldn't overclock at all. People that need rock stable stability don't overclock, business, schools or stuff like that never overclock.
I do encoding and mining on a daily basis, and will consistently bring a 12+ hour p95 stable overclock but not quite 24+ hour p95 stable on it's knees in minutes (just both of these programs aren't as consistent at detecting ram instability). If all I did was game I wouldn't even have an i7, or even an i5. I cannot have a crash at all during a stream, it'd lose me viewers quickly (so even if the risk is small, any risk is devastating), and I cannot have a crash when I'm at work or during the night during mining, or have returned HW errors which is very common on slightly unstable systems. I'd literally be losing significant money if my system was just ever so slightly unstable that I returned a single HW error a day.
So it's very, very important for me to be 100% rock stable. On top of that, I like to overclock, I like to know my system. Is it a crime that I want to know exactly the limits of a system I paid hundreds of dollars for? If I was just a GaMeR, of course, I would only use 1 hour of prime95, not even, and then tune an extra 200mhz on top of that.
And running p95 for 24 hours won't degrade your chip, just because you heard sin say it doesn't make it totally true, he also said running it for 24 hours at 80C+ would be an issue, over long term, not just a single stability test. I only hit ~85C during small fft, which is minutes 15-30 of a 24 hour prime95 run. The other ~23 hours of a prime95 run, I run in the mid-60's.
Even if degradation occurred, I wouldn't care. I'd drop the oc/raise voltage, take it a step back, and sell off the build in a few months anyways. Also, if a system were to degrade on stock, even from running p95 on stock cooling 24/7/365 (or you ran CPU mining, encoding, chess calculations, etc), that's grounds for RMA.
AIDA64 is hardly stable, it's an awful stress test if you need real stability, but it's fine if all you do is game. There is a reason why supercomputers take a good 2-3 weeks out every few months to run a straight stability test, because for certain programs and calculations, it's important. You don't tell the people who run CERN or TITAN that 'hey just run AIDA64 for a few minutes, you're good'.
I also don't want to be reinstalling my OS every 6 months because of corrupted registry errors.
As for people not overclocking if they need stability, the programs I use are unusable if I don't overclock, or, conversely, they bring even overclocked, top of the line CPUs to their knees. Even games like SC2 do not run smoothly on an overclocked 4.5ghz Ivy i7, because it only optimizes dual cores, so high IPC per core is very important, which you can only get by overclocking. If I didn't overclock, I'd have to spend over $600+ on CPU and RAM for weaker
performance, so it's definitely worth it to a value oriented, budget conscious streamer, encoder, and miner like myself.
As for businesses, schools, etc, not overclocking, that's largely due to the knowledge required, time it takes, and warranty issue. We all overclock because we just RMA if it breaks, no one would know it was overclocked, but if a large business or school did something like that, they'd have a lawsuit on their hands. A school isn't going to pay someone basically $20/hr to overclock their systems.Edited by Belial - 8/1/13 at 12:41am