I agree. Just tried Cinebench (which seems to be a favorite of the OP) and for a stress test, it's useless.
1. Temperatures are way lower than Prime95. Prime95 with Small FFT sends my CPU temp to ~83C for the hottest core. With Cinebench, the hottest core only hit 65C and only after several consecutive runs.
2. It's over in a few seconds. Seriously, it's not much better than simply being able to POST and boot into Windows.
3. Sure, you'll notice if it crashes your system completely, but you won't catch all rounding errors and other faults resulting from an unstable overclock.
The OP still has a very nice chip that is able to boot into Windows and survive 15 seconds of fairly high CPU load at such a high clock rate and low voltage. However it's not a system I would want to rely on for anything other than light web browsing in that configuration.
To achieve 24-hour Prime95 stability, the voltage would have to be increased significantly, which would send temperatures out of control, especially since Prime95 temps are almost 20C higher than Cinebench temps to begin with. I don't think 4.5 GHz would be unreasonable, maybe even 4.6, which is a few hundred MHz better than average (mine does 4.4, maybe more, but I'm more comfortable with the temps and voltage at 4.3).
I've never claimed it to be a stability test or that I was ever stable, not once in this entire thread. How much higher could you go on clocks just running cinebench compared to your prime tests, or how much lower could you go on voltage?
1. I already ran a AVX linpack for you, 80C at 4.7GHz -
http://forums.anandtech.com/showpost.php?p=35128884&postcount=24
2. Yes, it's a nice dirty checker, it's quick and consistent. It really helps gauge how changes you make are affecting the performance you're getting. That is actually the point, at least for me and why I used it. It fully saturates the cores, it gives me all the info I need from a test including if the tweaks I made increased or decreased performance. It's far more stressful than simply posting, or booting into windows, or validating, yet people do all of these because that's what overclockers like to do, which is what I am.
3. You'll see much of that in the performance you're getting, you can also see soft errors with software reporting.
Neither would I, hence I've never claimed to use superpi or cinebench for stability testing, nor have I claimed at any point "stable". I just got the chip, I'm playing with it "fun".
I don't play the game prime95, or encode with it either. Nor does anything I use run straight AVX code which loads up the Integer and FPU pipelines to the max effectively making your quad core a octacore. If that day comes, if software can utilize my cpu in the same manor as Prime95 does, and it's something I actually use than by all means I'll use Prime95 or most likely, the actual software to test stability.
Stable to you, to me, to everyone is different. As I said I haven't tested for stability yet, all I've done is play with my chip to see what it could do and to see if everything is ok with it. That was the point of this thread, it didn't seem right. I was looking to see if it was actually running those speeds, and at that voltage in the first place, as well as if it was scoring properly or if it was downclocking. Prime95 is worthless for this because it doesn't give you a score after 24 hours, it's worthless because the code path is linear as is it's instruction set. I stopped using Prime95 with Sandy Bridge, which 24 hours of Prime95 was unstable at idle, or when IG. That's not how I use my PC, so that's not how I go about determining what is stable for me and what is not.
I started doing my stability testing last night, I used Crysis 3 for this because of several factors. It's brutal on my GPU's, my typical voltage is completely unstable in Crysis 3. So what better place to test my cards with a chip that can actually push them? Better yet, Crysis 3 will run my i5 to the brink which is great, and it's use FPU and Integer and actual gaming code to do it, not a straight shot of 100% nothing but AVX turning my quad core into an octacore - if it was I wouldn't even bother overclocking it because it's performance would just be insane.
Is this good enough for you to call stable? No I'm sure it isn't, is it enough for me? Nope, need more testing, and realistically more just actual daily use. This thread isn't about stability, it never was meant to be. This is how I'm testing my gaming PC, with actual gaming use. Here I was at 4.2GHz, but as I went up higher I needed to increase my 7950's from 960 core to 1100 to really push the i5. Haswell effectively killed the need to test for max temps because Haswell will not crash if it gets too hot, it will throttle down and remain stable.
When I'm testing stability of my gaming system for gaming I use gaming, because I want to make sure that not only does it work when it's pushed to the max clocks at the max usage, but also for when it's not going full out. For when it's downclocking and undervolting because it doesn't need the extra clock speeds. I can't test these things with programs like Prime95 because it's never not at full load, it's never not at full voltage, it's never not at max clocks. This is of course just the start, I will find stable clocks in this game ranging from undervolted stock all the way as high as I can take it, then I'll do more games like BF3, and more programs like handbrake, and more idle usage like youtube and forums. To summarize for me stability means in all things *I do*, not just one program, doing one thing, at one speed - but this thread wasn't meant to be about stability and how you measure it.
did u just lord me?? im not a lord.. im an OVERLORD!
And yes, i know the cpu isnt in the same class, but this is haswell.
For the longest time.. everyone is going "Ivy-E faster... better... quadcore like a hexcore... "
Well... i almost believed everyone in thinking there quadcores are faster then my hexcore... :X
To my realization in this thread.. it would take your cpu at 5ghz+ on phase to equal what my 4.4ghz cpu does on water... :sneaky:
Indeed me Lord!
Keep in mind I just got it, and was just playing with it. There are a lot of things that effect performance in legacy code such as Cinebench, most noticeably here is RAM speed/timings.
This is at 4.4GHz, notice it's almost as fast as the 4.7GHz run from prior. If I could get my benching 2x2 kit to post I could do even better at the same clocks, but for now I'm just tweaking trying to maximize what I have.
That would wholly depend on your usage and workloads.
No, it would take my CPU more than that CPU, this is the $250 chip, that was the $350 chip. Mine doesn't have HT, it's just a flat 4 core CPU with no frills.