"no, prime fucks with the voltages. aida is fine"
I disagree, my experience with aida is a bad one, even completing 12 hours of stress testing i still get stability problems, but with prime 95 4 hours of stress testing seems golden.
That is the issue with stresstesting, though.
I've run multiple 8-hour sessions of Prime95, and had a core throw an error during the fourth time. Never experience any sort of problem otherwise. (Still, I bumped up the voltage, just to be safe.) Another friend of mine ran it for 30 minutes and called it good. No issues or errors. :\
I get what you're saying, but I've experienced the same issue with multiple processors from both amd and intel. I once stress tested aida for 4 hours, turned on prime 95 after that and it failed within 5 minutes. Prime 95 from my experience stresses the processor much harder, yes because of that there is a higher risk but if you know the limits of the cpu, you should be fine.
Also, I've been seeing many people with 5ghz overclocks with H100i coolers getting 68c, where as im getting over 10c cooler on the socket alone (55c) and 15c cooler on the cores (38c) with a similar cooler? Is this also due to the chip or a mix of things?
you can take the core temps from amd cpu´s with a grain of salt.
i had my cpu at 4.9Ghz, and it did fine, until the socket temp went over 65°C then it crap´d out on me lol.
Mix of things. Other heat sources may be having a lesser affect, and your processor may be running at a lesser voltage.
I would also check to see that you are running at the proper clock speeds, since you have AIDA.
EDIT: I have it pretty good authority that the extra voltage/clock is not in the consumer chips, only in the Xeon chips (Haswell-EP). Intel's spec is a bit ambiguous on the subject (I wish I could test it myself). So people claiming that some stress test over-volt their CPU should probably check their settings. Auto or Adaptive voltage settings in UEFI is not recommended.
Not this again. Prime can't change any voltages, it is impossible. What people need to understand is that Haswell CPUs have dynamic voltage depending on what instructions you run. It is also depends on the motherboard BIOS/UEFI. Has it set to auto? Does it over-volt anyway? Etc etc. Read the Intel spec, the Haswell CPU will increase the voltage when running certain instructions, FMA and AVX2 IIRC. If you already has it overclocked and/or raised the voltage, that might not be a good idea.
AMD CPUs does not do this. Although, many have observed that you need more Vcore to be AVX stable. Sometimes a lot more. At the same time if you're not AVX stable you usually have flaky and inconclusive results when benchmarking. The IMC (CPU/NB) usually also needs quite a bit of voltage to be stable. Good RAM is needed etc.
1 Like
Silicone is a compound, used as a sealant and formerly in breast implants, until they were found to cause cancer.
Silicon is the element used in electronics. Although it is "doped" with minute amounts of other semiconductors in order to slightly increase its conductivity, it is still elemental silicon and nothing like the compound, silicone.
Will somebody please make a porno titled "Silicone Valley" to clarify the difference between silicon & silicone for everybody? And in order not to make the confusion even worse, it should be written, produced, and directed by techies.
2 Likes
the FX chips are a bit screwey with prime95. it's mostly cause prime uses Floating point. the FX chips have waek FPUs, so it's still not a good stress test. you probably need to use multiple softwares.
EDIT: I have it pretty good authority that the extra voltage/clock is not in the consumer chips, only in the Xeon chips (Haswell-EP). Intel's spec is a bit ambiguous on the subject (I wish I could test it myself). So people claiming that some stress test over-volt their CPU should probably check their settings. Auto or Adaptive voltage settings in UEFI is not recommended.
Prim95 is more like three tests in one. You can stress FPU only, FPU+caches and a little RAM or FPU+caches+RAM. Small FFTs, In-Place Large FFTs or Blend/Custom (edit in 90% of your RAM). When you stress caches and RAM you stress the IMC (CPU/NB) pretty hard, usually an overclocked CPU has instability problems there making it score inconclusively in benches. I remember upping the voltage for the CPU/NB pretty high before I was stable even with good quality RAM. Had (still have) plenty of bad RAM that never got stable whatever I did. The RAM would work fine in older systems at lower frequencies, but not with a 8350 at even the rated freq with loose timings. Admittedly it was pretty cheap crappy RAM, I've given up on buying stuff like that since then.
I don't think Prime95 is screwey at all, it's just that getting a good and stable overclock has more to it than just upping the Vcore. Dialing in the CPU/NB and RAM can be finicky and time consuming. Especially when FSB overclocking, the best way to overclock IMHO as you get very nice performance. I think the "front-end" of FX CPUs are the weak part so to speak. Sure the FPU is weaker than on big Intel chips. But if the front-end can't even feed it, it's pretty moot how good it is. If an overclock seems a little flaky, running AVX instruction tests on it (like IBT with AMD AVX libs) usually makes it crap out as the CPU/NB and front-end isn't really stable.
AFAIK prime95 doesn't use AVX instructions on AM3+ CPUs. I've read some what the devs of prime have said about it, and it seems they don't care to work around the bugs. Piledriver has some nasty bugs concerning AVX instructions that make most devs avoid it (only Linpak seems to use AVX). Purportedly fixed in Steamroller and Excavator, but we are not getting them on AM3+ now are we. Hopes for Zen to be less buggy.
prime only messes with voltage on intel cpus
I find it to be the best program. Yeah it puts insult to injury when you have a nice oc and it's stable in everything but prime temperature and stability wise :D
So my 8320 at 4.0 ghz on stock voltage is pretty good then?