Anyone with a 3770K @ 5GHz on air?

[Klunt Bumskrint]

This is my target, anybody managed it?

If you did could you post your bios settings.

 

[Vectronic]

I think I've seen a few in other forums.

I can do 5.0GHz with my 3570K, but it's nowhere near usable since I have to keep CPU load to <70% otherwise it throttles (105).

 

[Borealis7]

more likely that Sandy-Bridges can do it on high-end Air, than Ivy-Bridges.

 

[VonDutch]

This is my target, anybody managed it?

If you did could you post your bios settings.

i can run 5.0ghz with mine,
i dont think a 3770K thats not delidded can tho
but theres always exceptions of course ..lol

i mean with "normal" cooling, like i have with mine, a aircooler that is (Scythe Mugen 2 Rev. B)
but as you can see, my chip isnt a good ocer, so im not running this 24/7
i settled for 4.8ghz at 1.420V vcore

 

[Klunt Bumskrint]

I'm fully stable at 4.5GHz with 1.23v so I think its possible. I tried just bunging the multiplier upto 50 with my existing settings (which took hours to fine tune) but it crashed and corrupted my windows installation.

Don't want to risk it again unless I have some some more info on what settings might work for 5.0GHzGHz

 

[Idontcare]

The silicon can definitely run at 5GHz, just as the silicon for piledriver can run at 8GHz under the right voltage and cooling conditions.

The question really is - can 5GHz operation be accomplished with 24/7 stability and at a voltage that isn't ridiculous (begging for chip degradation)?

Personally I don't expect 22nm IB chips to be able to survive 2yrs of operation at 1.4V. But it all depends on the duty cycle.

If folks are setting their rigs to run 1.5V and 5GHz but the processor sits idle 98% of the day then their chip will probably live for a decade. However, if they are setting it to 1.5V and 5GHz only to run DC or 10hr+ transcode jobs that peg the processor utilization at 100% for a significant fraction of every day then those chips can't be expected to last all that long.

For heavy duty cycles, 4.8GHz and 1.4V is probably your absolute max to be looking for.

 

[BrightCandle]

I wouldn't go above 1.35V, there have been quite a few early deaths at 1.4V already.

 

[Idontcare]

I wouldn't go above 1.35V, there have been quite a few early deaths at 1.4V already.

I believe it.

 

[cytg111]

I think I've seen a few in other forums.

I can do 5.0GHz with my 3570K, but it's nowhere near usable since I have to keep CPU load to <70% otherwise it throttles (105).

If there ever was a reason to delid++ etc, you're looking at it!

 

[sm625]

I believe it.

Where is the thread documenting these early deaths?

 

[pantsaregood]

i can run 5.0ghz with mine,
i dont think a 3770K thats not delidded can tho
but theres always exceptions of course ..lol

i mean with "normal" cooling, like i have with mine, a aircooler that is (Scythe Mugen 2 Rev. B)
but as you can see, my chip isnt a good ocer, so im not running this 24/7
i settled for 4.8ghz at 1.420V vcore

I think your overclock may be a victim of phantom instability. The error checking on the CPU is correcting the instability, but your performance is being hit in the process of it.

My reasoning behind this is that my i5-2500K at 4.4 GHz pushes 120 GFlops. Your CPU appears to be slower, despite the fact that at 5.0 GHz, it should be 15-20% faster due to Ivy Bridge's minor architecture improvements.

 

[VonDutch]

I think your overclock may be a victim of phantom instability. The error checking on the CPU is correcting the instability, but your performance is being hit in the process of it.

My reasoning behind this is that my i5-2500K at 4.4 GHz pushes 120 GFlops. Your CPU appears to be slower, despite the fact that at 5.0 GHz, it should be 15-20% faster due to Ivy Bridge's minor architecture improvements.

high oc isnt the only thing that makes gflops go higher,
and i have HT on, if i disable HT, my gflops would be at 125-130 gflops,
but i dont notice a big difference anyways between oc's, gflop wise
ram is important for gflops also

what do you mean with, "victim of phantom instability"?

 

[Vectronic]

If there ever was a reason to delid++ etc

Yeah I'm thinking about it. Reason I haven't is because the only TIM I have is the stuff that came with my cooler, ZM-STG2... which is barely good enough for IHS -> HSF.

Plus, I'd like a $200 buffer in case I screw up... at the moment, I have neither.

 

[john3850]

Go to you event viewer and check your WHEA-Logger events.
I thought I was good at 4800 until I checked my viewer last week.

 

[LagunaX]

I came close on air but don't want to get any higher temps with this one (also in the F/S forum):
http://i246.photobucket.com/albums/gg84/jhchang69/3770k249ghzMIVEIBT_zpse4d714f9.png

Maybe I'll try with my other 3770k that IS NOT for sale.

 

[pantsaregood]

"Phantom" instability is exactly as I described. Modern CPUs are equipped with quite robust error correction capability. If your CPUs error correction catches something and corrects it, you won't see errors when running stability tests. It will, however, drop your performance.

There's more to verifying stability than "oh, my PC didn't crash when I ran IBT."

 

[Klunt Bumskrint]

What's everyones thoughts on minimum volts for 5GHz?

 

[Idontcare]

What's everyones thoughts on minimum volts for 5GHz?

When I was generating date for this thread I was not able to get 5GHz stable to my requirements (5 passes of LinX with 14.2GB ram) at any voltage that I was comfortable with while delidded and using an H100.

4.9GHz operation required nearly 1.4V, itself a value I consider to be an effective death warrant for a 22nm IB CPU.

I did run some 5GHz benchmarking tests just for fun, but they weren't LinX stable. The voltage I used for the 5GHz runs was 1.45V. So 5GHz is going to take more than that if you want to be LinX stable.

 

[john3850]

What's everyones thoughts on minimum volts for 5GHz?

For daily use I would stay at or below 1.4v.
On sb 1.5v was too high for daily use.

 

[LagunaX]

Be happy with 4.8ghz and a SSD - it is plenty fast and sustainable for 24/7 for quite a while.

 

[VonDutch]

"Phantom" instability is exactly as I described. Modern CPUs are equipped with quite robust error correction capability. If your CPUs error correction catches something and corrects it, you won't see errors when running stability tests. It will, however, drop your performance.

There's more to verifying stability than "oh, my PC didn't crash when I ran IBT."

you mean whea errors?
like i said, gflops dont mean much, if what you said was true,
LagunaX would have the same "phantom instability" as me,
check his pic, its not much more then i have, and less then your 2500K

if i put more ram to ibt, my gflops would go up a bit,
if i run without HT, my gflops would be more then you have,
the difference between a few 100 mhz isnt that big either..


"There's more to verifying stability than
"oh, my PC didn't crash when I ran IBT."

what are you getting at?
i know theres more to it,
now you make me feel as if i dont know what im talking about?

 

[pantsaregood]

No, I don't specifically mean WHEA errors. Modern CPUs support error correction independent from the operating system. Overclocking can sometimes introduce computational errors that aren't severe enough to be visible to the operating system, but effectively slow the CPU down due to use of error correction.

As far as I know, the only way to detect this behavior is observation of performance. For example, I noticed my 2500K is "stable" at 4.7 GHz compared to 4.5 GHz, yet it consistently benchmarks lower. Adding voltage is enough to boost bench scores in that case.

Also, I apologize if you felt I implied you didn't know what you were talking about. It was not my intention. I was merely trying to point out that detecting instability is sometimes complicated.

 

[Ferzerp]

I think your overclock may be a victim of phantom instability. The error checking on the CPU is correcting the instability, but your performance is being hit in the process of it.

My reasoning behind this is that my i5-2500K at 4.4 GHz pushes 120 GFlops. Your CPU appears to be slower, despite the fact that at 5.0 GHz, it should be 15-20% faster due to Ivy Bridge's minor architecture improvements.

He's using "all" threads instead of "4" threads. For accurate readings on 3770k you have to use 4 threads and set affinity to all even numbered cores.

 

[pantsaregood]

You're absolutely right, HT can decrease performance in IBT. I would expect an i7-3770K at 5.0 GHz to push close to 140 GFlops. A 20% reduction in performance as a result of HT being turned on is rather extreme - if my memory serves me correctly, HT usually only reduces performance by a few percent.

 

[Ferzerp]

You're absolutely right, HT can decrease performance in IBT. I would expect an i7-3770K at 5.0 GHz to push close to 140 GFlops. A 20% reduction in performance as a result of HT being turned on is rather extreme - if my memory serves me correctly, HT usually only reduces performance by a few percent.

IIRC, at 4.4Ghz, I get 107-110 Gflops with it run the way he runs it, but 120-121Gflops if I set it up correctly.


He's getting the same at 5.0Ghz as I get at 4.4Ghz with the same settings.