View Single Post
  #4  
Old January 25th 17, 06:15 AM posted to alt.comp.hardware.pc-homebuilt,alt.comp.hardware.overclocking
B00ze
external usenet poster
 
Posts: 67
Default Overclock 6700K Max Temps Prime95 Small FFT

On 2017-01-24 04:21, Paul wrote:

[snip]

That heatsink seems to give pretty good performance here,
in terms of the degreesC/W.


Yeah, I bought it even tho I had a Hyper EVO in stock, but the heat is
so concentrated over a small area that I wonder if it really makes a
difference if you buy a more expensive heatsink over the cheaper EVO.

If the internal case air is at 35C, the Noctua added 20C (to get to Tcase),
you added another 25C case to junction (unknown, guess), the
Tjunction should be around 80C. Check and make sure the
computer case has a big enough fan on the back (to keep it at 35C).
If you notice that the CPU temp *drops* when you take
the side off the PC, that means your case cooling
sucks :-)


Case temp should be fine, it has 3 200mm fans + the back exhaust fan
(Thermaltake Urban T81 full-tower).

My case cooling isn't the greatest (since I fitted a
quieter fan on the back). I'm using the GPU as a proxy


That's the beauty of the case I got, since the fans are wide, they run
slower thus less noise.

for case air temp here. The hard drive is in the front of
the computer and hasn't budged. You can see the GPU temp
coming up a tiny bit, so I'd have to leave it running
for a bit longer, to get a true final temp. You can see
I could push this thing harder, but I'd have to fix
VCore somehow (too hot).


https://s28.postimg.org/ve04pgfl9/prime95_test.gif


I don't understand SpeedFan's graph; that line is around the 45 mark but
I know it's not 45C. And 1.088v is really low (for my CPU anyway, dunno
for your Ivy Bridge 6-Core). Strange that your HD0 temp goes up along
with the CPU, you'd think it would stay cool being in front of the case
away from the back exhaust...

That's cooled by a Noctua NH-D15 (with the center fan fitted).
I de-tuned the PC a bit when I got it, as the default
Asus settings were cooking the Vcore regulator. I turned
off the Turbo (which wouldn't make any difference in this
case, as it cannot Turbo if all cores are 100%). The measured
power, using a clamp-on ammeter, was 156W, before I turned
down the setting. That's the power feeding into VCore (ATX12V).
I have no easy way to measure the current flowing on VCore
itself. My clamp-on ammeter cannot fit around the copper
plane :-)


There is some bug with HWMonitor on my board, it keeps saying that
"Package" power is between 1W-2W; I wish it would read correctly, my
Watts must be through the roof when it hits 1.360v with all cores
running AVX2 in Prime95.

So my conditions aren't even close to a match for yours.
My Vcore heatsink was running 65C with the original settings
(with Prime95 as the test), so I couldn't leave the
thing that way. It probably won't go into thermal
runaway, but I'm not taking a chance. If only I'd
remembered to check the size of the VCore heatsink
before I bought the motherboard :-( Mistake.


Not OC'ed (i.e. at 1.200v) my CPU stays around 60C in Prime95 (and in
Intel Burn Test). I don't know why you "couldn't leave things that way"
- 65C is just fine, unless you plan on running Lin-pack 24/7 for the
next 12 years...

Take a picture and show the dynamics of your setup,
from "idle" to running the test for 7-8 minutes. Just
to give some idea what hops up first.


Prime95 Small FFT:

https://www.dropbox.com/s/u69v3u4479...1344v.jpg?dl=0

Prime95 128K FFT (this is as high as it goes, if I leave it running
longer it goes up a degree or two):

https://www.dropbox.com/s/odkzpw8yjz...1360v.jpg?dl=0

Things that matter:

1) Bent CPU.


Yeah, I thought of that, maybe my HS put too much pressure on the CPU,
but I'd have to take the thing apart to check :-(

2) Condition of the Intel TIM, between the silicon die
and the lid. Intel used low-temperatore solder between
the CPU and lid in the LGA775 days. To avoid "conflict
minerals", they went back to conventional dough-like
material on the newer processors.


Not sure it's to avoid conflict minerals - I read recently that they
found the lid would crack after hot/cold cycles if they used solder
instead of the goo they now use.

3) It would take a generous, gushing layer of AS5 to insulate.

On the "too lean" side, you should use an inspection mirror from
the side, and look for the color of the AS5 wetting the junction.
Depending on the size of the heatsink, it might not be possible
to eyeball it. I apply half a rice grain bit of AS, compress, verify
the spread diameter, then adjust the dose for the second and final
installation. Then, hold the inspection mirror around the side,
and look for a wetted junction. Mine don't generally gush all
over as a result. I've under-dosed at least once, and then
it was third-time-lucky.


I use the 2-Lines method from he

http://archive.benchmarkreviews.com/...1&limitstart=4

I used it once, then found out I'd put the heatsink in reverse (there
are little holes around the cooper posts, these I think should be facing
the fan, not the other way around,) so I removed it and did it again.
When I removed it, the paste was perfect, not too much, not too little,
spread nice and thin all over with very little run-off on the sides.

I stopped spreading it with a credit card, a long time ago :-)
That's just a waste.


Yeah.

I did some more googling, this time I added "Small FFT" to my terms, and
most people seem to agree that the latest Prime95 Small FFT which uses
AVX2 is just too punishing. If I use OCCT's "CPU" test (instead of it's
LinPack mode) then my CPU stays around 70C @ 1.344v, which I think is
fine; the only bench I have that makes LLC climb to 1.360v is Prime95.
LLC I think is supposed to LOWER the voltage when the CPU is really busy
so that temps stay constant; it's pretty cool to see MB manufacturers
switch it around and supply MORE volts when instead, so that the
overclock remains stable.

Best Regards,

--
! _\|/_ Sylvain /
! (o o) Memberavid-Suzuki-Fdn/EFF/Red+Cross/SPCA/Planetary-Society
oO-( )-Oo Compile, run, curse. Recompile, rerun, recurse.