Overclock.net › Forums › Intel › Intel CPUs › Haswell Overclocking Guide [With Statistics]
New Posts  All Forums:Forum Nav:

Haswell Overclocking Guide [With Statistics] - Page 494

post #4931 of 19539
Quote:
Originally Posted by filphil View Post

Question for you guys. How do you set your fan profiles when you stress test for stability? Is it acceptable to set all fans at 100% fan speed simply for stress testing since it puts it at a mostly unrealistic load and tune the fans for normal operating temps under 80C after?
w

I set all my fans to 100% while stress testing. And it depends on your motherboard on what you should use. For my ASUS board, I used FanXpert 2, but for my MSI board, I used SpeedFan. Also, unless you really feel the need to be prime, linx, etc stable, then go ahead,. However, x264 is more realistic on temps and allows you to overclock further if you are temp bound. Also, stability with x264 has not failed me yet. Sometimes even if I was prime95 stable, I would still BSOD in game.
post #4932 of 19539
I'm using Aida64 and prime95 at the moment. I may pick up x264 since I feel like using a multitude of stress testing programs would only benefit in tuning for overall stability.

I've picked up an i5 4670k and Sabertooth z87 for a really good price at my local microcenter and have managed these settings with a hyper 212 EVO:
CPU Multiplier - 45
CPU Cache Min/Max - 34/34
DRam Frequency - 1333
Vcore Voltage - 1.25v
Vcore Input Voltage - 1.8v
DRam Voltage - 1.5v
Edited by filphil - 11/3/13 at 6:17pm
post #4933 of 19539
if you have a 2400mhz memory kit...DO NOT run 2400mhz initially. Lower it to 1600mhz and try to see the max potential of the CPUkc0X8t
post #4934 of 19539
The stock speed of the ram I'm running is 1333mhz since that's what I have on hand. My 1600mhz kit is still in transit but it should be here at most by tuesday.
post #4935 of 19539
Quote:
Originally Posted by filphil View Post

I'm using Aida64 and prime95 at the moment. I may pick up x264 since I feel like using a multitude of stress testing programs would only benefit in tuning for overall stability.

I've picked up an i5 4670k and Sabertooth z87 for a really good price at my local microcenter and have managed these settings with a hyper 212 EVO:
CPU Multiplier - 45
CPU Cache Min/Max - 34/34
DRam Frequency - 1333
Vcore Voltage - 1.25v
Vcore Input Voltage - 1.8v
DRam Voltage - 1.5v

Darkwizzie we have a rare one here. Someone actually read the guide before asking a question.... Thanks mate! biggrin.gif You've got a nice chip there. I'm interested to see how far you can go. Although, you will eventually need to replace your cooling solution as I wouldn't push much further than 1.25v with a EVO. Great budget cooler, but definitely not a overclocking cooler. If you'd like a suggestion or two on what one to get to move forward. I'd be happy to help, but PM me so we don't spam this thread with non-sense not related to overclocking.
post #4936 of 19539
The hyper 212 was given to me by a friend who moved on to one of the corsair hydro series. I'm considering bring it up a notch to the Nh-d14 or the xspc kit with d5 pump. I'm waiting on the ram to come in before I continue with the overclocking of ram modules and uncore. At the moment the temperatures are hitting mid 70's with the current stress tests so I'll keep the vcore at it's current 1.25v to leave myself with some breathing room.
post #4937 of 19539
I'd go with liquid cooler. the D14 is nice, but you can only go so far. If you plan on pushing that chip as far as you can, then liquid cooling all the way.
post #4938 of 19539
Yeah I'm heavily considering the xspc kit. I'll be sure to chime in again with updates on my progress on the current cooling solution. Thanks helping me out! thumb.gif
post #4939 of 19539
You're definitely welcome! I'm interested to know what your chip can do considering your voltages for 4.5Ghz.
post #4940 of 19539
Quote:
Originally Posted by Darkwizzie View Post



So let's recap:
With Uncore 800mhz under core, we see a:
Loss of 0.75% in chess.
Loss of 0.2% in multiplayer BF3
Win of 0.5% in BF3 Campaign (BTW the BF4 is actually BF3, typo.)
Loss of 0.82% in Cinebench.
Loss of 0.6% in x264 Bench
Loss of 0.39% in SuperPi
Loss of 1.2% in Enemy Territory
Win of 1.27% in Runescape
Win of 1% in Oblivion
Win of 8.72% in Oblivion

Testing method:
Chess: Houdini 3, 9mb hash, starting position.
BF3 Multi: 64 player, crowded server, regular gameplay for entire round (BTW, during second round where I test x42 uncore, a few people left and I camped a little more. Just saying.)
BF3 Campaign: Second misson, following scripted NPC movement.
Enemy Territory: 30 vs 30, Fueldump.
Runescape: GE, World 3. Capturing FPS will stationary. Max detail, non HTML5. x4 AA Bloom enabled. (It seems to use CPU to do AA)
Oblivion 1: Walk out in the wild, through Oblivion gate, to town gate.
Oblivion 2: NPC combat in Imperial City. Several guards/NPC vs Umbra. Spawn 50 player copies and begin combat once Umbra dies.

Can you even call this a clear lead? No. If so, barely. Let me take out the outlier. That's a win on average of 0.132%. For DOUBLE the uncore difference. In reality a stock uncore is x38. That makes the average win to be half of 0.132%. Aka 0.066%.

Awesome!
You've proven that by overclocking uncore from stock we can expect an increase of 0.066% in performance once I take out the outlier!


I never said 'don't OC uncore', I just said 'don't expect noticeable real world results'.
If 800mhz drop = 0.132% loss in performance, than 100mhz of uncore drop = 0.0165% decrease in performance.  Also note that the tests favors uncore as most of my benchmarks are CPU benchmarks solely. Superpi? Chess? Cinebench? Bf3 which we know uses lots of CPU, on a 64 player server? Superpi? X264? Enemy Territory, CPU bottleneck? Runescape, single thread CPU bottleneck? Oblivion, single threaded CPU bottleneck? I can't make it any easier for you. Had I tried Crysis or something like that... 

Oh, and let's just say, a win of 0.132% is within margin of error.
I used to say and still sat a change in performance of 3% is barely noticeable if it is ( that's like a core multipier increased by 1 ), but here we're dealing with numbers like 0.066%. Let's suppose I'm at 60 FPS. And I get 0.066% in performance. Do you know what FPS boost you'd get? 0.0396 FPS. Good luck seeing a FPS difference there. When we review GPUs sometimes we see a FPS difference of 5-10%. Compare that to 0.066%. That's not even 0.1%.


Let's be generous and say +1 core multiplier = 2% increase in performance in these benchmarks, instead of 3%. That makes +1 uncore impact by THREE TIMES that of of overclocking your uncore (more specifically, from 3.8 -> 4.2) entirely. That also makes +1 core about 12 times larger impact than +1 uncore.

If you think my benchmarks are unfair, then you're certified delusional. I purposefully picked CPU dependent workloads. The only thing I didn't run was Linpack to prevent thermal overload (and I tried it, and yes, CPU meltdown). If you think the omitting of ONE benchmark destroys the credibility of the chart, you're also delusional. There are 3 possibilities if I included Linpack:
1) Higher uncore DESTROYS lower uncore. It's an outlier. THROWN OUT like the Oblivion result.
2) Higher uncore is slightly faster than lower uncore. Difference too small. Result still stands.
3) Higher uncore is slower than lower uncore. You cry foul.

Of course there is a reason why you cling onto Linpack like a lifeboat, because it's the one bench I'm not doing. Superpi stopped mattering when I started using it. Outliers stopped mattering when you realize it can only benefit your point. More tests stopped mattering when I'm doing it. But the irony is, no matter what result I get, my final result STILL STANDS. And if you think Linpack represents real world performance, you are also delusional.

You can't have it both ways. Either I keep all outliers or I remove them. Assuming Linpack would be an outlier. Assuming.
Your arguments make no sense and you flip flop around your position after I prove each point to be false.

So please.
Just stop.

Good you've wasted your time calling me delusional when was I originally said was that additional vRING was required for stability. Slick. It's not my fault you can read but not understand. I had clearly mentioned earlier that bottleneck or not ( where your tests have shown the latter ), additional vRING would be necessary for stability if upping the vCore didn't help. You went ahead and clung to the small bottleneck thing and went and wasted your time on it.

So basically, go back to my posts in the thread, read and think whether I'm "clinging" on to linpack and such. Also, if a particular stress test is stupid, you should probably tell reviewers and question them as to why they still use it.

As of right now, I'm done wasting my time responding to your posts about this, when you're clearly hung up on the wrong thing instead of something that would've helped people who couldn't get their setups stable. Also, if thinking calling me delusional is going to make your e-peen bigger and king of your own thread, by all means, please go ahead. Just saying, being a dick doesn't make you right; it just makes you a dick.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Haswell Overclocking Guide [With Statistics]