Mersenne Rig (take 2)

[simon389]

Originally Posted by fivemack

"not liking the newest version of the Torture Test since they feel it pushes their temps higher than normal"

I am not able rightly to apprehend the kind of confusion of ideas that could provoke such an attitude

Hah I know. Basically, I think the small FFT is the most intense torture test for CPU temps available, and many of them would rather have 95% stable +700mhz overclocks than 100% stable at something a bit lower, like +300mhz. They're attached to the validation the 20% overclock gives them.

In a way, I actually think its ok. After all, most of these hardware boys are really just pushing frames in first person shooters. Their computers don't actually need to be rock solid. But for us doing accurate math distributed computing it seems a very different story.

I got a somewhat stable overclock myself @ 4.3ghz. So I'm now dropping it to 4.2ghz (1.1v @ 66C load) to see if this lasts 36 hours on the Blend torture test.

[TheMawn]

Originally Posted by fivemack

"not liking the newest version of the Torture Test since they feel it pushes their temps higher than normal"

I am not able rightly to apprehend the kind of confusion of ideas that could provoke such an attitude

I think this is hilarious, too. The notion of "stable-for-what" has been lost to most. Some people used to run some heavy photo or video rendering job to stress-test their computer. People used to just play their damn game and see if it crashed. I still see the occasional guy who knows what he's doing suggesting that at the end of the day, you stability test your computer with whatever it is you're going to actually do with it.

Back when overclocking was a thing only for the geeks who are to geeks what geeks are to normal people, some actual effort went into it. Now there's a freaking button you can press and software does it for you! If the overclocking gets easier, all people want is for the stability test to get easier along with it. I'm sure once at least once per week on some of the overclockers' forums there's a new thread created by someone asking what the ultimate stability test software is (without previously looking it up on their own either because they're that kind of person) so that they can hit 'Start' and get a "Stable" or "Not Stable" answer in 30 seconds.

Simon got it pretty much on the dot: there's that sense of entitlement that's got everyone feeling that it's their right to get a 30% overclock and that any less is abuse. Some people complained that Intel's new 8-core monster is a slap in the face because its stock clocks are low and it overclocks poorly. Like Intel just sets stock clocks 1 GHz below normal just to give us the pleasure of setting our processors' "real" power ourselves.

[simon389]

One person at Overclock.net says I chose the wrong processor. He says the ivy bridge would have been better:

"4820k is an enthusiast grade Ivy Bridge-E and there are 6 and 8 core processors that run 12 and 16 threads ."

How is my 4690k better than the 4820k?

[Mark Rose]

Advantages: Lower power consumption; AVX 2 and FMA3, which make mprime/Prime95 run faster; and integrated GPU.

Disadvantages: Less RAM bandwidth per core.

[simon389]

Originally Posted by Mark Rose

Advantages: Lower power consumption; AVX 2 and FMA3, which make mprime/Prime95 run faster; and integrated GPU.

Disadvantages: Less RAM bandwidth per core.

But if I had a single computer using Prime95, having the latest 4690k is a faster computer than the fastest 4820k that has 2-4 additional cores?

Basically, for a dedicated Prime95 computer, did I purchase the right processor?

[VBCurtis]

Originally Posted by simon389

But if I had a single computer using Prime95, having the latest 4690k is a faster computer than the fastest 4820k that has 2-4 additional cores?

Basically, for a dedicated Prime95 computer, did I purchase the right processor?

Since you can get 2 4690k's for the price of one 4820k, yes. Asking "what does the most work?" needs a price cap, else we can aim you toward some 16-core Xeons for megabucks. If you suffer production envy toward someone's $1200 Haswell-E system, buy a second i5 barebones and know you saved money on two machines vs my one while doing more work too.

[axn]

Originally Posted by simon389

But if I had a single computer using Prime95, having the latest 4690k is a faster computer than the fastest 4820k that has 2-4 additional cores?

Basically, for a dedicated Prime95 computer, did I purchase the right processor?

4820K is a quad-core processor. So is 4690k.
4820K has quad-channel memory. 4690k doesn't.
4820K is 130W TDP. 4690k is 88W.

All of that means that both processors offer similar performance in P95, but 4690K is lot cheaper to buy and run.

http://ark.intel.com/compare/80811,77781

[LaurV]

Originally Posted by VBCurtis

Since you can...

Just to add to it, @simon: which additional cores? for P95 work, logical cores (HT) don't count. P95 is very optimized, taking full advantage of a physical core, you can't use it for a second task with the same priority.

[VBCurtis]

Originally Posted by axn

4820K is a quad-core processor. So is 4690k.
4820K has quad-channel memory. 4690k doesn't.
4820K is 130W TDP. 4690k is 88W.

All of that means that both processors offer similar performance in P95, but 4690K is lot cheaper to buy and run.

http://ark.intel.com/compare/80811,77781

Oops- my post referred to Haswell-E, which is the 5820 rather than the 4820. The price-performance tradeoff among 4 core CPUs is heavily tilted toward the fastest i5 available at any given time. You chose well.

[simon389]

Just to follow up on this whole process, I could not get a stable torture test at DDR 2800 at any speed (including stock). I asked the folks at overclock.net for some help, and one of them gave a great piece of advice, which is that these Haswells do not really play well with memory set so high. I dropped it to DDR2400 w/XMP and it was stable.

I then proceeded to run the P95 blend torture test at various overclocks. Despite a watercooled CPU that was at load 60-70C, the best I could get was 4.2Ghz. I then dropped the computer to 4.0Ghz @ 1.5v. It runs at 55C load (once and a while maxes at 59C).

So all of this work just to add a simple 100Mhz over the 4690K's 3.9Ghz and realize this CPU does not like DDR2800 :). No worries.

Anyways, it has reserved a bunch of LL doublechecks and each core is currently doing 50Ghz of work in 6 days. Is this good?

Second question: I set Prime95 to only get first time LL tests, but currently it is on doublechecks. Does this mean P95 needs my computer to prove its stable with doublechecks before it "trusts" me with first time LLs?

Thanks in advance,
Simon

[Prime95]

Have you tried DDR3-2666? It would provide a nice boost to throughput.

The DoubleChecks were probably reserved before changing the work type. No matter, as it would be a good idea to run some double-checks until you are thoroughly satisfied that you are stable.

Long-term I recommend going to the web site and selecting the CPUs web page. Click on your new CPU and set the "DC instead of LL percentage". I use 25, but 10 or 15 ought to be good enough. You'll occasionally run DCs and if you get too high a percentage of mismatches, you can take steps to improve stability. I also use that web page to set the option to email whenever the computer submits a suspicious result.