Comparing three 2600K processors - Voltage vs. Temperature vs. Stability

[mrjoltcola]

Just a little fun.

I've got three 2600Ks, two of which will go 5.0Ghz stable under Prime95 with Hyperthreading enabled, Maximus Extreme-Z with H100 cooler in a Corsair 600T case with fans in exhaust configuration, H100 pump on middle speed.

I took voltage and stability curves, and they max at:
Proc1: 5.0Ghz @ 1.490v / 67C - Batch# 3115B989 Costa Rica
Proc2: 5.0Ghz @ 1.440v / 72C - Batch# L132B629 Costa Rica (Reached stable 5.0 so did not increase... yet)
Proc3: 4.8Ghz @ 1.405v / 59C - Batch# L107B422 Costa Rica (Not stable at 4.9 even up to 1.500v unless I disable Hyperthreading)

Note how Proc1 runs cooler at 5Ghz, but runs out of stability steam faster. Stability is defined by passing multiple Prime95 4t & 8t passes, and running for at least an hour with no errors. I stop tuning when I either hit 5.0Ghz or 1.500v, whichever comes first. I expect Proc2 has more in its legs given that voltage.

I plan to post the curves overlaid on a single graph as soon as I get the data into a good graph maker, in the meantime, what are thoughts?

What is interesting to me:
1) Proc2 runs hotter, but stabler at lower voltage.
2) Given the above values, what is more detrimental to the Proc2, more voltage or more temp?

EDIT : Proc2 is stable at 5.1Ghz @ 1.460v now but 5.0Ghz is better for purposes of comparison to Proc1. I'm sure Proc3 will go stable at 5.0, but I will not overvolt it just to find out. I'd guess 1.550v would be needed.

 

[lehtv]

I agree, 1) is interesting. The 4 deg difference is hard to attribute to any single experimental error... but could it be a combination of the following: difference in how the thermal paste was applied, difference in ambient temperature, or the length of the stress test, or fan speed? Also, what were the core-specific temps?

 

[mrjoltcola]

Good point. I was thinking of reapplying paste and reseating cooler. I don't like the look of the contact "patch" on the CPU when I remove the cooler, it doesn't press all of the thermal paste out to the edges. Almost like the H100 might be concave or the CPU heat spreader is convex.

I'm not brave enough to try lapping the best processor I have (like IDC's thread of the year -> http://forums.anandtech.com/showthread.php?t=2181357 )

I didn't measure individual cores - 🙁 - and a long test to repeat with 3 chips. I may do it this weekend when I'm bored.

 

[Rvenger]

Wow those temps are interesting. This goes to show that no two heatspreaders are manufactured the same. Where were these 3 chips made at? I am wondering if the different locations have anything to do with this.


Temps would be more harmful for proc 2 but still well within the safe zone. I would say that since you are under 1.45v that is safe enough for a 24/7 overclock.

I would try reseating the heatsink to see if it would cut down below 70c.

 

[mrjoltcola]

Where were these 3 chips made at? I am wondering if the different locations have anything to do with this.

See original post, I have edited and added the Batch# and location.

I'll adjust tomorrow and rerun the 5.0Ghz test on Proc2.

 

[Rvenger]

My 2600k is a Costa Rica as well but my 2500k was Malaysia. They overclocked about the same but I think my 2600k overclocks just like your 3rd processor. My batch# is 3117B257 I did not try for 5.0ghz with this chip though, it seems like its hungry for voltage at the lower speeds. Is your processor 1 a newer one?

 

[mrjoltcola]

I bought Proc1 and Proc2 this weekend. Proc3 is from Newegg about 4 weeks ago. Does Intel use sequentially increasing batch numbers?

Luckily two family members went in for the 2600K Microcenter deal, so I've been able to test 4 processors this month. I'll add a 2700K to the mix when it comes.

Just want to point out, I don't run my rig at the max stable clocks. I run it at whatever it will do at approx 1.35v, and keep the high OC profile for fun. Proc2 is now replacing Proc3 in my rig, and it does 4.8 @ 1.360, which seems good enough for 24/7 even for a conservative.

As soon as I find a graph program I like, I'll post up the curves.

 

[blackened23]

Interesting post, +rep to you

 

[VirtualLarry]

It has been my experience that good clockers are leaky, and vice-versa. They clock higher with less volts, but run hotter. The Q6600 in my sig is a good example of that. Runs at 3.6 with very low volts (most people require something like 1.45v to get 3.6), but temps are horrible, 90C with a Tuniq Tower under OCCT 64-bit linpack sometimes.
Actual usage temps are less, so I just leave it at that.

 

[mrjoltcola]

I suppose a more useful chart may be for me to re-run comparison at highest common frequency and lowest common voltage. Stability is mostly about the weakest link among a billion transistors, so it really doesn't deserve an axis on the graph. I'll take runs with all chips @ 4.8Ghz and 1.415v and see what temps we get.

 

[Rvenger]

My 2500k was 12c cooler than my 2600k. This thing hits 70c under load @ 4.4ghz (1.34v) with a CM Hyper 212+

 

[tweakboy]

Wow proc 2 only needs 1.435v to get a stable 5Ghz. WOW You got a champ there. 1.490v is way high, 71c is ok , but Intel recommends 60's c as nominal so 71c although hot is not that bad. These CPU's will last forever even if the temps are always high.

Id like to put that 5Ghz piece under water and see if I can get 5.5Ghz with HT. thx gl I would get 20's idle and 40's c full load.

Just a little fun.

I've got three 2600Ks, two of which will go 5.0Ghz stable under Prime95 with Hyperthreading enabled, Maximus Extreme-Z with H100 cooler.

I took voltage and stability curves, and they max at:
Proc1: 5.0Ghz @ 1.490v / 67C - Batch# 3115B989 Costa Rica
Proc2: 5.0Ghz @ 1.435v / 71C - Batch# L132B629 Costa Rica (Reached stable 5.0 so did not increase... yet)
Proc3: 4.8Ghz @ 1.415v / 59C - Batch# L107B422 Costa Rica (Not stable at 4.9 even up to 1.500v unless I disable Hyperthreading)

Note how Proc1 runs cooler at 5Ghz, but runs out of stability steam faster. Stability is defined by passing multiple Prime95 4t & 8t passes, and running for at least an hour with no errors. I stop tuning when I either hit 5.0Ghz or 1.500v, whichever comes first. I expect Proc2 has more in its legs given that voltage.

I plan to post the curves overlaid on a single graph as soon as I get the data into a good graph maker, in the meantime, what are thoughts?

What is interesting to me:
1) Proc2 runs hotter, but stabler at lower voltage.
2) Given the above values, what is more detrimental to the Proc2, more voltage or more temp?

 

[mrjoltcola]

@Tweakboy - Yeh, I think it is a nice slab of silicon. I had to cherry pick from 4 processors to get it (luckily everyone in the family wants the 2600K combo deal from Microcenter, hehe).

Out of 4 total chips, I got 2 that went 5Ghz, and 2 that hit 4.8 with Hyperthreading. Both 4.8 chips will hit 4.9 stable without HTT. I've noticed a trend of about one multiplier and 4-5 degrees C with/without HTT.

As for my champ, it just went 5.1 @ 1.460v for last 30 minutes, and I can't crash it with Prime95, SuperPI, browsing, anything else. Typing this note at 5.1 with stuff running. Odd that it takes 1.440 @ 5.0, but only requires another 0.020v for 5.1Ghz, so it seems the curve isn't exponential so far for this one processor. I'd really like to see what it'll do at 1.50v but I'm not keen to put that much through it. Also 5.1 finally put the temp to 75C under load, which is above Intel Tcase of 72.6C for this proc, so time to back off.

What cooler do you recommend above the H100? I think I am going to dedicate a separate gaming rig to this chip so I can get a little crazy, and not put my work at risk. 🙂

 

[mrjoltcola]

My 2500k was 12c cooler than my 2600k. This thing hits 70c under load @ 4.4ghz (1.34v) with a CM Hyper 212+

SLK, a good portion of that is probably hyperthreading. I have a couple of 2600Ks that differ 6-7C with hyperthreading disabled/enabled. The gap gets wider the higher the clock. At 4.6Ghz I note about a 3-4C difference. A 2600K with no HTT is nearly identical to a 2500K as far as active silicon.