Further 2600k overclocked performance tweaks...

DamoTourettes

DamoTourettes

DamoTourettes

Associate
Joined
14 Dec 2010
Posts
495
Hi everyone,

Some of you might be familiar with a project I'm working on for work - I'm presently in the process of building ~50 overclocked machines for a datacentre build out with the following components:-

Asus Maximus IV Extreme mobo
2600k CPU
Noctua/crazy fan setup for cooling.

Basically I've managed to get couple of *totally stable* 2600k's running at 4.9Ghz with around 1.355 Vcore, and temps at load are high 50's, low 60's. This kind of configuration I'm more than happy with - however as I've mentioned before, what I'm looking to do is to get the very highest OC I can out of the machines which I can run at 24/7 stable.

The specifics of what the servers will be used for doesn't require the following:-

Hyper Threading
Speed Step

I'm just wondering what any other tweaks are that anyone has that might allow me to go Mhz higher whilst keeping volts/temps low - I've heard that reducing CPU PLL is good in keeping temps down, however I don't really know enough about overclocking to start to looking at things other than Vcore. At the moment I've just got BCLK set to 100 (as I've read running it higher is bad for the IMC on the chip which in turn affects ultimate clock speeds), but I'm wondering if anyone can recommend any further tweaks they use in overclocking SB?

I've been working with the idea that for 24/7 use (which these servers will run), 1.380v is the max vcore I'll go to, and provided I don't bounce much over 70 deg C I'm happy with the temps - anyone with some pointers?

Cheers in advance...
 
Remember that for your single-threaded raw performance, it won't only be about cpu speeds but also RAM latencies. Use SuperPi to see how memory timings can affect your crunching numbers.

I wouldn't be too concerned about the volts; none of these machines is expected to have a long happy life is it? You will also find that some chips simply go further than others; your 4.9 settings may not work with all while some may go much higher.
 
Any reason for your employer buying £300 motherboards?

There are plenty of P67 boards for a little over £100 that will overclock a 2600K just as well.

Your employer could have saved £10,000.
 
Mike - RAM I've been running at 6-8-6-24 1T of late which has brought down speeds, however real-world our application runs with most of the binary pinged to memory on the chip - with 8MB we rarely go outside this... OS is on one core, and the remaining 3 have instances of our application on them.

Surveyor - there's no reason we can't use other P67 boards - you're totally right.
 
DamoTourettes said:
Yeah, well... there's the problem - we can't run water in a data centre environment...
Click to expand...

It certainly has been done before. There was a thread on here a while back where the OP wanted to run highly overclocked servers for trading purposes.

To be honest once you have your PLL and VTT optimised then you just apply more vcore. As long as your temps stay under control then I don't see why you can't run more than 1.38vcore. All correct me if I'm wrong but I have yet to see any reports on XS of people seeing any serious degradation on these chips at sensible volts. I guess it depends on your scenario. Other than that, you may as above want to investigate faster ram?

Can you repost the details around your project?
 
beh* said:
It certainly has been done before. There was a thread on here a while back where the OP wanted to run highly overclocked servers for trading purposes.

To be honest once you have your PLL and VTT optimised then you just apply more vcore. As long as your temps stay under control then I don't see why you can't run more than 1.38vcore. All correct me if I'm wrong but I have yet to see any reports on XS of people seeing any serious degradation on these chips at sensible volts. I guess it depends on your scenario. Other than that, you may as above want to investigate faster ram?

Can you repost the details around your project?
Click to expand...

The guy that was running highly overclocked servers for trading purposes was me.

Can you help me with optimisation of PLL and VTT? I've got both set to "Auto" at the moment, and I'm wondering if further optimisation of them will allow me to reduce temps/vcore for the same OC or increase vcore for further overclocks.

As for RAM, as mentioned I'm running G.Skill RipJaws (2x2GB) @ 6-8-6-24 1T - 1600Mhz - 1.50v - do you think I could do better than that?
 
Great stuff - was checking that guide and they're saying that 1.40-1.425 vcore seems to be OK for 24/7 operation. Anyone running this kind of vcore with any comments? I've been running Intel Burn Test at 32 threads/Maximum for about 24 hours per run to prove stability and have had no problems with 4.8Ghz @ 1.355 vcore on a couple of chips, but according to the sticky in the CPU forum here, 1.380v seems to be the absolute top on SB. Would I be right in surmising that if I'm keeping temps in the 70's 1.40 - 1.425 vcore is 24/7 stable?
 
Quick Note regarding Voltage Scaling – Internal binning of both D1 and D2 parts we discovered consistent voltage scaling patterns.
1. For K series parts, the stock voltage supplied will allow for consistent overclocking generally up to a multiplier of 43x. There is potential for the multi to be raised to 44x depending on the load induced. This default voltage range be approx 1.240 to 1.260 under load.
2. Increased range between 44 to 47x multipliers will generally require a voltage range between 1.30 to 1.375V with an LLC recommended setting of high to ultra high.
3. Increasing the range between 48 to 50x multiplier will generally require a voltage range between 1.40 to 1.500 with a LLC recommended setting of ultra high.
4. Increased range between 50 to 52 (52 generally considered peak max multiplier except for rare 54x parts) will generally require a CPU voltage range between 1.515 to 1.535V with LLC at Ultra High and potential fine adjustments to the CPU skew range.
Click to expand...

That sums it up for me. Key points are: use speedstep, use offset voltage (this reduces vdroop on my board), use PLL overvoltage, use LLC and that you will need > 1.4 to reach stability at >4.8 (with hyperthreading and all cores enabled).
 
DamoTourettes said:
Great stuff - was checking that guide and they're saying that 1.40-1.425 vcore seems to be OK for 24/7 operation. Anyone running this kind of vcore with any comments? I've been running Intel Burn Test at 32 threads/Maximum for about 24 hours per run to prove stability and have had no problems with 4.8Ghz @ 1.355 vcore on a couple of chips, but according to the sticky in the CPU forum here, 1.380v seems to be the absolute top on SB. Would I be right in surmising that if I'm keeping temps in the 70's 1.40 - 1.425 vcore is 24/7 stable?
Click to expand...

I'm running that kind of vcore and to date haven't seen any evidence that those kind of volts cause CPU degradation especially given that my temps never get above 60 and I use speed step and offset voltage. With all cores and hyperthreading enabled I would be very surprised if you could get 8 hours Prime Blend stability. I was Lynx/IBT stable at 1.375 @ 4.8 but Prime Blend found instability in minutes.

As above it would be good if you could explain more about your requirements...
 
Last edited:
beh* said:
I'm running that kind of vcore and to date haven't seen any evidence that those kind of volts cause CPU degradation especially given that my temps never get above 60 and I use speed step and offset voltage. With all cores and hyperthreading enabled I would be very surprised if you could get 8 hours Prime Blend stability. I was Lynx/IBT stable at 1.375 @ 4.8 but Prime Blend found instability in minutes.

As above it would be good if you could explain more about your requirements...
Click to expand...

Thanks for your help thus far. The requirements that we have are (as per my other thread on the subject), highest possible 24/7 clock speed, no speed step, no hyperthreading. As we're implementing into a data centre environment we've been using server cases with BIG fans on them as noise isn't a problem, so we're basically just strapping big coolers to these boxes and going at it that way. So what you're saying is provided I can keep temps down around mid 70's then pumping more vcore (to around 1.4 - 1.425) whatever top end clock I get should be fine? Obviously with a water setup yourself the low 60's are a much better temp range to be working within...
 
This is what I run mine at although you have to remember that is definitely not 24/7. I'm lucky if I get to use my rig at load for more than 2 hours a week at present. Most of the time it's sat at 1600Mhz at very low volts lol. All I can say is that at 1.42 its rock solid at 4.8Ghz after 8 hours of Prime Blend which to be is the best test of stability. You should also use 1 hour of Lynx / IBT as a secondary test.

Perhaps ask this question in the folding section as I assume some of those nutters might be running at 4.8+ 24/7 loaded. It would be interesting to see if they have experienced any degradation.
 
You must log in or register to reply here.
Back
Top Bottom