Putting together final overclock

[Stealth3si]

The image below does not show the CPU Core Temps but am guessing the Cpu Core Temps to be about 10c or so below the CPU Temp and that should be very good at 4.0Ghz with not too great of a Vcore applied. Looks like a pretty good overclock actually.

That's right, the core temp sensors are borked when the CPU is unlocked @ 6 cores.

Core temps should be 15C degrees lower actually according to my CPU-locked load test @ 3.0Ghz:

That would make core temps in this case to be about 37C-40C.

It should also be similar when the CPU is locked @ 4 cores running the same temps.

That is excellent! To be able to get to 4.0 ghz on that low of core voltage with all the cores unlocked is not common. You have a bit of a cherry CPU there I would say. I could not get mine stable on six cores past 3.9ghz without exceeding 1.5 vcore by a considerable amount.

Thank you!

What you should try now is to see how far you can overclock it as a four core. More than likely it will go higher than 4.0ghz as a four core and since few applications can take advantage of six cores anyway you might find that actually gives you better performance than running it at six cores. You would see that to be true in games, for instance where a higher clock speed will give better performance that more cores at a lower clock speed.

Can I try to overclock the NB to 3000+mhz and RAM to 1700+mhz and tighten the timings before I see how far I can overclock at 4 cores? And any general tips on how to do this?

I want to be able to achieve this with perfect stability in Prime95 Blend for 24+ hours!

 

[RGone]

Can I try to overclock the NB to 3000+mhz and RAM to 1700+mhz and tighten the timings before I see how far I can overclock at 4 cores?

I want to be able to achieve this with perfect stability in Prime95 Blend for 24+ hours!

Sure do whatever you want to with it. All we were doing is making suggestions to get up and going. Beyond that we don't have a dog in the race. It is your system, run it how you wish.

'trents' mentioned that 6 core utilization is not that great yet in most games and that more Mhz with 4 cores, might actually work better for gaming, however your clock and use of the system is strictly up to yourself. Have fun and good luck, since we do not have to sign off on anything.

 

[trents]

RGone said what I was going to say in that it's your stuff, do what you want with it.

I have doubts about being stable with the ram at 1700 mhz, especially with tighter timings but, hey, if you want to try it, go right ahead. The limiting factor there may not only be the capability of the ram but also of the CPU ICM (Integrated Memory Controller) which is only rated for 1333 mhz. Also, benchmarking has shown that there is almost no performance gain to be had from running ram above 1333 mhz. Current CPUs, whether Intel or AMD, are not starved for memory bandwidth even at 1333.

Yes, the differential between CPU socket temp and core temp remains pretty constant whether the core are locked or unlocked or whether the CPU is stock or overclocked so the core temp can be calculated from the CPU temp when core are unlocked.

 

[Stealth3si]

Sure do whatever you want to with it. All we were doing is making suggestions to get up and going. Beyond that we don't have a dog in the race. It is your system, run it how you wish.

'trents' mentioned that 6 core utilization is not that great yet in most games and that more Mhz with 4 cores, might actually work better for gaming, however your clock and use of the system is strictly up to yourself. Have fun and good luck, since we do not have to sign off on anything.

RGone said what I was going to say in that it's your stuff, do what you want with it.

I have doubts about being stable with the ram at 1700 mhz, especially with tighter timings but, hey, if you want to try it, go right ahead. The limiting factor there may not only be the capability of the ram but also of the CPU ICM (Integrated Memory Controller) which is only rated for 1333 mhz. Also, benchmarking has shown that there is almost no performance gain to be had from running ram above 1333 mhz. Current CPUs, whether Intel or AMD, are not starved for memory bandwidth even at 1333.

Yes, the differential between CPU socket temp and core temp remains pretty constant whether the core are locked or unlocked or whether the CPU is stock or overclocked so the core temp can be calculated from the CPU temp when core are unlocked.

Thanks for the 'jumpstart'!

If I have any further questions or issues I'll post them in this thread. But I think I can take it from here without additional help.

Otherwise, I'll detail my final 6-core & 4-core overclocks once I get them perfectly stable by next week or more.

 

[trents]

You can also disable a core and go with a five core. Sometimes that's a good compromise that gives both a lot of cores and a high overclock.

 

[Stealth3si]

Final 6-core overclock!




Other than what's in the pic:

SB voltage = 1.20V

The following are enabled:

CPU LLC
CPU/NB LLC
Cool N' Quiet
C1E State
Q-Fan CPU/Chassis Function: Manual
Min. Processor State: 0% / Max Processor State: 100%

Currently working on 4-core overclock. Stay tuned....

 

[trents]

Hey, pretty good! Congrats! How long did you test it with Prime?

Keep in mind the Thuban core sensor is poorly calibrated on the cool side. You should probably add about 10c to those core temps.

 

[Stealth3si]

Hey, pretty good! Congrats!

thx!

How long did you test it with Prime?

34 hours! the pic shows it!

Keep in mind the Thuban core sensor is poorly calibrated on the cool side. You should probably add about 10c to those core temps.

okay. Core temps should still be under 55C degree on a relatively hot summer day.

 

[RGone]

You pretty golden it seems. Congrats to you. Maybe now you can even use the computer. Hehehe. Getting it done always seems a good thing when I got to that point.

RGone...ster.

 

[BeepBeep2]

I hate to burst a bubble but the 3:10 ram divider is faster.
If you can get near those settings using the lower ram divider and a higher HT Reference Clock you will have a little better performance.

Also, tRAS by definition should be CL + tRCD + tRP, and tRC should be no less than tRCD + tRAS.

As for what trents said, your temps are really fine. The readout in HWMonitor is the motherboard sensor, when you unlock cores on a Phenom II CPU the temperatures no longer readout from the CPU sensor.

 

[caddi daddi]

with his room temp going to 79f (79c stated) i would like to see a little more temp headroom built in to assure he stays below his 62c stated limit.

 

[BeepBeep2]

Unless he is running Prime95 24/7 I think he will be fine for daily use.

Usually you can run Prime for an hour or so and temps will continue to climb a few degrees in the hours after for a bit and settle back down depending on the test being run.

 

[caddi daddi]

he's got a good clock now.
what i have found with my chassis setup program that i run under linux is that i have to add two clicks of the + key to vcore from what i wind up with in prime95 undr windows to keep it from crashing and this uses up my temp overhead if i don't keep it wide.

 

[trents]

. . . As for what trents said, your temps are really fine. The readout in HWMonitor is the motherboard sensor, when you unlock cores on a Phenom II CPU the temperatures no longer readout from the CPU sensor.

By "CPU sensor" I assume you mean he core temp sensor which is built into the CPU die, as opposed to the CPU socket temp sensor. True, the core temp reading is no longer functional but he was trying to estimate it based on the delta between the socket temp reading and the core temp reading when it was locked.

 

[Stealth3si]

I hate to burst a bubble but the 3:10 ram divider is faster.
If you can get near those settings using the lower ram divider and a higher HT Reference Clock you will have a little better performance.

You're likely referring to #5 in the chart above in post #1...

to get near those settings I need to push beyond 1.40V on CPU/NB

Also, tRAS by definition should be CL + tRCD + tRP, and tRC should be no less than tRCD + tRAS.

I know but 16 is going to give better performance than 27.

i did some tests to show this.

tRAS = 16, tRC = 25

avg read
10775.7 MB/s

avg write
9090.8 MB/s

avg copy
12230.4 MB/s

avg latency
38.21 ns

------------------------
tRAS = 27, tRC = 36

avg read
10787.9 MB/s

avg write
9075.5 MB/s

avg copy
12307.2 MB/s

avg latency
38.76 ns

The ones in 'bold' have better results.

i do realize that tRAS 16 could introduce data corruption, though, if it is unstable...

 

[Stealth3si]

Here's the final 4-core overclock!



Other than what's in the pic:

HT Voltage = 1.25V
SB voltage = 1.20V

The following are enabled:

CPU LLC
CPU/NB LLC
Cool N' Quiet
C1E State
Q-Fan CPU/Chassis Function: Manual
Min. Processor State: 0% / Max Processor State: 100%