lets see some low volt OCs!!!

[grkM3]

I got bored with my sandy and Im now trying to see how stable I can get it with the least amounts of volts.

Here is my first try with avx enabled and 8 threads of linp.

1.336 volts for 4,600mhz

thats the best my chip will do scaling wise as it take 1.376 to get it to 4,700 mhz stable for 5 passes with 8 threads running and then I need 1.425 for 4,800.

I can boot windows and do everything without any stability problems with way less volts but its not intel burn test stable for even 1 second.If you can get 5 passes with 8 threads and have sp1 windows 7 with avx enabled your system is pretty much rock solid for daily use.

 

[Idontcare]

You need to be aware that stable does mean fast.

What I mean is that Sandy Bridge microarchitecture has some ECC checks in it that are designed to keep the chip from causing BSOD and so on at the expense of redoing calcs that show internal inconsistencies.

You can prove this to yourself with Linx. If you are fully LinX stable and providing enough volts to the CPU that ECC is not forcing calcs to be redone then you will see a max Gflops that seems reasonable (~102-104GFlops for 2600K at 3.8GHz on 4 threads).

As you decrease voltage, going for that "low voltage but stable OC" the GFlops in linpack will actually start to decline, but no errors are detected by LinX itself because the errors are being caught within the chip itself.

(this is also why some folks erronously think that LinX GFlops scaling dies off as you go to higher clocks, its really just the case where their OC itself is actually not robust)

So for a challenge like your thread you need to better define the rules of the game.

 

[grkM3]

As long as you can get 5 passes of the latest intel burn test with 8 threads and windows sp1 you are good to post.

Its funny you mention the gflops droping as I get way higher with 4 threads and hyperthreading off.

My 2500k gets higher numbers ect.

On the linx side my setup will lock up if its to low and I have never seen the numbers go down with less volts.

Hyperthreading kills my numbers at stock or overclocked

 

[blackened23]

Your desktop is so boring need more interesting stuff going on there!

 

[Idontcare]

On the linx side my setup will lock up if its to low and I have never seen the numbers go down with less volts.

It may be the case for you then that you've always been operating in the Vcc-starved regime of the curve, never having given your chip enough voltage to actually get above the ECC checks that reduce performance.

At 4.5 GHz, running 4 threads on my 2600K, here's my GFlops versus Vcc (measured with multi-meter):

Vcc........GFlops
1.190V....N/A (system BSOD's during LinX)
1.293V....N/A (LinX stops on error detected)
1.325V....95 GFlops
1.329V....102 GFlops
1.334V....108.2 GFlops
1.339V....116.6 GFlops
​

Voltages above 1.339Vcc at 4.5GHz yield identical GFlops.

Note I am "100% stable" at 1.325V, I can run IBT for 12 hrs and not crash or BSOD, I also don't crash with any other apps at that voltage.

But I'm clearly forcing the chip to operate in an unoptimal environment, the ECC is kicking in to take care of the internal inconsistencies at the expense of requiring the computations to be redone periodically (thus lowering IPC).

This is no different than what folks found with OC'ing Nvidia's ECC enabled GPU's, you can OC them to the point where they aren't crashing but they also aren't performing.

So you have to decide what you really want from your OC when minimizing the lower voltages. If stability is your only metric of concern then you will find the Sandy Bridge architecture to be quite accommodating, but if you are OC'ing because you want more performance to come from the OC then you need to put in a little more elbow grease and do your due diligence to confirm you aren't Vcc starving the chip.

 

[Shmee]

4.6 at 1.336 sounds pretty good to me lol. I wish I had really good chips. Though my Q6600 was pretty good.

 

[VirtualLarry]

It may be the case for you then that you've always been operating in the Vcc-starved regime of the curve, never having given your chip enough voltage to actually get above the ECC checks that reduce performance.

At 4.5 GHz, running 4 threads on my 2600K, here's my GFlops versus Vcc (measured with multi-meter):

Vcc........GFlops
1.190V....N/A (system BSOD's during LinX)
1.293V....N/A (LinX stops on error detected)
1.325V....95 GFlops
1.329V....102 GFlops
1.334V....108.2 GFlops
1.339V....116.6 GFlops
​

Voltages above 1.339Vcc at 4.5GHz yield identical GFlops.

Very insightful, IDC, I had no idea that SB behaved that way.

That sort of throws a wrench into the traditional style of "increase freqency, then increase vcore until stable" overclocking.

Btw, I somehow doubt that SB is actually re-doing calculations, what I suspect is that the cache (which is protected by ECC), is causing delays.

At least, I cannot see a way that SB could double-check the outputs of say, divider logic, without having a second set of silicon to check the results with.

 

[grkM3]

It may be the case for you then that you've always been operating in the Vcc-starved regime of the curve, never having given your chip enough voltage to actually get above the ECC checks that reduce performance.

At 4.5 GHz, running 4 threads on my 2600K, here's my GFlops versus Vcc (measured with multi-meter):

Vcc........GFlops
1.190V....N/A (system BSOD's during LinX)
1.293V....N/A (LinX stops on error detected)
1.325V....95 GFlops
1.329V....102 GFlops
1.334V....108.2 GFlops
1.339V....116.6 GFlops
​

Voltages above 1.339Vcc at 4.5GHz yield identical GFlops.

Note I am "100% stable" at 1.325V, I can run IBT for 12 hrs and not crash or BSOD, I also don't crash with any other apps at that voltage.

But I'm clearly forcing the chip to operate in an unoptimal environment, the ECC is kicking in to take care of the internal inconsistencies at the expense of requiring the computations to be redone periodically (thus lowering IPC).

This is no different than what folks found with OC'ing Nvidia's ECC enabled GPU's, you can OC them to the point where they aren't crashing but they also aren't performing.

So you have to decide what you really want from your OC when minimizing the lower voltages. If stability is your only metric of concern then you will find the Sandy Bridge architecture to be quite accommodating, but if you are OC'ing because you want more performance to come from the OC then you need to put in a little more elbow grease and do your due diligence to confirm you aren't Vcc starving the chip.

Its because you have hyperthreading enabled and your 2600k is trying to spread 4 threads across 8 cores.

try disabling hyper and see how much higher your chip will run at 4 threads.

I just ran 4.8ghz with latest intel burn test and got 122gflops with 4 threads and 126gflops with only turning hyper threading off.You are comparing linx64 number to my intel burn test.I have never seen my numbers change with voltage and I will run linx to see if it makes a difference.

changing memory size will also boost your gflops and I used the base min of 1 gb on my run.

 

[grkM3]

Here is my linx runs with 1.360 and 1.328 @4.6 with 4 threads and they are pretty much exactly the same.

and here is the same as above with the only difference is hyperthreading is on in the bios.

and here is the same with 8 threads,you can see how hyperthreading is killing performance because its seeing 8 real coded threads and cant use them all up.Hyperthreading is good for poor coded apps but once you throw 8 threads at it it chokes up.If I had a real 8 core chip with hypo off the gflops would be well over 200 at this mhz

 

[Sunburn74]

You know I have a 2600k at 4.4ghz and hyperthread enabled, but can't seem to score more than around 50ish gflops on linpack... however on other benches like cinebench etc it scores appropriately

can anyone explain whats going on?

 

[grkM3]

You know I have a 2600k at 4.4ghz and hyperthread enabled, but can't seem to score more than around 50ish gflops on linpack... however on other benches like cinebench etc it scores appropriately

can anyone explain whats going on?

try running it at 4 threads and shutting hyper off.You have to also make sure the version you have supports avx.

I should also note my ram is running at 2133mhz with tight timings.

 

[Sunburn74]

whats avx? I have the o.64 version of linx. My ram runs at 1600mhz with 999-24 timings

Best run I've managed is like 54glops which is not even close to your 76 at identical settings in linx...

 

[grkM3]

whats avx? I have the o.64 version of linx. My ram runs at 1600mhz with 999-24 timings

Best run I've managed is like 54glops which is not even close to your 76 at identical settings in linx...

Do you have windows 7 with sp1 installed? avx is an instuction set from intel that boosts gflops big time.

On my 75gflop run I have the chip at 4.6ghz with hyperthreading on and 8 threads with 4gb of ram selected(I have 8gb total)so make sure you checked 4gb becasue choosing less ram will show lower results but will be less stress on the cpu.

I also tested wil more and less volts ans I havnt seen what idontcare was saying with the gflops droping.I ran the chip at 1.45 volts and 1.320 and it scored the same,any lower at those mhz the system would crash.My ram is running 2133 at 7-9-7 1t so it adds a few flops.I get about 26000/sec in aida 64

 

[Sunburn74]

yeah i have win7 with sp1 installed.

I picked 4gb of ram. Still only 54 ish. Are you running in safe mode or something? Turning off power features?

 

[Mallibu]

yeah i have win7 with sp1 installed.

I picked 4gb of ram. Still only 54 ish. Are you running in safe mode or something? Turning off power features?

You need a LinX with AVX support. Just google it and you'll find many links to download...the score you have is normal because your current program doesn't use AVX

edit. Here you go: http://gigaflopd.com/downloads/linx/

 

[blackened23]

what cinebench score do you get with that setup OP

 

[nyker96]

It may be the case for you then that you've always been operating in the Vcc-starved regime of the curve, never having given your chip enough voltage to actually get above the ECC checks that reduce performance.

At 4.5 GHz, running 4 threads on my 2600K, here's my GFlops versus Vcc (measured with multi-meter):

Vcc........GFlops
1.190V....N/A (system BSOD's during LinX)
1.293V....N/A (LinX stops on error detected)
1.325V....95 GFlops
1.329V....102 GFlops
1.334V....108.2 GFlops
1.339V....116.6 GFlops
​

Voltages above 1.339Vcc at 4.5GHz yield identical GFlops.

Note I am "100% stable" at 1.325V, I can run IBT for 12 hrs and not crash or BSOD, I also don't crash with any other apps at that voltage.

But I'm clearly forcing the chip to operate in an unoptimal environment, the ECC is kicking in to take care of the internal inconsistencies at the expense of requiring the computations to be redone periodically (thus lowering IPC).

This is no different than what folks found with OC'ing Nvidia's ECC enabled GPU's, you can OC them to the point where they aren't crashing but they also aren't performing.

So you have to decide what you really want from your OC when minimizing the lower voltages. If stability is your only metric of concern then you will find the Sandy Bridge architecture to be quite accommodating, but if you are OC'ing because you want more performance to come from the OC then you need to put in a little more elbow grease and do your due diligence to confirm you aren't Vcc starving the chip.

very interesting indeed, i did not know sandy do this at all!! great info!

 

[grkM3]

very interesting indeed, i did not know sandy do this at all!! great info!

Just a heads up I didn't see any drop or gain with lower of higher volts and its also the fist time iv heard it mentioned also.

I get 9.7 in cine but that's my 24/7 run.

I'm sure I can do a lot higher at 5.2ghz and I did some test last night trying 1333 vs 2133 and I gained about as much as a 200mhz cpu overclock would do just from running faster ram settings.