i5 750 @ 4ghz

[netxzero64]

Good day to everyone.

I have an intel core i5 750 processer under asus p7p55d evo board

My questions is that what voltage should I set my processor to have a stable 4ghz OC?

currently my IMC voltage is set at 1.225v and memory at 1.625v @ 1950mhz which is a bit high as 1.6v should be at 2133mhz. Processor speed at 3.7ghz

Any suggestions would be appreciated.

 

[Saylick]

Hm... It's been a while since I looked into OCing my 750 but if I'm not mistaken, it looks like you have your BLCK set to 195 with a memory multiplier of 10 and a CPU multiplier of 19, hence 1950 MHz on the RAM and 3.705 GHz on the CPU.

Try raising your BCLK to 200 and set your CPU multiplier to 20. Turn off Turbo Boost or else the chip will turbo to 4.6 GHz when you only have 1 core loaded, at which your chip will probably be unstable.

You can try something around 1.2V to 1.25V Vcc for 4 GHz. You can always start with 1.25V to get the system to boot, and then after a couple of quick passes of LinX or IBT, you can slowly lower the voltage until you reach instability.
I think you should be able to lower your Vtt (or as you refer to it, the IMC voltage) as well. I don't recall ever needing a Vtt higher than 1.17V - 1.19V when I pushed my BLCK to 200.

 

[netxzero64]

I see.. I am not so familiar with the terminologies as this is the first time that I have an intel processor as my last one was a pentium 200mhz mmx.

Sometimes, with the memory speed, I get BSOD and I think this is because of the limitations of the memory controller of the 750. Correct me if I'm wrong on this sir.

I only entered bios last night and tried to play with the asus auto tune and it gave me auto on all voltages with a 3.705ghz as you guessed it and a 1560mhz on the memory speed. turbo is automatically turned off once I play with the multiplier for the processor.

I appreciate your comment sir and will try to play with that setting you suggested tonight

any more suggestions will be accepted and put to the test.

 

[Saylick]

Again, it all comes down to tinkering with the settings. Just remember that everything is related to the BCLK.

CPU Freq = BCLK x CPU Multiplier (higher means you need higher Vcc)
RAM Freq = BCLK x Memory Multiplier (higher means you need higher RAM voltage)
QPI Freq = BCLK x QPI Multiplier (higher means you need higher Vtt)

If you are concerned about the memory and/or the memory controller being the bottleneck, you can try 190 BCLK and a 21x CPU multiplier. You don't get exactly 4 GHz, but it's close (3.99 Ghz to be exact). This way, your memory controller and memory run slower. I recommend leaving the QPI multiplier at 32x to keep the uncore clocked low; from what I understand, it's typically the uncore that bottlenecks the OC - most don't push their BCLK beyond 200.

 

[Deders]

I managed to get 3.8 out of mine with 1.272-1.288 (as read by CPUz). I could push it further but the power to performance ratio isn't worth it for me yet.

You'll want to make sure load line calibration (LLC) is enabled before you start overclocking, this will even out the electrical signal so it doesn't crash at higher frequencies.

I've found that with mine, I turned the BCLK to 180, and the multiplier to 19 (apparently this generation overclock better with odd multipliers) and let the turbo do the rest, it generally clocks up to 38 (21x multiplier) and happily clocks down to 1.625GHz when idle.

I also had to raise the VTT a little to help even out the difference in voltage between the CPU and the memory.

I believe the memory shouldn't be more than 0.5v away form the VTT or it will damage the memory controller.

 

[netxzero64]

I understand. I am not worrying about the bottleneck because I don't think that I have greater bottlenecks than what I have with my previous rig which is an AMD Athlon II X3 unlocked to X4 with DDR2 memory modules. So I am very happy to be able to go up the DDR3 ladder now hehe..

I will take note of the QPI multiplier as you have mentioned. What do you mean by the uncore clocked low? I am sorry. I am not used to such jargons nowadays.

And by the way, the speedstep is enabled in my settings. C1E or any other halt states are disabled but the processor still downclocks on idle modes.

 

[netxzero64]

so the VTT or IMC voltage should be almost the same with the ram? is that what are you trying to say? sorry. my level of understanding is not that good on this type of discussion.

 

[Deders]

The uncore is the part of the CPU that holds pretty much all of what used to be the northbridge on a motherboard. In the 750's case it's the memory and PCIe controllers.

As for VTT/IMC, voltages it should be somewhere between the CPU and the memory voltage, but not more than 0.5v away from the memory voltage.

I found that the after arriving on a cpu voltage, raising the VTT a couple of notches was enough to keep my computer stable for a 24 hour run of prime64. Otherwise it would only last 23 hours.

I'll check what its actually set to in the bios in a moment and update this post.

Edit: my voltages as set in the bios are:

CPU 1.251v
VTT 1.208v
MEM 1.6v

Every utility I've used to read them from within windows gives different results.

 

[Saylick]

With the 750, there's the core voltage (Vcc) and there's the uncore (which should be Vtt). Uncore includes memory controller, L3 cache, PCIe, and other I/O.

Courtesy of Anandtech:

"The moral of the story is in Phenom and later in Nehalem, the processor is divided into two parts. Intel named them the core and the un-core. The "core" of these multi-core processors is made up of each individual processor core and its associated private caches (L1/L2). The "uncore" refers to everything else: PCIe controller, memory controller, DMI/QPI and the L3 cache.

The uncore isn't as critical for performance but is made up of a ton of transistors; roughly 400 million in the case of Lynnfield/Bloomfield (more if you count the PCIe controller). In order to save power, Intel uses slower transistors that have lower leakage for the un-core. As a result, the un-core can't clock up as high as the core and runs at a lower multiplier." - Anandtech

The most important part is that last sentence. Because the memory controller frequency is determined by the uncore frequency (which is that QPI frequency I was talking about earlier), you ideally want to make sure it stays low since it can't clock as high as the core. In other words, you might be able to get a 50% overclock on the core, but you might not be able to get a 50% overclock on the uncore. Just as a heads up, the stock uncore frequency is 2.128 GHz (or 2.13 GHz if you round), which is 16x the BCLK. For some reason, my motherboard displays this as a 32x multiplier which is why I suggested it to you earlier. If you board is like mine, where 32x is in reality a 16x, then leave it at 32x. If your board is different than mine, i.e. 16x is in reality 16x, then by all means please ignore my previous recommendation and please proceed to use 16x.

I personally have my BCLK set to 160 in order to keep the uncore clocked as low as possible. This results in a uncore/QPI frequency of 2.56 GHz, which is stable at a Vtt of 1.15V. I used to go 180 BCLK (uncore freq = 2.88 GHz), and that required a Vtt of 1.17V. 200 BCLK would probably require something around 1.2V for Vtt, but I somewhat recall that a Vtt > 1.17V is outside of Intel's spec. No risk, no reward, I guess.

As for the people saying your Vtt should be within 0.5V of the RAM voltage, I personally have NOT run my RAM at any values exceeding my Vtt + 0.5V. At the same time, those people probably have good reason to recommend keeping both voltages within 0.5V of each other. Whatever you do, do NOT run your RAM voltage at the same Vtt or vice-versa. 1.5V Vtt would most certainly result in failure of some sort (possibly a fried chip), and 1.15V for RAM would probably crash the computer if run at stock frequency.

Go with start with the numbers Deders has.

EDIT: Ah, did not see Deders is running at 3.8 GHz. Nonetheless, the numbers are still very reasonable. Try his numbers but with the core voltage (Vcc) cranked up a few notches.

 

[Deders]

Interesting , i've been running my VTT at 1.208 for several years but like I said before without actually measuring it with a £500 multimeter It's hard to get an exact reading of what it really is set to from within windows. MSI's own control centre reads 1.195.

I've also got the option to knock the uncore down a couple of hundred MHz but I've never had to use it.

 

[biostud]

Not all will reach 4Ghz stable. I settled with 3.8Ghz, also because after that the power consumption starts to increase drastically, with very little performance benefit.

 

[netxzero64]

thank you gentlemen for those ideas.. I am still comprehending what Saylick is trying to say with the QPI frequency but I will get to it once I get home from work. I will check if there are any multipliers for the QPI on my ASUS P7P55D EVO. The only thing that I will try to check is the QPI and the VTT voltage. I have set it at around 1.225 gee.. hopefully nobody uses the computer or something may fry the chip.

So for all of you guys, what is the best frequency to run the i5 then? 4ghz? or 3.6ghz to 3.8ghz for performance/watt efficiency with slight increase in power consumption?

 

[Saylick]

So for all of you guys, what is the best frequency to run the i5 then? 4ghz? or 3.6ghz to 3.8ghz for performance/watt efficiency with slight increase in power consumption?

Check this out. I linked the efficiency page, but the rest of the article is useful, too.

Note that THG decided to run a dynamic OC for some of their configurations, which is different than the static OC I presume you are going for. The points marked with "Turbo" are all dynamic OCs. Those without are static OCs.

With that said, 3.6 GHz to 3.8 GHz seems like a good place for a static OC. If you really want to squeeze out every last bit of power efficiency, you can aim for the 3.2 GHz Turbo dynamic OC.

Lastly, here is the page where they posted their settings for each OC they tested.

 

[netxzero64]

Dynamic OC is the OC with turbo right? 3.2ghz increase through just BCLK correct?

 

[Saylick]

Dynamic OC is the OC with turbo right? 3.2ghz increase through just BCLK correct?

Yes, and yes. A dynamic OC for a 750 is simply nothing more than leaving Turbo mode ON, bumping up the BCLK, and changing voltages + memory speeds to make sure that the chip is stable across all the frequencies the chip may hit.

In my opinion, one of the few advantages a static OC has over a dynamic OC is that you don't have to worry about setting up a variable voltage for the chip, but if you can get beyond that, you are good to go. Some people, like myself, are lazy and just leave the voltage on Auto so that it adjusts itself. Others prefer to optimize to the max and manually determine how much the voltage should vary.

 

[netxzero64]

I see. I don't want to leave it on auto because if I OC it to reach 4ghz, the voltage is at 1.5v+ as done on a review at overclock3D.. so it generates more heat and power.. but I will try dynamic OC once I am able to get my hands on the seidon 120m.. will just bump it up until 3.6ghz for dynamic OC.

 

[netxzero64]

@saylick

Here's what I got from last night's observation.

195 BCLK is now somewhat unstable. I made the VTT/IMC at 1.225v on bios, mems at 1.6125v @ 1950mhz. VCC is on auto. Then I reset everything back to default and just increased the BCLK to 145 and everything is on auto then no problem since. Is it because of the memory controller or the board that is making the OC unstable?

I also looked on the QPI frequency as I have no option for QPI multiplier on bios. From what I remember, I only have 3 options to choose from the QPI frequency (auto, 6ghz+ and 7ghz+). I did not played with them as I am afraid that I may damage the board so I left it on auto.

 

[netxzero64]

The uncore is the part of the CPU that holds pretty much all of what used to be the northbridge on a motherboard. In the 750's case it's the memory and PCIe controllers.

As for VTT/IMC, voltages it should be somewhere between the CPU and the memory voltage, but not more than 0.5v away from the memory voltage.

I found that the after arriving on a cpu voltage, raising the VTT a couple of notches was enough to keep my computer stable for a 24 hour run of prime64. Otherwise it would only last 23 hours.

I'll check what its actually set to in the bios in a moment and update this post.

Edit: my voltages as set in the bios are:

CPU 1.251v
VTT 1.208v
MEM 1.6v

Every utility I've used to read them from within windows gives different results.

this is your 3.8ghz setting?

 

[Saylick]

Hm. I'd have to say it's likely the memory controller/uncore that is causing your chip to be unstable. For my 750, I was never quite able to reach 100% stability at 200 BCLK; it would still crash and revert to stock settings every once in a while. 180 BCLK worked if I raised Vtt high enough. Every chip is different.

As for the QPI, I'd agree with you and leave it on Auto as well.

OCing can be a pretty monotonous process. After weeks of testing various settings, I finally settled with the following:

167 BCLK
8x Memory Multiplier => 1336 MHz RAM 7-7-7-20
20x CPU Multiplier => Turbo ON, Base Freq @ 3.34 GHz, 4.008 GHz @ 1 core, 3.841 GHz @ 2 cores, 1.503 GHz @ Idle
Vcc = Auto, it fluctuates depending on load, ~1.15V at base freq, haven't checked the core freq. when only 1 core is loaded
Vtt = 1.15V
RAM Voltage = 1.58V

 

[Deders]

this is your 3.8ghz setting?

Yep, apparently mine needs a little more voltage then most.

My BCLK is at 180 (181 in os), my multiplier is at 19x but it boosts up to 21. Had to leave my memory at 4x (8x on some motherboards) to get 1440 (cas7)

 

[netxzero64]

Hm. I'd have to say it's likely the memory controller/uncore that is causing your chip to be unstable. For my 750, I was never quite able to reach 100% stability at 200 BCLK; it would still crash and revert to stock settings every once in a while. 180 BCLK worked if I raised Vtt high enough. Every chip is different.

As for the QPI, I'd agree with you and leave it on Auto as well.

OCing can be a pretty monotonous process. After weeks of testing various settings, I finally settled with the following:

167 BCLK
8x Memory Multiplier => 1336 MHz RAM 7-7-7-20
20x CPU Multiplier => Turbo ON, Base Freq @ 3.34 GHz, 4.008 GHz @ 1 core, 3.841 GHz @ 2 cores, 1.503 GHz @ Idle
Vcc = Auto, it fluctuates depending on load, ~1.15V at base freq, haven't checked the core freq. when only 1 core is loaded
Vtt = 1.15V
RAM Voltage = 1.58V

I see. I have tested the maximum BCLK which is stable is 173 which is about 1730mhz on the memory with 14x or 15x on the processor multiplier as I am still using the stock cooler and afraid to fry the chip.

I will try to reach 180 BCLK and try to increase the Vtt to get a stable OC. I don't know how to check the turbo modes if it goes 1 core or 2 cores turbo as I am in game most of the time.

I will try to mimic your settings. but as of the moment my setting is at 145 BCLK and everything on auto. memory timings at 7-8-7-20 which I have set manually because 7-7-7-20 doesn't boot. lol!

@deders

I will take both yours and saylick's advise on this one and observe more so that I can really see the limits of my hardware.

Can a bios update decrease OC capability?

 

[Deders]

Best way to be sure what each component can handle is to clock the others down, so while you are testing the BCLK, clock the CPU and Memory down. Then when you have a stable motherboard, raise the CPU multiplier and voltages (You've already established your memory can run very high). That way you'll know which is the weakest link and the best way to deal with it.

 

[netxzero64]

Best way to be sure what each component can handle is to clock the others down, so while you are testing the BCLK, clock the CPU and Memory down. Then when you have a stable motherboard, raise the CPU multiplier and voltages (You've already established your memory can run very high). That way you'll know which is the weakest link and the best way to deal with it.

Fair enough. Will try to test that later tonight. Thank you so much for your advise. I learned a lot from you guys. It is really that complicated when OC'ing intel processors ehh? AMD systems are a lot easier to OC. XD

 

[Saylick]

I see. I have tested the maximum BCLK which is stable is 173 which is about 1730mhz on the memory with 14x or 15x on the processor multiplier as I am still using the stock cooler and afraid to fry the chip.

I will try to reach 180 BCLK and try to increase the Vtt to get a stable OC. I don't know how to check the turbo modes if it goes 1 core or 2 cores turbo as I am in game most of the time.

I will try to mimic your settings. but as of the moment my setting is at 145 BCLK and everything on auto. memory timings at 7-8-7-20 which I have set manually because 7-7-7-20 doesn't boot. lol!

Haha, yeah... I think part of the reason I was able to hit 7-7-7-20 was because I didn't push the memory frequency that high. In fact, my RAM is rated for 1600 MHz @ 1.65V but I found out that my RAM actually isn't all that good, hence lowering the frequency to 1336 MHz. If I decided to raise the freq. of my RAM, I'd have to loosen my timings for sure.

As for checking the turbo modes, I recall using TMonitor in conjunction with Prime95. I started up TMonitor, then loaded up P95 onto all cores. I would check the freq. of the chip as all cores were loaded, then cancel a P95 instance on 1 of the cores, check the freq. across all the cores (one core should be idling for the most part by now), then cancel another P95 instance, check the freq., rinse and repeat for all the cores.

Ideally, the Turbo Boost freq. should be in increments equal to your BCLK. I just checked with Anandtech and THG, and both indicate that the Turbo Boost bins are 1/1/4/4, i.e. +1 bin for 4 cores loaded, +1 bins for 3 cores, +4 bin for 2 cores loaded, and also +4 bins for one core loaded.

For some reason, my chip doesn't do that... I'm not sure if it has to do with my BIOS or whatever, but for my chip, it goes 1/1/3/4 with +4 bins for 1 core loaded, +3 bins for 2 cores, +1 bin for 3 or 4 cores loaded. So in my configuration, I have a BCLK of 167 and a CPU multiplier of 20, which produces the 3.36 GHz base clock. When 3 or 4 cores are loaded up, Turbo Boost kicks in and gives all cores a 167 MHz boost, equal to 1 bin, for a total of 3.507 GHz on all cores. At 2 loaded cores, Turbo Boost gives a 501 MHz boost, equal to 3 bins, for a total of 3.841 GHz on the two loaded cores. Lastly, at 1 loaded core, Turbo Boost gives a whopping 668 MHz boost, equal to 4 bins, for a total of 4.008 GHz on the single loaded core.

 

[netxzero64]

Haha, yeah... I think part of the reason I was able to hit 7-7-7-20 was because I didn't push the memory frequency that high. In fact, my RAM is rated for 1600 MHz @ 1.65V but I found out that my RAM actually isn't all that good, hence lowering the frequency to 1336 MHz. If I decided to raise the freq. of my RAM, I'd have to loosen my timings for sure.

As for checking the turbo modes, I recall using TMonitor in conjunction with Prime95. I started up TMonitor, then loaded up P95 onto all cores. I would check the freq. of the chip as all cores were loaded, then cancel a P95 instance on 1 of the cores, check the freq. across all the cores (one core should be idling for the most part by now), then cancel another P95 instance, check the freq., rinse and repeat for all the cores.

Ideally, the Turbo Boost freq. should be in increments equal to your BCLK. I just checked with Anandtech and THG, and both indicate that the Turbo Boost bins are 1/1/4/4, i.e. +1 bin for 4 cores loaded, +1 bins for 3 cores, +4 bin for 2 cores loaded, and also +4 bins for one core loaded.

For some reason, my chip doesn't do that... I'm not sure if it has to do with my BIOS or whatever, but for my chip, it goes 1/1/3/4 with +4 bins for 1 core loaded, +3 bins for 2 cores, +1 bin for 3 or 4 cores loaded. So in my configuration, I have a BCLK of 167 and a CPU multiplier of 20, which produces the 3.36 GHz base clock. When 3 or 4 cores are loaded up, Turbo Boost kicks in and gives all cores a 167 MHz boost, equal to 1 bin, for a total of 3.507 GHz on all cores. At 2 loaded cores, Turbo Boost gives a 501 MHz boost, equal to 3 bins, for a total of 3.841 GHz on the two loaded cores. Lastly, at 1 loaded core, Turbo Boost gives a whopping 668 MHz boost, equal to 4 bins, for a total of 4.008 GHz on the single loaded core.

I see now. Maybe I can push my frequency to 6-7-6-18 if I increased my ram voltage?