i5-4690k. I r noob attempting overclock.

[layarion]

i5-4590k currently OC'ed to 4.5@1.310 volts. Small Prime FFT test hit 90C, only 76C on "1344" stability test. (Cooler is NH-D15)

8GBx2 1600Mhz PNY RAM.

Tower: HAF-X, default number of fans that it ships with.

Gigabyte G1. Sniper Z97

GTX 1070 (not overclocked)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Last Sunday I got my NH-D15 in the mail, and installed it immediately. Then i went to this site: http://overclocking.guide/gigabyte-g1-sniper-z97/ and followed the instructions in the video. I am completely new to anything overclocking so I just kept my head low and just followed the instructions on the vid exactly. Save the the memory profile, because i don't know what to do with that.

 

Like the video suggested I disabled all turbo boost, and power saving features. Then i started by just raising the multiplier to 43, and voltage to 1.250 and manually set the other voltage settings to exactly what was seen in the vid.

 

I ran the latest version of Prime95 at 1344 to 1344, FFT in place, and 15 minutes and let it run for 30 minutes. Was fine, and i jumped to a multiplier of 46 and a voltage of 1.325 and the system wouldn't even start windows. So far, messing with only the multiplier, core voltage, and literally nothing else i seem to have (form 30 minute test mind you) a stable of 4.5 at 1.310v...though i think i could get away with just 1.3v for 45.

 

Running the stability 1344 test, tempts didn't go above 76C, but if i specifically push for heat and use Small FFT on prime95 one of my cores maxes out at 90C.

 

so two questions:

1) i was told to not let the tempts go above 90C, and i ask: Should i base that limit on the stress test tempts, or the Heat test tempts? because i can get more voltage room if i go off the stress test version.

2) Is there anything more to do than raise the multiplier and core voltage?

3) Do i need to do anything with my RAM or Ring volts/speed?

 

basically what more should i be doing before doing the 21 hour stress test?

 

update: while rendering a video on adobe premiere the hate stays very tame at 65C

Edited June 27, 2017 by layarion

[ARikozuM]

@done12many2 @MageTank know more about overclocking than I ever would.

 

[LooneyJuice]

Just a suggestion. Disabling power saving features are for people who strictly bench as I see it. If you're going to be using a daily overclock, it's much much better to leave all the C states on in your Bios and let the chip throttle and drop voltage when idle. It really won't hurt performance for day-to-day tasks, and it'll save you a heck of a lot of power, heat and lifespan in the long run. 

 

Some tend to disable power saving features for stability, but honestly, if the power saving features are what knock your OC out, it was pretty marginal to begin with imo. You have a decent enough board with a good enough VRM to hold its own.

 

Edit: Brainfart, hit save too soon.

 

Also, small FTT test is worst case scenario for temps. I'd use it for measuring absolute cooling performance, but 90C is still toasty. More realistic loads will be in a safer temp margin, but you'll have to be weary of ambient temps.

 

Additionally, voltage per clock will vary quite a bit on an individual chip basis. So if you've found the lowest stable voltage already for your core clock, that's pretty much all you can do. 

 

I don't know how much you looked into this too, but have you followed recommended settings regarding LLC (Load Line Callibration)? Insufficient voltage when ramping up can cause an otherwise stable-looking OC to crap itself.

[MageTank]

4 minutes ago, layarion said:

i5-4590k currently OC'ed to 4.5@1.310 volts. Small Prime FFT test hit 90C, only 76C on "1344" stability test. (Cooler is NH-D15)

8GBx2 1600Mhz PNY RAM.

Tower: HAF-X, default number of fans that it ships with.

Gigabyte G1. Sniper Z97

GTX 1070 (not overclocked)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Last Sunday I got my NH-D15 in the mail, and installed it immediately. Then i went to this site: http://overclocking.guide/gigabyte-g1-sniper-z97/ and followed the instructions in the video. I am completely new to anything overclocking so I just kept my head low and just followed the instructions on the vid exactly. Save the the memory profile, because i don't know what to do with that.

 

Like the video suggested I disabled all turbo boost, and power saving features. Then i started by just raising the multiplier to 43, and voltage to 1.250 and manually set the other voltage settings to exactly what was seen in the vid.

 

I ran the latest version of Prime95 at 1344 to 1344, FFT in place, and 15 minutes and let it run for 30 minutes. Was fine, and i jumped to a multiplier of 46 and a voltage of 1.325 and the system wouldn't even start windows. So far, messing with only the multiplier, core voltage, and literally nothing else i seem to have (form 30 minute test mind you) a stable of 4.5 at 1.310v...though i think i could get away with just 1.3v for 45.

 

Running the stability 1344 test, tempts didn't go above 76C, but if i specifically push for heat and use Small FFT on prime95 one of my cores maxes out at 90C.

 

so two questions:

1) i was told to not let the tempts go above 90C, and i ask: Should i base that limit on the stress test tempts, or the Heat test tempts? because i can get more voltage room if i go off the stress test version.

2) Is there anything more to do that raise the multiplier and core voltage?

3) Do i need to do anything with my RAM or Ring volts/speed?

 

basically what more should i be doing before doing the 21 hour stress test?

Ok. First of all, 1344k and 2688k are your vcore tests. Small FFT's (48k for example) are so hot because they are small enough to fit in your cache, and absolutely hammer your CPU. It's unrealistic heat, so don't worry about it. 1344k is also hotter than what you will see when gaming and general use, just not as hot as small FFT or Linpack MKL. How hot 1344k gets, is dependent entirely on how much ram you throw at it. We recommend going no higher than 75% of your total ram. So if you have 16GB (16x1024MB = 16384, we recommend using 12288 if ram in Prime95. Leave "in-place" unticked. Also, this takes several hours to verify general stability, not just 30 minutes.

 

As for your questions:

1. You define that answer. If you are trying to test for absolute worst possible scenarios, then base it on small FFT's. If you are looking for real-world thermals, base it on the larger FFT's. We only use small FFT's as a means to stress our thermal solution, not necessarily to stress our CPU itself. We use it to see the absolute max heat our coolers can dissipate, so that we have a realistic understanding of what we can do with 24/7 loads. 

2. You can try to tighten that voltage up for lower thermals. You must understand that once you hit the voltage wall (the point in which it requires far too much voltage to achieve the next clock step), it's time to back off. This is especially true with someone experiencing the thermals that you are seeing. Rather than try to achieve a higher CPU speed, settle with what you have, and work on refining it for a better longevity. Lowering your voltage/current into the CPU, will in turn lower your thermals, which will help prevent thermal runaway and aid in your CPU's lifespan. 

3. Cache speed? Not really, unless you are trying to achieve slightly higher bench scores. Ram speed on the other hand is very important for minimum framerates. Check my sig for a guide on that if you are interested in taking that on. Be warned: overclocking your ram will further increase your CPU package temps. This may interfere with your current stability, as the hotter your silicon gets, the harder it is to maintain stability at the same voltage. All the more reason to work on bringing those thermals down with a more refined vcore.

 

There is a Prime95 guide within my ram guide to tell you how to stress your ram. If you have questions on specific voltages and timings, let me know and I'll provide more details if need be.

[layarion]

No i don't know anything about LLC and i dunno if my board has it. that does lead me to another question in regards to your bench marking comments. If i hit a stable baseclock of 4.5 on all cores, do i bother turning turbo on? like does it go above that, or do i lower the other cores to allow 1 core to go to 4.5 or could they all stay at 4.5 but allow one to go above 4.5?

[LooneyJuice]

34 minutes ago, layarion said:

No i don't know anything about LLC and i dunno if my board has it. that does lead me to another question in regards to your bench marking comments. If i hit a stable baseclock of 4.5 on all cores, do i bother turning turbo on? like does it go above that, or do i lower the other cores to allow 1 core to go to 4.5 or could they all stay at 4.5 but allow one to go above 4.5?

As was stated above, there are way more knowledgeable people around here, but what I usually suggest - if you're comfortable with it - is trying to find a guide pertaining to your setup (Board + CPU Generation). Devil's Canyon is very well documented by now, and I'm sure there are video tutorials for your specific board. There are many combos and finding guides tailored to yours is way more convenient.

 

Speaking of:

 

Just for reference, depending on your board, there are a number of LLC levels which you can use to offset the voltage droop when a CPU ramps up. It's almost essential in most cases to increase the setting as you draw more current when overclocking. Most overclocking boards have it, yours should too. Unless you go by suggested LLC values for your board, there's a bit of trial and error, but usually, again, it's well documented for most CPUs.

 

Now, regarding turbo etc. I wouldn't necessarily mess with running higher clocks per number of active cores, makes things a bit more complicated. I'm assuming you just want a straightforward daily overclock. In which case, you do what you've already done, you follow what @MageTank said, check out the video, and you also want Speedstep and C-States enabled in your Bios. Additionally you want to go to your windows power plan and set a min and a max frequency if you've previously pegged them at 100%.

[MageTank]

11 minutes ago, layarion said:

No i don't know anything about LLC and i dunno if my board has it. that does lead me to another question in regards to your bench marking comments. If i hit a stable baseclock of 4.5 on all cores, do i bother turning turbo on? like does it go above that, or do i lower the other cores to allow 1 core to go to 4.5 or could they all stay at 4.5 but allow one to go above 4.5?

Your manual overclocking will override turbo boost. All cores will run at your defined CPU multiplier. Since your default turbo table is 3.9ghz, 4.5 is already well beyond that, and it will not apply an additional turbo on top of your overclock.

 

Someone can correct me if I am wrong, but I don't remember if haswell has a per-core multiplier. It's been a long time since I overclocked on Haswell. If it does, you can then define that scenario in which one core runs faster than the rest, but I wouldn't really recommend it. If all cores are capable of 4.5, then you are better off running all cores at 4.5. If you seriously wanted to improve overclocking, I'd recommend investing in a delid tool, and some liquid metal TIM. https://rockitcool.myshopify.com/ https://www.amazon.com/Coollaboratory-Thermal-Compound-Processor-Heatsink/dp/B0039RY3MM

 

This will dramatically drop your temperatures, with some seeing improvements up to 20C. Meaning if you are currently hitting 76C, you could see temperatures as low as 56C (though, your mileage may vary). If you are interested in going that route, I have some information in an older post of mine regarding the steps I took, along with some precautions: 

@LooneyJuice is correct about LLC, and I completely neglected to mention that very important setting. Be warned when adjusting LLC: The more aggressive the LLC, the more vcore Prime95 will request from the CPU. At LLC level 2 on my board, I draw an additional .020mv over what I manually defined for vcore. For example: 1.35v becomes 1.37v under Prime95 stress. Just keep that in mind, and use hwinfo64 to monitor vcore when testing. This will show you how aggressive each LLC setting is under load. Find what works best for you.

[ARikozuM]

1 minute ago, MageTank said:

Someone can correct me if I am wrong, but I don't remember if haswell has a per-core multiplier.

It does. Running one core (4690K) at 46 rather than 45 since it had lower temps than the rest of the cores. Making sure it's stable is the hard part as with any overclocking.

[MageTank]

Just now, ARikozuM said:

It does. Running one core (4690K) at 46 rather than 45 since it had lower temps than the rest of the cores. Making sure it's stable is the hard part as with any overclocking.

Yeah... I was full ITX in the Haswell days, and I don't recall my cheapo ITX board having that option. Granted, my ITX board also completely lacked VRIN as well, so I suppose I shouldn't base any claims on that motherboard, lol. 

[LooneyJuice]

4 minutes ago, MageTank said:

-snip-

I think I remember both Ivy and Haswell having custom turbo multipliers per active cores. Not sure you could peg a core specifically, it just depended on utilization AFAIK.

 

Edit: Never mind, the fresh posts didn't show up for me, derp.

 

Also, @layarion, yeah, just as a precaution, 'cause I've heard some people do this. Don't get too enthusiastic and max out your LLC thinking you'll get past your OC wall, that can kill your chip pretty quick if like MageTank said you run a stress test and the board just cranks up the voltage in response to the load. It's the same premise as tweaking voltage, be as frugal as possible.

[ARikozuM]

4 minutes ago, MageTank said:

Yeah... I was full ITX in the Haswell days, and I don't recall my cheapo ITX board having that option. Granted, my ITX board also completely lacked VRIN as well, so I suppose I shouldn't base any claims on that motherboard, lol. 

I'm sure it could also depend on the BIOS as well. 

 

Edit: Also, on [Devil's Canyon], Intel was kind enough to put an asterisk next to the best core that could be read by the BIOS.

[LooneyJuice]

2 minutes ago, ARikozuM said:

I'm sure it could also depend on the BIOS as well. 

 

Edit: Also, on Haswell, Intel was kind enough to put an asterisk next to the best core that could be read by the BIOS.

 

Oh, wasn't that just on Haswell-E?

[ARikozuM]

Just now, LooneyJuice said:

Oh, wasn't that just on Haswell-E?

Devil's Canyon, specifically.

[LooneyJuice]

2 minutes ago, ARikozuM said:

Devil's Canyon, specifically.

Oh, very nice. Was it board specific by any chance? Built a couple of Devil's Canyon rigs for friends and I can't recall much. Can't say I spent all that much time on them after though.

[ARikozuM]

Just now, LooneyJuice said:

Oh, very nice. Was it board specific by any chance? 

Should be all of them. As far as I can recall, all of the MSI and Asrock [Z97] boards showed the asterisk.

[layarion]

ok thank you all for the valuable info and especially the clear answers to my questions. I will need time to digest what i read, and it might be awhile before i try RAM overlcoking and longer still if i ever try deidling (because that scares me). Actually i might try overclocking the RAM soon, because it seems important...but for now i need to just absorb all this before doing more. still, i'll post when i clean up that voltage and settle on the cpu setup.

 

No i wasn't interested in improving benchmark test, i was more or less trying to figure out what it was and if it was important. Since it seems not to be, i'll leave the ring voltage where the video had me leave it.

[LooneyJuice]

1 hour ago, layarion said:

 Actually i might try overclocking the RAM soon, because it seems important...but for now i need to just absorb all this before doing more. still, i'll post when i clean up that voltage and settle on the cpu setup.

Just a word of warning, if you're doing it for actual real world performance, Mem OC at least for DDR3 may net you marginal gains, if anything depending on your use case. I know it can help minimums in some cases as @MageTank mentioned, but it won't make or break your machine. And as he also mentioned, it will additionally heat up the CPU package, and can also lead to some instability even with a previously stable overclock. Check around for memory benchmarks for the stuff you use if you can find them. Sure, you'll see scaling for synthetic benchmarks, but you shouldn't expect any night and day differences. That being said, if you feel like going through with it for the tinkering, by all means. Just, as usual, when it comes to component lifespan, excessive heat and voltage are the killers, memory is only passively cooled by promoted airflow and does not have any safeguards, so again, super-frugal with voltage.

 

Other than that, good luck, and if you encounter any massive hiccups, someone around here will be able to get you out of the hole I'm sure.

[MageTank]

Just now, LooneyJuice said:

Just a word of warning, if you're doing it for actual real world performance, Mem OC at least for DDR3 may net you marginal gains, if anything depending on your use case. I know it can help minimums in some cases as @MageTank mentioned, but it won't make or break your machine. And as he also mentioned, it will additionally heat up the CPU package, and can also lead to some instability even with a previously stable overclock. Check around for memory benchmarks for the stuff you use if you can find them. Sure, you'll see scaling for synthetic benchmarks, but you shouldn't expect any night and day differences. That being said, if you feel like going through with it for the tinkering, by all means. Just, as usual, when it comes to component lifespan, excessive heat and voltage are the killers, memory is only passively cooled by promoted airflow and does not have any safeguards, so again, super-frugal with voltage.

 

Other than that, good luck, and if you encounter any massive hiccups, someone around here will be able to get you out of the hole I'm sure.

It's entirely dependent on the amount of overhead. If you are IO bound, then memory overclocking will yield some tangible results. I've personally seen improvements of up to 15% in my minimum framerates in specific titles, with a 6700k and GTX 1070 at 1440p. Now, this won't hold true for every title, and if you are GPU bottlenecked at all times, it won't make a difference at all. If you can afford a stable XMP kit, then it's probably safer to go that route instead of trying to manually overclock blindly. Diminishing returns also kicks in pretty hard. For DDR4, anything beyond 2800 felt like diminishing returns. While it does scale beyond that, it's not to the same degree as going from 2133 to 2800. With DDR3, anything beyond 2133 gave little to no benefit. This is especially true for DDR3 2666, as that strap is nearly impossible to make stable with 2DPC, and very few multi-rank kits exist at that speed. It's far too much stress on the IMC, and the tertiary timings needed to make it stable completely defeat the purpose.

 

TL:DR? Memory overclocking does help with gaming, but it will depend entirely on the game, and your system configuration. Invest in a decent XMP overclock if you can afford it, or look towards my guide if you plan on tweaking ram manually. While it might seem difficult at first, everyone will eventually get the hang of it. This goes double for Ryzen.

[ARikozuM]

Just now, MageTank said:

Diminishing returns also kicks in pretty hard. For DDR4, anything beyond 2800 felt like diminishing returns.e.

DDR3 pricing was terrible at the time. 

 

Sweet spot for price-performance DDR3 was 1866/10 whereas DDR4 is 3000/15.

[LooneyJuice]

3 minutes ago, MageTank said:

It's entirely dependent on the amount of overhead. If you are IO bound, then memory overclocking will yield some tangible results. I've personally seen improvements of up to 15% in my minimum framerates in specific titles, with a 6700k and GTX 1070 at 1440p. Now, this won't hold true for every title, and if you are GPU bottlenecked at all times, it won't make a difference at all. If you can afford a stable XMP kit, then it's probably safer to go that route instead of trying to manually overclock blindly. Diminishing returns also kicks in pretty hard. For DDR4, anything beyond 2800 felt like diminishing returns. While it does scale beyond that, it's not to the same degree as going from 2133 to 2800. With DDR3, anything beyond 2133 gave little to no benefit. This is especially true for DDR3 2666, as that strap is nearly impossible to make stable with 2DPC, and very few multi-rank kits exist at that speed. It's far too much stress on the IMC, and the tertiary timings needed to make it stable completely defeat the purpose.

 

TL:DR? Memory overclocking does help with gaming, but it will depend entirely on the game, and your system configuration. Invest in a decent XMP overclock if you can afford it, or look towards my guide if you plan on tweaking ram manually. While it might seem difficult at first, everyone will eventually get the hang of it. This goes double for Ryzen.

Yeah, this sort of granular explanation I can totally get behind, Totally agree. And I know it varies between generations as well. Sandy IIRC didn't scale all that much past 1600. Thanks for the write-up though, much appreciated.

[LooneyJuice]

4 minutes ago, ARikozuM said:

DDR3 pricing was terrible at the time. 

 

Sweet spot for price-performance DDR3 was 1866/10 whereas DDR4 is 3000/15.

I so wish that still held true for DDR3 here. I had a kit go poop on me and the 16GB 1600mhz kit I nabbed cost be the same as decent DDR4 16GB kit, regardless of where I got it from. 

[MageTank]

1 minute ago, ARikozuM said:

DDR3 pricing was terrible at the time. 

 

Sweet spot for price-performance DDR3 was 1866/10 whereas DDR4 is 3000/15.

For those on a strict budget, they are better off learning how to overclock ram. It's not easy, even for someone that knows the tricks of the trade, but it's free (assuming your platform supports it) and it's a great experience once you get the hang of it. Even the worst DDR4 Micron IC's can still hit 3000mhz. I've tested about 6 different Micron kits, and not a single one failed to hit 3000 at 1.35v. Samsung, even their worst kits hit 3200+ in my tests. The same is said of Hynix. For DDR3, the best we had was Samsung HYK0 and Elpida "Magic Memory". There were some higher end Hynix IC's, but they came about during that extreme DRAM price hike, and were unobtainable without paying a hefty premium. 

 

My point is, they don't make kits that can't OC. Even the worst of the worst can improve if you take the time to do so. So if price:performance is your #1 goal, it's best to learn how to do it the hard way, lol. I've been looking into Thaiphoon burner to see if I can create my own XMP's on various kits for people to test out. While it's difficult to make IC's for every IMC/motherboard trace topology, it would at least show people how I go about my tweaking. If I had the spare cash, I'd invest in budget motherboards across every manufacturer, and do it that way. That way, if I can guarantee an SPD profile to work on the worst boards, I know for certain they will work on better boards. The only variable that would exist beyond that, would be the IMC. 

[LooneyJuice]

5 minutes ago, MageTank said:

For those on a strict budget, they are better off learning how to overclock ram. It's not easy, even for someone that knows the tricks of the trade, but it's free (assuming your platform supports it) and it's a great experience once you get the hang of it. Even the worst DDR4 Micron IC's can still hit 3000mhz. I've tested about 6 different Micron kits, and not a single one failed to hit 3000 at 1.35v. Samsung, even their worst kits hit 3200+ in my tests. The same is said of Hynix. For DDR3, the best we had was Samsung HYK0 and Elpida "Magic Memory". There were some higher end Hynix IC's, but they came about during that extreme DRAM price hike, and were unobtainable without paying a hefty premium. 

 

-snip-

Was kinda my point with this one too. Was not going to bother paying a silly premium over old memory with a marginally faster XMP. Not trying to be stingy, just my shit is old and tired and not really worth it since it won't be used past this rig. Ergo, I'll probably mess with it at some point, but the CMOS jumper is under the second card so I'm suffering from severe do-it-tomorrow-itis.

[MageTank]

1 minute ago, LooneyJuice said:

Was kinda my point with this one too. Was not going to bother paying a silly premium over old memory with a marginally faster XMP. Not trying to be stingy, just my shit is old and tired and not really worth it since it won't be used past this rig. Ergo, I'll probably mess with it at some point, but the CMOS jumper is under the second card so I'm suffering from severe do-it-tomorrow-itis.

Who needs a CMOS jumper, when you can just unplug the PSU and hold that power button for 30 seconds?

 

EDIT: Could use my old trick of plugging your reset button into your CLEAR_CMOS jumper. That works wonders too. I mean, who even uses the reset button to reset their PC anymore?

[LooneyJuice]

Just now, MageTank said:

Who needs a CMOS jumper, when you can just unplug the PSU and hold that power button for 30 seconds?

Ain't nobody got time fo' dat!

 

Also, I don't wish to be too forward, but you made me go through your guide you b**tard! Neat read, me likey.