Overclock.net › Forums › Intel › Intel CPUs › Kaby Lake Overclocking Guide [With Statistics]
New Posts  All Forums:Forum Nav:

Kaby Lake Overclocking Guide [With Statistics] - Page 220

post #2191 of 2845
Quote:
Originally Posted by becks View Post

For me as PSU XTR - XFX has been great so far...no fails since

Seasonic OEM I think smile.gif I remember them being fairly cheap a few years ago and being recommended a lot.
Quote:
Originally Posted by dirtyred View Post

I know EVGA does alot of rebranding but in general I'd say they aren't a great brand when it comes to power supplies. The ones you mentioned are super expensive, for the same reason why Superflower is usually damn expensive (subjective). That's why I recommended SeaSonic as it's a brand you can trust most of the times and the price is right. Had a Corsair (CX 430 or something similar, was about 15 years ago) failing on me and dragging half the system with it. Got myself a SeaSonic that still works in my old (now my parents) PC and got one in my current one as well and wasn't expensive. For $60 you won't find many others that perform on this level. Leadex PSU's are generally $35-40 more expensive so it's not really in this price range.

Corsair rebrands everything, they make nothing themselves. I've got a HX 650 V2 which is based off a Seasonic G platform, although slightly worse because it lacked a couple of extra PCI-E cables (I bought it because I wanted something blue at the time). I think mainly they use CWT now, but supposedly they design their own at the higher end and get them to make it or so theyve said. I dont follow PSU's religiously but its well known that very few of the major vendors make their own. Seasonic, FSP, and Superflower sell under their own brands while also building for others that I can recall of the top of my head at the more decent end of the spectrum.

EVGA, Antec, Bequiet, Fractal Design, Corsair etc etc are all rebranding different OEM's sometimes using several for different price points and changing up as time goes on.

Basically ignore the branding completely, you need to read the reviews or ask around in forums to find out whats actually inside.

I agree Seasonic branded PSU's are almost always very good, but often you can get something cheaper thats as good, and sometimes its also made by Seasonic under a different brand name, like the old XFX XTR series. tongue.gif
post #2192 of 2845
Hi there! I'm new to overclocking and I have a quick question.

I have a 7700k and a Noctua D14. I'm trying to get a stable overclock of 4.8 if possible.

I noticed in the OP that someone with the same cooler was able to get 4.8ghz with a Vcore of 1.24. Unfortunately I have BF1 crashing at 1.29 Vcore with 4.8ghz overclock. Does this mean I just got a potato chip?

My temps are low 80's in stress tests (prime 95 27.9) with 1.29, so I'm not sure I can really push it any further than that.

I have XMP enabled on the motherboard for my 3000mhz ram. My mobo has 2 options for LLC - Auto or Mode 1 (which is apparently more aggressive). I have that set to mode 1.

Any suggestions?
post #2193 of 2845
Quote:
Originally Posted by rodubbs View Post

Hi there! I'm new to overclocking and I have a quick question.

I have a 7700k and a Noctua D14. I'm trying to get a stable overclock of 4.8 if possible.

I noticed in the OP that someone with the same cooler was able to get 4.8ghz with a Vcore of 1.24. Unfortunately I have BF1 crashing at 1.29 Vcore with 4.8ghz overclock. Does this mean I just got a potato chip?

My temps are low 80's in stress tests (prime 95 27.9) with 1.29, so I'm not sure I can really push it any further than that.

I have XMP enabled on the motherboard for my 3000mhz ram. My mobo has 2 options for LLC - Auto or Mode 1 (which is apparently more aggressive). I have that set to mode 1.

Any suggestions?

I've got myself a new cooler yesterday, the Phanteks PH-TC14PE which is just like the D15 in terms of performance (but not ugly). Since yesterday I've been trying to get stable 4.8 GHz and it was failing horribly until 1.28 V using adaptive voltage with LLC 5. But I noticed it wasn't failing under 100% load. So I thought to myself maybe it's failing because while not under full load, it's not getting 1.28 V (and wasn't, because I monitored Vcore closely). So I switched from adaptive to manual constant voltage and tried again starting from 1.24 V. I had to go for 1.26 V because even with LLC 5 it drops to 1.248 V but now it seems stable. I've only tested so far for 2 hours with RealBench but it passed so it's promising and will do an 8 hours test when I have time.

Low 80's are fine but Prime95 27.9 uses AVX instructions so it generates more heat. Test with v26.6 as it doesn't support AVX and you probably won't really use it unless you do 3d rendering or something special. Prime95 any version usually generates more heat then any real life scenario would. I recommend either v26.6 or preferably RealBench (uses AVX but generates less heat).

TL;DR: Go with manual constant voltage of 1.26 - 1.28 V and use LLC 5 is you have Asus. Try to get the most stable and closest voltage to what you've set in BIOS under load. Use Prime95 v26.6 or RealBench.
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
post #2194 of 2845
Appreciate all the feedback thus far.

1. The power supply -- I agree 750W is probably sufficient. I've heard EVGA makes some good power supplies, and this one reviewed fairly well, but it is semi-modular. As I said I seriously doubt I'd ever go SLI, but I'm also the type that won't upgrade for the next best thing. So, down the road, as price pushes GTX 1070s down -- if there is actually a need for my video performance (which I just can't see right now), it would be nice to know I'm not on the edge due to the number of drivers I'm running and my overclock. I'm torn, and I haven't decided just yet. I'm probably leaning 60/40 right now to installing the G3 and keeping the BQ if I decide to keep the second mobo. The plan would be down the road to just build a second Kaby Lake machine using the Fata1ity mobo, mis-matched RAM, and the PowerSpec case. It really isn't that bad of a case for a standard build at all.

2. I'm kicking myself a bit and wishing I'd have gone for the MSI. I really like the Fata1ity, and actually wanted the Taichi in the first place. That said, it looks like Asrock has some work to do with BIOS updates to get their memory QVL expanded a bit. MSI has a lot more on their QVL, and they seem to be a few days ahead on BIOS and driver updates. However, the Taichi included a $15 rebate for purchase and I believe an additional $10 if I review it on Newegg. So, apples to apples I'm looking at about a $100 difference between the MSI and the Taichi. Or, as stated previously, could just stick with the Fata1ity. I really want to run 32GB of RAM (I do use it), and the mis-matched pair issue leaves me in a place of either getting another set of Geil Forzas or Ripjaws for around $115-$125. Perhaps I should just do that and return the Taichi. Then I'd have a really good shot of having all 4 sticks working at the rated XMP profile and still be $50-$60 ahead which I could put towards my eventual purchase of an additional NVME drive. With RAM and SSD prices so high right now -- it may just be worthwhile to wait. Only caveat is that if you believe the rumors it could be a year or more before prices go back down, and they could actually go up MORE in the interim. Is it worth it to wait and lose a year of what I actually want to do with this system? That's the question I have to answer.

3. One point I didn't mention about the Hyper EVO vs AIO Corsair -- the Hyper EVO is blocking a RAM slot, and I have to really work to get the 4th stick under it. That means I have to have the fan edged up about 1.5 inches reducing my cooling ability. I added a very quiet fan on the other side of the EVO mounted lower to pull to make up for the gap that was created from needing space for the RAM. I've seen you can mount the EVO sideways, but I don't know how that impacts cooling. My research hasn't lead to a definitive answer. A bit frustrating. My hope was that I could achieve a stable 5.0GHZ OC as well as not have the RAM space issues.

If I had to make a decision right this second I'd say I'm going to do the case transfer, use the Taichi (2 extra SATA slots and better VRM vs the Fata1ity), and keep the Fata1ity for a build late this year or sell it. Main issue, however, is that I don't think I'd recoup very much of what I'd want to get for it because I don't have the mobo box or anything like that because this was a pre-built from Microcenter. So i'm strongly leaning toward holding on to it, and wait and see how far Kaby Lake gets pushed down by Ryzen competition. Because I want the RAM slots to be unencumbered I'm going to install the AIO water cooler. I'm just going to wait on the RAM and see if any BIOS updates down the road help out or if the Taichi for some shocking reason can actually run the 4 sticks at their default XMP. Believe me -- I tried a million different timings and settings but they are simply not stable above 2900MHZ when grouped together no matter how much voltage or how loose I make the timing. What's infuriating is that the two RipJaws by themselves or the two Geils by themselves run the 3200 XMP profile just fine and pass Memtest with zero issues. Once I throw all 4 sticks in -- even if I loosen the timings more than the loosest set (the Ripjaws are 16-18-18-38) I simply can't reach 3200.

Thanks so much to everyone that has responded so far. I'm still going to think about it this weekend before I move forward. I'm trying to finish up the 24 CAT6 runs I just ran in the house and transfer from my Netgear switches and Edgerouter to a Ubiquiti Unifi Security Gateway, Unifi 24 port switch, and a couple of Unifi AC Pro APs. That's been a long-term project that is now in its second month. With kids and everything else just not enough time! Appreciate all the feedback, and anyone else that wants to chime it in more than welcome to do so.

Forecast
post #2195 of 2845
Quote:
Originally Posted by dirtyred View Post

I've got myself a new cooler yesterday, the Phanteks PH-TC14PE which is just like the D15 in terms of performance (but not ugly). Since yesterday I've been trying to get stable 4.8 GHz and it was failing horribly until 1.28 V using adaptive voltage with LLC 5. But I noticed it wasn't failing under 100% load. So I thought to myself maybe it's failing because while not under full load, it's not getting 1.28 V (and wasn't, because I monitored Vcore closely). So I switched from adaptive to manual constant voltage and tried again starting from 1.24 V. I had to go for 1.26 V because even with LLC 5 it drops to 1.248 V but now it seems stable. I've only tested so far for 2 hours with RealBench but it passed so it's promising and will do an 8 hours test when I have time.

Low 80's are fine but Prime95 27.9 uses AVX instructions so it generates more heat. Test with v26.6 as it doesn't support AVX and you probably won't really use it unless you do 3d rendering or something special. Prime95 any version usually generates more heat then any real life scenario would. I recommend either v26.6 or preferably RealBench (uses AVX but generates less heat).

TL;DR: Go with manual constant voltage of 1.26 - 1.28 V and use LLC 5 is you have Asus. Try to get the most stable and closest voltage to what you've set in BIOS under load. Use Prime95 v26.6 or RealBench.

Thanks for the response! I do use Handbrake which uses AVX, but that's about it. Unfortunately my MSI board has 2 LLC options "Auto" and "Mode 1" (Mode 1 apparently being more aggressive). Even at 1.29 (manual voltage) it was not stable at all, but seems to be 100% stable at 1.3 which is probably running too hot (BF1 is running at 80 degrees).

Quick question - I've heard of the 'Silicon Lottery' and that essentially not all chips are made equal. Does that mean that some chips are essentially able to be stable at lower voltages than others? At this point I'm under the impression that the ability to be stable at a given voltage/overclock isn't necessarily related to heat - too much heat is just a product of using more voltage you can handle?

The reason I ask is that Amazon has offered to swap my chip for another one (it's a new build). Is there the potential that I could get a higher overclock on the one they send me? I did some experimenting on it and the one I have and was unable to post at 5GHZ with a 1.42 vcore...
Edited by rodubbs - 4/29/17 at 6:30pm
post #2196 of 2845
Quote:
Originally Posted by rodubbs View Post

Thanks for the response! Unfortunately my MSI board has 2 LLC options "Auto" and "Mode 1" (Mode 1 apparently being more aggressive). Even at 1.29 (manual voltage) it was not stable at all, but seems to be 100% stable at 1.3 which is probably running too hot (BF1 is running at 80 degrees).

Quick question - I've heard of the 'Silicon Lottery' and that essentially not all chips are made equal. Does that mean that some chips are essentially able to be stable at lower voltages than others? At this point I'm under the impression that the ability to be stable at a given voltage/overclock isn't necessarily related to heat - too much heat is just a product of using more voltage you can handle?

The reason I ask is that Amazon has offered to swap my chip for another one (it's a new build). Is there the potential that I could get a higher overclock on the one they send me? I did some experimenting on it and the one I have and was unable to post at 5GHZ with a 1.42 vcore...
Yep heat is one of the factors that keeps the oc potential lower. Thats why many do delliding.
But in your case if it wont post 1.4v vcore at 5ghz , its not the heat its the chip itself...
Even if you dellid it you would get a big temps drop and maybe it will post at 5ghz at 1.4v , but it wont be stable .

My bet that you have found this chips almost maximum potential.
post #2197 of 2845
Quote:
Originally Posted by MaKeN View Post

Yep heat is one of the factors that keeps the oc potential lower. Thats why many do delliding.
But in your case if it wont post 1.4v vcore at 5ghz , its not the heat its the chip itself...
Even if you dellid it you would get a big temps drop and maybe it will post at 5ghz at 1.4v , but it wont be stable .

My bet that you have found this chips almost maximum potential.

Yeah, I figured it was a bad one. It actually came rattling around loose in the box (which was also rattling around in its own box). Thanks Amazon.

I think I'm going to take them up on it. They offered to have a new one here tomorrow and I can mail the old one back after I swap them. Kinda feel like a dick wasting a perfectly "good" chip, but I bought a new cooler for it and was hoping for more than 4.7...
post #2198 of 2845
Quote:
Originally Posted by rodubbs View Post

Thanks for the response! I do use Handbrake which uses AVX, but that's about it. Unfortunately my MSI board has 2 LLC options "Auto" and "Mode 1" (Mode 1 apparently being more aggressive). Even at 1.29 (manual voltage) it was not stable at all, but seems to be 100% stable at 1.3 which is probably running too hot (BF1 is running at 80 degrees).

Quick question - I've heard of the 'Silicon Lottery' and that essentially not all chips are made equal. Does that mean that some chips are essentially able to be stable at lower voltages than others? At this point I'm under the impression that the ability to be stable at a given voltage/overclock isn't necessarily related to heat - too much heat is just a product of using more voltage you can handle?

The reason I ask is that Amazon has offered to swap my chip for another one (it's a new build). Is there the potential that I could get a higher overclock on the one they send me? I did some experimenting on it and the one I have and was unable to post at 5GHZ with a 1.42 vcore...

Yes, sadly the silicon lottery is a thing and it exists everywhere: Nvidia (all chips are the same architecture but they disable some cores and sell them as lower tiers like 1070 instead of 1080), AMD. Heat is a factor in stability. BTW, what you can do try to lower the temps and getting more headroom is to lower the DRAM voltage because between 1.2 - 1.35 V the CPU's temp difference can be 10 or more degrees. I have a G.Skill 3200 RAM that needs 1.35 V for 3200 16-16-16-36-2T but if I do 3000 16-16-16-28-2T then I can use 1.2 V and gain about 0.05 V for Vcore, so I can go from 1.25 V to 1.3 V resulting about 200 MHz. You can try to use AVX offset if temps are too high and go for 5 GHz only for non-AVX instructions (most apps). AVX is much faster anyways then SSE so even with an offset of 2-3 you'll still be ahead.

Not all chips can reach 5 GHz According to Asus:
  1. 20% of samples are stable with Handbrake/AVX workloads when running at 5GHz CPU core speeds.
  2. The AVX offset parameter can be used to clock 80% of CPU samples to 5GHz for light workloads, falling back to 4.8GHz for applications that use AVX code.


Quote:
Originally Posted by rodubbs View Post

Yeah, I figured it was a bad one. It actually came rattling around loose in the box (which was also rattling around in its own box). Thanks Amazon.

I think I'm going to take them up on it. They offered to have a new one here tomorrow and I can mail the old one back after I swap them. Kinda feel like a dick wasting a perfectly "good" chip, but I bought a new cooler for it and was hoping for more than 4.7...

Well, if you have the money then buy a couple (maybe from different sources), test them out and keep only the best one. Then repeat if you still have the patience.
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
post #2199 of 2845
Quote:
Originally Posted by dirtyred View Post

Yes, sadly the silicon lottery is a thing and it exists everywhere: Nvidia (all chips are the same architecture but they disable some cores and sell them as lower tiers like 1070 instead of 1080), AMD. Heat is a factor in stability. BTW, what you can do try to lower the temps and getting more headroom is to lower the DRAM voltage because between 1.2 - 1.35 V the CPU's temp difference can be 10 or more degrees. I have a G.Skill 3200 RAM that needs 1.35 V for 3200 16-16-16-36-2T but if I do 3000 16-16-16-28-2T then I can use 1.2 V and gain about 0.05 V for Vcore, so I can go from 1.25 V to 1.3 V resulting about 200 MHz. You can try to use AVX offset if temps are too high and go for 5 GHz only for non-AVX instructions (most apps). AVX is much faster anyways then SSE so even with an offset of 2-3 you'll still be ahead.

Not all chips can reach 5 GHz According to Asus:
  1. 20% of samples are stable with Handbrake/AVX workloads when running at 5GHz CPU core speeds.
  2. The AVX offset parameter can be used to clock 80% of CPU samples to 5GHz for light workloads, falling back to 4.8GHz for applications that use AVX code.
Well, if you have the money then buy a couple (maybe from different sources), test them out and keep only the best one. Then repeat if you still have the patience.

Quick question on the AVX offset - I've been using a constant voltage and did try running handbrake with a -2 offset. I definitely noticed the clocks were 200MHz lower, but I didn't see a noticeable difference in temps. Will lowering the clock speed with the same manual voltage still lower temps?

Another thing... since my board basically doesn't offer LLC (Auto vs Mode 1) will that limit me as well? Apparently Mode 1 means more LLC (MSI is kind of cryptic about this) but I definitely saw voltage go down under heavy load.
post #2200 of 2845
Quote:
Originally Posted by rodubbs View Post

Quick question on the AVX offset - I've been using a constant voltage and did try running handbrake with a -2 offset. I definitely noticed the clocks were 200MHz lower, but I didn't see a noticeable difference in temps. Will lowering the clock speed with the same manual voltage still lower temps?

Another thing... since my board basically doesn't offer LLC (Auto vs Mode 1) will that limit me as well? Apparently Mode 1 means more LLC (MSI is kind of cryptic about this) but I definitely saw voltage go down under heavy load.

AVX instructions are very demanding. It needs more power and voltage then SSE resulting in more heat. If you use AVX offset then the CPU will need less voltage to be stable. So if you're feeding constant voltage (let's say 1.28 V) and without AVX offset it crashes, then with AVX offset you should be stable. In other words if you're stable on a set voltage running a software that doesn't use AVX instructions (all games, most apps) and you found the upper limit of voltage to stay below 80 degrees (some random spikes above 80 is acceptable) but running AVX apps you crash just use AVX offset.

Your board seems to be a limiting factor but you can work around it. If you can't control Vdroop then you will crash. To solve this you need to set a higher Vcore to have headroom for Vdroop and be stable. Asus and Gigabyte does have LLC options with different levels but all they do is compensate for Vdroop automatically. For example if I set 1.28 V and LLC 5 then my Vcore is mostly around 1.264 V under full load and sometimes 1.248 V. Lower LLC settings will allow even more Vdroop, higher ones will results in Vboost. So there's no way to always get the same Vcore that you've set. So it doesn't matter what Vcore you set, what matters is what you're getting under full load. LLC settings just allow you more control.
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i7 7700 K @ 4.7 GHz ASUS Prime Z270-K Gigabyte GeForce GTX 1060 WINDFORCE OC 6G  G.Skill Ripjaws V 16 GB 
Hard DriveCoolingOSMonitor
Samsung 940 EVO 256 GB Phanteks PH-TC14PE Windows 10 Pro Dell U2312HM 
PowerCaseMouse
Seasonic M12II-620 EVO Fractal Design Define C Razer Taipan 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Kaby Lake Overclocking Guide [With Statistics]