Need help OCing my 1650v3

[SomeGuy133]

So I have
ASRock x99 WS MB
64GB 2133 ECC Memory
1650v3

I have
input voltage at 1.8v
vcore 1.35v
uncore 1.35v
load level 1 or whatever it is called
blck freq at high (supposedly this helps)

I believe that is everything I have changed.

I can get to 4.4GHz stable but 4.5 crashes. This counts leaving cache at 3.5GHz

Everytime I try to OC cache it freezes

I tried 4.2GHz core and 4/4.2GHz cache and it freezes every time I try to change it in intel XTU. I also tried change both at once and changing core first then changing cache.

Any thoughts on why I can't OC cache?

Am I missing any settings?

Also everything is static voltage until I find max OC. So there is no issue with offset voltages because I am not using offset. I am using static for now. (40-50w idle TDP lol)

I plan on doing a combination of static voltage and offset once I finish finding max OC.

Any advice and help would be awesome!

UPDATE:

Still can't get cache to OC and 4.4 GHz is unstable at 1.3v at high load like Prime95 but stable at F@H (48 hours plus) but crashes in prime in a few minutes) at 1.35v Prime95 stable for at least 30 mins. Didn't need it on longer for when testing fans and cooling.

Update I increased
input 1.9
core 1.3
cache 1.3

and got 4GHz stable

I increased
input 1.9
core 1.375
cache 1.375

and crashed at 4.2GHz so it seems inout is limiting factor. Is input safe at 2v? or should I keep it at 1.9v?

I increased
input 1.9
core 1.4
cache 1.30

and seem at first glance to be now stable at 4.5 GHz. I tried putting cache at 4 Ghz and 1.4v and it crashed.

So uncore still seems to be really unstable anyone have thoughts? I was at least able to get 4GHz core and 4 GHz cache but that is it so far.

Is there a setting being missed? Does it due with ECC RAM? Is input too low?

EDIT: I tried
input 1.9
core 1.4
cache 1.4

and 4.2 GHz and it instantly froze but 4 GHz works on 1.375/1.4v so maybe input needs to be higher but I am feeling really iffy about these voltages or is this normal and safe? Or is it something else?

 

[SomeGuy133]

bump anyone know why my cache can't OC and if I am missing anything?

 

[SomeGuy133]

bump. Anyone know why cache always causes a crashes? What am I missing?

 

[TheEschaton]

bumping this thread because if anyone who actually knows can answer, then I think I could learn something. I have no idea why you can't OC cache; in fact cache OCing is not something I'm very familiar with at all; I had always assumed it overclocked along with the rest of the CPU these days.

 

[SomeGuy133]

bumping this thread because if anyone who actually knows can answer, then I think I could learn something. I have no idea why you can't OC cache; in fact cache OCing is not something I'm very familiar with at all; I had always assumed it overclocked along with the rest of the CPU these days.

normally it OCs perfectly fine and matches the core but every time I touch it even with increased voltage it just freezes so I don't know if I got a bad chip or I am just missing a setting.

It doesn't have to due with ECC RAM does it? Does cache OC change something with RAM? I thought it was independent but maybe I missed something in regards to how it plays with RAM?

 

[TheEschaton]

normally it OCs perfectly fine and matches the core but every time I touch it even with increased voltage it just freezes so I don't know if I got a bad chip or I am just missing a setting.

It doesn't have to due with ECC RAM does it? Does cache OC change something with RAM? I thought it was independent but maybe I missed something in regards to how it plays with RAM?

Like I said earlier, I don't have a clue. Are you OCing via the clock or the "FSB" (not sure what it would be called on Haswell-E)?

 

[SomeGuy133]

Like I said earlier, I don't have a clue. Are you OCing via the clock or the "FSB" (not sure what it would be called on Haswell-E)?

I am using multipliers. I can go up to 44x on the core but on the cache I can't increase it even by 1. I.E. 35-36x just freezes.

I'll test it again later when I got the NAS on again but I am done for today with tinkering.

 

[TheEschaton]

Take a look at uncore overclocking, as I believe the cache is considered part of the uncore:

http://hwbot.org/news/11337_der8auers_guide_for_haswell_e_4ghz_uncore_for_all_motherboards/

 

[LigTasm]

Your voltages are low for an average Hex, I would try getting the 4.4ghz core stable before attempting to mess with the cache again. For the speeds you have I would try 1.400V Vcore and 1.95 or 2.00V VCCIN. With 64GB I would also strongly consider bumping up the VCCIO voltage, you're probably stressing the IMC pretty hard.

Cache really has very little effect on anything anyways, why are you so set on overclocking it?

 

[SomeGuy133]

Your voltages are low for an average Hex, I would try getting the 4.4ghz core stable before attempting to mess with the cache again. For the speeds you have I would try 1.400V Vcore and 1.95 or 2.00V VCCIN. With 64GB I would also strongly consider bumping up the VCCIO voltage, you're probably stressing the IMC pretty hard.

Cache really has very little effect on anything anyways, why are you so set on overclocking it?

its 4x16Gb RDDIMMS with ECC so the IMC shouldn't have an issue write?

Also I saw most people get 4.4GHz with 1.3-1.35v and 1.4 Ghz was on the extreme that people said its not really a good idle for 24/7 use....are they mistaken is 1.4v kosher now?

Are there other names for the VCCIO voltage because some boards label shit weird.

find it odd that 4.4GHz on a hex is higher then 1.3-1.35v for an averable chip.

All of Silicon Lottery ships use less then 1.35v for 4.5-4.7 GHz but they do use fairly high input

Take a look at uncore overclocking, as I believe the cache is considered part of the uncore:

http://hwbot.org/news/11337_der8auers_guide_for_haswell_e_4ghz_uncore_for_all_motherboards/

thanks but i am nearly positive pll mod is not needed for this.

 

[TheEschaton]

its 4x16Gb RDDIMMS with ECC so the IMC shouldn't have an issue write?

Also I saw most people get 4.4GHz with 1.3-1.35v and 1.4 Ghz was on the extreme that people said its not really a good idle for 24/7 use....are they mistaken is 1.4v kosher now?

Are there other names for the VCCIO voltage because some boards label shit weird.

find it odd that 4.4GHz on a hex is higher then 1.3-1.35v for an averable chip.

All of Silicon Lottery ships use less then 1.35v for 4.5-4.7 GHz but they do use fairly high input



thanks but i am nearly positive pll mod is not needed for this.

I was referring to the parts of that article that dealt with the aspects of the OC that did not deal with pll modding. I agree that is not needed

 

[SomeGuy133]

I was referring to the parts of that article that dealt with the aspects of the OC that did not deal with pll modding. I agree that is not needed

I put in green and bold what I got so far and a couple more questions that need answering

ah ok gotcha

Update I increased
input 1.9
core 1.3
cache 1.3

and got 4GHz stable

I increased
input 1.9
core 1.375
cache 1.375

and crashed at 4.2GHz so it seems inout is limiting factor. Is input safe at 2v? or should I keep it at 1.9v?


4.5GHz core is unstable at 1.375v with no added uncore so I will try 1.4v and 1.9 input but I think input ATM is the culprit. I want to get 4.5 for Photoshop and stuff and 4.2 daily. I try uncore again when i get that 4.5GHz. I think uncore is just a bitch for some reason.

What is safe max input voltage? I see some going 1.95-2v but is that safe 24/7?


4.5GHz 1.4v core and 1.9v inout so far stable for a few mins. haven't touch cache yet. I am using max heat test in prrime 95

my H100i GTX is not keeping up with CPU lol. I am getting 92C peak with water cooler but I can drop it a couple degrees. Rad is being full saturated so I could increase fans to 3500 instead of 2500 rpms to drop temp a couple of degrees but the pump and transfer in not enough for 200-220 W TDP

EDIT: yea 1.9v input and 1.4v core seems to be stable for a decent bit at 4.5GHz. I'll do a long benchmark down the road but I think i found the spot that is stable or at least reasonably stable. I don't have time ATM to get super into it. I want to try that uncore again before I call it a day.

NM screw uncore for now because it is just a pain. I could use some more advise before I mess with uncore.

 

[LigTasm]

2.000V is fine, both of my 5820k needed it to run over 4.3-ish. It has a much bigger effect if you're running offset voltages which I assume you want to do after you manaully dial in the voltage.

Your temps are way too high, are you using AVX enabled Prime95? You'll have a very, very hard time running that with Haswell-E. Just google Prime95 and Haswell and you'll find a multitude of problems. I've started using Asus Realbench and x264 encoding to test Haswell/Skylake, you can find a ton of articles about this if you do a quick search.

As far as uncore, it doesn't help much but you don't need to match it to your core clock. Increase your I/O voltage (sometimes called system agent, I'm not familiar with ASRock BIOS). You've got a lot of RAM, and the 16GB DIMMs may be causing your stability issues. I don't think I've seen anyone else tyring to OC with large amounts of ECC installed before, nor 16GB modules of any kind.

 

[Hotar9978]

The 1650 v3 is the Xeon equivalent of the 5930k yes?

First you need to back off on that cache voltage. Earlier guides recommend 1.35v as max but now after a year people are finding their processors degrading. Use 1.25v max for cache.
Take whatever you can get out of 1.25v, probably not much because your board doesn't have an oc socket. Not that it matters because cache have no real world performance impact.

Input voltage is what kills chips, stick under 2.1v. It's also the make or break voltage that'll get your chip stable. More doesn't necessarily means better, each chip favors a certain voltage. My chip wants 1.93v and any less or more it'll become unstable.

Vcore doesn't really matter so start with the 1.4v or whatever your cooler can handle and work your way down after finding your stable input voltage.

As a final note you should have real expectations. True 4.5ghz stable is uncommon for Haswell-E hexacores. Don't be disappointed if your chip isn't capable.

 

[SomeGuy133]

i heard cache affects a lot of things like ADIA or something...forget what it was called.

yea i tried ASUS real bench and thats a weak bench mark. it only draws like 160 W TDP when Prine95 from their site is getting 200-220w TDP.

I'll stick to 1.3v cache then but can you find some sources for 1.25 cache being the new max? that seems really low.

That means basically 0 OC to cache for me if what your saying is true :/

I'll try in out of 1.95v next then 2v and then stop there.

how do i know if this has the OC socket? I thought all modern MB had that now as standard?
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157536

 

[Hotar9978]

Search the OCN Haswell-e owners thread and hwbot, it's been discussed before. On mobile right now and the search function is broken for me. As for the OC socket it's an Asus exclusive for the most part and your board doesn't have it. Last time I checked non oc socket boards tops out around 3.8ghz for the cache, which you whould be able to hit with 1.25v easily.

For CPU only stress testing I like prime95 small fft. 99% of the problems are weeded out within the first two hours and if it passes it should be able to do the 24 hour blend run no problem. For unstable overclocks it should outright fail within ten minutes, then as you get closer to stability individual threads will start to crash. Threads crashing is a very good sign and means you're close to your ideal voltage.

 

[SomeGuy133]

Alright I'll take a look into that in a bit but I also have these voltages that I haven't touched

PCH PLL voltage
CPU I/O
ME
PCH
DRAM (won't touch that because it'll fuck up RDDIMMs i assume)

Trying
input 1.95
vcore 1.3 +.1v
cache 1.25


EDIT: yea the H100i GTX AIO is eating shit lol. It can handle 170-200w but breaking 200 W TDP temps skyrocket. I am getting 95 C at 220+ W TDP lol

 

[SomeGuy133]

man so many people can get 4.5 Ghz and I can't even touch it with 1.4v and 1.95v :/ This sucks. I guess 4.4Ghz it is :/ Pity, I really wanted 4.5. I also can't get 3.8 cache on 1.25v so I am just going to have to rock 1.375v core and 1.25 cache and 4.4 and 3.5 :/

 

[GotNoRice]

EDIT: yea the H100i GTX AIO is eating shit lol. It can handle 170-200w but breaking 200 W TDP temps skyrocket. I am getting 95 C at 220+ W TDP lol

Odd. I'm running an H100i GTX with my 5820k @ 4.5Ghz, 1.4v, and it's rare for me to surpass 80C. Sounds like your issue, really, is that you're just too set on Prime95. There are a number of threads out there already that detail why Prime95 is not good for recent processors such as Haswell-E. You are creating an unrealistic scenario that not only fails to represent any real situation that your processor will ever face, but is also actually dangerous for your CPU. What's the point? My previous 2500k @ 5Ghz would fail when running Prime95 at that speed, yet I ran it like that for over 4 years without a single BSOD during gaming or other real-world usage.

What are your temps with something a bit more down-to-earth, like say, the built-in CPU-Z stress test? Also, do you run the pump in quiet mode or performance mode? What fans do you use with the H100i GTX, and what setting(s) do you use for those?

 

[SomeGuy133]

Odd. I'm running an H100i GTX with my 5820k @ 4.5Ghz, 1.4v, and I can't recall the last time I even saw 80C. Sounds like your issue, really, is that you're just too set on Prime95. There are a number of threads out there already that detail why Prime95 is not good for recent processors such as Haswell-E. You are creating an unrealistic scenario that not only fails to represent any real situation that your processor will ever face, but is also actually dangerous for your CPU. What's the point? My previous 2500k @ 5Ghz would fail when running Prime95 at that speed, yet I ran it like that for over 4 years without a single BSOD during gaming or other real-world usage.

What are your temps with something a bit more down-to-earth, like say, the built-in CPU-Z stress test? Also, do you run the pump in quiet mode or performance mode? What fans do you use with the H100i GTX, and what setting(s) do you use for those?

http://hardforum.com/showthread.php?t=1879864
see that thread for my review of the cooler with various high performance fans. Remember I am pulling 220 W TDP. What TDP are you hitting? 170w TDP is easy for the cooler it is when you hit 200+ TDP that it craps out.

I used to use intel burn test on my 920xm and 3920xm so I figured Prime95 was actually only moderately challenging. Prime95 is considered hard? :/ IBT was so much worse. I heard there was a new version and I was going to give that a shot :/

I thought the ROG realbench was laughable for stress testing :/

 

[GotNoRice]

I used to use intel burn test on my 920xm and 3920xm so I figured Prime95 was actually only moderately challenging. Prime95 is considered hard? :/ IBT was so much worse. I heard there was a new version and I was going to give that a shot :/

Prime95 has been constantly updated over the years. I believe the core of the issue is AVX2, which is supported by Haswell and newer processors. Newer versions of Prime95 use AVX2. Apparently AVX2 is so demanding, it can cause some processors to throttle even at stock. See this post by an ASUS tech for more detailed info:

http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club/2390#post_22900116

The newer versions of Prime load in a way that they are only safe to run at near stock settings. The server processors actually downclock when AVX2 is detected to retain their TDP rating. On the desktop we're free to play and the thing most people don't know is how much current these routines can generate. It can be lethal for a CPU to see that level of current for prolonged periods.

Remember I am pulling 220 W TDP. What TDP are you hitting? 170w TDP is easy for the cooler it is when you hit 200+ TDP that it craps out.

I'm not sure about TDP, as I usually just focus on temps. Are you doing anything besides Prime95 that actually pushes 200+ TDP? What are you at during something like a run of 3dmark or a different stress test?

 

[SomeGuy133]

Prime95 has been constantly updated over the years. I believe the core of the issue is AVX2, which is supported by Haswell and newer processors. Newer versions of Prime95 use AVX2. Apparently AVX2 is so demanding, it can cause some processors to throttle even at stock. See this post by an ASUS tech for more detailed info:

http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club/2390#post_22900116




I'm not sure about TDP, as I usually just focus on temps. Are you doing anything besides Prime95 that actually pushes 200+ TDP? What are you at during something like a run of 3dmark or a different stress test?

ROG realbench only requires 170w at those settings and my temps are fine as stated before. It is just the cooler can't keep up with those kinds of draw. 220w TDP is pretty large and not surprising. F@H only gets like 170 W TDP IIRC

from that review
50C for 170 W
67-80C for 200-208w TDP

 

[GotNoRice]

ROG realbench only requires 170w at those settings and my temps are fine as stated before. It is just the cooler can't keep up with those kinds of draw. 220w TDP is pretty large and not surprising. F@H only gets like 170 W TDP IIRC

from that review
50C for 170 W
67-80C for 200-208w TDP

Ok, but i'm not asking about the cooler, i'm asking about your overclock. Is Prime95 the only stress test which causes TDP to go that high? If so, that is probably just due to how Prime95 uses AVX2 and not really indicative of your true overclocking limit.

Also, you could try downloading Prime95 26.6, which I believe is the last version without AVX2. Compare the temps/TDP to see the effect of AVX2
http://www.mersenneforum.org/showthread.php?t=15504

 

[SomeGuy133]

Ok, but i'm not asking about the cooler, i'm asking about your overclock. Is Prime95 the only stress test which causes TDP to go that high? If so, that is probably just due to how Prime95 uses AVX2 and not really indicative of your true overclocking limit.

Also, you could try downloading Prime95 26.6, which I believe is the last version without AVX2. Compare the temps/TDP to see the effect of AVX2
http://www.mersenneforum.org/showthread.php?t=15504

as I said F@H and ROG realbnech have 170-180 TDP vs prime95 has 220

 

[LigTasm]

as I said F@H and ROG realbnech have 170-180 TDP vs prime95 has 220

That is because of the AVX utilization. You obviously know it all already and don't have any intention of trying to learn or break your preconceived notions, so just do what you want.

 

[SomeGuy133]

That is because of the AVX utilization. You obviously know it all already and don't have any intention of trying to learn or break your preconceived notions, so just do what you want.

I am finding the max stable clock for any case. I can't have it crash when I am in the middle of something or crash 30 hours into an encode. Plus I have always used programs like these. I am honestly shocked that people recommended lower standards for testing stability. The norm used to be IBT and old prime, which was weaker then the current one was frowned upon. Asus Real Bench seems as weak as F@H and thats not even challenging for a CPU so I am just a little lost on why Realbench is considered a good bench when it is on par with F@H. And most people used to use like IBT to test OCs to make sure they would be solid for F@H.

 

[LigTasm]

I am finding the max stable clock for any case. I can't have it crash when I am in the middle of something or crash 30 hours into an encode. Plus I have always used programs like these. I am honestly shocked that people recommended lower standards for testing stability. The norm used to be IBT and old prime, which was weaker then the current one was frowned upon. Asus Real Bench seems as weak as F@H and thats not even challenging for a CPU so I am just a little lost on why Realbench is considered a good bench when it is on par with F@H. And most people used to use like IBT to test OCs to make sure they would be solid for F@H.

Your concept on how stressful a bench is seems to be about power draw, which is ultimately not going to get you far in stability. Feel free to type Haswell+AVX into google and do the research yourself.

Its actually kind of funny reading this stuff, you go to any successful business in the world right now and you'll find the phrase "because thats the way we always did it" to be extremely frowned upon, and for good reason.

 

[Hotar9978]

The thing I've always found odd is that the difference between regular and avx2 stress testing on the hexacores averages only about 30w more. I mean really, will 30w more break the camel's back and degrade processors?

The octacore variants on the other hand can draw up to 450w from the stock 140w. It has been show time and time again that the octacore variants cannot run or pass prime at all for any long period of time without hitting 100c+

 

[SomeGuy133]

Your concept on how stressful a bench is seems to be about power draw, which is ultimately not going to get you far in stability. Feel free to type Haswell+AVX into google and do the research yourself.

Its actually kind of funny reading this stuff, you go to any successful business in the world right now and you'll find the phrase "because thats the way we always did it" to be extremely frowned upon, and for good reason.

lol. no the point is we used to always use the most stressful program to find any spot of instability and now its like lets us a moderately challenging program and call it stable -_-

I ran F@H for 2 days and was stable or at least I didn't run into an issue (doesn't mean I would never run into an issue but I ran prime95 and presto crash. If i get small FFT stable for an hour I know I'll be good in pretty much anything for days so encoding won't be an issue. Realbench ran for 15 mins like nothing because it is not a challenging program. I ran Prime95 and presto crash....only took seconds. Again, looking for complete stability not hey as long as it is mostly stable I am ok with that. I can't afford to waste 30 hours on an encode and find it crashed and have to start from scratch or be working on a massive PS project and it crash. On my main desktop that is a gaming rig hell yea! I go for max clocks that are mostly stable but not my NAS/Server/PS/Encoding machine plus this will be running SnapRAID for my NAS and having a sketchy stable machine for SnapRAID is just stupid. Same with my future Ark server. I am not playing around with instability. Desktop gaming rig...hell I run my desktop rig fairly sketchy in terms of stability. I can't even keep F@H stable on it but I am good because I can get max out of PC for what I need it for but again that is my desktop not a productivity machine.

The thing I've always found odd is that the difference between regular and avx2 stress testing on the hexacores averages only about 30w more. I mean really, will 30w more break the camel's back and degrade processors?

The octacore variants on the other hand can draw up to 450w from the stock 140w. It has been show time and time again that the octacore variants cannot run or pass prime at all for any long period of time without hitting 100c+

Yea I don't think 220 W TDP is unreasonable for 6 core CPU because I have seen reviews run the 8 core at 400 W TDP so I see no reason why this would be bad for it as long as you can keep it cool. I am more concerned about voltage then TDP.

EDIT: and IIRC handbrake uses AVX right? that is a program my 1650v3 will be using a lot and I need it to be stable for that.

 

[LigTasm]

lol. no the point is we used to always use the most stressful program to find any spot of instability and now its like lets us a moderately challenging program and call it stable -_-

And can you explain in detail how you know which ones are moderately challenging and which ones are not? Do you know which instruction sets each different program likes to hit, and how they relate to stability in the applications you want to use? If you really want to get picky, do you run a linux livedisk for some of the stress tests available on that platform that are better for certain subsystems than what we have available to Windows?

Again, your focus is entirely on power consumption, which is fine if you're trying to test your cooling solution but don't care about actual stability. I've seen machines that can run P95 blend for 24 hours straight and lock up trying to do a Windows update.

 

[SomeGuy133]

And can you explain in detail how you know which ones are moderately challenging and which ones are not? Do you know which instruction sets each different program likes to hit, and how they relate to stability in the applications you want to use? If you really want to get picky, do you run a linux livedisk for some of the stress tests available on that platform that are better for certain subsystems than what we have available to Windows?

Again, your focus is entirely on power consumption, which is fine if you're trying to test your cooling solution but don't care about actual stability. I've seen machines that can run P95 blend for 24 hours straight and lock up trying to do a Windows update.

it is one of many tests...you think prime95 is the only one I will test? It is one of the first tests to use to check voltages because it requires lots of power. Which tests do you think will show lack of voltages first? IBT, ROG Realbench, Prime95, AIDA, F@H, TS, or something else? Do you really think Realbench and TS and F@H are going to give you the best first impression that you found the right voltage? BTW that was a rhetorical question.

Once I can at least pass Prime95 and AIDA. I know I found a solid base for voltages and then its worth the effort to test the other ones. The other ones are not going to be big determiners of what voltage is best. They will help weed otu instability but they are not the first ones to use to find the right voltage. You are literally going through the process backwards by using Realbench and F@H first then trying Prime/AIDA.

 

[LigTasm]

Its become obvious by now that you just want someone to tell you your chip sucks because you already know everything there is to know about overclocking Haswell-E.

 

[SomeGuy133]

Its become obvious by now that you just want someone to tell you your chip sucks because you already know everything there is to know about overclocking Haswell-E.

whatever dude. If that helps you sleep better at night. I still got no clue what those other voltages do or if they even matter. Someone was saying increase voltage on VOCCT orr something early for the memory controller but I don't know what that is in my motherboard.

Looking through the thread again as I play around a bit more and It was you that mentioned VCCIO and I am not sure if I should mess with that because if the ECC RAM. Is it safe to increase voltage on ECC RAM or am I increasing only a voltage in the CPU and not the actual RAM? I am unsure if I even need to do that because isn't that the whole point of RDDIMMs is so that they don't stress the actual memory controller?

EDIT: Doing 8 hour realbench now. I didn't crash on Prime but it is too hard to run because of temps. 150-165 W TDP and max temp was 69.