Overclock.net banner

1 - 14 of 14 Posts

·
Registered
Joined
·
12 Posts
Discussion Starter #1
Hi folks,

I recently did my first OC - followed the guide from Linus Tech Tips YouTube video since it's the same as my CPU. It went well, but I have a few questions.

My system:
Intel i7-5820k
ASUS x99-A
Noctua NH-D15 air cooler
Corsair RM750i PSU
32 GB Corsair Vengeance Lpx DDR4-2666MHZ (8 x 4GB sticks)
ASUS GTX 660
2 HDDs, 2SSDs, 2 external drives

I leave my system on all day and use it most of the day, and I also do a lot of video work (editing in Premiere, mograph in AE, compositing in Nuke, CG in C4D, simulations in RealFlow, etc.) which is why I went for a system with a lot of RAM and a 6-core CPU.

QUESTIONS:
1) I recently overclocked the system and, following Linus' guide, started off with the CPU core at 45 and the voltage at 1.3. I ran the AIDA64 test and had a couple cores touch the low 80s - I understand that to be not great.
Q: Is it considered inadvisable to keep these settings just if the CPU touches the low 80s, or if it AVERAGES in the 80s? In the former, I might as well keep it at 45 and 1.3, otherwise I'll have to lower it, especially given the long render times I sometimes do on the machine.

2) Deciding to leave the system at 1.25v, I next turned on the XMP ram profile at 2666 - this changed the cpu strap to 125. Deciding to play it safe, I put the CPU core at 35 (35 x 125 = 4,375). Ran AIDA again and the temps averaged in the 70s, two cores (that I could see) went into the 80s.
Q: I'd rather stay with a slightly lower Mhz and know I'm not risking anything with my temps - are these settings good for my "final" settings? You can find idle and load temp screenshots below.

3) I did a bit of Googling and it seems that the XMP profile for the RAM makes little to no difference - 2-5% in a few cases, in some cases no difference. Is it better to just leave the RAM XMP off and know I can boost my CPU core speed a bit to get my MHZ up from 4.375 to 4.5, both at 1.25v? Then again, maybe I won't even notice the difference for the ran RAM at 2666 instead of 2111 in something like After Effects?

4) Until my OC I've never heard my Corsair RM750i fan boot up - it only does it when it's under a certain percentage of output. Previously the fan was never on, even during 100% CPU-utilization Cinema4d renders, which I would sometimes run for 12+ hours. Given that the fan now DOES turn on with my OC settings, is it going to be safe if I continue doing long 12+ hour, 100% CPU renders? Is this safe for consistent use? Say, one or two long renders a week for the next few years?

FINALLY, QUESTION IN TITLE:

5) After the OC I've noticed that most of the time I boot up my computer in the morning it doesn't boot - just goes to a black screen, no BIOS, nothing. I then flip the on/off switch on the power supply and the machine boots fine. Maybe 20% of the time this doesn't happen and it will boot on the first press of the power button. Needless to say, this has me worried, and I suspect it has something to do with the OC. The power supply is new - roughly 9 months old - and I understand it's a very good one so I doubt that's the problem. Any advice for what's going on here?

Any advice would be appreciated. Thanks!

PS

Here are some benchmark stats for my new machine that I'm comparing to my old i5-750 - please take a look and let me know if these benchmark results are in-line with what I should expect from this OC:

Computer Benchmarks I5-750 / I7-5820 3.3 / I7-5820 4.37 XMP

CINEBENCH R15 61.41 ; 261 / 140.62 ; 1008 / 152.85 ; 1258
2m36s / 0m42s / 0m32s

3D MARK

Firestrike 4213 / 4871 / 4924

Sky Diver na / 16534 / 17139

Cloud Gate 10093 / 20892 / 23557

Ice Storm 91986 / 113652 / 132375

UNIGINE

Low 87.2 ; 2196 / 99.5 ; 2506 / 99.8 ; 2513

Ultra 32.2 ; 811 / 33 ; 831 / 33.5 ; 844


 

·
Crayon Evangelist
Joined
·
7,609 Posts
^ Cold boot is the hardest to get through. If you did not notice it happening until after you set XMP, I suspect that you are failing RAM training and might need to manually tune VCCSA...

VCCSA is kind of tricky, it is not like core voltage that you can just always add more for better stability, too much can be as much a problem as too little. When tuning VCCSA, or messing with RAM/Cache in general, it is a good idea to completely power down after BIOS changes. This helps mimic the conditions of your cold boot problems. Go slow with the voltage, cold boot without warm boot issues means it is probably pretty close.

As far as the temps go, I would rather see you stick with the 125 strap, and the lower overclock and hopefully lower voltage and lower temps. With the 8+ hour renders you do, a change in ambient temperatures, or dust build up over time, or a temporary air blockage (clothing, animals, dishes, whatever else you may sit on/infront or your rig) can send your temps much higher. Better safe than sorry with a render machine IMO. Gaming rigs I feel have less need for mission critical stability and temperature considerations. Overclocking into the danger zone may take years off of the CPUs life compared to the few minutes it may take off of a long render.

I have an X99-A and a 5820k as well, great combo for not allot of scratch!

Good luck!
thumb.gif
 

·
Registered
Joined
·
12 Posts
Discussion Starter #3
Thanks for the feedback!

Regarding your suspicion that it's the RAM, that kind of makes me want to just disable the XMP entirely - maybe go back to 1.25v and 4.4x100.

Given what I'm using the computer for, how much of a benefit - if any - do you think the XMP is giving me? If it's on the order of 1 - 2% then I don't mind turning it off entirely.
 

·
Crayon Evangelist
Joined
·
7,609 Posts
As far as I understand it, Core is king, and will net better performance in practically every scenario over the equivalent overclock on RAM. However, in heavily multi threaded workflows, like rendering, RAM will have an impact, and getting a good core +RAM overclock is really the way to go. Since VCCSA barely has any effect on heat production inside the CPU, you should be able to get both without too much worry of overloading the CPU. VCCSA is not really all that hard to tune. in your case it is only unstable for that first boot, so again...its close, tune it and have the best of CPU and RAM overclocks.

Also, the 2666 RAM divider is mighty strong on the 100 strap. If you have a rock solid 44x on the 100 strap, you may just try manually setting the RAM using the specs provided by the manufacturer.

Is that 2 kits combined? I dont see a 32GB 8x4 LPX kit with a quick google search, they are all 4x8GB.
 

·
Registered
Joined
·
12 Posts
Discussion Starter #5
Thanks for the help!

Yes, it's two kits of 4x8GB for a total of 8 sticks.

Thanks for the insight, but I don't really have any idea how to modify VCCSA (or what it is). How should I modify it, and in what direction? What's the easiest "bang for your buck" solution I should go for here - again, I'm not against just disabling XMP, but am willing to keep it on if you really think it's worth it. Thanks!
 

·
Crayon Evangelist
Joined
·
7,609 Posts
Filling all slots puts stress on the IMC. Mxing kits, even kits of the exact same make/model/quality puts strain on the IMC. Overclocking Core adds strain on the IMC (faster R/W/C requests).

Honestly, I am not surprised you cant run XMP. While 32GB is not particularly difficult, those kits are rated to run at XMP at 1/2 the density you are trying to run with. I have watched a few people on this forum struggle with it. You will have to tune them manually to get them to run. VCCSA is your best bet, along with maybe some additional VDIMM, or at worst running JDEC standards.

VCCSA will be in the BIOS under "AI Tweaker" tab, it can also be called System Agent. Do NOT go over 1.200, and preferably under 1.15. Stock should be right around .900, so I would try between .900 and 1.050 to start with.

I do think it is worth taking ram higher than 2133 for sure. I suspect if you are willing to tune it and keep with it, you can get both kits to run at XMP speeds, but it is going to be more difficult than just switching from auto to XMP.
 

·
Registered
Joined
·
12 Posts
Discussion Starter #7
Thanks - I just changed the CPU System Agent voltage from Auto (believe it was 0.81ish) to 0.95. Should I now do some more stress tests, or is this only modulating how reliably it starts up? I.e. will temps change at all with this new setting? I'm thinking I'll leave it at this setting for a day or two and see if the startup issue is solved.

When you say my RAM kit is rated to run at half the density I'm doing, are you referring to the fact that I have two 16GB kits? I just assumed you could combine the two and XMP would not be affected - and yes, they are the same type of ram, although I think one kit has a CL number that's 1 higher or lower.
 

·
Crayon Evangelist
Joined
·
7,609 Posts
Did you power all the way down after changing the System Agent Voltage? That is an important step, because your RAM is having issues passing cold-boot memory training, and a reboot does not go through memory training to the same degree it does during a full power cycle.
 

·
Registered
Joined
·
12 Posts
Discussion Starter #9
Thanks! Yes, just did a cold boot - turned system off, then flipped switch on back of power supply off for about a minute, then booted up.

Would you suggest I now do some stress tests again? Also, regarding my original question above, is it bad if a core just HITS a temp in the 80s, or if it AVERAGES a temp in the 80s? For my previous settings when I was doing the tests (results above) the averages were in the mid 70s, but a temp on a core would sometimes hit 80 then go back down. Is that bad, or does not matter?
 

·
Crayon Evangelist
Joined
·
7,609 Posts
There is not a hard line answer to that question. Hitting over 80 on a couple cores during a stress test is not terribly uncommon. I would pay more attention to temps an hour or two into a long render, anything touching 80 I would reduce clock and voltage.

So no, I dont believe your temps are out of control. While I dont put much stock in Aida64 for a stress test, it certainly puts out the heat when you check the FPU box, and you likely wont see above 70 in real world rendering or gaming with temps in the low 80s with FPU testing. That FPU gets HOT!
 

·
Registered
Joined
·
12 Posts
Discussion Starter #11
Well this morning I had quite a scare - tried booting the computer up and it wouldn't start - as normal - so I hit the power switch on the back. But this time it STILL didn't boot after I hit the power switch. I had to do it three times, waiting 20+ seconds between each reset, until it finally started. Really freaked me out because I started wondering - what if it just won't boot? Without it going into post I'm stuck!

Thankfully eventually it did boot - I changed CPU System Agent voltage from 0.95 to 1.05 and did a cold boot - worked fine, but we'll see how it is tomorrow morning.

Another question that's been puzzling me: I have the strap at 125 and the cores at sync all cores and set to 35. This should give me 4.375, which it does when I do a stress test or high CPU activity and monitor it in Win 10. However, under the ASUS bios "Hardware Monitor" on the right side of the screen it says "Ratio = 33x" and frequency 4125. Shouldn't it say 4375? And why does it detect the ratio at 33 instead of 35? Is this nothing to worry about, or have I set up something wrong?
 

·
Crayon Evangelist
Joined
·
7,609 Posts
33 is the default mumtiplier for a 5820k, it could just be reading it wrong under that tab, as long as it is running as expected, I wouldnt worry about it. But I really dont have a whole lot of experience with the 125 strap, I run 3200 ram on the 100 strap. You may also need to increase VDIMM right along side System Agent. ASUS actually allots for two different types of VDIMM, boot and eventual. The eventual voltage can be found in the timing menu. You might try setting the VDIMM on the main AI tweaker page higher, and since you know the RAM will run once booted, you change the eventual voltage to what your currently running. So if you are using a 1.2V kit you would want:

VDIMM on AI Tweaker Page :1.35
Eventual VDIMM in timing page:1.2

There is another work around as well, which really is NOT the right way, and that is to enable ATTEMPT FAST BOOT and ATTEMPT COLD FAST BOOT in the memory timing menu.

I would really try to tune it rather than use a work around.
 

·
Registered
Joined
·
12 Posts
Discussion Starter #13
Once again, thanks a lot for the help.

Yesterday when I booted my computer in the morning it went to the standard black screen (system agent at 1.05 at this point), then just shut itself off and rebooted itself to another black screen. I did a cold power boot and it worked fine, but that really freaked me out.

I've decided to just abandon the XMP altogether and set everything back to the regular OC - went with 1.25v, 100 strap, and multiplier of 44x. Sorry to have you help me with all that XMP stuff only for me to abandon it, but I'm just more comfortable this way, and I'm willing to be the difference is negligible.

Going to do another stress test to confirm it's good, but it should be fine. Do you have another stress test you'd recommend over AIDA?
 

·
Crayon Evangelist
Joined
·
7,609 Posts
No worries! You weighed the benefit vs. difficulty and made an informed decision. It is your PC, do what is best for you!

Aida is fine if you use the FPU test in conjunction with the rest. The thing about FPU testing is that it gets hotter than you can ever expect your CPU to actually get in 99% of all common workloads. Aida without the FPU test is kind of junky. Something like ASUS RealBench is a bit more 'real world' testing without hammering the FPU into oblivion. I am a huge fan of using x265 encodes for stability, because it is realistically the hardest thing that MY system will ever encounter, and HWBot has a nice GUI-wrapped version of x265 that is point and click instead of command line.
 
1 - 14 of 14 Posts
Top