Raja: OC guide revision recommendations

[HalloweenWeed]

This thread is a result of my discussion W/Raja in another thread regarding the updating of his two Overclocking (OCing) thread stickys on this forum. He asked me for specifics, and it has occurred to me I may not be the only one W/good opinions on how Raja should update/revise his OC guides. So I made this thread so all of you could chime in without derailing the other thread.


Raja,
First, there's no mention of Vreg downclocking. If you have a WB on your Vreg, this is not an issue. But for me, I watched my Vreg external temp W/a Raytek, and every time it got >81*C on the hotspot, the LinX Gflops dropped significantly. This was a limiting factor in my CPU OC. You should have a discussion of this in at least one of your guides. Since the RIVE does not pass the Vreg temp info, most ppl have no idea this is happening, and some believe they have made a sucessful higher OC when in fact their CPU is being throttled by the board, due to Vreg overtemp.

Second, VCCSA is crucial to not only your memory speed, but the CPU core speed as well. As you raise your multi, you need more VCCSA, despite the unchanged memory speed. According to your guides, one would think that once you have a working VCCSA for your memory speed you could leave it alone. If you raise VCCSA just a little it allows faster core speeds without raising Vcore and thus cooler CPU temps. It is better to raise VCCSA to get your OC, than raise Vcore - up to a point - and you do that when you get the same signals (such as BSOD 101) as to raise Vcore. The method would be to raise VCCSA when otherwise indicated to raise Vcore, then if that don't work reset VCCSA and then try raising Vcore. I personally would suggest a limit of 1.4V or so - conservatively.


Third, there is nothing about the use of CPU Vcore in the UEFI BIOS guide. Ppl come here from using X58 OC, and think it is similar, and it is not. They automatically select manual Vcore and begin OCing. Many do not know what the implications of that are. I am repeatedly having to explain it to them when I help them W/their OC. They do not know they can get to 4.8GHz W/3930K (and a Vreg WC) with variable voltage, nor do they know (sometimes)
offset = variable
and
manual = fixed.
You should explain it to them in the UEFI BIOS guide. I was able to OC to 4.6GHz W/my 3930K and auto variable Vcore, just by raising VCCSA (and the other settings to remove or increase the current limits). A lot of these ppl asking for help have already raised their Vcore to 1.4V at this point, because that's what they thought they had to do.


My other suggestion is more precipitated from personal experience, so it is more of a soft suggestion, as I am not totally sure of the accuracy:

Ai Overclock tuner: Neither "auto" nor XMP should be used when OCing. I know you did not recommend to use XMP, and you recommended using "manual." But ppl still try to use them when OCing, they are not getting the message. It may be acceptable to use XMP to find your memory settings at stock CPU speed, then hard-select them in "manual" mode - providing they have the newer (2012) XMP tables programmed in their memory modules for SB (quad-channel) use. When the OC is somewhat unstable, upon BIOS POST the settings can get messed up if not set on "manual," causing a worse instability, possibly causing a corrupt Windows build. I think you need to stress this a bit more as ppl are not getting the message.


Thank you Raja, and we do appreciate your help.

[Arne_Saknussemm]

First I reckon there is no mention of CPU cooling either but people should know about that...the guide is more about the BIOS options than a cover all A to Z OC guide.

Second I'm pretty sure it says in there to keep vccsa within 0.3 of vcore and 0.6v of VDIMM so raising vcore would logically lead to raising VCCSA. The thing about VCCSA overclocking I really am not sure about. In fact I have seen BSOD codes that with sandy bridge were associated with VCCSA turn out to be remedied not with VCCSA on SB-E but with vcore leading me to think that it's importance is very much memory stability related and not so much CPU OC. I would appreciate it if you could point me to some backup for the VCCSA theory. I presume it has worked for your system but I have not seen much evidence for this anywhere else. Indeed I have seen plenty of recommendations to limit VCCSA to 1.1 or below...

Third...maybe this could be clearer...but the info is there to be found if you want it....people should be encouraged to read and find things out for themselves a bit I reckon

[marcmartyn]

It would be a very good guide. By now, it has been tested a lot.
With that clarification of some doubts about it, we would have many things clearer. and we could go further with our processors.


To infinity and beyond.

[HalloweenWeed]

Arne Saknussemm wrote:
Third...maybe this could be clearer...but the info is there to be found if you want it....people should be encouraged to read and find things out for themselves a bit I reckon

This is Raja's sentiment as well. I am just trying to reduce the number of major causes of OC failure that ppl are asking for help with.

About the VCCSA, OK, I have looked and I see what you mean Arne. I seem to be the only one touting this, though I have only raised my VCCSA to 1.25V. That said, I am willing to concede the VCCSA part. For those few that might be interested: Back last December 2011, nobody on this forum could figure out how to OC the RIVE despite the guides and the YouTube vid. Half could not even get their rigs stable (a BIOS update eventually remedied this). There were appeals for help here on the ROG forums, but nobody knew how to help. So I had to ask the EVGA forums for help, and that's where I learned about raising VCCSA. You can read about our OC's back then here (if interested):

EVGA forums: X79 3930K Overclock discussions

Even though we were on the EVGA forums most of us initially on that thread had RIVEs, or would have the RIVE soon. We didn't get a good OC until approx Dec. 19. Several ppl here on this forum now was there. Much thnx to mdzcpa, xerozen and loveordie (liveordie on this forum). mdzcpa speaks of the Vreg throttling issue, the Asus ROG RIVE OC video, Raja, and Shamino here:

EVGA forums: X79 3930K Overclock discussions - Post #86


An excerpt:

"The OC guide contradicts JJ's video in a few important areas. Most importantly in the memory tweaking area, and in the proper use of voltages. JJ's video recommends to overvolt the snot out of everything. Although this can get the average Joe User up and running, optimum clocks will be reached using a much less ham fisted approach. Overvolting beyond what is needed just creates heat and throttling conditions on this warm running X79 platform."

Nicely put mdzcpa. I was led astray by the Asus ROG RIVE videos too. I suppose this is more just FYI than any evidence for any argument. I just thought you might like to know. It took a long time (months) before this ROG Rampage forum built up to a level of support comparable to the EVGA forum.


So you see, I am not just speaking for my own results - you can see these on that EVGA thread. I have seen others with improved results using my technique(s), but none with extreme cooling. I didn't know until recently that I was the only one arguing the use of VCCSA to reduce Vcore.

[Arne_Saknussemm]

HalloweenWeed wrote:
I seem to be the only one touting this,

Copernicus was in the same position....didn't make him wrong;) I didn't ask for backup in a doubting way.... was genuinely interested. I have played about with this a bit on my system and it hasn't got me anywhere but that is my system.

Thanks for the links I'll have a read through...

[marcmartyn]

Hi HalloweenWeed

I have read the previous post from EVGA. According to the last question you asked me in a previous post. My results in LINX with 4.8 MHz and memory to 2133 is 100 Gflops. I see that your results are from 198 to 200. I'll try different settings to see the results. I worry so low ranking.
I read that may be the temp of the VRM or is the version of Linx AVX which offers 196 Gflop?
Mine is 0.6.4 Linx normal, not AVX

[HalloweenWeed]

marcmartyn wrote:
Hi HalloweenWeed

I have read the previous post from EVGA. According to the last question you asked me in a previous post. My results in LINX with 4.8 MHz and memory to 2133 is 100 Gflops. I see that your results are from 198 to 200. I'll try different settings to see the results. I worry so low ranking.
I read that may be the temp of the VRM or is the version of Linx AVX which offers 196 Gflop?
Mine is 0.6.4 Linx normal, not AVX

Well, I see two possible reasons for your low LinX Gflops. 1st, if you downloaded LinX from the normal sources the "binaries" are two years old or more. The LinX version is just the interface, the LinPack binaries are up to 11.0.1.005. You can update them by going to the Intel site, and downloading the latest Linpack for Windows, then copying the exe file to your LinX directory. Or you can use the download that I host:
LinX.html

The other thing is that >180Gflops is with HT off. With HT on you should be getting 110-125Gflops. That means yes, you may also be experiencing the Vreg downclocking. You can tell two ways (once you update your binaries):
1. The Vreg takes a bit to heat up to the downclock level. About 10min. in my case, but could happen sooner or later depending on your exact conditions - such as higher voltages for instance.
2. By reducing your OC. If the Gflops go up when you reduce your OC, then it would indicate downclocking.

GL.

[cx-ray]

Just FYI I get 161-162 Gflops with a 3960X @ 4.6Ghz HT enabled. Running LinX + latest binaries and max optimal mem for the application of 9216 MB. If you score considerably less there's something not quite right with your stability during the test.

[Raja]

Hi HW,

Thanks for the suggestions:

1) Arne answered the part about VCCSA. It's quite simple, as you increase CPU core speed the IO increases, which means the memory controller has to pass more transactions. EIther way, when you keep the VCC~VCCSA delta, you end up raising VCCSA up on high OCs anyway.



2) There is a guide in the Maximus section for overclocking using Offset mode. I can just add a link to that. It's what I usually do when someone asks or needs to know. All that happened was I did not put the hyperlink into the word "here" under the Vcore description in the R4E guide. The hyperlink is there now.

http://rog.asus.com/forum/showthread.php?2162-Overclocking-Using-Offset-Mode-for-CPU-Core-Voltage&go...


3) I think I did mention somewhere the need for actively cooling the board. I can prolly add something to make it a bit more clear if I feel it is required. On the R4E the VRM has two-stage protection - thermal and current. The first step throttles the CPU. at second step (after x time) board will shutdown. Prolly didn't go into much detail, but I don't see too much confusion now, and if there is we usually answer it on the forums here.


This is just a minor gripe of my own I have found and am saying out aloud here on the forum. The more I put in the guides, the less people read them. People's attention spans (especially those not well versed) are horrifically short. Those guides are considered too advanced by folks already. I think I'll have to diversify the number of guides for various mindsets with the next platform launch - if I have time. With that said, it is nice to see that some people read the guides and help others. It shows progress.

Thanks for the feedback 🙂

-Raja

[H3avyM3tal]

Can anyone elaborate on the VCC VCCSA delta subject?