Quote:

Honestly, I don't know. You would have to have CPU-z open while running the benchmark and then see if the clockspeeds jump above 4.0 GHz. It could be possible that this happens and that it had an influence on scores.

Quote:


OH! shit... hmm. I thought that my settings were simply the MAX clocks that turbo would allow. I will look into that tonight when I make it home, after a few beers in the office

Bean 3930K on RIVE @4.5ghz - 16gb Gskill Ripjaw Z - 780 Ghz Edition - Corsair AX850 - assorted SSD's and Mech drives.

Well, it doesn't necessarily have to be the case, that the scores are inaccurate, but in case of 4GHz efficiency runs, I would try to make sure, that the CPU always stays at 4.00 GHz. Also imagine, you were to compete in a competition and you'd have to prove, that your CPU didn't pass the 4GHz mark, that's much harder, when you don't fix the clocks.

I have decided to submit my results to HWbot, probably also a good way for me to keep record of all my progress. my first submission for the 4GHz 32m superpi
http://hwbot.org/submission/2461431_

I also search for all 3930K 4ghz 32M superpi submissions. Of 5 guys that have done it, I am currently second. The guy in first place has a HUGE 17.116 second advantage over my score. I suppose this is my goal. whether or not my CPU and RAM will allow that is my new mission

linky - Current top spot.

Bean 3930K on RIVE @4.5ghz - 16gb Gskill Ripjaw Z - 780 Ghz Edition - Corsair AX850 - assorted SSD's and Mech drives.

Nice Threat Bean

Did Everyone have a good weekend?? I managed to force 1.59VCORE through my poor Air cooled CPU at one point

After working on my ram more and more, fiddling with timings and frequency, I ended up with 1166.7MHz CL9-11-10-20 1T with tRFC at 100 clocks.
This gave me a nicely improved time of 8m 22.289s on SuperPi 32M.


By this time I was a bit tired of 4GHz, so I decided to turn some cores off and up the cpu itself.
Less than an hour later, I was booting 4.82GHz on 2 cores with 1185MHz CL9-11-10-20 1T with tRFC at
100 clocks. This gave me a new overall personal record of 7m 04.118s. To achieve these clocks, I had
set a BLCK Strap of 125, and upped the FSB to 127, having a multi of 38 gave me my 4.82GHz, and using
the Strap plus upped FSB allowed me to extract a little more out of the Ram.


Being able to boot and bench these speeds with absolutely no issue made me want to get to 2400MHz.
I made that my mission. I laxed out the timings to CL11-13-12-32 and tried to boot 2400MHz,
having down clocked the cpu for better stability (I read online that higher memory clocks are
achievable with a lower cpu overclock). The sucker simply didn't like it and I reset the BIOS again,
this was about the 20th reset for the afternoon.
I figured it was strange that it wouldn't boot those settings, it was only 15MHz over the old 1185MHz,
and I had even laxed the timings.. I decided then to go for CL10-12-11, suddenly it wanted to play ball
Of course, 15MHz more with looser timings was not good enough for me, so I decided to reboot straight away and go for CL9.
I threw in 9-11-10-20 tRFC 183. Booted straight to windows and screenshotted my victory.


Now, of course, I had to bench. I ended up with a bit of a disappointing 8m 22.167, only 0.122s faster than when the ram was at 1166.7MHz.


I am actually really quite unsure as to why I got this result. It was the last proper result of the night. I had limited myself to 1.8VDIMM, and felt I couldn't get much more out of it. It was also getting quite late by this point, and I was being beckoned for cuddles

I had a flick through my screenshots this morning to try and look through what I had done. Higher frequencies with worse results than with lower freqs, lowering the tRFC from 183 to 100 gave me better results at all frequencies. Lowering the tRAS from 20 to 19 gave me much worse scores (almost 2 seconds), although, lowering the rRAS from 27 to 20 yielded almost 1.5 seconds of improvement. I think that this memory overclocking thing is a fine balancing act, it's something that needs to be tested until you find optimal settings, not simply as tight as you can go, but timings that 'fit' with each other. Having discussed with Christian Ney again, he confirmed much of what I had guessed, He does though, have a much better grounding in this field and was able to give me some more advice. One of which was for VDIMM "since those are gskill pcb is good so prolly for your benching you can go 1.9" When I get the next opportunity to clock, I will give 1.9VDIMM a shot and see whether it yields more clocks and/or tighter timings. Another piece of information I found very useful was this: "tRFC you can go low but will be limited by memory I dont have formula in head tRAS = CL + tRP + tRCD most of the time basically all timings are linked" Another short formula that I can relate to. It then makes me think that my tRAS is simply wrong. I will try using that formula and compare it to my current setting of tRAS 20.

One more thing that I noticed whilst perusing screenshots was; on my previous fastest run at 1166.7MHz, my 'rated FSB' (what cpuz says) is set to 5GHz, on my current fastest run at 1200MHz, the 'rated FSB' is only 3.2GHz. Again I consulted Chris and he suggested that it could be the IMC (Intergrated Memory Controller) clocks. I'm not sure how big of an impact this plays, but given the name, I think it could affect the overall performance.

To conclude, I had many funs I'm not sure how much real progress I have made, but I think I understand a bit more about memory overclocking and a few more of it's finer points.
For my next overclocking session, I plan to try and find the Max that my ram will do at 1.8VDIMM regardless of CPU clock speed, and then I will apply 1.9VDIMM to see whether that gives me more options with regards to higher freqs/tighter timings.

As always, thanks for reading!!
Feel free to discuss, ask questions, slander, anything you like

Bean







3930K on RIVE @4.5ghz - 16gb Gskill Ripjaw Z - 780 Ghz Edition - Corsair AX850 - assorted SSD's and Mech drives.

Wow, that was one long post. Thanks for writing down all your thoughts. It's really interesting to follow your progress!

Rewarder says my Words
Thump up

Hey Guys

Just a short (famous last words...) post/update. I know I'm yet to delve into GFX benchmarks, but I am
about to undergo a rather huge overhaul of my system. Where there was once a 6990+6970, there is now only
empty PCI-e slots. The cards are now wrapped in cardboard (I threw out all my antistatic bags when I moved flats)
and with me at work. I am moving from that Trifire setup to SLi'd GTX-780 DirectCU II OC's.
I'm glad I'm doing it now, before I've started any GFX benchies, I'm also glad because it'll some needed performance in modern games at my native 1440p on my U2711

GFX benchmarks are a little way off at the moment, but yeah, just a quick update to say my system is changing. Something a little more modern, and for me, the biggest point is that I now no longer have
ANY AMD products Goodbye RedTeam...

Bean

Attach file:

---

  (589.60 KB)

3930K on RIVE @4.5ghz - 16gb Gskill Ripjaw Z - 780 Ghz Edition - Corsair AX850 - assorted SSD's and Mech drives.

Sweet more mining power incoming :O