ROG Team Super Pi Challenge Oct 18 - Nov 17 2012

[HiVizMan]

To celebrate the formation of the new ROG Supreme OC HWBOT team. I think it is only appropriate that we have a little benching challenge to get things really started. This challenge is only open to ROG Supreme OC members.

Not a member of ROG Supreme yet, join the team now. Register Here

Submissions open - 18 October 2012
Submissions close - 17 November 2012 12:00 GMT

And what better benchmark than Super Pi.

Download here

Rules here

Challange wallpaper downloaded here

The Challenge - ROG Super Pi efficiency

Post your best 32 mil and 1 mil times.
Both submissions must be run at the same CPU frequency.
Memory frequency is capped at 2400 MHz. Not one MHz more. 🙂
Open to AMD and Intel users.
Both submissions are rated for efficiency and the efficiency rating for each benchmark combined.
The winner has the best efficiency total.

Requirements
Air and Water cooling only, for memory and CPU - No sub zero cooling allowed.
A photo of your benching rig must accompany your submission.
Contest is only open to ROG Supreme OC team members - your entry must include your HWBOT profile link. 🙂
Each submission must have the wallpaper that Chris designed for this challenge.
Each submission must comply with the HWBOT rules for submission.
Each submission must have the motherboard tab of CPUz open.


Prize

There is a CPU or kit of Ram up for grabs.

The CPU will be either a:

Intel® Core™ i7-990X Processor Extreme Edition (Brand New and never used)
or
Intel® Core™ i5-2550K Processor (binned 5.5GHz) not used for HWBOT submission will be certain silver cups in all 2D benchmarks and CPUz

I will let the winner decide which prize they need or want the most.

[Zka17]

Have no idea about your display settings...

But, regarding the SuperPi, you have too many things running in the background... also you have SpeedStep enabled and/or Windows is not running in High Performance mode...

[Skull6]

It is what it is. As this is a work computer, I'm happy with the settings as they are. The next desktop I build...things will be different.

Thanks.

[Zka17]

He-he, I'm not benching on my work computers either... 😄

[HiVizMan]

I will update scores and the leader board in them morning. GMT morning.

[Skull6]

I do so hope to see "Skull6" supporting all the others on the list(s) 😉

[Zka17]

I got an idea! What if I would post the major steps of my memory tweaking for SuperPi? I think, it could help those who lost their faith... hope, HiVizMan will agree with this post... 🙂 maybe, even add the ERs...

So, I started with setting the CPU to 4200MHz and selected the max RAM speed allowed in this challenge, 2400MHz. I found that selecting the CPU bus speed:DRAM speed ratio as 100:133 gives better results... Then went ahead and set the primary timings - in this case, I stopped at 9-11-10-24/1T... and the results are shown in my post #495 in this thread...

Just to summarize:
Super Pi 1M: 8sec 672msec
Super Pi 32M: 7min 38sec 094msec

Then started to tweak the Latency Boundary... first changed the "Auto" to 14 and decreased from there - in my case it stopped at 5 (intentionally I didn't want to touch the voltages yet, all this is at DRAM voltage of 1.65V)

Then the secondary timings were the next...
tRDD went from 5 to 4
tRFC from 128 to 100

At this point these were my results:

Super Pi 1M: 8sec 688msec



Super Pi 32M: 7min 37sec 391msec



The 1M is not too good, but a difference already shows in 32M... 🙂

Continued with
WRT from 14 to 10
tRTP from 6 to 4
tFAW down to 16 (unfortunately, I forgot to note the initial value)
tWRT from 6 to 4

New results:

Super Pi 1M: 8sec 687msec



Super Pi 32M: 7min 37sec 157msec



Then started to play with the third timings...
tRWDR to 4
tRWSR to 4

Super Pi 32M: 7min 36sec 953msec - went below 7min 37sec!



The 1M didn't changed, so not posting again...

tRRSR to 4
tWWSR to 4
CLK Period down to 2

Super Pi 1M: 8sec 672msec



Super Pi 32M: 7min 36sec 922msec



At this point, any more tweaking was not helpful anymore (again, didn't touched yet the voltages), so went ahead and tweaked the OS...

In System Properties, I set the Processor to "Background Services", memory usage to "System cache" and virtual memory to 512 initial and 512 max size on D drive...
Then, in msconfig/Boot.ini tab set maxmem to 600 and clicked pcilock...
The same msconfig, but Services tab: "Disable all"
Restart

Open Super Pi, but don't start yet... ctrl+alt+tab opens Task Manager, there right click on superpi.exe and set priority to "high" and affinity to CPU #1, then end explorer.exe - and run the 1M:

Super Pi 1M: 8sec 672msec - didn't really helped...



Restarted the whole system, repeated the things in the Task Manager and run 32M:

Super Pi 32M: 7min 33sec 219msec



This is an improvement! :cool: Shaved of almost 5sec from my initial 32M time... - hope, this little demonstration helps...

Now will sleep a little bit and tomorrow will start over with tighter primary settings! :cool:

[asipamn]

do oc'd the ram speed affect so much? my vengeance has rating of 1600MHZ with 1.5v...i set it to 1680 @ 1.65v. is it too dangerous?

[Zka17]

It's not dangerous... max it won't boot...

The thing is that the Vengeance is the lower quality RAM of Corsair... but you still should be able to OC it!

You OC'd it at 1680... - that means you used the BCLK only? Why not a higher step from the memory frequency? Click on that and you should see higher speeds too...

However, from your submissions here and on HWBOT I observed that you have 4 RAM modules, but in your CPU-Z shows only 12GB... - that's meaning that one of your sticks is not working...

For this challenge, I would remove two sticks and run only the remaining two in the red slots... less sticks is meaning less stress on your CPU's memory controller, so, better chances to go higher frequencies...

[asipamn]

yes i oc'd from the bclck because when i choose from the the memory frequency it failed to boot. i have no choice but to oc from the bclock. the 12ram detected is corect. i recently upgrade my system from 4 to 12gb ( 2x2gb 2x4gb). should i remove the 2x4gb just using 4gb total ram?

[Alex-Ro]

u should use the ram that clocks the best,try 2x2,see what it does,try 2x4 and see that too.