AMD FX-8350 Super Pi 32 Mil 9m19.828



The last three months I have been competing in the Pro Cup Q4 contest. I tend to bench for me, and my results are for my enjoyment and pleasure alone I hardly ever submit scores. I like the tweaking side of things, the OS set up for specific hardware combinations is a bit of a hobby. My team mate for this contest, 8 Pack, is an exceptional over clocker, he can get an extra 100MHz out of hardware that others can not. He also benches to have the world’s top scores and is constantly posting top score. So you could say we would make a good team because of those differences, and we do.

You can go have a look at his impressive results here.



However Ian is more of an Intel kind of guy, and I am a bit of a Fanboi about the processors and the brand. Add to that Super Pi 32mil is the Blue Ribbon of 2D benching and I was hooked when Ian asked me if I could do that leg of the Pro Cup.
This is a story about that journey.





This journey would not have been possible without one man’s brilliance. The Stilt has single handily changed the landscape of AMD benching. Without The Stilt and his body of work no sub 10 minute Super Pi 32 mil scores would be possible I do not think. Thank you sir and long may you continue to explore.

You can read up about this HERE

The set up.

AMD FX8350 8 Core Processor
ASUS ROG Crosshair V Formula-Z
Bios 1701
Corsair Dominator Platinum
PSU Enermax
CPU LN2 Pot – Custom built by Thomas Keating Ltd.
OS – Windows XP Pro (SP3)
Hitachi 1TB mechanical hard drive
LN2 Memory Pot

The first thing to do when going sub-zero is to make sure your motherboard can take the cold and survive an extended period of benching.



This is what my boards look like when I start to prep them. I paint them with liquid plastic to make them condensation proof. Much cheaper than Liquid Electric Tape and has the same conductivity factor as LET.

Artist Eraser is the next part of the process, I block up the one set of dimm slots.





Next is the paper towelling. Here I use the industrial paper towel rolls and will use an entire roll in a week when benching different platforms. So I buy them in bulk, pallets of 48 in fact. Makes my wife go crazy because out house is really small.



That is about it you are now ready to bench.




There is a whole heap of stuff that goes on before any good score is produced. All most folks see is the finished product, the score. That is just part of the story.


What you do not see of often hear is the days and weeks that get eaten up in the binning process, in the testing of OS phase once you have found two or three good processors. And the other hardware binning that is needed too.


Let us start with the processor. I binned over 20 CPU’s for my first good processor. Then an outstanding bencher from Germany, Roman came along with his first CPU and benched at over 8GHz blowing my submitted score and all my backups out the water. That meant going back and starting all over again. This time I tried not only 8350’s but samples of all the AMD processors that were available. Yeah that includes the APU products too. Until finally I found one CPU that would let me bench close to where he was able to run 32 Mil. This process of searching and binning had taken nearly three weeks.

How I bin. I set up my motherboard, the Crosshair V Formula-Z, with a single profile that includes voltages for the processor and memory. Then with the multi at 20x and set a high FSB of 300 I expect each CPU to boot into OS at 6GHz with all cores active. I then use TurboV to increase the multiplier by one each time, without changing any of the other setting values. I keep the two best CPU’s that were able to clock the highest at that set voltage. This is repeated time and time again until that one golden CPU is found, or not as is the case most often.

Different CPU pots were tried to see which was best and in the end I went with the pot that Trevor Walker technical director of Thomas Keating Ltd designed and fabricated for me. The pot is fabricated from 12mm copper rods and a solid round bar base and not machined from one chunk of copper, held together by cryogenic glue. These guys make stuff for the aerospace industry. Love your work dude I owe you a pint big time.

What you may not know is that I also bin motherboards. I tried with the Crosshair V Formula and the Z versions, 12 in total. I kept three of those twelve as performance was better by some measurable mark in Super Pi efficiency. One CHV and two CHVZ boards.



Just some of the boards, I am too ashamed to show you the others as they are just all over the floor. I am not a tidy bencher at all and tend to take over the entire downstairs area.

The same goes for the memory by the way, sadly I damaged my best set of memory two days before producing the final result, my back up for the contest otherwise I am sure I could have improved on the efficiency of the run. None the less I did have backups and had done the preparation work in determining what worked and what did not work.

The whole process of the contest was heaps of fun and stress in equal parts. The first few sessions I could not get the one board and processor to run dual channel memory.


9.42 run.




I was able to improve on that run with a different board but still not get the pesky dual channel to work. And that was costing me big time.


Finally I got the combination right, only to damage the memory sticks as I said earlier.


I am quite embarrassed to say that the excitement got to me that day when it all clicked into place. You see I forgot to set CPUz to read the correct core. For those of you that do not know with AMD you can overclock a particular core higher than the others. PSCheck 3.4.1 is the application that allows you to do this, and you can change multiplier of one core. I then use TurboV to change the FSB. Thing is CPUz reads core #0.

So on this run below I was actually benching at 7990 MHz according to my calculations, give or take a 10 MHz. Not sure I made 8GHz benchable but must have got darn close.




Now that I had a level that I could bench at I saved the profile in my TurboV and started to work on my memory. At this point I realised my benching kit was damaged and had to turn to plan B. But at least I had done most of the hard work.




This is the final outcome. My final back up for the contest and one that I did not need to use.



My best run and the product of three months’ work, over 600 litters of LN2. And speaking of numbers, I must have run the benchmark at least 2500 times over that 3 months. It was all consuming and I had very little time for any of the other things that were going on in my life at the time.

This is the method I used.

Boot into OS
TurboV and Super PI in the start menu

Get Super PI 32 mil ready to run

Set frequency of strongest CPU core
Run stop bat file
Wait for 3 to 5 minutes and do your wazza
Wazza is 3 x the size of my maxmem well spotted Arne. 🙂
D --> C pause 10 seconds C ---> D pause seconds D ---> C 45 seconds pause.

Run benchmark.

Repeat over and over again changing FSB and hope that it does not fall over because that mean new OS. After any blue screen ditch the OS and clone from ISO found that a blue screen or any freeze impacted on the times measurably.













In conclusion special thanks to Mansfield Cryogenics for their willingness to drive down South and replenish my two Dewars, and of course ROG for the great motherboard design. Sadly I cannot thank AMD as I had to buy all my own processors and took a huge hit selling them on. But hey that is the price you pay for the sport you love.