Also, we do not have any further technical information on exactly how ACC works. We have had several discussions with AMD but nothing concrete to report so far. There are some valid theories about what AMD is doing behind the big curtain, but nothing substantiated yet. Throwing caution to the wind, I would say inter-register fine tuning is occurring along with some other pipeline and clock optimizations. But I could be completely wrong (would not be the first or last time) if AMD ever opens the curtain for us to take a peek inside.
Foxconn has provided a new BIOS update tonight for the A79A-S motherboard that allows us to switch off the TLB fix as our preliminary results with the 9600BE have been a bit strange. Also, we have production release RAID 5 drivers to test in order to answer a few storage performance questions and we will take a quick look at 8GB memory clocking on this board in our update.
In regards to our 9600BE testing to date, our results have been unusual to say the least. We easily took our CPU up to 3.1GHz on the Foxconn board, but we have not noticed any dramatic differences with ACC enabled. As we approached the 3.1GHz range, our voltage requirements were 1.4825V with ACC enabled and 1.525V without it enabled. However, no differences in core clock speeds were noted on this 790FX/SB750 equipped board at these voltage settings. It was not until we installed this CPU in our 790FX/SB600 board that we noticed a true clock speed difference as our CPU sample would not go above 2.8GHz with 1.525V on the core. We found it strange that ACC did not really affect our core clocks as it had on our B3 processors. So, either the SB750 chipset provided a boost (not likely, but who knows at this point) or it's the board/BIOS combination from Foxconn. We are still testing on other 790FX/SB600 boards to discover any differences, but so far it appears this CPU is hitting a 2.8GHz wall on the Gigabyte, MSI, and DFI boards.
We decided to do one more test and that was to increase CPU core voltage to a maximum 1.575V (does that void the warranty?, editor - probably so) on the Foxconn board and see where that would take us with this CPU. With ACC disabled we hit 3.2GHz and thought that would be our limit based on earlier results. After some fine tuning of the cores with ACC enabled, we hit 3.33GHz as shown in the screenshot below. We would never run at this voltage on a 24/7 basis, but with the help of some serious cooling assistance from our CoolIT Systems Freezone Elite, we ended up with a platform that has been stable looping various tests for the past couple of days. That said, with the TLB fix enabled, performance has not been good so we are looking forward to what the results will be with the fix disabled.
We finally get our first 1GHz overclock on a Phenom but wish it did not take so many volts. Additional images are in the gallery.
No comments:
Post a Comment