The results were as good as expected, but one of his obvious disappointments was the front side bus - the same as I experienced with my old QX6700 sample.
Last week, I also got a QX6800 and tried to run it on an ECS NForce 680i board which, being an actual Nvidia mobo, looks & feels exactly like the EVGA unit Theo uses.
That same board, with QX6700, never went above FSB1200 to boot Windoze, no matter what I set in the BIOS (even up to 1.5V FSB) or what prayers I recited in front of it.
With the recent P24 BIOS rev, and assisted by the Zalman CNPS9700 monster, one of the most powerful copper air heatsinks around, the QX6800 sample in my hands fared a bit better - I could get 3333 GHz CPU with FSB 1333, the same as the QX6700 gave on Striker Extreme, at 1.36 V CPU and 1.4 V FSB / NB voltage settings.
I then updated the board with the brand new P25 BIOS, and decided to set the multiplier to 416 x 8, with effective FSB 1667 - the same as my other reference system, the Xeon 3220 UP on Asus Striker Extreme with something like 11 fans inside. I didn't change any voltages this time, just tried the setting - not expecting it to actually work.
Well, it did.
The QX6800 booted Windows at the very same settings as the cherry-picked Clovertown UP X3220, i.e. 3333 CPU, FSB 1667, and GEIL 2 x 1 GB DDR2-1160 CL4 memory, runnng here at DDR2-833 CL 3-3-3-5 at 2.2 volts. So, I was in Windows, ran Sandra tests, fine - 8.7 GB/s Sandra memory test is awesome for an Intel Core 2 platform. However, after around 5 minutes, the system was down.
Immediately upon that restart, the check in the BIOS temperature monitor revealed that, while the CPU ran at respectable 52 deg C (not bad at all for a 160++ W TDP on air), the Nforce North Bridge was at 73 C - and that was after the restart. And, that was 1.4 V FSB and NB voltage, not even 1.45 V like Striker Extreme setting.
Was a high FSB a problem? I changed the settings to 333 x 10, i.e. 3333 CPU and FSB 1333 - the Sandra ran again fine, I started 3DMark06 CPU test, it went down within 15 seconds. Another BIOS temperature check? CPU temperature 51 C, North Bridge at 72 C - no difference.
Then I moved the system on a pedestal facing my lab room aircon DIRECTLY - the air was blowing directly at the open system innards. FSB back to 1667, guess what - the thing completed the 3DMark06, whole test this time. I also ran Cinema4D and Povray renders as good stability gauges, just in case. Upon (proper) restart, the NB temperature was 65°C.
The last round was leaving the system to boot, do the full memory test and then go to BIOS temperature monitor, leaving it for 10 minutes, this time without air-conditioning in Singapore weather (read: London's last hot summer readings). On FSB 1333 setting, the NB heated up from 63°C to 68°C in that time, while, on FSB 1667, the NB was aroused from 64°C to 70°C in the same short period - while in both cases, keep in mind, practically IDLE. As you can see, the Nforce 680i is hot and hotter, no matter what FSB you run it on, and it needs something to take that heat away. If the FSB makes no difference, then Nvidia simply has a very hot potato chip on its hands - a semicon process shrink may be what the good doctor prescribes here.
By the way, if keeping the FSB at 1600, and upping the divider to 9, I could boot Windows at 3.6 GHz (400 x 9) and run Sandra, at 1.375 V CPU voltage. 3DMark06 CPU crashed there, I guess a bit more voltage is needed for the CPU here. This is where the QX6800 solves the multiplier-limit problem of the XeonUP part - and, of course, the multiplier here can go way above 9...
Summary: Some, but seemingly not all, of the QX6800 parts really do have that fantastic FSB performance of the XeonUP parts combined with the unlocked multipliers for sky high overclocking potential - part of the reason, I suspect, is due to them possibly coming out of the same bin as the high-FSB Clovertown parts for Apple Mac Pro.
If using Striker Extreme, with its elaborate heat pipes and zillion fans (one on each copper block), you're fine at FSB 1667, maybe higher - at least I was for the past month or two. However, the ECS / EVGA (read: Nvidia) generic 680i board design is woefully undercovered heat-wise. Theo will be running it in water cooled mode soon, and I feel even more than that may be needed as we aim for a reliable 2+ GHz FSB operation. µ
Sign up for INQbot – a weekly roundup of the best from the INQ