Well, after we encountered all the issues on the night of the Swiftech Challenge we decided to come back on Wednesday to try it with gear that we knew would be more stable. Gabe from Swiftech showed up with a spanking new EVGA 680i mother board, and Charles (FUGGER) contributed two sticks of tested and true Corsair XMS Pro DDR2 8000 ram. He even allowed us to lap his retail QX6700 quad core CPU we had previously used the night before. Everything else seemed stable.
Gabe from Swiftech, Free Cable Guy and me.
Everyone else - Jin, IFMU, Sgt Vipor, Brad, Tony from OCZ, as well as Jeorg and Tom from Enermax were out watching Vince (K|ngp|n) and Charles overclock Charles’ X6800 to 5100mhz and nearly crack 10 seconds on a Super Pi 1M run. That is chronicled elsewhere so I’ll spare you. Kevin, OnePageBook (another of the top overclockers in the world) was also invited but something came up last minute and he had to decline.
First we set about to lapping FUGGER's X6800 so he could get on with attempting WR's: ho-hum
Gabe & Kris lapping away:
Lapped to 300 grit:
Lapped to a perfect reflection:
Then we lapped our test QX6700, FUGGER's pride and joy and mounted it into the test bed:
Our new and improved stable test bed & water loop:
We then mounted the Apogee GT with the smaller gasket (flat base). Gabe touched it up with 1000 grit to clean up a few marks that remained after the previous night's testing. We were much more finicky and exacting during the mounting process this time. We used regular nuts and turned them down to barely contact the springs of the mounting kit, then turned all four nuts down four turns using a diagonal star pattern. We fine tuned the final position of each nut with a micrometer ensuring each nut was turned down to within 0.012" of each other. We loaded the v-core to 1.5 volts and overclocked the Kentsfield to a modest 3.466 ghz per core. We used CPU-Z to ID the proc and CPU speeds; Core Temp Beta v0.94 to monitor temps; and CPU Burn 5 to load the cores.
Apogee GT in pieces:
Mounted in the new and improved test rig:
We ran it for about 30 minutes taking readings along the way.
Apogee GT loaded:
After 30 minutes it seemed to have settled in and we took reading of:
Ambient=23.3c C0=65c C1=65c C2=65c C3=65c=65c average loaded less ampient of 23.3c yielding a delta T of 41.7c
We were impressed in how even the temps were across the cores with the lapped cpu. But, that run's delta T with a stable test bed under more optimal conditions did not fare any better than the previous night. Scratching our heads we mounted the D-Tek FuZion to the test bed using the same exacting mounting method we used on the Apogee. We reused the Apogee studs, springs and nuts for this test.
FuZion's base showing the recessed inner pin grid:(pic borrowed from another thread as my pic came out blurry.)
D-Tek FuZion back in the new test bed:
Initial loaded temps:
We ran the FuZion for about 15 minutes after which we felt it had settled in nicely. We noted the following temps:
Amb=24.0 C0=65c C1=62c C2=67c C3=63c=64.25 less ambient temp of 24.0c yielding a delta T of 40.25c
These were actually a little worse than the night before. We had a little more voltage on the CPU and we figured that might account for the lesser performance from the night previous. It still beat the Apogee GT with a flat base by almost 1c.
We decided to remount the FuZion using the D-Tek mounting studs and springs as the studs were significantly longer and springs were longer and stiffer than the Swifty mounting bits. We did another run of 30 minutes after which we felt the block had settled in its performance. We noted the following temps:
amb=24.7 C0=63c C1=61c C2=63c C3=62c=62.25 avg loaded temp less ambient of 24.7 yielded a delta T of 37.3
Now that was on par with last nights performance! We were already to call the FuZion the winner when Gabe said he wanted to run the Aprgee one more time with the bowed base. So, we reassembled the Apogee GT with the thicker gasket and when the base was properly torqued down it bowed the base slightly. We remounted the newly configured block using the same careful procedure as the previous mountings. We used the D-Tek studs but we used the Swiftech springs. We ran it for about 22 minutes but could've taken readings after only one minute. It's performence was nothing short of spectacular! We noted the following readings:
Apogee GT bowed base loaded temps:
amb=24.4 C0=63c C1=56c C2=56c C3=61c=59c avg loaded temp less ambient temp of 24.9 yielding a delta T of 34.1c
Wow!! The Apogee with the bowed base outperformed the FuZion by 3.2c!
So, we concluded that under these conditions the Apogee GT was the clear winner with a Kentsfield style die. Under less extreme conditions especially where a higher flow block is needed to balance an otherwise restrictive water loop then the D-Tek FuZion would be a logical choice. In the end they are both killer blocks.
We think we might be able to get the outer cores (C0 and C3) to have the same performance as the inner cores (C1& C2) by somehow expanding the contact patch of the bowed part of the base. That could effectively yield delta T temps of 32c with a Kentsfield style quad core die. Swiftech will be working on this for sure in the near future.
I will tell you that you will get that additional 2c delta on a dual core as the cores of a Conroe are concentrated in the middle of the die. I am putting together a Conroe based test bed to test these two blocks along with another as-yet-to-be-released block from Swiftech with a dual core CPU to see how each block performs in that setting. I should have everything set up in 7-10 days and data gathered within 3-4 days after the rig is set up. I'll let you know how they fare.
We wandered out to see how Charles and Vince were doing with overclocking his X6800 to 5.1ghz. We happened on Vince asleep at the wheel while benching the rig @ -92c loaded.
The Comatose Overclocker: