Foxconn MARS: Lab Update

Memory Testing and Overclocking


The Foxconn MARS P06 BIOS improves the Everest latency and Sandra unbuffered scores by a minimal amount, but more importantly the memory voltage has decreased from 2.340V to 2.295V for our settings regardless of 2GB or 4GB configurations. The memory voltage requirements are still too high when compared to the other P35 boards. The main culprit for the higher voltages is the very aggressive performance level (read delay) settings that Foxconn utilizes to improve the stock performance on the board.

After receiving numerous emails on the subject, we have included DDR2-800 settings for this board at stock speeds. Latencies are up to l5% worse than the DDR2-1066 configuration with Sandra unbuffered numbers coming in about 8% lower. While these differences will show up in most synthetic benchmarks, testing with actual applications showed very little differences, if at all, between the two settings at stock CPU speeds.

Unlike our initial test results with the P03 BIOS, the 4GB configurations on this board actually perform slightly better than the 2GB setup. We did not need to change voltages either. Performance differences while overclocking were minimal with the 4GB configurations being able to run the same timings and voltages as the 2GB setups. However, the P06 BIOS would not POST with our 2x2GB or 4x2GB memory configurations. Foxconn is investigating this problem currently.

Q6600 9x266 - DDR2-1066:

P03 BIOS - 2GB:
P06 BIOS - 2GB:
P06 BIOS - 4GB:
Q6600 9x266 - DDR2-800:

P06 BIOS - 2GB:
P06 BIOS - 4GB:
Q6600 9x435:

P03 BIOS:
P06 BIOS:
Q6600 8x490:

P03 BIOS:
P06 BIOS:
In the Q6600 overclocking results, with the P03 BIOS we were able to run on the 266 strap with a final 9x435 setting with the CPU at 3.914GHz and memory at DDR2-1086. We reached the same 9x435 setting with the P06 BIOS but had to run the memory at DDR2-870 at 3-4-4-10 1N for absolute system stability. The P06 passed all of our benchmarks except for Enemy Territory: Quake Wars at the same settings as the P03. We are still investigating this error.

We were able to match the same 8x490 settings with the P06 BIOS while improving our memory timings at DDR2-980 from 4-4-4-10 2N to 4-4-3-10 1N (tRFC down to 36 from 42) with voltages decreasing from 2.340V to 2.295V. Latencies are basically even with read rates improving around 1% indicating other sub-timings were loosened at the expense of tighter memory settings.

The P03 BIOS required voltages for the CPU at 1.4875V, Northbridge at 1.898V, CPU VTT at 1.575V, and memory at 2.340V. The P06 BIOS required voltages for the CPU at 1.4625V, Northbridge at 1.763V, CPU VTT at 1.500V, and memory at 2.295V. Vdroop was excellent with an average drop around .02V~.04V with this particular CPU. We have seen Vdrop/Vdroops as high as .06V with our QX6850.

Our results with the P06 BIOS mirrored that of the P03 with a final FSB setting of 495 with our E6550. Our E6400 reached the same FSB setting of 520. We were unable to reach the advertised 575FSB that Foxconn has shown. Any attempt to boot the board past 530FSB resulted in three beeps and a return to stock settings, even with a CPU that generally does around 560FSB.

As far as the P06 BIOS is concerned, it still completes a shutdown and then quick startup procedure after any changes are made to the FSB or memory settings. If the BIOS has to recover from a failed overclock setting, all of the Gladiator section settings are reverted back to default settings. It is advisable to setup basic overclock profile settings in order to quickly set the board back to a manual configuration. We are still having a few problems utilizing the 4:5 memory ratio when overclocking, either it works at a particular setting or it does not. Foxconn is still tuning the BIOS at this time but the P06 production release is fairly solid, especially for a first effort by Foxconn in this market sector.