AS WE SAW earlier in the week here, Intel's D5400XS "Skulltrail" is possibly the most powerful - as in row computing and I/O - desktop board for this spring at least. Except for one big complaint - FB-DIMM memory.

Yeah, FBD-800 modules, especially those cooler, low-power 1.5 volt units, aren't that easy to find yet. We noticed many web sites benchmarked Skulltrail using just two of these - whatever others tell you, there is a BIG benefit from running this (and any other dual-Xeon Seaburg) mobo with all four channels populated. The benefit is sizable enough to change the board's suitability for high-FPS gaming or other memory-intensive apps quite a few notches up.

Unlike the usual desktop mobos, this is NOT a dual-channel board where just putting in the first two DIMMs will usually be faster than going on all fours. All Seaburg-based mobos including Skulltrail are dual-FSB with four DIMM channels which, if set up fully in - usually default - interleaved fashion, will deliver a boost to almost all benchmarks, including gaming ones.

We made use of Skulltrail's FBD basic latency modification capabilities - just the four base parameters in the Beta BIOS - to get the best out of the memory on hand, as FSB overclocking doesn't go that well on this board yet. Not to mention that command rate and Trd chipset latency settings are still not there. Here's the web's first comparison of different FBD modules on Skulltrail.

We had two different quad-module FBD-800 sets here: one is from Nanya, another from Micron. Both are engineering samples - the black Nanya one actually from mid-2007, with declared 5-5-5-11 basic latency figures. OK, not bad for an FB-DIMM at this clock. The Nanya DIMM is a pure 1.5v product - both DRAM dies and the AMB advanced memory buffer on the module run at 1.5 volts. The blue Micron DIMMs are rated at 1.8 volts instead, resulting in slightly more heat and power use.

Micron

Nanya

We ran both DIMM kits under exactly same speeds and BIOS conditions, without any overvoltage or such. First, we pushed the BIOS settings as far as we could. On both sets, 5-4-4-9 was the maximum. Note that this BIOS rev, compared to the last one, seems to ignore the last (Tras min) parameter, and always keeps it as 12, increasing the latency a bit - 93ns vs 89ns in Sandra XII random latency test.

Both modules ran under Windows XP64 on 4GHz/FSB1600 dual CPU setting, with Sandra XII SP1 as well as 64-bit Linpack DP 10000x10000 test run.

Here are the bandwidth results - remember, all speed and voltage settings same:

Micron bandwidth

Nanya bandwidth

Interesting? We're talking about nearly eight per cent difference here, even though the latency stays the same at 93ns in both cases.

In Linpack, interestingly, the roles reversed - the average of three test runs on Nanya gave us 97.4GFLOPs, while the Micron three-test run gave us 96.7GFLOPs. Most importantly, each run on Micron was consistently a little slower than on Nanya. And, oh yes, we needed to run the CPUs at 1.425 volts compared to usual 1.3875 to get Linpack completed in a stable fashion - it is usually the most demanding test run.

Notice that Micron DIMM set consumes some five watts more than the Nanya one, too - according to Sandra estimates at least. During the tests, both sets ran hot despite a high-speed, 8cm Thermaltake fan blowing 3,000 rpm air at them right on top - but Nanya felt a little cooler. In BIOS hardware monitor, the DIMM area temperature showed 41C for Nanya and 45C for Micron in our airconless 'tropical extreme test' room at 32C room temperature.

In summary - before you proclaim a platform memory-crippled, unsuitable for gaming or whatever, check what and how much memory you're running, and how you configure it. Also, even with the same voltage, speed and latency settings, the performance of different modules may vary widely in synthetic benchmarks. As for me, it's time to see if there's any effect in real apps. µ