ScienceMark 2.0

Since computers are computers, and they were originally developed to handle increasingly complex scientific calculations, it is only appropriate that we benchmark the very latest and greatest in computing with a tool called ScienceMark. Really though, ScienceMark 2.0 is probably the best collection of CPU processing tests that there is. The benchmark suite has your processor do everything from calculating 1536x1536 matrices to cryptographic calculations and simulating molecular dynamics.

We were very happy with the comprehensiveness and clarity of the results we achieved with ScienceMark in our Core 2 Duo review. So much so, that we decided to utilize it again. This time, however, we have added a memory/cache performance test to get an idea of the throughput rates and cache latency of the processor/memory interface. First, the overall test results.


Here you can see right off the bat that at the high speeds, the two processors are pretty close. You can also see that we got the same weird results with the QX6700 at stock and 2933MHz as we did in PCMark05. We still don’t have an explanation for these scores, considering system settings were literally identical except for the extra 266MHz on the CPU. Regardless, both CPUs performed extremely well in this test, and, as expected, since ScienceMark is not a multi-threaded application, the QX6700 offered no tangible performance increases.

Moving on to the memory tests, we recorded the values of latency of 256 byte strides through the memory and L2 Cache, as well as the total realized memory bandwidth. These numbers should be interesting, as they will give us insight about the performance hits that may or may not have been incurred while mounting two dies on one processor package.


It is pretty clear from these results that there was indeed a latency hit involved with putting the two processors on the same package. This is similar to what we saw with the Pentium-D line of processors, which were essentially two Prescott dies bolted together, much the same way the QX6700 is two Conroe dies bolted together. There was also a significant difference in memory throughput between the two processors, which was pretty surprising considering the RAM settings were constant throughout stock and 2933MHz testing. This could be a possible reason for some of the weird numbers we have been getting previously. Whatever the case, these performance hits seem to vanish when the processor and memory are overclocked significantly.