Page 8 - Multi Core testing: Dhrystone, memory & Queen
DryStone CPU test
We make use of a multi-threaded Dhrystone test, which basically is a suite of arithmetic and string manipulating programs. Since the whole program should be really small, it fits into the processor cache. It can be used to measure two aspects, both the processor's speed as well as the optimizing capabilities of the compiler. The resulting number is the number of executions of the program suite per second.
You can see all the way on top the QX9770 processors from the Skulltrail system. Now look all the way to your right and compare it with the E6600 or FX-62 processor. This is a multi-threaded (uses all CPU cores) test, and it shows for sure.
Both SANDRA and Everest have excellent tools to measure memory bandwidth with. The FB-DIMM results however are really disappointing. We're only showing one test with a synthetic benchmark (Sandra) yet through out all tests that passed our eyes, this performance is consistent. For example check out Everest:Everest Ultimate Read MB/s Write MB/s Copy MB/sSkulltrail (Dual QX9770) 4127 4608 5430
Though write and copy speeds are pretty okay, I tend to believe that the choice for using FB-DIMM memory on an enthusiast platform was probably not the best one. You can achieve much better results with for example PC2-9200 DDR2 memory. Everest would show you similar performance results, PCMark would show you similar performance results as well. This will have a tiny effect on overall performance. Enough said, let's move on.
Queen CPU test
This simple integer benchmark focuses on the branch prediction capabilities and the misprediction penalties of the CPU. It finds the solutions for the classic "Queens problem" on a 10 by 10 sized chessboard. At the same clock speed theoretically the processor with the shorter pipeline and smaller misprediction penalties will attain higher benchmark scores. For example -- with HyperThreading disabled -- the Intel Northwood core processors get higher scores than the Intel Prescott core based ones due to the 20-step vs 31-step long pipeline. However, with enabled HyperThreading the picture is controversial, because due to architectural bottlenecks the Northwood core runs out of internal resources and slows down. Similarly, at the same clock speed AMD K8 class processors will be faster than AMD K7 ones due to the improved branch prediction capabilities of the K8 architecture.
CPU Queen test uses only the basic x86 instructions, it consumes less than 1 MB system memory and it is HyperThreading, multi-processor (SMP) and multi-core aware and thus is a multithreading CPU Benchmark with MMX, SSE2 and SSSE3 optimizations.
An excellent test to get some synthetic performance numbers. To the left in yellow let's focus on that Q6600 processor, to the right the two QX9770 CPUs on Skulltrail. Back in the early nineties this performance only could be found in super computers.