A Preview of Intel’s Bensley Platform (Part II)

Pages: 1 2 3 4 5 6 7

Cinebench CPU Rendering

Cinebench is a freely available benchmarking suite based on MAXON’s CINEMA 4D software, that provides both CPU and graphics benchmarks. Since most servers eschew graphics cards, only the CPU portion will be presented. The CPU test renders “Daylight”, a 620×620 scene using Maxon 4D’s ray-tracing engine. Daylight has 35 light sources, with 16 shadow maps, according to the documentation. The CPU benchmark can run in both a serial mode, and MP mode. In MP mode, the Bensley system will use 8 threads, and the Nocona system 4 threads. Ray-tracing is an extraordinarily parallel task, and the documentation indicates an 80-90% speed up for the second CPU, and around 10-20% for multithreading on the Xeon. The benchmark is pre-rendered at a small size to move all the data to memory, so there should be no disk accesses during the benchmark. The benchmark can run in two modes, serial and SMP; the latter supports both multithreading and multiple processors. See www.cinebench.com for more details or to download the benchmark.

Figure 1 – Cinebench CPU Performance The standard deviations for the Nocona runs were 0.152 seconds for serial and 0.1 seconds for SMP. The Bensley standard deviations were 0.288 seconds, and 0.1 seconds respectively. The Nocona system scaled by a factor of 2.16 from serial to SMP (i.e. execution time for serial was 2.16 times the SMP execution time), while the Bensley system scaled by a factor of 3.51. The Bensley system is slightly slower in single processor performance, likely due to the increased latency of the FB-DIMMs, and the slightly slower clockspeed. However, in SMP operation, which is the relevant comparison, the Bensley system is 50% faster in terms of execution time. This seems low; since rendering is an embarrassingly parallel activity, the improvement should be closer to 70-80%.

Pages: « Prev   1 2 3 4 5 6 7   Next »

Discuss (143 comments)