Mac Pro Nehalem Tests: Scaling/stacking Images

2009-06-01 • SEND FEEDBACK •
Related: bandwidth, Mac Pro, Macs, memory, memory bandwidth

This page evaluates programs that scale images to larger sizes and/or focus stack or similar operations.

Want a top-performing Mac Pro pre-configured and pre-tested? See this page.

Genuine Fractals 6.0 PERMALINK

Excellent!

In spite of using 16-cores most of the time, there is no evidence of memory bandwidth being a limitation, with identical times for the 16GB and 12GB configurations.

The MP08 took 36% longer than the MP09, but a 3.2GHz unit would take only 18% longer (estimated). This is odd given 16 (virtual) cores, but the numbers are what they are, and were very consistent across multiple trials. Employed cores do not necessarily mean efficient cores (like government workers), but in this case memory bandwidth is clearly not an issue, so perhaps virtual cores (HyperThreading) just doesn’t deliver its promise with some workloads.

Memory configuration eg memory bandwidth has no effect on running time here, with the 12/16/24GB configurations all taking the same amount of time.

Single vs dual CPU: the single-CPU MP09 took 85% longer than the dual-CPU MP9, no surprise given the use of all available cores.

Genuine Fractals 6

PhotoZoom Pro PERMALINK

Single-threaded

Compute intensive, and completely single-threaded, so it can use only a single core. This test generates a 2.33GB 16-bit TIF by scaling a file 400% linearly (16X).

The MP08-2.8 takes 24% longer than MP09 here. Again we see that a 3.2GHz MP08 (estimated) compares very favorably.

Single vs dual CPU: the single-CPU MP09 took 11% longer than the dual-CPU MP9, even though PhotoZoom Pro is single-threaded. This might result from the lower memory bandwidth.

PhotoZoom Pro Scale Image

Helicon Focus 4.0.1 PERMALINK

Very good!

Usage pattern is full utilization of all 16 cores on the MP09, followed by mostly idling them for about the same period of time. Only towards the last 1/4 of the process do all 16 cores start to be utilized more consistently.

The 2.8GHz MP08 takes 6% longer here. It appears that the 2008 3.2GHz Mac Pro would actually be faster than the 2.93GHz MP09!

Time for the MP08-2.8 is estimated based on previous results (390 sec with v4.0.0), due to a version change in Helicon focus which resulted in a ~20% speedup when retested.

Single vs dual CPU: no contest— not only is the single-CPU MP09 82% slower than the dual-CPU MP09, it’s much slower than the older 2.8GHz MP08.

Helicon Focus stack images

Conclusions PERMALINK

The MP09 is looking good here, but it’s not showing much of an improvement (or any) over the previous generation 3.2GHz model.

See How to Choose and Buy a Mac.