Comparing an 8-core to 4-core Mac Pro
by , 2:25 PM EDT, April 13th, 2007
Bare Feats conducted some performance tests, comparing the processor performance of an 8-core Mac Pro to a 4-core Mac Pro. A factor of two was not always obtained, and technical issues may be the cause.
As one would expect, the results varied depending on the OS and software's ability to properly exploit eight cores. In addition, there was some concern about the Mac Pro's memory bandwidth.
"Our 8 core (dual quad-core) Mac Pro arrived ahead of schedule. We stayed up late last night testing it against the 4 core (dual dual-core) Mac Pro running at the same clock speed (3GHz). We used the same 8 memory modules (total of 16GB) and identical models of hard drive. The 8 core had the Quadro FX 4500 graphics card while the 4 core had the Radeon X1900 XT, but this first round of testing is strictly CPU intensive," the report stated.
In the very first test with Cinebench 9.5, all eight cores were clearly used in a render, exhibited by screen shot of the 8-way venetian blind effect. Even so, the resulting rendering time was only 45% faster than the 4-core Mac Pro.
Again, in a Photoshop and Aperture test, the Mac OS X Activity Monitor showed all eight cores in action, but the time to complete the task was barely faster than the 4-core Mac.
One explanation might be the memory bandwidth, according to Lloyd Chambers. "Memory bandwidth is inadequate for 8 cores. It�s already a limiting factor with the current quad-core 3.0 GHz Mac Pro. Memory copy speed is at best 2.9GB/sec on the Mac Pro, in spite of Apple�s highly misleading claims of 21.3 GB/sec (�maximum processor bandwidth of up to 21.3 GB/s��bandwidth is a bit more than double the memory copy speed)."
Another problem may be Tiger's ability to deal with so many potential threads. Mr. Chambers concurred with the possible problem, "the OS does not know that it should keep a thread on the SAME core instead of swapping it around to any one of the 8 cores. When a thread moves to a different core-group, the cache has to be reloaded on the new core. Since Intel quad-cores are two Core 2 Duos "duct-taped" together, the problem is worse than it would otherwise have been. When this thread/core shift happens repeatedly, it makes the cache ineffective and floods the memory bus with activity."
It may be possible, in hindsight, that the 8-core Mac Pro was designed with better thread handling of Leopard in mind. In any case, in the future, developers will likely be able to better exploit Leopard's capabilities and refine their apps to exploit the 8-core Mac Pro.