[LAU] New workstation | DAW pc
meissner.fritz at gmail.com
Thu Sep 9 16:36:19 UTC 2010
On 5 September 2010 21:18, rosea grammostola <rosea.grammostola at gmail.com>wrote:
> Thanks for the info. After some reading and discussion, the i3 or i5 seems
> to be good options indeed.
> After reading this: http://www.anandtech.com/show/2901/7 and considering
> my budget
> It looks like I better go for a i3 540 or something close to that, compared
> to a i5...
> My perspective on the processors is a little different. If you go for the
Arrandale i5-750 you get 4 cores (but no hyperthreading or GPU) instead of 2
cores with hyperthreading plus on-chip GPU. You can see the equivalent
Anandtech review for Arrandale at http://www.anandtech.com/show/2832 . If
you compare the benchmarks in the review you quoted, the i5-750 has a
Sysmark score of 217 versus 204 for the i3-540, which is only a 6%
difference. However, if you look at multiple benchmarks (see Tom's hardware
is for i5-530, not i5-540 but is still relevant)) the differences are
often much greater. In particular, if you look at the Adobe Photoshop Image
processing benchmark, which is the only one that I know is definitely using
parallel processing on all available cores, then the 4 core processor is
twice as fast (unsurprisingly). My expectation is that as time goes on more
and more compute-intensive applications (such as audio and video encoding
and compiling) will start to use multiple cores fully, so it makes sense to
buy for the future.
Another factor is memory bandwidth; assuming that some audio work is memory
intensive (in particular playback using large sample banks) then I think
this is important. The i5-750 has a memory bandwidth of 16.9 GB/s vs. 11.6
for the i3-530 (from the comparison chart); this is a 46% improvement.
Compounding this is the fact that the onboard GPU is using main memory as
video memory, so that memory bandwidth is also being shared with all the
video data, which could eat up a large chunk of the capacity. So the i5 must
come out way ahead on memory bandwidth available for the CPU.
Thirdly, you can get cheap fanless video cards with far better performance
with the onboard GPU and no uncertainty about whether they work with the
real-time kernel (but ask what models are tested, I remember hearing that
some of the proprietary drivers also caused problems it the RT kernel).
My philosophy is to let the CPU do processing, and the video card handle
video with its own separate memory that doesn't compete with the CPU;
combining the two can only lead to compromises.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Linux-audio-user