Finding Bottlenecks

Hey all,
Can anyone help me with regard to the bottleneck on this system?

I've got a query with regard to a non-gaming build that is designed for GIS systems. (Geographic Information Systems).

Now the eejits that write the software say that it benefits from multi-core processors but it's a single threaded application. In layman's terms, this means that the single thread won't be affected by background stuff.

What the GIS system is doing is really similar to google maps satellite view with the exception that it's overlaying a LOT of polygons and other data.

The polygon data itself is in a simple table form with the polygons all referenced to geographic coordinates so there must be some sorting/searching/querying going on.

The total imagery in use is over 30Gb.


Zooming in or out knocks my system for between anywhere from 3-30 secs.

I'm running an athlon x2 6000+ overclocked to 3.3Ghz on air. It's the older 125W processor with the larger L2 cache)
I'm simply increasing he base clock for overclocking - no voltage tweaks as it doesn't seem to change the frequency that I can operate the machine at complete stability. For a MSI 790XT-G45 entry level motherboard it's not too bad.

It seems that if I change the memory timings, the machine does still boot and run but it seems slower (anyone experienced this) and Passmark memory scores don't improve (or get worse).

Everest scores show weak performance in terms of memory on this system (2Gb DDR2 800 running at 825 Mhz) compared with other similar Memory modules. The memory is set as hyperthreading etc. latencies are 6/6/6/18/24 @412 MHz on CPU-Z which is a little over stock values.

Memory scores on Passmark however give me fair value over the competition.

Storage duties are carried out by a raid 0 partition across 2 samsung spinpoint F3 drives giving me an AVERAGE throughput of 250MB/s.

How can I increase the performance of this PC noticeably?

To give you an idea of the general performance I'm getting - an AutoGK backup of a movie (so the kids won't scratch it) takes 40-50 mins at 60% quality 640 pixels wide.

cb
3 answers Last reply
More about finding bottlenecks
  1. It's hard to know what to recommend. You'd expect a single threaded program to be CPU bound.
    Is there anyway you can borrow time on a wider range of systems and do a little benchmarking of your own?
    You could also under-clock your own CPU to get more insight on how the software performs.
  2. Thanks WR2 - hadn't thought of that. Just change the cpu multiplier from 15x to 7.5x and see if it takes twice as long.
    Methinks that it's going to be string sorting performance on the polygon land parcels and decompression on the image pyramids - both point to an Intel I3 being better.
    Lastly though would be memory performance. Just change the mem clock multiplier there to test bottleneck - again a version of your idea.
    cb
  3. WR2 said:
    It's hard to know what to recommend. You'd expect a single threaded program to be CPU bound.
    Is there anyway you can borrow time on a wider range of systems and do a little benchmarking of your own?
    You could also under-clock your own CPU to get more insight on how the software performs.


    Tested a few things already:
    Half memory speed clock at the same timings: next to no difference
    Analysis with Sysinternals Process Explorer: for a long period of time, there's activity in the I/O Bytes history limited to 53MB/s while CPU stays around 10%
    Maybe I'm wrong but doesn't this mean that this application (for which there is no replacement) seems to be limited by an abnormally slow I/O performance?
    Everest for instance rates the memory at 5900MB/s.
    This really has me baffled...
    cb
Ask a new question

Read More

CPUs Bottleneck