Performance discrepancy!
For some time I have had a Dell Inspiron 1720 Notebook with Intel T7500 (2 x 2.2 GHz) and 2GByte RAM running Vista32 Home Premium which was able to easily outperform my aging desktop when it came to processing my Canon 5D raw files with Capture One 4.x taking about 12 seconds per image. Capture One 4.5 Pro on this notebook still processes in approximately the same 12 seconds per image.
Now, over the last two weeks, I have upgraded my desktop to an Intel Q9550 (4 x 2.83GHz) and 4GByte RAM and completely re-installed Windows XP and Capture One 4.5 Pro. Based on my Notebook experience, I would have expected processing times for my 5D raw files of about 4 - 6 seconds. In fact, my 5D images are still taking about 12 seconds to process. What is going on?
Wade
Now, over the last two weeks, I have upgraded my desktop to an Intel Q9550 (4 x 2.83GHz) and 4GByte RAM and completely re-installed Windows XP and Capture One 4.5 Pro. Based on my Notebook experience, I would have expected processing times for my 5D raw files of about 4 - 6 seconds. In fact, my 5D images are still taking about 12 seconds to process. What is going on?
Wade
0
-
IIRC, unless you are running XP Pro version.... you might be using no more than 2 of those cores.
Which could put you in the same time frame for processing.
And adding more processors doesn't scale directly.
( Double the processors doesn't cut the time in half. )0 -
[quote="Steve" wrote:
IIRC, unless you are running XP Pro version.... you might be using no more than 2 of those cores.
Which could put you in the same time frame for processing.
Indeed I am using XP-professional and I can see all four cores bursting into life.
However, I believe that a four core processor is still only counted as 1 processor and thus, even on XP-Home, you should be able to utilise all four cores.[quote="Steve" wrote:
And adding more processors doesn't scale directly.
( Double the processors doesn't cut the time in half. )
Whilst this is also true, I was not comparing like for like in respect of everything except the number of cores.
The desktop machine has twice the memory bandwidth of the notebook.
Each core runs at 2.83GHz on the desktop instead of 2.2 GHz on the notebook
There are four cores on the desktop and only two on the notebook.
The desktop uses a RAID 0 disk array with approximately 5 x the throughput (both reading and writing) of the notebook which uses a single 5400rpm disk.
Both machine support SSE, SSE2, SSE3 and SSSE3. In addition the desktop supports SSE4.1 (although this is probably not relevant).
In every other respect that I have tried, the desktop machine is considerably faster than the notebook (although these tests were somewhat subjective). Photoshop CS, for example opens and initialises in ~10 seconds on the desktop and ~20 seconds on the notebook the first time around (when not cached) and ~5 seconds on the desktop and 15 seconds on the laptop on second and subsequent times opened ( when it is in the cache)
Putting all of this together, I don't think that I was unreasonable expecting considerably higher performance than I am getting. Whilst I will admit that I was hoping for something in the region of 4-5 seconds, I would have been happy with 6 seconds and not duly unhappy with 7 or 8 seconds.
Surely since I am not seeing any performance improvement at all moving from the notebook to the desktop, something is wrong. I now wish to find out what.0 -
Wade, although you have not yet explained how you measured the development time, I can give you a suggestion to help you find out what is going on.
What you could do for a start is monitor the behavior of both computers during developing your raw files from the Task Manager, Performance tab. You can observe utilization of all processor cores and determine how CO 4 is using them. Next, you will also notice that a few seconds of high processor utilization is followed by low utilization. This is the time the file is written to disk. About half of total interval time of processing and storing a file is used by the disk. A faster processor could be compensated for by a slower disk, for example.
This is how I measured processing speed of raw processors on several computers: I took a batch of eleven (11) images, no adjustments, put in processor batch, start batch and don't touch the computer till it's finished (switch off screen saver). Next I write down the time stamp of the last and first image of the batch, subtracting the values, convert to seconds and divide by ten (10!). A 12 seconds interval is a good value for 5D raw files development on your dual core machine.
It is correct that the processor count (number of processors) supported by Windows XP and later is relating the number of physical processors (1 dual core is 1 proc.). The number of cores or hyperthreading threads determines the number of logical processors. This way, XP Home, which supports only a single physical processor, should be able to support all cores a single quad core processor.0 -
Paul
Thanks for your comment.
Firstly, I measured the time to process each image by processing a small batch on each machine (with identical raw files and settings files to eliminate processing variation) and manually timing the batch. A little mental arithmetic and I have an average time per image. Since errors in the manual timing are divided among the number of files in the batch, this method is pretty reliable - certainly whilst the processing times are around 12 seconds.
Secondly, I do indeed see the processor utilisation go up to 100% on all cores whilst the actuall processing is occurring followed by a dip whilst the file (in my case a JPG) is being written out to disk. On the laptop the file writing takes about 2-3 seconds with the procesor activity falling to about 30~40% during the write period. On the desktop in comparison, the file write time is probably slightly lower - but not much say about 2 seconds - but the overall time is only very slightly faster.
You state that 12 seconds for a 5D file on a dual core machine is a good time - I can't agree more - I was delighted when I first got this notebook! However, the same 12 seconds on a much faster quad core machine appears to be somewhat less exciting.
The point you make about the disk performance also contributing to the overall time is also understood. However, the desktop machines disk performance is so very much greater than that of the notebook that I don't think that thiS could explain the performance balance between the two.
Wade0 -
I thought I would follow up this somewhat old thread since the situation has now changed.
Up until Christmas my desktop machine behaved as described in the opening post of this thread - taking around 12 seconds to process a 5D (mk1) file.
After Christmas, however, without conciously changing anything and still using CO4PRO 4.5.1, the machine started taking just under 6 seconds per image: Very much inline with the 4-6 seconds that I was originally expecting.
I have subsequently uninstallled 4.5.1 and installed 4.6 and now the time taken to process 21 images is 1 minute 47 seconds or just a smidgen over 5 seconds per image.
The time to process 5D images on my laptop is still about 11 or 12 seconds.
Well I don't know what has happened but something has changed very much for the better! The comparative times between my two machines are now much more in line with what I would expect.0 -
Thanks for the update Wade. Interesting to see what Christmas can do for a photographer's computer! HoHoHo. 0 -
By any chance did you run a defrag utility between then and Christmas? I have to say, this is one odd thread. 0 -
I use a somewhat old but still functional (just) version of Diskkeeper to perform schedule background defragging when the machine is idle. So the answer is 'Yes the disk has been defragged since Christmas' - however it was unlikely to be heavily fragmented before Christmas either.
What may have changed is the anti-virus. Over the Christmas period I was away for nearly a week and so when I came back there were updates to be applied - I use Avast. It may be that there was a program update or someother feature of the virus signature files that may explain it (I don't know since the updates are often performed when I am away from the machine) although I do not see how since I use the same anti virus solution on the laptop as well.
Finally, there have been a number of Windows updates to XP on the desktop (and Vista on the laptop) over the last month which I have applied (without ill effect) over the last week but I noticed the significant improvement sometime before these updates were applied so although they may just possibly be able to share (with the CO 4.6 update) some of the credit for reducing the 5D file process time from 6 seconds to 5 seconds, they can't be a contributing factor to the initially observed fall in process time from 12 seconds to 6.
Anyway, hopefully this thread serves to illustrate that CO4 can usefully utilise a 4 core processor provided that disk IO is not a limiting factor.0
Post is closed for comments.
Comments
8 comments