Best Computer System for Proccessing

[Starry Eyes]

I am using an W7 HP Pavilion Laptop w 8G ram that is 4 years old for my work. In looking to the future what should I be looking for in a computer system to replace it? Reasonably priced of course!

My AstroBin page
http://www.astrobin.com/users/Starry-Eyes/

[solo05]

I am also waiting to replace my netbook for in the field use and then processing the results.

I have seen somewhere on here Ivo answer this question in a somewhat vague way. By that I mean just in general terms without naming a brand etc.

His top tips seemed to be the processor with the most amount of cores and as much ram as you can throw at it. An SSD instead of HDD would also help.

Now comes the kicker, AMD have just released the Kaveri processors and HP have even stuck one in a notebook (HP EliteBook 755).

Pity they haven't realised the full potential of the chip as they have only put 4Gb of RAM and a 500GB HDD along with a lower resolution screen.

Question to Ivo now as to how Startools uses the available cores. Can it use the 10 cores of the A10 PRO-7350B with Radeon R6 Graphics to any better effect than say the new Intel i7 ?

I have put the 2 slides from AMD's site below to show some comparisons. I do not know if this means AMD are better than Intel for what we are trying to achieve i.e. faster startools processing or not.

The speed test table has obviously been weighted to show AMD in the best light, but the results for heavy graphics use do look very promising.

AMD APU Slide_2-900-90.jpg (81.52 KiB) Viewed 13190 times

AMD APU Slide-900-90.jpg (129.56 KiB) Viewed 13190 times

[admin]

Hi all,

The original 'vague' recommendations still stand;

Lots of memory and lots of CPU cores.
If you have lots of memory, run StarTools from a RAM disk (on Windows) or move your /tmp folder to a RAM disk (*nix). The speed increase will be significant (as this allows StarTools to keep its Tracking data in memory at all times).

StarTools still does not rely on your graphics card, as the landscape is still too fragmented and the hardware does not allow fast communication with system memory (also see post elsewhere). GPU's are great for vastly accelerating simple operations/algorithms, that need to be carried many, many times on the same data. However when results need to be transfered to system memory, or new data needs to be transfered to the GPU for processing, performance incurs a pretty big hit. StarTools' algorithms make the latter 2 scenarios very common occurrences, mainly due to the Tracking feature (it shifts around *huge* amounts of data). As a result, 'GPU accelerated' module performance would not increase processing speed much, or may even be slower.

There are interesting developments going on to alleviate the memory transfer problem, but solutions to this are not common place yet in hardware.

The speed increase you will get from increased amounts of CPU cores will diminish however, as, here too, memory will become a bottleneck (cores will start have to wait for access to the memory bus).

To make matters even more interesting, this all depends on the algorihtm/module used - some are more memory intensive, some are more CPU intensive, etc.

[Rkonrad]

I, too, am upgrading. I've got a i3-4130@3.4x2 with 8 gigs memory on linux. It's running very slowly and I need to bin most images so processing doesn't take too l long. Upgrading processor is too expensive but I may buy more memory to a total of 24 gigs. Would this be a good option or should I wait until I can afford a better processor? Also with my linux system, would it make sense then to run everything from a ram disk? (I noticed you mentioned that that would be an advantage on a windows system). Thanks in advance. I'm really enjoying your programme!

[admin]

Upgrading your memory will definitely help, as will the inclusion of an SSD. Especially if you use approx 1/2 of your RAM for your /tmp/ folder (which is what I do on two of my machines that have 16GB total), your system will fly (and not just in StarTools). The i3 is somewhat of a bottleneck, and an i5 with 4 physical cores would help somewhat for some modules. That said, the RAM and SSD would be a good place to start!

[u2pilotjt]

Ivo:
Just saw this thread and I'd like to ask you a clarification question from the earlier posts.

I'm running 16Gig of RAM and have RAMdisk available. So if I simply run StarTools from on the RAMdisk (8Gigs worth), will that do the trick? OR, do I also need to move my temp folder to the RAMdisk as well? I'd like the simplest route and I'm not concerned about the computer's speed at any other time; just for StarTools.

Oh, and would it be best to move the image being processed onto the RAMdisk with StarTools before opening it? I assume it would.

Thanks;
JT

[admin]

Hi,

Ivo:
Just saw this thread and I'd like to ask you a clarification question from the earlier posts.

I'm running 16Gig of RAM and have RAMdisk available. So if I simply run StarTools from on the RAMdisk (8Gigs worth), will that do the trick?

On Windows, you'd have to run StarTools from the RAM disk, but on Linux or MacOSX the tmp folder needs to reside on the RAM disk instead

If running a recent version of Ubuntu or Linux Mint, you could simply add the following to /etc/fstab;

Code: Select all

# Make /tmp folder an 8GB RAM drive
tmpfs  /tmp      tmpfs  size=8000M,mode=0777 0 0

Oh, and would it be best to move the image being processed onto the RAMdisk with StarTools before opening it? I assume it would.

StarTools loads the full image into memory and keeps it there. It doesn't need access to the original file after that. The thing that will speed up StarTools immensely is speeding up read/write access to the .trk files; the Tracking data mining information. These files get updated and consulted regularly, and since they can be multiple GB in size, fast access is a huge help.

Hope this helps!