Author Topic: Performance Testing  (Read 2743 times)

Offline sreilly

  • PixInsight Padawan
  • ****
  • Posts: 791
    • Imaging at Dogwood Ridge Observatory
Performance Testing
« on: 2015 December 31 12:17:50 »
I have been wondering about the swap drive setup and had both a Samsung 500GB SSD 840 EVO that had a folder of PI_Swap along with a few other folders. Free area on the drive is 449 GBs. I also had a Western Digital 3 TB drive that had a swap folder and is mostly an empty drive. I tested with both active and found my swap drive numbers lower when both were used compared to just the SSD. So now I'm wondering if there is any advantage to multiple swap drives. I also found a boost when I opened the swap folder and deleted all the contents which wasn't all that much. I guess I really don't understand the full use concept of this in PI. Wouldn't the swap drive be emptied when the program closed? Same with using the Cache. I sometimes fine that the processes are slow to start while reading the cache which won't have any affect on the current process. What is the proper way to clear the cache and can it be done in PI?

The whole issue is about upgrading the computer or if I have the current computer maximized for processing. My latest Benchmark test today is attached as a PDF file saved in Office 2016. This is with the single swap drive as the SSD. PI is set to use all threads. Would  new build be justified or are these respectable ratings? I've looked at 6 core i7 CPUs but between the cost of the CPU and MB (2011v3) it's north of $800 alone and I'm not sure gain that much. Not to mention the new DDR4 memory and loss of PCI expansion slots. These MBs seem geared at gaming and not much else. This computer is my home office computer (e-mail, Office 365, etc) and most all of my image processing. With the STL-11002M camera the raw files are 5-21 MBs and often 30-50 at a time for calibration, alignment, and combining, saved as FITS images and then the final RGB/Lum is saved as 16 bit TIFF before JPEG for posting on my website.

Any suggestions are welcomed.
Steve
www.astral-imaging.com
AP1200
OGS 12.5" RC
Tak FSQ-106ED
ST10XME/CFW8/AO8
STL-11000M/FW8/AO-L
Pyxis 3" Rotator
Baader LRGBHa Filters
PixInsight/MaxIm/ACP/Registar/Mira AP/PS CS5

Offline sreilly

  • PixInsight Padawan
  • ****
  • Posts: 791
    • Imaging at Dogwood Ridge Observatory
Re: Performance Testing
« Reply #1 on: 2016 January 06 08:36:04 »
So here's where the new research has taken me, I obsess over upgrading as I don't do it that often, I looked at the i7-5020K CPU vs the i7-6700K CPU. The i7-6700K is clocked at 4GHz and is a quad core processor that has 8 threads. The i7-5020K is a 6 core/12 thread processor clocked at 3.3GHz. Now maybe it's a bit harder to figure than my approach but my thinking is that the only program other than possibly the OS, Windows 10 Pro 64, so greater overall performance improvement would likely be seen with the i7-6700K. But from the aspect of PixInsight using all cores and threads it would seem the i7-5020K would maybe eek out a slightly higher improvement. The difference is in the motherboards. The i7-5020K requires a LGA2011-V3 socket motherboard and from what I've seen of most of those they are aimed mostly at gamers. Almost all had 3-4 PCI Express 3.0 x16 slots that would not get used. I use one high quality video card and would prefer to have PCI slots available for add-on cards. If this computer down the road would end up out in the observatory I end up with few to no ways to add serial cards for the mounts, focusers, flat panels, and so on. From what I've seen so far it would likely be a pairing of i7-6700K/Z170-A from Asus or i7-5020K/Asus X99-A. Both would need 32GBs of DDR4 RAM, use the existing SSDs I already have.

Suggestions and opinions always welcomed. Building computers has gotten way more complicated that the days of just choosing a CPU based on cpu speed. Remember Celeron vs Pentium, Core 2 Duo Dual Core vs  Core 2 Duo Quad Core, pretty much based on rated MHz/GHz ratings and # of cores. Seems now everything is based on gaming needs and at 61 I have no intent of "gaming" other than the Solitaire/Mahjong windows games. I have enough distractions to keep me busy. Amazing how you loose a ton of time looking at Facebook.

Thanks
Steve
www.astral-imaging.com
AP1200
OGS 12.5" RC
Tak FSQ-106ED
ST10XME/CFW8/AO8
STL-11000M/FW8/AO-L
Pyxis 3" Rotator
Baader LRGBHa Filters
PixInsight/MaxIm/ACP/Registar/Mira AP/PS CS5

Offline NGC7789

  • PixInsight Old Hand
  • ****
  • Posts: 391
Re: Performance Testing
« Reply #2 on: 2016 January 06 11:33:51 »
The whole question of if such and such an upgrade is worth it really boils down to how much waiting can you tolerate between processing steps. When I upgraded from my 2009 iMac certain processes were taking several minutes to complete and it was becoming impossible experiment with different parameter variations for these processes. Once my upgrade was complete I achieved a 10 fold increase! So those same processes were now taking about 30 secs or better.

That was with an i7 4770k processor overclocked to 4.2Ghz. Since that time there are newer and faster processors available like the i7-5820k (not 5020k by the way). These processors are benchmarking at 15-25% better but now that would only about to 5 or 10 secs more saving so another upgrade does not appear worth it for me.

Based on your benchmark I think you will enjoy noticeable improvements from any of the upgrades you are suggesting. Obviously the i7-5820k will perform better than the i7-6700k (6 cores is better than 4) but only you can say if the extra cost is worth it. I wouldn't get hung up on particular motherboards being for "gaming" as this is really a marketing term. Also gaming is among the most demanding for a computer so it's not bad for you MB to be good for gaming even if you never even plan to play solitaire.

For the swap questions I would first say not to get hung up too much on it. The main (only?) value of PI swap is for holding undo/redo states. A large fast swap with allow you to toggle back and forth between many undo/redo states quickly. A single swap on an SSD is almost certainly fast enough. Introducing a RAM disk can increase performance at the expense of total swap size (total swap size is the smallest swap source times the number or sources). If your SSD is sufficiently large I don't see the value of adding a rotational drive to the swap as it will only slow it down (as you have noted). Optimizing the number of threads for best performance is a nice tweak but not really necessary.

 

Offline sreilly

  • PixInsight Padawan
  • ****
  • Posts: 791
    • Imaging at Dogwood Ridge Observatory
Re: Performance Testing
« Reply #3 on: 2016 January 06 11:57:15 »
I agree with your reply. My hangup is the design of the new motherboards (LGA2011-V3) have few options for expansion cards and more of an emphasis on multiple video cards which is wasted to my use. And their higher costs doesn't seem warranted short of needing those "extras". Looking for that socket board with good expansion abilities is getting very difficult. I try to look at future use of the computer as time goes on to further its ability to do other things such as become an observatory computer.
As a retired person I guess I'm looking to maximize my dollars.

Thanks for the reply,

Steve
Steve
www.astral-imaging.com
AP1200
OGS 12.5" RC
Tak FSQ-106ED
ST10XME/CFW8/AO8
STL-11000M/FW8/AO-L
Pyxis 3" Rotator
Baader LRGBHa Filters
PixInsight/MaxIm/ACP/Registar/Mira AP/PS CS5

Offline NGC7789

  • PixInsight Old Hand
  • ****
  • Posts: 391
Re: Performance Testing
« Reply #4 on: 2016 January 06 12:23:36 »
I'm not sure what motherboard you are looking at but the PCI Express slots can accommodate any compatible card not just graphics cards. Again wording about dual (or more) graphics cards is really just marketing.

As for DDR4, most would view it as a plus because of it's performance advantages. But of course this has to be weighed against cost, especially if you already have DDR3 RAM you can use.