New PC: biggest bang for the buck?

[bitli]

Thanks, indeed I had to scroll horiontally to find it. I feeld stupid, but at least I was not the only one :-)

19.2 sec (with console), 18.3 sec (console closed) on an HP Compaq 6730b,
a portable with Core 2 DUO P8700 @2.53 GHz 4.00GB
Vista 32 bits, a few other programs were present (its Windows, after all).

Not bad IMHO for a portable computer. I just ordered a Core I7 with 12GB,
I hope this will be worth the additional expense.

-- bitli
NOTE: This was with the old icon set.  The new icon set results in 16 secs (with console).

[bitli]

And for the fun:

Dell Inspiron 3600 (Pentium III 900 Mhz running at 600 Mhz), 2GB memory, Windows XP.
Slightly older version of Pixinsight, but results in 56 secs (first bench) and 46 second (so called parallel,
but I have a single cpu).  Sql server in the background...

The fan seems to run at 1Ghz when PI is running :-)

Slow but not unusable.
However I was totally unable to process APN sized images with this machine, except very basic operations, .
-- bitli

[vicent_peris]

Dell Inspiron 3600 (Pentium III 900 Mhz running at 600 Mhz)

900 MHz running at 600 MHz?? Why??


V.

[avastro]

New set of icons:
7.924s console open
Win Vista 32, i7 920 @2.67ghz, 4Go RAM DDR3 1333,500Go SATA2
MSI X58C.

Big improvement! 

Antoine

[Silvercup]

Hi:

Well, I feel stupid too. I was on Desktop 2.

First Process Icon set  17,129 s. Console opened.
First Process Icon set  16,888 s. Console opened.


New Process Icon set 11,887 s. Console opened.
New Process Icon set 11,200 s. Console closed.

Intel Core2 Quad Q9550 2,83 GHz 8Mb Bus 1333 MHz Windows Vista Home Premium 64 Bits

 
Silvercup.

[georg.viehoever]

Here a couple of measurements, all with PI 1.5.9.561 on different Hardware and OSes. Summary:
-- i7-860 blows your mind away.
-- But 4 real cores+HT give speedups of "only" 2.3 (which is a quite reasonable speedup from my experience with programming)
-- Windows7-x64 does parallel processing much better than Vista-x32
-- Windows7 and PI get slower over time
-- Running Fedora11-x64 inside a virtual machine on Vista-x32 is slow.

Georg

All measurements below are medians from 3 measurements each.
- On Toshiba Satego X200 Laptop, Intel Core2Duo@2GHz, Chipset PM965, 4 GB RAM PC2-5300(333MHz)
-- On Win7-x64 and PI-x64 that has been running for quite a while:
--- Benchmark 31.3 seconds
--- BenchmarkParallel 22.7 seconds
-- On freshly booted Win7-x64 and PI-x64:
--- Benchmark 27.8 seconds
--- BenchmarkParallel 19.36 seconds
-- On freshly booted Vista-x32 and PI-x32:
--- Benchmark 28.34 seconds
--- BenchmarkParallel 23.9 seconds
-- On Fedora11-x64 Linux running inside VirtualBox inside Vista-32 (yes, thats possible, but gives only 1 core to Linux)
--- Benchmark 75.5 seconds
--- BenchmarkParallel 52.1 seconds

- On Desktop with i7-860@2.8 GHz (4 cores+HT), Asus P7P55D, 8 GB RAM PC3-10700H (667 MHz), Win7-x64, PI-x64
--With PI defaults:
--- Benchmark 13.46 seconds
--- BenchmarkParallel 7.0 seconds
-- With Parallel Processing Disabled in Global Settings:
--- Benchmark 19.68 seconds
--- BenchmarkParallel 16.14 seconds

[sleshin]

HP Pavilion Elite Desktop
Win 7 Ultimate 64 bit
Intel i7 quad core(975) 3.33 GHz
RAM: 12 GB DDR3-1066MHz
HD: 320GB @ 10K rpm

Benchmark 11.5 s
BenchmarkParallel  6.43 s

[Yuriy Toropin]

Intel Core 2 Quad Q9650 (@3GHz), 8Gb RAM (@1066MHz), 1Tb Samsung HDD,
Win 7 Ultimate (x64 bits),
PI 1.05.09.0561 eng (x86-64)

It takes 13.9sec (console open) to proceed - old set.

New BenchmarkParallel PSM
7.2 sec w/console opened, on average, individual runs took 6.8...7.8 sec,
6.9 sec on average with console closed

[Niall Saunders]

Firstly, my PC has not been switched off (all I EVER do is switch the monitor off, and even that doesn't happen that often) for over two months now. I have NEVER defragmented a hard drive - at least not this century!! I have had DOZENS of applications open and closed, so I am BOUND to have had a few 'memory leaks' as a result.

I am also running TWO instances of PI, the first of which has been running (endlessly) since the computer was last re-booted (two months ago, when I had to change the UPS batteries). The first instance has seven workspaces open, over fifty images open, and HUNDREDS of process icons all over it.

So, I decided - to be safe - that I would just open a second instance and load the benchmark data into that. I loaded the standard PSM, and then 'merged' the second PSM in beside it. Then I loaded the RGB.fits image (in fact I ended up loading many, many 'copies' of the image, whilst I repeated the application of the PSM container to a new image each time).

The following are mean averages of three applications of each process, to a fresh copy of the image each time (without ever clearing any previous images, and all with the console opening during, and closing after the process run)

Code: [Select]

The 12-step process "Benchmark_M74.psm"          :- 14.775 sec

The 19-step process "Benchmark_M74_parallel.psm" :-  8.894 sec

This was for PI v01.05.09.0547 eng (x86_64) running on an Intel Core 2 Quad CPU (Q9550), running at 2.83GHz, with 8Gb RAM (at 800MHz, I think - certainly nothing 'special'), and all under Windows Vista 64 Home Premium SP1 (using a 7200rpm 320GB HDD) [total cost was WELL under £1000, 18 months ago]

That seems to compare well with Georg's results for an i7-860/2.8GHz/8Gb machine (13.46 and 7.0 seconds)

I don't understand why Silvercup's results - for a very similar machine - are apparently 'slower' than mine, given that his FSB speed is MUCH higher

Even Antione's i7-920 (albeit only 32-bit OS) is not 'much' faster

My 'first thoughts' are that the i7 chip is not THAT much faster - and my machine was NOT purchased as a 'screamer', it was just the most 'sensible use' of a £1000 (FIXED) budget that I had available at the time. It was certainly NOT purchased for 'games' (I have NEVER played ANY game on it !!), it WAS purchased with PI in mind, hence 2x320GB internal and an upgrade from 4Gb to 8Gb last November (with a further 3 x 1TB dual RAID NAS devices on the LAN).

An interesting test. Now we need someone to compile a proper table of results

Cheers,

[georg.viehoever]

Niall,

This was for PI v01.05.09.0547 eng (x86_64) running on an Intel Core 2 Quad CPU (Q9550), running at 2.83GHz, with 8Gb RAM (at 800MHz, I think - certainly nothing 'special'), and all under Windows Vista 64 Home Premium SP1 (using a 7200rpm 320GB HDD) [total cost was WELL under £1000, 18 months ago]

Could you try the same with the latest PI version (1.5.9.561)? I wonder if this makes a difference.

Georg

[Niall Saunders]

I would LOVE to - but that probably means 'closing down' my existing PI session - and the sheer number of processes and images now terrifies me - hence the reason I have been pleading with Juan for a 'Save ALL' option

Give me a day or so to think about it - I have come over all hot and sweaty just at the very though of NOT having PI running on my machine (like someone else once said, my darling wife actually thinks that PixInsight actually IS the Operating System on this PC )

Cheers,

[georg.viehoever]

Hi,

some additional information about the effect of threads and priorities. Summary:
-- even the "Hyperthreading"  (threads>4) gives some benefits
-- thread priorities do not influence performance (at least on my desktop that has plenty of performance. On my Laptop, TimeCritical will freeze Media Player).
Details below.

Georg

System: Desktop with i7-860@2.8 GHz (4 cores+HT), Asus P7P55D, 8 GB RAM PC3-10700H (667 MHz), Win7-x64, PI-x64
Measurements with BenchmarkParallel, median of 3 measurements.
- Modified the number of threads in Global preferences/Parallel Processing
-- default threads (use all available): 6.83
-- 8 threads (equivalent to all): 6.88
-- 7 threads: 6.89
-- 6 threads: 7.22
-- 5 threads: 7.69
-- 4 threads: 7.97
-- 3 threads: 8.61
-- 2 threads: 10.07
-- 1 thread (=parallel deactivated): 16.2
- Influence of thread priority
-- TimeCritical default=: 6.68
-- Lowest: 6.82
-- Adiitional Load (youtube video) made no difference

[georg.viehoever]

...
--- BenchmarkParallel 7.0 seconds
...

Disableing the console window (setting delay to 10 seconds in Global preferences/Miscellaneous Process Settings) reduces this to 6.24 seconds. The console consumes an amazing amount of time...

Georg

PS: Setting the delay to 0 makes the window appear immediately (contrary to bubble help description).

[georg.viehoever]

Here a couple of measurements, all with PI 1.5.9.561 on different Hardware and OSes. Summary:
...
All measurements below are medians from 3 measurements each.
- On Toshiba Satego X200 Laptop, Intel Core2Duo@2GHz, Chipset PM965, 4 GB RAM PC2-5300(333MHz)
...
-- On Fedora11-x64 Linux running inside VirtualBox inside Vista-32 (yes, thats possible, but gives only 1 core to Linux)
--- Benchmark 75.5 seconds
--- BenchmarkParallel 52.1 seconds
...

Another measurement on the same laptop, this time Fedora11-x64 running inside VirtualBox inside Windows7-x64 (giving it 2 of the 2 cores):
--- Benchmark 45.3 seconds
--- BenchmarkParallel 24.8 seconds.

Still considerably slower than native Win7-x64, but much better than virtual on top of Vista32.

Georg

[bitli]

Version 01.06.00.592 on Windows 7 64 bits,
On Core 7 93 @2.8 GHz0, 12 GB, a single 1 TB Sata drive.

Using the parallel icons.
7.5 second when run on the first time (with console)
6.6 seconds when run a second time (with the console), 6.4 without the console.

Looking at the console log, I see that on the first execution, I have:

ProcessContainer: Processing view: RGB
Writing swap files...
15.00 MB/s

While on second and third execution it is 55MB/s and 65MB/s. But the time is not logged, so I do not know if this explains the difference.
The other operations do not log any information that could help finding how they contribute to the processing time.

I am a little bit surprised that PI bother to write to the swap file when it has probably 10GB available. But I am pretty happy with the performances.

-- bitli