New PC: biggest bang for the buck?

[jdupton]

To continue the benchmark results and add a data point on PixInsight v1.8, I just ran a trial on the system I built over the weekend.

M74_Benchmark = 6.552 sec  (Average of three runs)
M74_Parallel      = 2.964 sec  (Average of three runs)

System:
CPU  = i7-3930K @4.17 GHz
Mem = 16 GB DDR3 @1333 MHz
Disk  = Samsung 840 Pro SSD
OS    = Windows 8 Pro

I built this system for photo editing and CAD work.  PixInsight will be a real pleasure on it compared to my five year old desktop system.

John

[georg.viehoever]

Is this due to PI1.8, or your PC hardware. Can you make the same run with PI 1.7?

[Andres.Pozo]

PI 1.8 feels much faster than 1.7 and the benchmarks confirm it.

Nearly two years ago I wrote a message with the benchmark results in my new (then) machine: http://pixinsight.com/forum/index.php?topic=1052.msg20590#msg20590

These are the results in the same machine using 1.8RC5:

                        PI1.8RC5   Old version
Benchmark_M74:          7.54s      20.140s
Benchmark_M74_parallel: 3.93s       7.176s

[jdupton]

Since this was a freshly built system, I did not try v1.7 on it.  I simply installed v1.8 and ran the benchmarks so I cannot do a fair comparison of speed between the two versions.  I did notice that my old desktop had an improvement in speed with v1.8 but did not benchmark as did Andres.  (I am not sure I saw a 2x improvement, but it did seem "snappier".

Sorry, I don't have a direct comparison between the two versions yet.  I do however, have PI installed on my imaging laptop and am still at v1.7 there.  I can benchmark that system again and then upgrade to v1.8 to get a direct comparison data point to go along with the comparison Andres performed.

John

[Andres.Pozo]

I don't remember the version that I used for the benchmarks two years ago, but PI 1.7 was not announced until June of 2011, so I probably used a 1.6.x version.

[georg.viehoever]

PI 1.8 feels much faster than 1.7 and the benchmarks confirm it.
...

That would be a very impressive performance boost achieved by Juan. I know how difficult this is from my daytime job.

[atudorica]

Hi,

   I'm wondering why the benchmark performance I get is very low compared to the results you posted here. I get around 40 seconds for both processes. The thing is that I am using 16 physical processors which can run 2 threads each (AMD Opteron 6220 at 3 Ghz) and insane amounts of RAM (128 Gb) and disk space (>70 Tb of SSDs in RAID 0, I/O rate of about 1100 MiB/s).
   When first starting PI 1.8, I get the messages:

 * Parallel processing enabled: Using 16 logical processors.
* Thread CPU affinity control enabled.
* PSM AutoSave enabled. Auto-save period: 30 seconds.

** No running X11 compositing manager has been detected. Some core GUI functionality will be reduced or unavailable.

     Do you think that the missing X11 would have an impact on performance? Everything moves a bit in slow motion.

     This is quite interesting as playing with very large images seems not to be problematic, most processes take only a few seconds in images as large as 30k x 30k.

   Thanks,
Alex

[ruediger]

** No running X11 compositing manager has been detected. Some core GUI functionality will be reduced or unavailable.

Are you running PixInsight with X11 forwarding to a client computer?
I have a similar setup like yours (12 Xeon Cores, lots of RAM, SAN storage, virtualized environment), but the benchmark's running time is about 5 times slower than on a MacBook Pro. The speed difference is mainly because of the amount of screen updates the benchmarks does and my way of forwarding X11 via ssh -X.
In other tasks the virtualized PI machine is running much faster than the MacBook Pro.

Rüdiger

[Carlos Milovic]

My results with PI 1.8 RC7 at home (AMD Phenom II 1055T @2.8GHz, 12Gb RAM, SSD) with Fedora 18 + KDE:
B74: 8.676s
B76_parallel: 6.976s

There is a definitive improvement to previous test in the single core process (before, 10.496s), even when at that time F15 enabled me to use the CPU at 3.3GHz (overclocking). The parallel test is slighty slower (before, 6.511s). Either way, is a good result, considering the slower CPU, and that I have not fine tunned the OS to get more speed, as I did at that time (disabling services, etc).

[atudorica]

Are you running PixInsight with X11 forwarding to a client computer?
I have a similar setup like yours (12 Xeon Cores, lots of RAM, SAN storage, virtualized environment), but the benchmark's running time is about 5 times slower than on a MacBook Pro. The speed difference is mainly because of the amount of screen updates the benchmarks does and my way of forwarding X11 via ssh -X.
In other tasks the virtualized PI machine is running much faster than the MacBook Pro.

   I am indeed running PI through a ssh -X connection to a remote server and I think that you are right, this is the "problem". I'll try to find how to increase the ssh bandwidth (it's limited by default, if I understood that well) and I'll let you know what I found.

    Thanks,
Alex

[georg.viehoever]

I have found that using VNC can give much better performance compared X11 forwarding)if the network connection is slow.
Georg

[Andres.Pozo]

The console slows the processes. The effect in a remote session probably would be worse. You can try to rerun the benchmarks hiding the console.

[atudorica]

The console slows the processes. The effect in a remote session probably would be worse. You can try to rerun the benchmarks hiding the console.

   I've tried that, but it does not significantly change the result for some reason...

[atudorica]

I've added the -Xc option when connecting through SSH to my server and the interface moves much faster now, there is no noticeable delay when I'm opening a menu or writing a command. The benchmark runs now 3 times faster with the console opened and 4 times faster with it closed.

[dmax11]

Just saw this and boy do i love benchmarks!

M74 Parallel: 4.493 seconds
M74 : 7.584 seconds

running latest release version of PI

hardware on my machine is
i5-2500k overclocked to 4.4 ghz (used to be 4.5, became unstable after a year so i lowered it to 4.4) (Watercooled, Antec unit)
16 GB DDR3 running in dual channel mode at 1600 mhz
Mainboard is an Asus P8P67 Pro
Video cards if it matters is two Nvidia 560 Ti's in SLI
hard drives I run OS and main programs off of are 2x Seagate Cheetah 15K.5 SAS drives in Raid 1 via an LSI controller card (used to have 3 in Raid 0, man it was fast but one became unrelaible so i rescued my data and went to Raid 1 with the remaining two drives to be safe until they die, they are retired server drives from a large tech company and have massive amounts of use put on them when i got them lol)
OS Windows 7 64 bit.

looks like PI loves CPU power and Ram as one would expect