Page 1 of 40 12311 ... LastLast
Results 1 to 15 of 623

Thread: 11.5's BenchmarkMarbles.lws - share your machine's render time here

Hybrid View

Previous Post Previous Post   Next Post Next Post
  1. #1
    Red Mage Celshader's Avatar
    Join Date
    Feb 2003
    Location
    Burbank, California
    Posts
    1,817

    11.5's BenchmarkMarbles.lws - share your machine's render time here

    In the LightWave_11-5_Content folder, you will find a Dave Jerrard benchmark scene for modern processors here:

    Code:
    LightWave_11-5_Content\Rendering\Benchmark_Marbles\BenchmarkMarbles
    Set your content directory to the "BenchmarkMarbles" folder at the end of the above path, then load "BenchmarkMarbles.lws" into LightWave 11.5 and hit F9. Share your results and your machine specs here.

    Here is the 1280x720 image:



    Here's an example of render times on an AMD FX-8150:




    Specs on the machine responsible for the screengrab above:

    Jennifer Hachigian's machine at home:

    Render time: 2h 15m 22s
    CPU: AMD FX-8150
    RAM: 32GB DDR3 1866 (PC3 14900) 10-11-10-30
    Overclocking: None
    OS: Windows 7

    Other known render times:

    Dave Jerrard's machine at home:

    Render time: 57m 57s
    CPU: dual AMD Opteron 6272
    RAM: 32GB DDR3 SDRAM ECC Registered DDR3 1333 CL9
    Overclocking: None
    OS: Windows 7

    Dave Jerrard's machine at work:

    Render time: 1hr 47min
    CPU: Intel x5650
    RAM: 6GB (unknown type)
    Overclocking: None
    OS: Windows 7
    Attached Thumbnails Attached Thumbnails Click image for larger version. 

Name:	Bench_FX-8150__11_5.jpg 
Views:	413 
Size:	126.6 KB 
ID:	110982   Click image for larger version. 

Name:	Bench_FX-8150__11_5_Render.jpg 
Views:	574 
Size:	240.8 KB 
ID:	110983  
    Last edited by Celshader; 02-02-2013 at 01:38 PM. Reason: Dave just told me the exact minutes/seconds of his benchmark results at home.
    Jen's 3D -- LightWave stuff.
    Jen's 2D -- my comic book.

    Python is my smashing board. LightWave is my S.M.A.K.

  2. #2
    Studio Animator DonJMyers's Avatar
    Join Date
    Aug 2003
    Location
    Los Angeles, CA
    Posts
    575
    Jen I noticed the version of this scene sent as a sample has shading samples and light samples at 1. This makes renders fast but pebbly. But your render looks great. Did you change those settings and, if so, doesn't that make the benchmark times messed up?

  3. #3
    Red Mage Celshader's Avatar
    Join Date
    Feb 2003
    Location
    Burbank, California
    Posts
    1,817

    Lightbulb

    madno, that is one amazing machine you have there.

    Quote Originally Posted by DonJMyers View Post
    Jen I noticed the version of this scene sent as a sample has shading samples and light samples at 1. This makes renders fast but pebbly. But your render looks great. Did you change those settings and, if so, doesn't that make the benchmark times messed up?
    I did not change the settings. I loaded up the scene as-is from the 11.5 content directory and hit F9. What happens when you render this benchmark on your machine?

    Dave Jerrard tends to throw the burden of render cleanup on the back of Adaptive Sampling. That is why he set the Light/Shading Samples to 1, and the Rays Per Evaluation for Monte Carlo to 4. With the Minimum Samples set to 3, the scene actually starts with a base level of 3 Light/Shading Samples -- 1 Light/Shading sample for each of the 3 camera rays. The scene also has a base level of 12 RPE for the Monte Carlo Radiosity -- 4 RPE for each of the 3 camera rays initially fired per pixel.

    With a Threshold of 0.01 and a Maximum Samples of 192, the camera is forced to fire up to 192 rays per pixel to clean up the initial mess. It will keep firing rays until the difference in brightness between each pair of pixels falls below 0.01, or until 192 rays have been fired for each pixel -- whichever comes first. Each additional camera ray takes an additional 1 Light Sample, 1 Shading Sample, and 4 Rays Per Evaluation. A pixel that hits the maximum 192 camera samples will also have 192 Light Samples (1 Light Sample x 192 camera rays), 192 Shading Samples (1 Shading Sample x 192 camera rays) and 768 Rays Per Evaluation (4 RPE x 192 camera rays).

    The Ray Cutoff (Render Globals > Render > Ray Cutoff) is just as finicky as the Threshold in this scene -- a value of 0.01. This is the "Threshold" setting for Ray Recursion. The Ray Recursion Limit is set to 12, but the rays will stop recursing if the difference in brightness between the last two bounces was less than 0.01 or until they stop hitting a reflective/refractive surface. If anyone want this render to take even longer, set the Ray Cutoff to 0.0 to force all rays to recurse up to 12 bounces until they stop hitting a reflective/refractive surface.
    Last edited by Celshader; 02-03-2013 at 10:10 AM. Reason: Added more tech info...
    Jen's 3D -- LightWave stuff.
    Jen's 2D -- my comic book.

    Python is my smashing board. LightWave is my S.M.A.K.

  4. #4
    Studio Animator DonJMyers's Avatar
    Join Date
    Aug 2003
    Location
    Los Angeles, CA
    Posts
    575
    Quote Originally Posted by Celshader View Post
    I did not change the settings. I loaded up the scene as-is from the 11.5 content directory and hit F9. What happens when you render this benchmark on your machine?
    My screw up Jen. I lied... I haven't really run the whole benchmark yet so I thought it would be as pebbly at the end as the beginning. When I rendered a limited area ALL 192 TIMES it came out smooth. Ooops!

    Oh well, it was worth screwing up just to read your detailed explanation of what happens.

  5. #5
    Carbon fibre dongle® 50one's Avatar
    Join Date
    Jan 2011
    Location
    Honeycomb hideout
    Posts
    2,384
    Time : 2h 6m 39s

    Click image for larger version. 

Name:	renderfest.jpg 
Views:	604 
Size:	407.4 KB 
ID:	110994


    Machine Spec(@ home):
    - i7 2600K non-overclocked 3.4GHz
    - 16GB Corsair DDR3 Vengeance
    - Asus P8Z77-V LX, Intel Z77
    - Win7 / x64

    Was doing other things in the background during the render tho(Zbrush & modo & browsing the web).

    Must admit that this release seems to be the one of, if not the most stable LW up-to-date, remember times when exiting app without saving the scene was causing the app to go into 'freeze' mode.
    Last edited by 50one; 02-02-2013 at 01:11 PM. Reason: should be 39s instead of 59s

  6. #6
    Cow Orker cagey5's Avatar
    Join Date
    Feb 2003
    Location
    Manchester UK
    Posts
    3,273
    1Hr 50m 25s.
    Intel i7 3770 @ 3.5Ghz no overclocking,
    win 7, x64
    8 Meg Memory
    Forum Tips
    If English is your first language, you should know it's 'losing' and not 'loosing'!

  7. #7
    Banned
    Join Date
    Aug 2007
    Location
    Sydney, Australia
    Posts
    846
    hardware as below is current. Time to re-equip.
    3hrs 5secs
    Last edited by BokadCastle; 02-02-2013 at 07:01 PM.

  8. #8
    Spinny Spinland's Avatar
    Join Date
    Apr 2005
    Location
    Utica NY, USA
    Posts
    2,014
    Heh. I knew I would probably be close to last in this, but it was fun to check out where I stood.

    Mac Mini with 2.0 GHz quad core i7, 16GB of RAM.

    Click image for larger version. 

Name:	mini_stats.png 
Views:	565 
Size:	19.5 KB 
ID:	111021

    Here's my (pretty weak) render time:

    Click image for larger version. 

Name:	marbles_time.png 
Views:	1006 
Size:	518.7 KB 
ID:	111022

    And the final render:

    Click image for larger version. 

Name:	marbles_benchmark.png 
Views:	469 
Size:	1.75 MB 
ID:	111023
    “There may be people that have more talent than you, but there’s no excuse for anyone to work harder than you do.” ~ Derek Jeter

    iMac Pro
    "iCan" Mac Pro
    MacBook Pro
    Multiple Mac Minis

  9. #9
    obfuscated SDK hacker Lightwolf's Avatar
    Join Date
    Feb 2003
    Location
    Stuttgart, Germany
    Posts
    13,613
    For the heck of it:
    2h 6m 12s (with a bit of work being done at the same time).

    i7-2600K - slight OC to 3.6GHz (up 100Mhz from stock), 4 Cores + 4HT
    32GB, GeForce 560Ti, homebuilt

    Edit: Amazing to see that the 6-core with OC is almost as fast as Dave's 32 core machine.
    Last edited by Lightwolf; 02-06-2013 at 04:12 AM.

  10. #10
    Red Mage Celshader's Avatar
    Join Date
    Feb 2003
    Location
    Burbank, California
    Posts
    1,817
    Quote Originally Posted by Lightwolf View Post
    For the heck of it:
    2h 6m 12s (with a bit of work being done at the same time).

    i7-2600K - slight OC to 3.6GHz (up 100Mhz from stock), 4 Cores + 4HT
    32GB, GeForce 560Ti, homebuilt

    Edit: Amazing to see that the 6-core with OC is almost as fast as Dave's 32 core machine.
    Dave's cores run at 2.1Ghz, though, not 3.6Ghz.

    A better comparison would be my own FX-8150 3.6Ghz machine, which has four Bulldozer modules on the processor (eight integer cores + four shared 256bit FPUs). The shared 256bit FPUs are supposed to be able to execute two 128bit instructions at a time, but in our tests it seemed like the speed boost was more like 1.6x instead of 2x.

    LightWolf's total Ghz = 3.6 x 4 FPUs = 14.4Ghz
    est. 33% HT speed boost - the above x 1.33 = 19.152Ghz
    2h 6m 12s render time

    Jen's total Ghz = 3.6Ghz x 4 shared FPUs = 14.4Ghz
    The above x 1.6 instructions at a time = 23.04Ghz
    2h 15m 22s render time

    So, your $319 3.6Ghz processor easily bested my $180 3.6Ghz processor, even though my processor executed more instructions per second.
    Jen's 3D -- LightWave stuff.
    Jen's 2D -- my comic book.

    Python is my smashing board. LightWave is my S.M.A.K.

  11. #11
    Axes grinder- Dongle #99
    Join Date
    Jul 2003
    Location
    Seattle
    Posts
    14,732
    Quote Originally Posted by Celshader View Post
    So, your $319 3.6Ghz processor easily bested my $180 3.6Ghz processor, even though my processor executed more instructions per second.
    Time to introduce a dollar-based metric in these benchmarks. Surely render-farm managers must.
    They only call it 'class warfare' when we fight back.
    Praise to Buddha! #resist
    Chard's Credo-"Documentation is PART of the Interface"
    Film the cops. Always FILM THE COPS. Use this app.

  12. #12
    obfuscated SDK hacker Lightwolf's Avatar
    Join Date
    Feb 2003
    Location
    Stuttgart, Germany
    Posts
    13,613
    Quote Originally Posted by jeric_synergy View Post
    Time to introduce a dollar-based metric in these benchmarks. Surely render-farm managers must.
    Yes, but then you also need to factor in the complete system price (which will be a lot more equal since most components cost the same) and then compare the price difference.

    I.e. a difference of 140 bucks sounds like a lot if the performance differences is, say, 30-10% only. However, if the base system costs you, say, US$1500 then the difference in price is in tune with the difference in performance.

    For a larger installation, power use is crucial as well (direct and indirect cost: electricity and cooling).

    Cheers,
    Mike

  13. #13
    obfuscated SDK hacker Lightwolf's Avatar
    Join Date
    Feb 2003
    Location
    Stuttgart, Germany
    Posts
    13,613
    Quote Originally Posted by Celshader View Post
    Dave's cores run at 2.1Ghz, though, not 3.6Ghz.

    A better comparison would be my own FX-8150 3.6Ghz machine, which has four Bulldozer modules on the processor (eight integer cores + four shared 256bit FPUs). The shared 256bit FPUs are supposed to be able to execute two 128bit instructions at a time, but in our tests it seemed like the speed boost was more like 1.6x instead of 2x.

    ...

    So, your $319 3.6Ghz processor easily bested my $180 3.6Ghz processor, even though my processor executed more instructions per second.
    I computed something similarly, but trying to find out the normalised render time: I.e. one physical core of a certain machine at 1Ghz. Essentially by multiplying the time by the amount of (physical) cores and Ghz. Precisely to find out the speed per clock (lower time = better).

    In my case, it'd be 1814 minutes running on a single core at 1GHz (including the HT boost), 3626 minutes assuming that one physical core+HT is two cores.
    Dave's machine would be 3939 minutes, yours 3888 (assuming 8 cores, especially since LW doesn't use 256bit fp instructions and the cores can thus split the shared fpu).
    Two double check, the E5-2687W mentioned earlier would take 1636 minutes (assuming to automatic overclocking, stock speed of 3.1Ghz). Assuming that it automatically overclocks by a bin (like mine does) it'd be around 1700 minutes, which is (not) surprisingly close to the performance of my i7 (it's the same gen CPU).

    However, that also implies that Intel has the lead when it comes to the instructions per clock, one way or another (which is consistent with the background information that I've been reading).

    Cheers,
    Mike

  14. #14
    Red Mage Celshader's Avatar
    Join Date
    Feb 2003
    Location
    Burbank, California
    Posts
    1,817
    Quote Originally Posted by Lightwolf View Post
    assuming 8 cores, especially since LW doesn't use 256bit fp instructions and the cores can thus split the shared fpu
    We've rendered the benchmark at 16 threads on Dave's machine (one thread per FPU) and 32 threads (two threads per FPU). Instead of twice the performance, it's more like 1.6x the performance.

    The speed boost is there, but it is not the same as 32 independent FPUs would have delivered.

    I also botched my description above -- my 3.6Ghz system executed more total clock cycles than your 3.6Ghz system, not more instructions. The Intels definitely do more per clock cycle than the AMDs. However, AMD systems are cheaper to build in the States and have much more forgiving upgrade paths, so Dave and I are sticking with the AMD platform at this time.
    Jen's 3D -- LightWave stuff.
    Jen's 2D -- my comic book.

    Python is my smashing board. LightWave is my S.M.A.K.

  15. #15
    gettin all wavy rwhunt99's Avatar
    Join Date
    May 2004
    Location
    Osceola, IN
    Posts
    213
    Spnland; I was noticing your render, it seems different from mine and others. The marble on the left has little color in it, and you seem to have much more I don't know what you call it (clear glass) on the edges on the left and right marbles. Wonder what causes that?
    Last edited by rwhunt99; 02-06-2013 at 07:12 PM.

Page 1 of 40 12311 ... LastLast

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •