Download Trial Buy now
Log in or Sign up for the chat.

Search  | Login  | Register

Post new topic Reply to topic  [ 72 posts ]  Go to page Previous  1, 2, 3, 4, 5
Offline
User avatar
Indigo 100
TonyD wrote:
Hi, I ran 2 different setups on my rig...
Thanks!
Tony D.

unbelievable
what's the hell!!
what's the system specification?


 Profile  
 
Offline
snorky wrote:
TonyD wrote:
Hi, I ran 2 different setups on my rig...
Thanks!
Tony D.

unbelievable
what's the hell!!
what's the system specification?


Hello, Win7 64 on Asrock X79 Extreme 11. Titan X on mobo.
Use a pair of Amfeltec splitters for the Titan Z arrangement, and USB Risers for the additional 2 Titan X arrangement.
BIOS won't boot with > 13, so run either of these 2 arrangements (external PSU) depending on strategy.

The Benchmark shows render speed but doesn't actually show real-life process time (ie including OpenCL scene build time, acceleration to top GPU speed...). So it's a funny thing, how it measured them at parity, it just so happens to capture them at 3 tile building passes. However, a 12 Titan Z, arrangement on a larger render (High res, many samples, >3 tile-building passes .igs) would be 'faster', whereas in building a scene quickly at low res and lower samples, and reaching top render speed, 3 Titan X's would be 'faster' (ie for a frame sequence .igq), .

In other words, if IR could build a scene and have the rendering GPUs reach top speed with 12 TitanZ as fast as with 3 TitanX, 12 TitanZ would be faster in all cases. But since with 12 Z it builds OpenCL and reaches top speed slower, the 3 Titan X is actually much faster at pumping out frames. Perhaps more PCI 1x cards slows scene building and reaching top render speed down.

I will note that 4.0.45 BETA was improved to have OpenCL scenes with multi-GPU build much faster than 4.0.44 BETA did!
:)

-Tony D.


 Profile  
 
Offline
User avatar
TonyD wrote:
snorky wrote:
TonyD wrote:
Hi, I ran 2 different setups on my rig...
Thanks!
Tony D.

unbelievable
what's the hell!!
what's the system specification?


Hello, Win7 64 on Asrock X79 Extreme 11. Titan X on mobo.
Use a pair of Amfeltec splitters for the Titan Z arrangement, and USB Risers for the additional 2 Titan X arrangement.
BIOS won't boot with > 13, so run either of these 2 arrangements (external PSU) depending on strategy.

The Benchmark shows render speed but doesn't actually show real-life process time (ie including OpenCL scene build time, acceleration to top GPU speed...). So it's a funny thing, how it measured them at parity, it just so happens to capture them at 3 tile building passes. However, a 12 Titan Z, arrangement on a larger render (High res, many samples, >3 tile-building passes .igs) would be 'faster', whereas in building a scene quickly at low res and lower samples, and reaching top render speed, 3 Titan X's would be 'faster' (ie for a frame sequence .igq), .

In other words, if IR could build a scene and have the rendering GPUs reach top speed with 12 TitanZ as fast as with 3 TitanX, 12 TitanZ would be faster in all cases. But since with 12 Z it builds OpenCL and reaches top speed slower, the 3 Titan X is actually much faster at pumping out frames. Perhaps more PCI 1x cards slows scene building and reaching top render speed down.

I will note that 4.0.45 BETA was improved to have OpenCL scenes with multi-GPU build much faster than 4.0.44 BETA did!
:)

-Tony D.


I think, in essence, it's partly due to longer build time, but also due to short passes not being suitable for many GPUs. The GPUs need to sync and sometimes wait for each other at the end of every pass, and can really start to stretch their legs only when the passes are longer. The devs are well aware of this.


 Profile  
 
Offline
Oscar J wrote:

I think, in essence, it's partly due to longer build time, but also due to short passes not being suitable for many GPUs. The GPUs need to sync and sometimes wait for each other at the end of every pass, and can really start to stretch their legs only when the passes are longer. The devs are well aware of this.


Seems like a lot of material IDs (ie individual materials) will slow down the render time, as well as the build time. If you have a model with one master mat, my goodness, Indigo whips through each pass blazing speed on OpenCL (like 1 second!) So, assuming if all GPU ran at same speed, it seems some slowdown factors will be:

-Higher Model Resolution
-Greater # Individual Materials
-More GPUs being fed the data

-Tony D.


 Profile  
 
Offline
User avatar
Slower rendering with increased model and material complexity is sort of expected from any renderer. :) Regarding build time, it's afaik the number of material types in your scenes that matter the most - Indigo has to compile the material types.


 Profile  
 
Offline
Oscar J wrote:
Slower rendering with increased model and material complexity is sort of expected from any renderer. :) Regarding build time, it's afaik the number of material types in your scenes that matter the most - Indigo has to compile the material types.


Regarding what OscarJ stated about build and rendering speed...
Wonder if this would be possible - a 'Freeze' mode, to:
(1) collapse all individual geometries into a unified mesh, as opposed to handling independent geometry objects.
(2) create a single material map, as opposed to multiple material IDs

As analogy, in multi-track audio, for rendering, a lot of times in a DAW (digital audio workstations) you can 'freeze', or lock, a track's intricate details (ie EQ, delay, chorus, effects), to reduce the rendering time of the final mixdown. Basically, you create a mini render of the track before the final mix. It allows the final mix to go faster.

When OscarJ mentioned the # materials needing a compilation, perhaps we could "ask" Indigo to 'freeze' the render prior to rendering such that it assigns everything otherwise modifiable as 'frozen' (locked), and then after build when it goes into rendering, it will allow the GPUs to go full tilt on the single mesh & material map.
BLAZING SPEED POSSIBILITIES?

-Tony D (with inspiration from OscarJ :P )


 Profile  
 
Offline
User avatar
Wait for the next beta... :)


 Profile  
 
Offline
Oscar J wrote:
Wait for the next beta... :)


ME WANT NOW!
ME WANT NOW!

Okay :)


 Profile  
 
Offline
Ono-Sendai,
idea - Maybe future benchmark we could set # passes?
Even if only available in an offline mode (ie not for uploading/submission to benchmark page), it would be helpful to have an organized comparative metric to see under different rendering scenarios how a GPU arrangement performs on the test scenes.
Thx and Regards,
Tony D.


 Profile  
 
Offline
User avatar
Ok, here goes my wooden pc benchmarks

Indigo Benchmark v4.0.46, Windows 64-bit build.
AMD FX(tm)-6100 Six-Core Processor (AuthenticAMD) - Bedroom: 0.227 M samples/s
AMD FX(tm)-6100 Six-Core Processor (AuthenticAMD) - Supercar: 0.488 M samples/s

Indigo Benchmark v4.0.46, Windows 64-bit build.
Quadro 4000 (NVIDIA Corporation) - Bedroom: 0.240 M samples/s
Quadro 4000 (NVIDIA Corporation) - Supercar: 0.794 M samples/s


 Profile  
 
Offline
I only see my CPU, but not my GPU (GTX 1080)


Indigo Benchmark v4.0.46, Windows 64-bit build.
Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz (GenuineIntel) - Bedroom: 1.141 M samples/s
Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz (GenuineIntel) - Supercar: 2.462 M samples/s


 Profile  
 
Offline
User avatar
Can not submit new results. Got this error massage:

Error: <!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML 2.0//EN"> <html><head> <title>302 Found</title> </head><body> <h1>Found</h1> <p>The document has moved <a href="https://www.indigorenderer.com/benchmark/benchmark_submit.php">here</a>.</p> <hr> <address>Apache/2.4.10 (Debian) Server at indigorenderer.com Port 80</address> </body></html>


 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 72 posts ]  Go to page Previous  1, 2, 3, 4, 5



You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  

Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group