Luxmark V3 and Render/Gaming GPU Compute Efficiency

Toggle sidebar Toggle sidebar
A

artivix

Member
May 5, 2014
56
0
0
luxmarkv3c5c2c.jpg

LuxMark v3.x

LuxMark is a OpenCL cross-platform benchmark tool and has become, over past years, one of the most used (if not the most used) OpenCL benchmark. It is intended as a promotional tool for LuxRender and it is now based on LuxCore, the LuxRender v2.x C++ or Python API available under Apache Licence 2.0 and freely usable in open source and commercial applications.

http://www.luxrender.net/wiki/LuxMark#Binaries
Click to expand...

When tech review sites have comparisons for rendering and compute speed, it looks like current AMD Hawaii/Tahiti GPUs have dramatically higher efficiency over Nvidia Maxwell/Kepler GPUs.

Considering that AMD will be the foundation for all the major consoles and that gaming will involve more GPU compute, it would seem that AMD could take the overall efficiency crown with their next gen 3xxx graphics cards when every variable is taken into account.

http://www.luxmark.info/top_results/Hotel/OpenCL/GPU
 
Last edited:
3

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Shhh! We don't talk about compute when comparing efficiency of the 2 brands. ;)

Seriously though, it seems that nVidia has a more specialized arch for gaming that has traded all around performance for lower power usage. AMD is making higher performing all around chips but in doing so require more power.
 
H

Hitman928

Diamond Member
Apr 15, 2012
6,614
12,127
136
bystander36 said:
In previous versions, it shows the 980 at the top, and that version didn't even show a 980 in there, but a 970 showed up. I'm not sure if that is a good cross-section or not.

http://wccftech.com/nvidia-maxwell-gtx-980-opencl-benchmarks/
Click to expand...

The wccftech links to phoronix http://www.phoronix.com/scan.php?page=article&item=nvidia_gtx980_opencl&num=5

They're using v2.1 whereas the tables in the op show the results for the latest version 3.1. I would probably trust the latest version results more but I'm not familiar with the program, I don't do render stuff.
 
myocardia

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
Why do you have the word gaming in your thread title? What you linked has zero to do with gaming.
 
K

kagui

Member
Jun 1, 2013
78
0
0
the old test even showed more advantage to amd requiering 8 titans to surpass 4 hd7970
http://www.luxrender.net/luxmark/

but that results when was nvidia was hurting their own customers with restricting opencl performance, and the send results was broken so you cant send new results and see the changes nvidia have been doing to their driver.

so with the new test, you can see they are taking opencl more serious, even when they reduced double presicion to 1/32. at least you could use more of your nvidia card
 
el etro

el etro

Golden Member
Jul 21, 2013
1,584
14
81
3DVagabond said:
Shhh! We don't talk about compute when comparing efficiency of the 2 brands. ;)

Seriously though, it seems that nVidia has a more specialized arch for gaming that has traded all around performance for lower power usage. AMD is making higher performing all around chips but in doing so require more power.
Click to expand...

Whoa! At least one understands it!


Bravo! :D:D:D

(is not irony, i'm totally agreed with you).
 
Fallen Kell

Fallen Kell

Diamond Member
Oct 9, 1999
6,148
503
126
el etro said:
3DVagabond said:
Shhh! We don't talk about compute when comparing efficiency of the 2 brands. ;)

Seriously though, it seems that nVidia has a more specialized arch for gaming that has traded all around performance for lower power usage. AMD is making higher performing all around chips but in doing so require more power.
Click to expand...

Whoa! At least one understands it!


Bravo! :D:D:D

(is not irony, i'm totally agreed with you).
Click to expand...

I actually wouldn't say "AMD is making higher performing all around chips". It is more like, AMD is supporting GPGPU on their consumer class cards. Nvidia simply segragates their products into 2 lines, consumer which mainly need frames per second for displaying rendered computer graphics with some boost in performance of a small set of massively parallel data computations (typically in video rendering and processing), and their professional lines which focus on GPGPU. AMD simply has one product line (yes, they have a "professional line" but it is mainly just added support).

Nvidia also flat out performs poorly with OpenCL because they don't really support it. They prefer to support their proprietary CUDA, and rightly so considering they developed it to begin with and OpenCL came about afterwards in an attempt to have a "standardized" platform that works on "all" GPUs. AMD embraced OpenCL because it let them get into a market which Nvidia had created. Nvidia simply sees it as a competitor to their own product, but begrudgingly is starting to support it as they can understand that developers would much rather write their code once and not need to have multiple code branches for dealing with different hardware.
 
Last edited:
3

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Fallen Kell said:
I actually wouldn't say "AMD is making higher performing all around chips". It is more like, AMD is supporting GPGPU on their consumer class cards. Nvidia simply segragates their products into 2 lines, consumer which mainly need frames per second for displaying rendered computer graphics with some boost in performance of a small set of massively parallel data computations (typically in video rendering and processing), and their professional lines which focus on GPGPU. AMD simply has one product line (yes, they have a "professional line" but it is mainly just added support).

Nvidia also flat out performs poorly with OpenCL because they don't really support it. They prefer to support their proprietary CUDA, and rightly so considering they developed it to begin with and OpenCL came about afterwards in an attempt to have a "standardized" platform that works on "all" GPUs. AMD embraced OpenCL because it let them get into a market which Nvidia had created. Nvidia simply sees it as a competitor to their own product, but begrudgingly is starting to support it as they can understand that developers would much rather write their code once and not need to have multiple code branches for dealing with different hardware.
Click to expand...

GCN is just stronger with compute loads. That's why I said stronger all around. Nothing touches Hawaii's 1/2 DP in nVidia's pro line, for example.
 
Makaveli

Makaveli

Diamond Member
Feb 8, 2002
4,916
1,504
136
Last edited:
F

FiendishMind

Member
Aug 9, 2013
60
14
81
IIRC Maxwell is neck and neck with Hawaii in DirectCompute performance based on ComputeMark benches. Seems like Nvidia just has poor OpenCL performance but that's their fault for not properly supporting it.
 
3

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Makaveli said:
http://en.wikipedia.org/wiki/AMD_Radeon_Rx_200_Series

Double precision performance of Hawaii is 1/8 of single precision performance,[43] Tahiti is 1/4 of single precision performance, others 28 nm chip is 1/16 of single precision performance.

http://i60.tinypic.com/2n03rbc.png

http://i57.tinypic.com/23ucyon.png

http://i57.tinypic.com/2zedky1.png
Click to expand...

The AMD FirePro W9100 graphics features and benefits include:

2.62 TFLOPS of peak double-precision GPU compute performance - AMD FirePro W9100 graphics cards feature ½ rate double precision1, enabling more than 2TFLOPS of peak double precision performance on a graphics processor for the first time. Unlike competing cards that are not optimized for double precision, AMD FirePro W9100 workstation graphics cards offer the most double precision performance, completing compute-intensive tasks faster than ever believed possible.3
5.24 TFLOPS of peak single-precision GPU compute performance - Helps speed up time required to complete single precision operations used within Simulations, Post Production and Effects, Signal Processing, Video Transcoding and Digital Rendering applications where high performance takes precedence over accuracy.

LINK:

Looks like 1/2 to me.
 
3

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
naukkis said:
It's 1/2 but consumer-grade versions are crippled to 1/8.
Click to expand...

What does that have to do with the capability of GCN? Do you see Fermi or Kepler, or even Maxwell able to do 1/2 DP, consumer or pro? No. Look at this thread. It shows GCN's superior compute capability (which is what it's about). All I said is that it's only gaming that nVidia beats AMD in efficiency.

That's the problem when people (not you) throw all of these strawman arguments in. Someone (like you) comes in a couple of posts later and they end up no knowing what the original post was about.
 
el etro

el etro

Golden Member
Jul 21, 2013
1,584
14
81
Fallen Kell said:
I actually wouldn't say "AMD is making higher performing all around chips". It is more like, AMD is supporting GPGPU on their consumer class cards. Nvidia simply segragates their products into 2 lines, consumer which mainly need frames per second for displaying rendered computer graphics with some boost in performance of a small set of massively parallel data computations (typically in video rendering and processing), and their professional lines which focus on GPGPU. AMD simply has one product line (yes, they have a "professional line" but it is mainly just added support).
Click to expand...

This. Here.

3DVagabond said:
GCN is just stronger with compute loads. That's why I said stronger all around. Nothing touches Hawaii's 1/2 DP in nVidia's pro line, for example.
Click to expand...
 
G

gamervivek

Senior member
Jan 17, 2011
490
53
91
Didn't realize that Kepler didn't do 1/2 DP or that Hawaii broke through 2TF barrier, AMD should be making some serious money in the pro division with such an advantage.
 
A

artivix

Member
May 5, 2014
56
0
0
myocardia said:
Why do you have the word gaming in your thread title? What you linked has zero to do with gaming.
Click to expand...

The 3D graphics software like Blender that utilize render plugins like Luxrender also have provision for gaming and simulations with tools like Bullet Physics 3 that use GPUs running OpenCL:

(From wiki)

Features --

Rigid body and soft body simulation with discrete and continuous collision detection

Collision shapes include: sphere, box, cylinder, cone, convex hull using GJK, non-convex and triangle mesh

Soft body support: cloth, rope and deformable objects

A rich set of rigid body and soft body constraints with constraint limits and motors

Plugins for Maya, Softimage, integrated into Houdini, Cinema 4D, LightWave 3D and Blender and import of COLLADA 1.4 physics content

Optional optimizations for PlayStation 3 Cell SPU, CUDA and OpenCL [3]

The Bullet website also hosts a Physics Forum for general discussion around Physics Simulation for Games and Animation.

At AMD Developer Summit (APU) in November 2013 Erwin Coumans presented the Bullet 3 OpenCL Rigid Body Simulation.

http://en.wikipedia.org/wiki/Bullet_(software)
Click to expand...
 
Makaveli

Makaveli

Diamond Member
Feb 8, 2002
4,916
1,504
136
3DVagabond said:
The AMD FirePro W9100 graphics features and benefits include:

2.62 TFLOPS of peak double-precision GPU compute performance - AMD FirePro W9100 graphics cards feature ½ rate double precision1, enabling more than 2TFLOPS of peak double precision performance on a graphics processor for the first time. Unlike competing cards that are not optimized for double precision, AMD FirePro W9100 workstation graphics cards offer the most double precision performance, completing compute-intensive tasks faster than ever believed possible.3
5.24 TFLOPS of peak single-precision GPU compute performance - Helps speed up time required to complete single precision operations used within Simulations, Post Production and Effects, Signal Processing, Video Transcoding and Digital Rendering applications where high performance takes precedence over accuracy.

LINK:

Looks like 1/2 to me.
Click to expand...

Thanks for the link I thought DP was the same with the Pro cards and the consumer ones.
 
L

Lepton87

Platinum Member
Jul 28, 2009
2,544
9
81
How does GK210 perform in compute workloads? I haven't seen any tests of it yet. I couldn't even find the information about its die size I wonder if it's at all bigger than GK110 or how many shaders it physically has. Current implementation on the Tesla K80 only has 2X13SMX'es(2496) Such a big unknown. I
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom