Full Kepler?

[lambchops511]

Hi All.

I have a GTX460 right now and its not meeting my needs and am looking to upgrade. What is the latest rumour out on the full Kepler? Or should I just grab the GTX680.

How is the power consumption of GTX680, is it comparable to the GTX460, will it be able to handle the load? The PSU is Antec NEO ECO 520C 520W.

 

[toyota]

it probably wont be until very late this year or early next year before we get Big Kepler for desktop. and you can look at the tons of gtx680 reviews to see how power consumption stacks up. and you need to list ALL of you other specs as power supply needs are not just based on video card. and while you are at it go ahead and list the resolution you play at too.

 

[lambchops511]

CPU is i5-760 / motherboard is some Intel standard thing.

Not using it for gaming, using it for compute. Would GTX580 be better? I am using float32 not double64 tho, all the benchmarks I saw was float64 GTX580 better, but unclear for float32.

Or is GTX670 a good buy? It is a good $200 cheaper...

 

[Smoblikat]

CPU is i5-760 / motherboard is some Intel standard thing.

Not using it for gaming, using it for compute. Would GTX580 be better? I am using float32 not double64 tho, all the benchmarks I saw was float64 GTX580 better, but unclear for float32.

Or is GTX670 a good buy? It is a good $200 cheaper...

Im pretty sure they gimped compute on the 680. I hear AMD is better for compute.

 

[Denithor]

GTX 580 better than GTX 680 for compute.

7970 even better than GTX 580.

 

[Don Karnage]

Expect 2nd week of december for GK110

 

[Nintendesert]

Expect 2nd week of december for GK110

D:

 

[hawtdawg]

Since China isn't going to be buying 10 million GPU's like they were going to, maybe we'll see big Kepler earlier than expected.

 

[Dark Shroud]

Since China isn't going to be buying 10 million GPU's like they were going to, maybe we'll see big Kepler earlier than expected.

And maybe it can be delivered by Santa Clause.

 

[lambchops511]

GTX 580 better than GTX 680 for compute.

7970 even better than GTX 580.

even for single fp?

 

[Arkadrel]

even for single fp?

Yes even for Single FP.

http://www.theinquirer.net/inquirer...hed-amds-mid-range-radeon-hd-7870-gpu-compute

Wow! The claim of beating the HD7970 goes right into the thin air, it seems. Nvidia's new GPU is beaten by the Radeon HD 7970 by an order of magnitude here in double precision floating-point, as well as nearly twice in ordinary single precision floating point. One is speechless here. Even the Radeon HD 7870 with its restricted double precision floating-point still outperforms the GTX680 by a noticeable margin in this department,...

​

http://forums.developer.nvidia.com/...-even-slower-than-a-gtx480-when-using-cuda/p1

I've tested several Apps in the GPU Computing SDK, such as the GrabCutNPP. Surprisingly I found the GTX680 is even slower than my old GTX480 (about 0.9x). Why could this happen? In contrast, the test on 3DMark11 reported that the GTX680 is 2x faster.

re:
It could be because the GTX 680 double-precision floating point calculations are purposely slowed down to 1/24th of single-precision ones; see my post on this:
http://forums.developer.nvidia.com/devforum/discussion/7126/gtx-680-performance

http://news.softpedia.com/news/AMD-Radeon-HD-7970-Much-Better-than-Kepler-in-OpenCL-270895.shtml

In single-precision floating point performance, AMD’s Radeon HD 7970 card achieves a 42% higher performance than Nvidia’s Kepler and 155% over Fermi.

In double-precision floating point performance, AMD’s Radeon HD 7970 card obtains a 478% higher performance than Nvidia’s Kepler and 293% over Fermi.

A 7970 spanks a 680 when it comes to compute tasks.

 

[Don Karnage]

Can someone list some "Compute" tasks for me

 

[The_Golden_Man]

Can someone list some "Compute" tasks for me

I also wonder about this... I use my GTX 670's for gaming... What excactly do you use Compute tasks for?

 

[Arkadrel]

Can someone list some "Compute" tasks for me

http://en.wikipedia.org/wiki/GPGPU

Applications:


The following are some of the areas where GPUs have been used for general purpose computing:

• MATLAB acceleration using the Parallel Computing Toolbox and MATLAB Distributed Computing Server,[14] as well as 3rd party packages like Jacket.
• k-nearest neighbor algorithm[15]
• Computer clusters or a variation of a parallel computing (utilizing GPU cluster technology) for highly calculation-intensive tasks:
  • High-performance computing clusters (HPC clusters) (often termed supercomputers)
    • including cluster technologies like Message Passing Interface, and single-system image (SSI), distributed computing, and Beowulf
  • Grid computing (a form of distributed computing) (networking many heterogeneous computers to create a virtual computer architecture)
  • Load-balancing clusters (sometimes termed a server farm)
• Physical based simulation and physics engines (usually based on Newtonian physics models)
  • Conway's Game of Life, cloth simulation, incompressible fluid flow by solution of Navier-Stokes equations
• Statistical physics
  • Ising model
• Lattice gauge theory
• Segmentation – 2D and 3D
• Level-set methods
• CT reconstruction
• Fast Fourier transform
• Tone mapping
• Audio signal processing
  • Audio and Sound Effects Processing, to use a GPU for DSP (digital signal processing)
  • Analog signal processing
  • Speech processing
• Digital image processing
• Video Processing[16]
  • Hardware accelerated video decoding and post-processing
    • Motion compensation (mo comp)
    • Inverse discrete cosine transform (iDCT)
    • Variable-length decoding (VLD)
    • Inverse quantization (IQ)
    • In-loop deblocking
    • Bitstream processing (CAVLC/CABAC) using special purpose hardware for this task because this is a serial task not suitable for regular GPGPU computation
    • Deinterlacing
      • Spatial-temporal de-interlacing
    • Noise reduction
    • Edge enhancement
    • Color correction
  • Hardware accelerated video encoding and pre-processing
• Global illumination – ray tracing, photon mapping, radiosity among others, subsurface scattering
• Geometric computing – constructive solid geometry, distance fields, collision detection, transparency computation, shadow generation
• Scientific computing
  • Monte Carlo simulation of light propagation[17]
  • Weather forecasting
  • Climate research
  • Molecular modeling on GPU
  • Quantum mechanical physics
  • Astrophysics[18]
• Bioinformatics[19][20]
• Computational finance
• Medical imaging
• Computer vision
• Digital signal processing / signal processing
• Control engineering
• Neural networks
• Database operations[21]
• Lattice Boltzmann methods
• Cryptography and cryptanalysis
  • Implementation of MD6
  • Implementation of AES[22][23]
  • Implementation of DES
  • Implementation of RSA[24]
  • Implementation of ECC
  • Password cracking[25][26]
• Electronic Design Automation[27][28][29]
• Antivirus software[30][31]
• Intrusion Detection[32][33]
• Bitcoin peer-to-peer currency relies on a distributed computing network for performing SHA256 calculations where GPGPUs have become the dominant mode of calculation

^ Im sure thats just top of the iceberg.

For everyday personal usage, theres a million small programs that make use of the GPU.

 

[3DVagabond]

Personally I would use it for making "test renders" of my models. Doing so, at good quality, is extremely time consuming with a CPU. While final rendering is still better quality with a CPU, GPU rendering allows for very good quality (not excellent) in a very short time. I haven't upgraded yet to a GCN card, but plan on it when I have some extra cash just for that reason.

 

[SlowSpyder]

I use my video card for gaming, primarily. But I think the only reason GPU advancements happen at the rate they do is because Nvidia and AMD see the potential of GPGPU. GPGPU likely has a big future, GPU's and consequently PC gaming advances because of that.

I wonder how much money Nvidia and AMD pour into a new GPU, from start to finished silicon ready to go on a board. I wonder if PC gaming alone could support that anymore?

Anyway, I think we're all looking forward to big K. I have a feeling it's gaming performance will not be a huge advancement, but it's GPGPU performance will. I'm sure it'll be faster overall by a decent amount, but my hunch (based on nothing) is that it won't be a huge leap in gaming performance.

 

[Nintendesert]

I also wonder about this... I use my GTX 670's for gaming... What excactly do you use Compute tasks for?

It's stuff people do when they don't pay electricity bills.

 

[Homeles]

It's stuff people do when they don't pay electricity bills.

This is so ridiculously ignorant...

 

[3DVagabond]

It's stuff people do when they don't pay electricity bills.

Well, some would say that it's people who don't pay electricity bills who spend thousands of dollars on a PC to just play games. That's probably not any more accurate, but it's a reasonable parallel argument.

 

[Silverforce11]

I recently visited a company called BGI based in Beijing/HK. At the time, they were expanding server racks for the entire floor of a building, with raw GPU computing to process massive sequencing and proteomic data. They do this service for profit, its fast and efficient, and most likely not possible/profitable with CPUs. It's an amazing sight to see so many GPUs put to good use.

Now with the advant of sosphisticated mass-spec analysis for biology research, all these fancy equips need to be backed up by raw calculation power, something CPUs just fall flat on their face in terms of per/area, /w and $ etc. Many institutions have small server racks just for these purpose.. not to mention giant super computers for theoretical physics.

So statements like "GPGPU/HPC etc becoming big in the future".. no, it's already big. It's just going to get massive.

AMD may have better hardware currently but I fail to see news on their penetrance in this market, and it's pretty damn obvious why they haven't: drivers and software support. Without a big dev support push like what NV has, they will continue to fail regardless of how good their hardware is. So until they win a huge deal for the next super computer, color me unimpressed with their efforts.

edit: Personally I think the approach of NV is perfect. Gaming GPUs DO NOT need compute and dp throughput. Leave those costly (die space and TDP) features in cards dedicated for that market.

 

[Dark Shroud]

Can someone list some "Compute" tasks for me

Any video game that makes use of Direct Compute. Also a hell of a lot of us do BitCoins.

 

[Arkadrel]

Can someone list some "Compute" tasks for me

For fun I went to lookup (googled) a few OpenCL (specific) applications:
(I tried not to include any Debuggers, Benchmark tools, Emulators, software library or API's that use OpenCL ect)



List: (below)

AlgoryX Agx Physicis Simulation

Abaqus FEA (used in the automotive, aerospace, and industrial products industries. Product is popular with academic & research institutions)

Adobe Photoshop CS6

Adobe Flash

ArcSoft Panorama Maker Pro 5 or 6 (Stitch photos and videos into 3D panoramas)

ArcSift ShowBiz 5.0.1

Autodesk Maya - Physics Plug-in

Battlefield 3 (Uses tile-based lighting through DirectX 11 Compute Shaders)

Blackmagic-design's DaVinci Resolve

Brown Deer's N-Body Simulations (motion of Particles subject to a force due to particle-particle interations between all particles in the system)

Brown Deer's Finite-Difference Time-Domain Solvers for Modeling Velocity-Stress Wave Propagation

Bullet Physics 2.80 (engine for Game Physics Simulation)

Corel VideoStudio Pro X4 (HD video-editing software with DVD and Blu-ray authoring using OpenCL for rendering)

Cryptohaze (OpenCL accelerated password auditing tools for security professionals)

CyberLink PowerDirector 9+ (Video Editor)

CyberLink PowerProducer 5.5+ (Create Compelling Video and Slideshow)

CyberLink MediaEspresso 6.5+ ( Ultra Fast Universal Media Converter)

Darktable (Virtual lighttable and darkroom for photographers with OpenCL acceleration on the GPU)

Dolphin (Nintendo Wii/GC Emulator)

EDEM (element modeling for particle flow simulation of bulk material handling)

Elcomsoft Phone Password Breaker

Elcomsoft Wireless Security Auditor

Fusion by EyesON (an image compositing software program)

Final Cut Pro X (Uses OpenCL for accelerated rendering)

Geoweb3d (Advancing geospatial visualization and analysis)

GIMP (2.8rc1+) (Photo editing and image manipulation)

Handbrake (Video transcoding for video scaling, color space conversion, and lookahead function)

HydrogenAudio's FLACCL v0.2+ (lossless Audio Compression)

Ikena ISR (Real-time enhancement for full motion video via OpenCL)

ImageMagick (Software suite to create, edit, compose, or convert bitmap images)

Indigo Renderer (photorealistic renderer which simulates the physics of light to achieve near-perfect image realism)

LAMMPS Molecular Dynamics Simulations

MAGIX Movie Edit Pro MX Plus (Video editing, special effects and export for AVCHD)

MainConcept’s OpenCL™ H.264/AVC Encoder (Video encoder)

Mathwork's MATLAB (programming environment for algorithm development, data analysis, visualization, and numerical computation)

Movie Gate 3.11+

Movie Studio HD 11 (Video creation, editing and publishing)

MotionDSP vReveal (Photo/video analysis to fix video shake, poor lighting, or incorrect color using OpenCL acceleration)

MSC Nastran (Finite Element Analysis (FEA) solver for simulating stress, dynamics, or vibration of real-world, complex systems)

MuseMage (photo editing software)

OPTIS's THEIA RT (True physically correct real-time rendering and interaction)

Phaseone Capture One 6.0 (Image Software)

Phone Password Breaker (Forensic access to encrypted information stored in popular smartphones)

PhotoMonkee (Image editing with OpenCL-accelerated filters)

PowerDirector (HD video editing software using OpenCL)

PowerToy Particle & Fluid Simulations Application

ratGPU (Ray Tracing Renderer)

ReconstructMe (3D scanning app reconstructs mesh in realtime from Kinect/Xtion depth image)

RLIPS (R Linear Inverse Problem Solver) (R package for solving large overdetermined linear inverse problems with rotations made in parallel)

StagePresence (snips a person’s image out of any background as a virtual “green screen” )

Sony Vega's Pro 10 & 11

Sony Movie Studio HD 11

TotalMedia Theatre 5 (OpenCL-accelerated all-in-one media player for videos, Blu-ray, DVDs, and AVCHD)

Triton Ocean SDK (lets developers add open-ocean and shallow water effects to their virtual environments)

Unified Color The HDR Expose 2 (Photo editor / enhancer)

Vantage Transcode (uses OpenCL to accelerate video processing and H.264 encoding, with more sophisticated algorithms for scaling and deinterlacing and full 16-bit 4:4:4:4 processing)

Vegas Pro 11 (Integrated content creation environment for video and broadcast professionals for video processing, playback and rendering.)

Viewdle (OpenCL optimized facial recognition software)

VLC media player (Open-source based media players using OpenCL for stabilization and de-noising)

WinZip (16.5+) (By using OpenCL, WinZip is able to deliver dramatically faster compression speeds, processing multiple files in parallel)

Wireless Security Auditor (Uses OpenCL to try recover WPA/WPA2-PSK text passwords in order to test the security of wireless networks)

Wolfram Mathematica 8 (for areas such as linear algebra, image processing, financial simulation, and Fourier transforms)






^ Remember this is just OpenCL (and just what I could find with a brief look around on google).
Im sure if you add Nvidia's CUDA, or AMD's APP/STREAM and DirectCompute to the list, it would grow alot bigger.

 

[Arkadrel]

So statements like "GPGPU/HPC etc becoming big in the future".. no, it's already big. It's just going to get massive.

A8-3850 (lano) takes 22,7 secounds without OpenCL, just useing the CPU
i5-2410M (Sandy Bridge) takes 11,9 secounds without OpenCL, just useing the CPU

turn on OpenCL

A8-3850 (lano) takes 0.28 secounds with OpenCL.
(this A8-3850 goes from being half speed of the sandy bridge, to being 42x times faster)



There are tons of programs where OpenCL does something simular like this.
Going from 22,7 secounds -> 0.28 secounds is a HUGE performance improvement.


Yes I do think GPGPU has a future, and its starting to show up in lots of things.
Your internet browser uses it, Flash uses it, Winzip uses it, Video Players use it, Games are starting to use it.
Video/Picture editing software,... lots of stuff that many people use daily.

 

[lambchops511]

Yes even for Single FP.

http://www.theinquirer.net/inquirer...hed-amds-mid-range-radeon-hd-7870-gpu-compute

​

http://forums.developer.nvidia.com/...-even-slower-than-a-gtx480-when-using-cuda/p1

http://news.softpedia.com/news/AMD-Radeon-HD-7970-Much-Better-than-Kepler-in-OpenCL-270895.shtml


A 7970 spanks a 680 when it comes to compute tasks.

7970 may have better theoretical performance, but until ATI/AMD ups the developer support, I will be sticking w CUDA and its ecosystem of libraries available.

Your graph is FP64, I only care about FP32. The Vector calculation benchmarks are non-conclusive, I am more interested in Matrix-Vector calculations for FP32.

 

[Arkadrel]

If you only care about 32bit (FP) Floatpoint point performance it looks something more like this:

42,3% differnce.

but until ATI/AMD ups the developer support, I will be sticking w CUDA and its ecosystem of libraries available.

Yeah programming languages are like that, if you learnt CUDA I can understand why you want to stick with something your formilur with. Or if your useing a program that only works with CUDA currently.

That said CUDA is dying.

Intel/AMD/NVIDA/Samsung/IBM/ARM (TI and Qualcomm ect) (and more)
Are all behinde OpenCL, vs Nvidia only for CUDA.


You said this:

Not using it for gaming, using it for compute. Would GTX580 be better? I am using float32 not double64 tho, all the benchmarks I saw was float64 GTX580 better, but unclear for float32.

Thats why people have been saying get a 7970, instead of a 680.
I guess they just persumed you cared about performance, but if your a nvidia only kinda guy, then the 680 is the fastest 32 FP card they have.

However for any programs that are double precision workloads, the 580 will be faster than the 680.