The [[Offical]] ATI vs NVIDIA thread

[rancid_elf]

Well I was not a fanboy of either as for me hardware is hardware. The better hardware with driver support at the time gets my money.

HOWEVER. due to nVidia tactics and antics with things like preventing physx if no nVidia gpu is used to display ( and yes I was one of the people thought purchased an Ageia physx card on release day ) I no refuse to fund this company.

Microsoft was hated by many and actually sued over preventing 3rd party web browsers working in windows and to me nVidia's actions are the same.

Quality wise one day nVidia are better and then the next Ati are better. But lets face it there is never too much in it. The new cards are always close in performance and quality to really argue over so for me it has to come down to the companies polotics and nVidia suck.

So bottom line is Ati all the way for me and all nVidia cards I now own are used to rest my coffee cups on.

 

[Rroff]

Microsoft was hated by many and actually sued over preventing 3rd party web browsers working in windows and to me nVidia's actions are the same.

Not really - that would be like nVidia preventing you using other physics libraries when PhysX was installed.

 

[Hoodlum]

I think you will find I said HD films
your link is for opencl.

OpenCL is not mentioned once on the entire page. OpenCL is also not supported on any of those cards listed. (CUDA is not supported on most of them either).

This was also linked on the previous article under " GPU Acceleration in Adobe Premiere Pro Third party hardware compatibility"

Did you you mean the CUDA HD video encoding plug in? Because if so they both provide a plug-in for GPU acceleration. I have no idea how they compare, though.

Sources:
Nvidia: http://techreport.com/discussions.x/15571
AMD : http://www.brightsideofnews.com/new...-releases-adobe-cs4-plug-in-for-radeons!.aspx

Thats rubbish - CUDA adoption in industry and education isn't insignificant and is fast growing theres 100s of projects (of all sizes) that are publicised and lots more that aren't.

Really? And how many extremely popular games support Physx? Yep, the majority is far and away gaming, where it is the standard. I realise they are trying to move away from this, though.

CUDA is preferred over Open CL due to:

Better support
More mature and stable feature set
Only having to debug against one platform
Better documentation
Better performance (currently) for many commonly used features
Larger availability of (quality) framework and blank slate projects
The level of intergration with visual studio

I'm guessing you're not using OpenCL on the 5xxx series then. The performance is quite superb.* OpenCL is pretty trivial to debug as well, but it is true that you have to do it. The other points are all true, visual studio is the biggest one IMO, extremely important.

*Source: http://sisoftware.co.uk/?d=qa&f=cpu_vs_gpu_proc&l=en&a=

The Radeon 4xxx series is badly suited to OpenCL period and the older Nvidia driver is incredibly slow but is improving, this could cause the performance issues you are speaking of "beta driver 195.39 provides a 30% gain over the previous 190.89 version!"*

*Source: http://www.bealto.com/mp-gpu_flops.html

Another benchmark: http://oscarbg.blogspot.com/2009/11/gpu-compute-benchmarks.html

Sorry I couldn't find any more. I may post my own at some time if people are interested.

I wouldn't say Open CL is dead in the water - many people I talk to would prefer to use it if some of the above points weren't true - but currently CUDA is fast becoming the industry standard and Open CL is losing popularity in commercial useage. I would say your viewpoint is skewed due to being too close to the scientific community where Open CL took up ground faster before CUDA was really anything to talk about.

Can't really deny this, probably true, it definitely gets used; but It is also true that you can easily "miss" the projects using it - except for games where it is very significant.

 

[fizzle-187]

I had an ati 9800pro which was terrible and blew up. I then changed to a 7800GS which ****ed all over the ati. And the drivers are superb.

I prefer Nvidia over ati.

 

[Rroff]

Really? And how many extremely popular games support Physx? Yep, the majority is far and away gaming, where it is the standard. I realise they are trying to move away from this, though.

I'm not quite sure what your saying here... but even if we take the list of games supporting PhysX (around 100) its far far shorter than the public list of projects using CUDA (well over 600) - and only about 4-5 of those games in that list use hardware PhysX so the rest aren't even touching CUDA.

CUDA is taught in mainstream IT courses in more than 300 Unis worldwide and I know of atleast 2 with large scale deployments that are used beyond just learning - so I bet theres a good few more.

 

[jigger]

I had an ati 9800pro which was terrible and blew up. I then changed to a 7800GS which ****ed all over the ati. And the drivers are superb.

I prefer Nvidia over ati.

Of course a 7800GS is faster than a 9800pro its three generations newer.

The 7800GS was competing against the x1950pro...

 

[Lightnix]

CUDA is too limited from a development ubiquity point of view compared to OpenCL to truly flourish in the commercial marketplace. Why? CUDA is only useful on high-end Nvidia graphics hardware. OpenCL is useful everywhere, not just on modern graphics hardware. OpenCL will take advantage of just about any parallel device you can point it at, provided you write a driver for it. This doesn't just encompass GPUs, but multicore CPUs (X86, ARM A9, Cell, PowerPC, etc.), DSPs, audio processors and so on. CUDA might be the standard in the academic and business segments, but it will be slaughtered in the consumer segment because the market always gravitates to the hardware-independent standard. Think Glide vs. Direct3D, OpenAL vs. EAX, the list goes on really.

The fact is, though, that Nvidia knows this. That's why they're supporting it, too. They're not stupid enough to make the same mistake that 3dfx did with Glide, make it their only real focus and then tack on the hardware independent standard for feature completeness. Either way their predicament forces them into a bit of an own-goal scenario - if they lag behind in OpenCL to make CUDA seem better, chances are they'll just look bad at compute workloads. If they do well in supporting OpenCL, it'll make CUDA seem insignificant by comparison.

CUDA might even have another couple of years as the leader, but ultimately it will meet its fate at the hands of OpenCL or DirectCompute for the reasons I've listed above, it's inevitable. And of course few people have heard of OpenCL at the moment, it's still a new standard and implementations have just barely started to pop up where CUDA has been about for a couple of years.

 

[Hoodlum]

I'm not quite sure what your saying here... but even if we take the list of games supporting PhysX (around 100) its far far shorter than the public list of projects using CUDA (well over 600)

That in terms of scale CUDA has had a *massive* impact on gaming, it is extremely widespread. More so than its other uses; you cannot play games and not be consistently confronted with PhysX.

CUDA is taught in mainstream IT courses in more than 300 Unis worldwide and I know of atleast 2 with large scale deployments that are used beyond just learning - so I bet theres a good few more.

Do you have any details? I have to admit I have not gone out of my way to learn about every individual use of CUDA but I do have an interest in GPU computing - both personally and professionally. I am hearing *a lot* more about it now only because of Fermi (which could massively increase it's use in the scientific community - far easier to support). Unfortunately as it isn't out yet we can't trial it. We will be when it is released, maybe i'll post some benchmarks then too.

OpenCL will take advantage of just about any parallel device you can point it at, provided you write a driver for it. This doesn't just encompass GPUs, but multicore CPUs (X86, ARM A9, Cell, PowerPC, etc.), DSPs, audio processors and so on.

And this is the other reason we are using OpenCL and not CUDA. Being able to use the thousands of processors available to us in addition is not an "optional" feature. It is required.

 

[D.P.]

CUDA is too limited from a development ubiquity point of view compared to OpenCL to truly flourish in the commercial marketplace. Why? CUDA is only useful on high-end Nvidia graphics hardware. OpenCL is useful everywhere, not just on modern graphics hardware. OpenCL will take advantage of just about any parallel device you can point it at, provided you write a driver for it. This doesn't just encompass GPUs, but multicore CPUs (X86, ARM A9, Cell, PowerPC, etc.), DSPs, audio processors and so on. CUDA might be the standard in the academic and business segments, but it will be slaughtered in the consumer segment because the market always gravitates to the hardware-independent standard. Think Glide vs. Direct3D, OpenAL vs. EAX, the list goes on really.

The fact is, though, that Nvidia knows this. That's why they're supporting it, too. They're not stupid enough to make the same mistake that 3dfx did with Glide, make it their only real focus and then tack on the hardware independent standard for feature completeness. Either way their predicament forces them into a bit of an own-goal scenario - if they lag behind in OpenCL to make CUDA seem better, chances are they'll just look bad at compute workloads. If they do well in supporting OpenCL, it'll make CUDA seem insignificant by comparison.

CUDA might even have another couple of years as the leader, but ultimately it will meet its fate at the hands of OpenCL or DirectCompute for the reasons I've listed above, it's inevitable. And of course few people have heard of OpenCL at the moment, it's still a new standard and implementations have just barely started to pop up where CUDA has been about for a couple of years.

From your list your Missing D3D vs OpenGL,

what happened to OGL in the end? It lost to a proprietary, closed standard.

OpenCL is going the same way. CUDA is taking off in a big, big way. I know of several projects in my university using CUDA, and have read several publications which used CUDA, and been to several seminars where GPGPU was used on a Tesla platform.

 

[Parsley]

That in terms of scale CUDA has had a *massive* impact on gaming, it is extremely widespread. More so than its other uses; you cannot play games and not be consistently confronted with PhysX.

What that tacked on thing a very significant proportion of people don't care about? It isn't in as many games as you would think. The CUDA vs OpenCL thing isn't my issue, I honestly don't have much background info on either, but PhysX really isn't *massive*.

 

[Lightnix]

From your list your Missing D3D vs OpenGL,

what happened to OGL in the end? It lost to a proprietary, closed standard.

OpenCL is going the same way. CUDA is taking off in a big, big way. I know of several projects in my university using CUDA, and have read several publications which used CUDA, and been to several seminars where GPGPU was used on a Tesla platform.

The key point is hardware independence, this has nothing to do with operating system limitations or access to source code.

 

[Hoodlum]

From your list your Missing D3D vs OpenGL,

what happened to OGL in the end? It lost to a proprietary, closed standard.

Excellent example! OpenGL is still the standard outside of games, it was not developed for games. Direct X is a laughing stock outside of gaming (It doesn't scale and lacks features).

Example 1: Films
Example 2: 3d rendering
Example 3: Cad

What that tacked on thing a very significant proportion of people don't care about? It isn't in as many games as you would think. The CUDA vs OpenCL thing isn't my issue, I honestly don't have much background info on either, but PhysX really isn't *massive*.

But it is widely used there, you are exposed to it all the time. No where else is it that pervasive. That was my point.

 

[Rroff]

CUDA might even have another couple of years as the leader, but ultimately it will meet its fate at the hands of OpenCL or DirectCompute for the reasons I've listed above, it's inevitable. And of course few people have heard of OpenCL at the moment, it's still a new standard and implementations have just barely started to pop up where CUDA has been about for a couple of years.

DirectCompute with the broader - but still predictable range of targets - and MS clout behind it might overtake it.

That in terms of scale CUDA has had a *massive* impact on gaming, it is extremely widespread. More so than its other uses; you cannot play games and not be consistently confronted with PhysX.

You can use software PhysX on systems without CUDA - its only hardware PhysX that you have a dependancy on CUDA - and only around 4-5 games so far do that.

Do you have any details? I have to admit I have not gone out of my way to learn about every individual use of CUDA but I do have an interest in GPU computing - both personally and professionally. I am hearing *a lot* more about it now only because of Fermi (which could massively increase it's use in the scientific community - far easier to support). Unfortunately as it isn't out yet we can't trial it. We will be when it is released, maybe i'll post some benchmarks then too.

Most of the info is readily available on CudaZone.

 

[JimAroo]

I like both companies, they both make good cards. I just buy whatever offers best value for money with my budget.

 

[dangerstat]

Sorry I've been a bit busy guys hope it's all been non spam / troll filled in here

 

[deuse]

Did you you mean the CUDA HD video encoding plug in? Because if so they both provide a plug-in for GPU acceleration. I have no idea how they compare, though.

Yes and thank you for putting me right(I must sleep tonight)

http://www.guru3d.com/news/nvidia-finalizes-cuda-20-photoshop-plugin/

There are other plugins.

I do use Arcsoft TotalMedia Theatre Platinum now that does use Cuda very well.

 

[Hoodlum]

You can use software PhysX on systems without CUDA - its only hardware PhysX that you have a dependancy on CUDA - and only around 4-5 games so far do that.

I am aware you can use the software without CUDA but I have been assuming that Nvidia cards accelerate this. Does this mean they do not (in general)? Only a few very specific titles? That seems a bit.....underwhelming.

Most of the info is readily available on CudaZone.

Thanks i'll check it out.

Yes and thank you for putting me right(I must sleep tonight)

http://www.guru3d.com/news/nvidia-finalizes-cuda-20-photoshop-plugin/

There are other plugins.

I do use Arcsoft TotalMedia Theatre Platinum now that does use Cuda very well.

Np, I'd heard of that one briefly before I think

 

[Rroff]

So am I right in assuming almost all of those games using physX are not hardware accelerated with an Nvidia GPU? Only a few very specific titles? That seems a bit.....underwhelming.

The list of PhysX titles that use CUDA for all/some feature levels is less than 20 and from a quick glance only 4 of those titles actually need GPU acceleration for the highest physx level - the rest can also be run on CPU (which doesn't use CUDA) with minimal if any slowdown.

 

[Hoodlum]

The list of PhysX titles that use CUDA for all/some feature levels is less than 20 and from a quick glance only 4 of those titles actually need GPU acceleration for the highest physx level - the rest can also be run on CPU (which doesn't use CUDA) with minimal if any slowdown.

I too stand corrected

Really this just strengthens my view that OpenCL will obsolete CUDA then. Being able to use all CPU cores and newer GPUs from all vendors (at the same time) with one fell swoop seems like too much of a boon even for the home market to ignore. It will also run on both GPUs on a SLI / Crossfire setup.

 

[kylew]

Excellent example! OpenGL is still the standard outside of games, it was not developed for games. Direct X is a laughing stock outside of gaming (It doesn't scale and lacks features).

Example 1: Films
Example 2: 3d rendering
Example 3: Cad



But it is widely used there, you are exposed to it all the time. No where else is it that pervasive. That was my point.

3DS Max uses Direct X over OpenGL by default so it can't be that bad.