TR Forums

I'm looking to do my first computer build this summer. I've been doing a lot of research, but as I've never even owned anything bigger than a laptop before, I don't really know anything. So, I'd like to ask a few dumb questions, if you don't mind.

1. I've read that the HDMI ports on newer graphics cards can output audio, due to a "audio controller" on-chip. Does this compute the audio directly on-chip, or does it copy it from a audio buffer from the motherboard's audio processor? (If that's not complete nonsense.) And does this affect other on-chip performance at all?

2. What features do graphics cards actually have? I read a lot of marketing mumbo-jumbo, and some pretty cool stuff about OpenCL and the like, but not terribly much that I've managed to understand.

3. What computing languages could I expect to be able to compile/run on a GPU? I could think of some highly parallel stuff I could come up with on say, Python. Is there a runtime compiler for Python that's GPU-based? Also, using my GPU for compute tasks, would that slow down Windows substantially?

4. Quick Sync is simply AMD's new VCE on an Intel IGP, right? Would Virtu's stuff still be useful in that light, or is that the only real feature of Sandy Bridge's IGP? And is that likely to change for Ivy? Are there any benefits to having an Intel IGP activated/running at all if I were already planning on discrete graphics?

5. And lastly, I don't have much of a library of computer games. So, what have you guys been enjoying recently? Indie or published games both welcome.

Thanks a lot for your responses!

There are no dumb questions! Just dumb people. And dumb answers (but hopefully this isn't one).

1. HDMI: In a newer graphics card with HDMI, there typically is a built in audio output device (see here for Nvidia cards, ATI is similar: ftp://download.nvidia.com/XFree86/gpu-h ... io_support)

2. When it comes to OpenCL and/or CUDA let me put it to you this way: If you don't already know exactly WHY you want those features before you even start shopping for the card, then ignore them as they are mostly
buzzwords. Despite the massive amount of hype that these terms generate on this website, I would say that < 1% of the people here ever use them in a serious way (i.e. for work) and only about 10% use them for hobby type activities like Folding at Home. Unless this describes you, they are pretty meaningless.

3. Computer Languages: The only real "languages" that a GPU speaks are shader languages, which look a lot like C, but are not in fact C. OpenCL and CUDA attempt to present a higher-level interface for languages like C and C++ for more general computational work that isn't specifically graphics. If you are interested in learnign this stuff, that's great but it's pretty esoteric (many full-time programmers who are very experienced never touch the stuff).

4. Quick Sync is simply AMD's new VCE on an Intel IGP, right? More like VCE is similar to QuickSync.... it's there for transcoding video formats in hardware. Believe it or not, standard GPUs (even very powerful ones like the 7XXX GPUs from AMD) are pretty lousy at doing transcoding. The media engines like QuickSynce and VCE do the hard work with a small amount of silicon that is dedicated to the task.

5. There's a lot of games out there. A quick & fun one that actually has excellent graphics is Oil Rush (http://oilrush-game.com/) It has excellent graphics and runs on Linux natively to boot.

1. The HDMI audio connection typically shows up as a second sound card. It is independent of the motherboard audio device.

2. If it is raw graphics performance you care about, look at online benchmarks. Beyond that, you have to decide whether additional features matter to you. Some people want massive multi-head configurations (Eyefinity). Other people might want to do dual cards in SLI/Crossfire. OpenCL (and its close relative CUDA) are methods for programming GPUs to do computations that would normally be handled on the CPU; unless you have a specific application that benefits from massive parallelism that can be coded in OpenCL/CUDA, there's no real benefit.

3. See #2. OpenCL and CUDA. Python has serious issues with multi-threaded operation, and can't even make proper use of a multi-core CPU, let alone a GPU. If you want to speed up your Python scripts, get a CPU with faster single-thread performance, or run multiple parallel Python scripts (if the parallelism of your application is coarse-grained enough to make this practical).

4. Dunno, haven't been following that.

5. I don't game much these days.

modulusshift wrote:

2. What features do graphics cards actually have? I read a lot of marketing mumbo-jumbo, and some pretty cool stuff about OpenCL and the like, but not terribly much that I've managed to understand.

I guess the most important one is display output, and it'll depend on your monitor. Third-party vendors can differ from the reference design in this case, offering (for example) 2x DVI-I + 1x DisplayPort outputs instead of a 2x miniDisplayPort + 1x mini-HDMI + 1x DVI-I on the reference card.

Third-party vendors also usually slap on their own coolers to the GPU PCB, instead of recycling the reference cooler. HIS/PowerColor has always sold some sort of liquid-cooled version of a graphics card, usually the top-dog AMD models. MSI meanwhile has this air cooler on some cards that spins its fans in reverse for a short time, supposedly for cleaning purposes.

4. Quick Sync is simply AMD's new VCE on an Intel IGP, right? Would Virtu's stuff still be useful in that light, or is that the only real feature of Sandy Bridge's IGP? And is that likely to change for Ivy? Are there any benefits to having an Intel IGP activated/running at all if I were already planning on discrete graphics?

For you to use Quick Sync, you'd have to use a quite limited number of applications for video conversion - CyberLink's MediaEspresso comes to mind. I haven't yet heard of a free utility that leverages the Quick Sync silicon; I can confirm that neither HandBrake nor AnyVideoConverter take advantage of it.

Ivy Bridge has more of the same on the Quick Sync end, except that it does the job faster than Sandy Bridge CPUs did - due to the faster HD Graphics 4000 IGP. AnandTech just published a review on that.

5. And lastly, I don't have much of a library of computer games. So, what have you guys been enjoying recently? Indie or published games both welcome.

Thanks a lot for your responses!

I'm relatively new to PC gaming, but I knew I always wanted to try Mass Effect, which I'm in the middle of now. Rather old, but rather good too. I'm putting off ME2 and ME3 until later, no hurries.

chuckula wrote:

There are no dumb questions! Just dumb people.

I think it's "there are no dumb questions, just dumb people asking questions".

modulusshift wrote:

2. What features do graphics cards actually have? I read a lot of marketing mumbo-jumbo, and some pretty cool stuff about OpenCL and the like, but not terribly much that I've managed to understand.

3. What computing languages could I expect to be able to compile/run on a GPU? I could think of some highly parallel stuff I could come up with on say, Python. Is there a runtime compiler for Python that's GPU-based? Also, using my GPU for compute tasks, would that slow down Windows substantially?

5. And lastly, I don't have much of a library of computer games. So, what have you guys been enjoying recently? Indie or published games both welcome.

2/3) Nvidia has better Linux drivers and CUDA. AMD has Eyefinity and workable Linux drivers. That's all I've got for features.

I was thinking about this last night. I just bought a new ThinkPad with the Optimus feature, and I was wondering what I could do with the Nvidia card since I'm not really going to stress the graphics subsystem.

Nvidia is the only vendor that supports both OpenCL and CUDA, since CUDA is a Nvidia specific technology. It's easy enough to swap a video card out of a desktop, so I would say get a Nvidia card if you want to mess around with both of them.

OpenCL is more cross platform, and as such, it's more abstracted then CUDA. It's like OpenGL except for computations. It can run on a regular CPU, x86 or ARM, or on a GPU from either AMD or Nvidia, but I'm not sure how portable it is. Each hardware vendor has it's own stack, and I'm not sure how that effects things. As far as languages goes, you'll need to use one that's supported by GCC 4.1+, MinGW 4.4+ (GCC for Windows), Visual Studio 2008-2010 Pro, or ICC 11.x, per the AMD documentation.

CUDA is more specific then OpenCL. It's for Nvidia GPUs, which could make it more portable since there is one toolkit from one vendor, and since it is more focused, it has low level and high level APIs. It's been around longer, so there is more support for it. Wikipedia lists Python, Lua, Ruby, Java, Perl, Fortran, Haskell, MATLAB, IDL, and Mathematica, in addition to C/C++, as programming languages that can be used with CUDA.

AMD: Getting Started with OpenCL
http://developer.amd.com/zones/openclzo ... arted.aspx

AMD APP SDK: System Requirements & Driver Compatibility
http://developer.amd.com/sdks/AMDAPPSDK ... ility.aspx

Nvidia: GPU Computing Documentation
http://developer.nvidia.com/nvidia-gpu- ... umentation

5) KDE has a really nice version of Mahjong, and I'm still trying to consistently win at Hearts on Windows. I'm an enterprise geek, and I don't really have time to game. The last game to tempt me was Forza Motorsports 4.