Hey guys, so I'm looking to build a powerful development machine that will be able to last me 5 years. NVIDIA just released its GTX 980. It looks like it's more powerful than the GTX 780 ti AND much cheaper. My friend was saying that I should just get two GTX 670s and crossfire them (I'm new with what crossfire means). It'll be cheaper, and more powerful than a single GTX 980. He says that the only big difference is that the GTX 980 is good for 4k gaming. Now, when I hear 4k, I think of higher texel and fillrates. I'm not going to worry about 4k for another few years, and focus my efforts on Full HD resolution. Due to this, I should have 4 times as much bandwidth and processor cycles left over for my fullscreen projects I develop. This makes me think that as a developer, I can run more sophisticated postprocessor shaders in my own game engine as well as more postprocessors simultaneously.

Not only that, but 4GB of VRAM is more than enough texture memory for my projects. I'm not working on anything like Watchdogs are Uncharted 4 lol.

Also, I don't think Blender's Cycles renderer supports multiple GPUs yet. My renders would be quicker if I'm using a single GPU that has all the power instead of 2 when it only has access to 1 of them. I mean, I won't be running other applications requiring the GPU outside of PhotoShop for the most part. Getting 2 GPUs means I'll need a motherboard that supports 2 GPUs. Honestly, I think it'd be cool to add a second GTX 980 in a year or two, or something even more powerful for its dollar.

What do you guys think?

2 GPUs is a nightmare...

There are a small handful of advantages to dual GPUs but not worth the cost and headaches IMO.

Building a machine to last 5 years is also not something I recommend usually. It's cheaper and easier and lets you be more reactive to market changes to build a cheaper machine and upgrade it every year or two. You won't even need a powerful graphics setup in the near future if you're just starting out out an indie dev. Buy what you actually need, not what you guess you might be cool a few years from now.

Crossfire is AMDs name for it. nVidia's name for it is "SLI". Both names just mean two+ GPUs working together.

Drivers will try to automatically make existing games make use of multiple GPUs... but they won't be able to provide the theoretical 2x speedup.

To get the best performance, games need to specifically be written with multiple GPUs in mind (and use vendor-specific SDKs).. So in some games you might get great performance, but in others you might get the same performance as if you just had one GPU...

Personally, I would avoid it in a developer machine, unless you're specifically planning on implementing multi-GPU rendering in your own game.

If you're going to spend on a new machine... RAM and SDDs. Then CPU. Those will affect your day-to-day far more than a high-end GPU would. A $200 GPU is all you need until you have prod you need more.

My friend was saying that I should just get two GTX 670s and crossfire them (I'm new with what crossfire means)

Wrong answer. I am a strong believer that you ONLY use multi-GPU in two situations:

1) No single GPU can get the performance you need at a reasonable price

2) You already have one of the GPUs and are looking for a cheap upgrade

In a world that the 980 exists, the idea of combining two GTX 670s is a sick joke. (PS, Crossfire is AMD's branding of dual GPU. NVIDIA calls theirs SLI.)

The GTX 780 Ti is about a hundred bucks cheaper than the 980. It is, however, about a hundred bucks more expensive than the 970, which has a similar capability level and a newer feature set. I recently went through the same debate and went with the 970, though the 980 is decidedly more powerful, but will come at a steep price tag -- 570 USD at the very lowest from what I've seen. You can get a 970 for about $330.

I've never run a dual-GPU setup, but everything I've heard echoes what Promit said above. You COULD try putting together two 970s for only a slightly higher price than a single 980, but I don't think many applications properly make use of dual GPU setups.

I'm not so sure I subscribe to this idea. A new CPU means (in a couple years, like you said) a new socket, which means a new motherboard. Right now, this will likely also mean new RAM. That alone is like half the cost of a new PC. Thankfully right now, there aren't any other new interfaces on the horizon that I know of, so you won't have a situation like my last PC which had IDE->SATA adapters everywhere.

What's important is "riding the curves" of the new interfaces.There's a lot of intermingling going on when it comes to interfaces, and you can easily wind up having to replace a lot of stuff. As long as you're smart about where you spend (and that doesn't always necessarily mesh with what gives you the most bang for the buck), you don't have to worry about it.

I'm not so sure I subscribe to this idea. A new CPU means (in a couple years, like you said) a new socket, which means a new motherboard. Right now, this will likely also mean new RAM.

It depends on how you spec it initially. For example, my old home machine started with an i5-750 and 8 GB of memory. Adding another 8 GB was trivial a few years later, and I'm about to put an i7-860 into it as it moves into other duties (replaced by an X99 monster). Notice what happened here - the 860 was available at the time I built the machine, I just wasn't willing to pay the price for one. If I'd specced it with the high end chip to begin with, there would have been no upgrade prospects. As it is, I have to buy the CPU used. The upgrade path never changed in five years, and it only exists because I bought a lower end part to begin with.

I would say that if there's one thing you want to be open to upgrade later it's the GPU. Buying one of the better available CPUs and motherboards available today will last several years without problem, and the GPU is very easy to switch out when new GPUs with 2x the performance are available in a couple of years. Doing multi-GPU just to get there today is meaningless unless there's a specific current game you want to increase your FPS in for a particular detail setting, or if you're developing something already today that you target the top-of-the-line in two years from now with.

It's likely not pure performance that will be the most important graphics-aspect for most developers when new technologies arrive, but having all the latest features. A D3D9-class GPU today that is 2x faster than the latest D3D11 GPU would be nice for playing a D3D9-game, but not much use for a developer who wants to target D3D11. The same thing will happen again in less than 5 years.

The second thing to keep open for upgrade is RAM, where standard RAM sizes grow quite quickly and will probably continue to do so, but that should be trivially upgradeable in the coming years if you get DDR4 where both speed and size per module are likely to grow rapidly, 128 or 256 GB RAM won't be that expensive in a couple of years.

It also depends on what you aim to do with the pc. If you're gonna make the new pacman 3d it's a bit overkill (lol). Seriously, what are your plans?