Best Gaming Video Cards for the Money

Tom's Hardware has decided to take a step back with their latest video card review. Instead of wowing their audience with in-depth benchmarks they head right for what someone reading a review really wants, an opinion of the best bang for the buck. From the article: "So if you don't have the time to research the benchmarks, or if you don't feel confident enough in your ability to make the right decision, fear not. We offer a simple list of the best gaming cards on offer for the money."

Comments

Summary

[LoverOfJoy]

Best PCIe Card For Under $100 - Radeon X800 GTO 128MB

Geforce 7300 GT GDDR3 (second choice/tie?)

Best PCIe Card For ~$140 - Geforce 7600 GT

Best PCIe Card For ~$200 - Radeon X1900 GT

Best PCIe Card For ~$250 - Radeon X1900 XT 256MB

Best PCIe Card For ~$340 - Geforce 7900 GTX

Radeon X1900 XTX (second pick)

Best PCIe Card For ~$500 - Geforce 7950 GX2

Best AGP Card For Under $100 - Radeon X700

Geforce 6600

Best AGP Card For ~$125: 3 Way Tie - Radeon X1600

Geforce 6600 GT

Radeon X800 GTO 128MB

Best AGP Card For ~$130 - Geforce 7600 GS

Radeon X1650 PRO

Best AGP Card For ~$175 - Geforce 7600 GT

Best AGP Card For +$200: None (Honorable Mention: Gainward Geforce 7800 GS+ silent 512)

It looked like nearly every card one at whatever price they sell at. A category for $125 (a three way tie there) and a category for $130? It's ridiculous. 7 pages worth.

Re:Summary

[BrookHarty]

What they dont say, an Nvidia 7800 GT is twice as fast as an ATI 800, 140 bux or 100 bux, 40 bux buys a lot more power.

I like toms hardware video card graphs to help quickly show how a card stacks up.
http://www.tomshardware.com/2005/12/02/vga_charts_ viii/page16.html

Re:Summary

[Damon Cleeve]

I wrote the article, and I can tell you I didn't sit on the fence unless there were legitimate close performance runs. The problem is, with different architectures some cards will win some benches and other cards will win others, it's difficult to difinitively say one is better than the other when they trade blows based on which game you're playing. The reason there's 7 pages of article is because there's so damn many categories. And for the record, I admit the AGP section I submitted was screwed up a bit. The $130 AGP section should have been omitted, that was a leftover from an old version... I've been doing this article on the forums for the past year and a bit. The $130 AGP category should have been the 7600 GS, X1650 PRO, and X800 GTO... which all do perform very closely.

Re:Video cards

[pe1chl]

Have you ever tried playing HD video on a 40-60 dollar card?
Those cards often cannot even drive monitors at 1920x1200 or 1920x1080, especially over DVI.

I had an nvidia FX5200 before, but it cannot play video at 1920x1200 fullscreen. Something goes haywire, probably because it is overloaded.

Re:sarcasm

[grammar fascist]

I'm running Xorg-air + Compiz on an Intel 945GM, and it's very snappy. (Of all the plugins, only Blur makes it crank.) It doesn't even drain the battery more than regular 2D compositing.

No, I can't play Quake IV on it, but I do have wobbly windows that stick to each other. :D Quake 3 and its generation all run fine.

Take that, integrated graphics naysayers.

Re:I like this one

[julesh]

I wonder how much longer the CPU will be able to keep up with increases in GPU speed. Beyond a certain point, returns will be small or non-existant.

Look at it this way: the current 'hot' CPU, the Core 2 Duo, has a bus connection that allows it to transfer 1066 Mwords of data per second. Typical applications require a complete refresh of vertex buffer data for each frame. Even for a really well optimised application that runs mostly out of cache, the CPU's likely to need to hit system memory several times for each vertex it outputs, so it's probably putting at most 400 Mwords of vertex data per second into the GPU's buffers (at 100fps, that's around 2 million vertices per frame, which is quite a lot).

The card quoted has 8 vertex shaders running at 650MHz, so it can already afford roughly 25 cycles per vertex, which is probably more than enough to perform any reasonable transformation on those vertices.

But then it's the pixel shaders and texturizers that get really stressed in most applications. This card has 24 of each. Per frame, that allows the same application 156 million pixel shader cycles and the same number of texturizer cycles. The highest resolution monitor I'm aware of has a max resolution of 2560x1600. That's roughly 4 megapixels, meaning that the shaders get 39 cycles per pixel. Given that these beasts are vector processors (i.e. they can process R, G, B, & A in a single cycle), that's just about enough to perform any realistic transformation on the pixels.

Yes, I think there are applications for faster GPUs. And certainly, improving the speed of the memory attached to the GPUs will continue yielding improvements for a while yet -- there's simply no way 1600 MWord/s memory access speeds can keep up with data transfer requirements to all of the 72x650MHz pipelines on this card. But I'm not sure how many generations of card we'll see before they match the performance of even the most demanding application current generation CPUs are capable of instructing them to perform.

And for gaming applications: there's already enough power in these GPUs to process as many vertices as the CPU can provide in any exotic way you can find a realistic need for, and produce high-resolution textured, realistically lit, bump-mapped, fogged, rasterized output overlayed with transparency over static controls, HUDs and background images at the highest resolution supported by 99% of monitors.

What more do you want?

Re:sarcasm

[damiangerous]

122 fps.

On a slightly lower end machine (with the 945GM video), 83 fps

Re:we require more minerals

[Breakfast Pants]

It is quite clear to me from your post that you are running linux and you are running Google Earth with software GL. Check to see if drivers are available for your on-board chipset--if it is intel they likely are.