GeForce GTX 280 Review, and Why Everyone Should Care

Click for live photos gallery,

NVIDIA is releasing two new high-end graphics cards called GeForce GTX 280 and GTX 260 (official photo gallery), both are part of the GTX 200 Series. The timing is perfect: it will allow the NVIDIA to sweep contracts for the second half of the year. In the GPU world, performance is the name of the game, and this is not a small update. Engineers have improved many aspects of the GPU, increasing efficiency and performance wherever they could, in addition of adding raw processing power by integrating more 240 stream processors* (versus 128 previously) along with additional internal memory and to optimize performance. Let's cut to the chase: here is the minimum that you should know about the GTX 200 Series:

Highlights

• The fastest single-GPU card
• 240 Stream processors (1.3Ghz, GTX 280)
• 1GB of Memory
• 60% less idle power
• N-SLI supported (limited only by the motherboard and chassis)
• 2 Dual-Link DVI (think two 30" displays)
• Requires a 6-pin and an 8-pin PCI-E power connectors
• Not DX 10.1

The Board

Not surprisingly, the board is big and if you look at our photo gallery, you will see that it barely fits in my Cooler Master Wave case. You will need a good power supply ("PSU", $100-$300 depending on models) to power it. I chose a Cooler Master 750W, but what's important is that you have one 8-pin PCI-E and one 6-pin PCI-E power connectors. Although a 750W PSU would be a better choice, it should run with a 500W PSU (depending on what else is in the PC) but in any case, get one that has a 120mm (or larger) fan, because it is quieter. In my setup, the only issue that I have is that the PCI-E connectors are sticking out, making it hard to close the case.

Graphics Performance

Yes, it outperforms the Radeon HD 3870 X2 by a good margin (about 2x in 3DMark Vantage and 1.5x in Unreal Tournament 3) in high-resolution with the best image quality. That is because in these conditions, the GTX 280 can use its massive bandwidth. However, the difference could be less dramatic when using lower-resolutions or lesser quality settings (filtering, anti-aliasing...). I'll spare you the 12-page review filled with benchmarks, (but we'll be adding pointers to some at the bottom of this review later).

The point is: it's the fastest single-GPU solution out there.

HPC: Badaboom Review (H.264 Encoder)

BadaBOOM from Elemental Technologies is an Windows Vista application that converts a video file from one of many formats to H.264, using the stream processors of the GPU to perform High-Performance Computing (HPC). BadaBOOM converts a media file (audio, video) an order of magnitude faster than a CPU-driven converter. We have seen it convert to H.264 at about 120 frames per seconds using a good to very good quality setting. Anyone who has compressed video files knows that this is uber-fast!

High-Performance Computing isn't new: this is the second generation of GPUs with a computing-friendly architecture (the first one being the GeForce 8 Series). However, this is the first time that anyone with an MP3 player, a camescope or a PC TV recorder could find a GPU to be really useful (for something other than games or scientific apps). Needless to say that YouTube and the likes should jump on the bandwagon. Designers will be pleased to learn that Adobe is already working on a GPU accelerated version of Premiere. Hopefully Photoshop will get the HPC treatment too.

Conclusion

The GeForce GTX 200 Series is technically impressive. Today, it simply has the best single-GPU performance and the best architectural efficiency. However, the real star is the fact that High-Performance Computing is going mainstream. In time, even low-end graphics processors will have enough processing power to outperform a high-end, many-core CPU in specific tasks such as video compression. Heck, some people are even looking at performing virus pattern matching using the GPU! In cold cash, this means that a $50 or $100 upgrade towards a GPU could yield a gigantic boost in perceived performance, making the GPU even more relevant to have a computer. HPC works on all GeForce 8 GPUs (and beyond).

As always, if you have some questions, please drop a comment, I'll try to reply while I have the board installed.

*Note that NVIDIA, Intel and AMD all have different definition for what "cores" and “stream processors” are, so they are not directly comparable.

     Become a fan!