Nvidia GeForce GTX 460: The Fermi We Were Waiting For

GeForce GTX 460 At 1 GB And 768 MB

Editor's Note: As part of the GeForce GTX 460 launch, we have 20 Just Cause 2 keys for Steam. Make sure you read through to the last page of this review for contest details and the link to enter!

“Honestly, I’m not sure what Nvidia was thinking with this one. Surely, its competitive analysis team ran these very same benchmarks and found the GeForce GTX 465 and Radeon HD 5830 trading blows. Surely, the same group of folks hopped online and saw Radeon HD 5830s selling for $220, going as low as $199 with rebates. How, then, did they decide that $279 was a good starting point for suggested pricing?”

That was the conclusion I reached after spending a week with Nvidia’s GeForce GTX 465. I was right on the money, too.

Nvidia is hammering that paragraph home for me today with a pair of new cards that achieve what the 465 couldn’t. The GeForce GTX 460 1 GB and the GeForce GTX 460 768 MB both offer much better value.

After seeing three cards (GTX 480, 470, and 465) center on the GF100 graphics processor, we finally have a true derivative part called GF104. Nvidia could have very well cut GF100 in half for the GeForce GTX 460 launch, yielding a part with 256 shader cores, 32 texture units, and a 192-bit memory bus. It would have performed well enough, and it would have been an evolutionary step down from the GTX 465 sporting 352 shaders, 44 texture units, and a 256-bit bus. Such a part would have had to go up against AMD’s Radeon HD 5770 at best, though.

Instead, we’re looking at a re-designed chip that employs the Fermi architecture, but sports a different arrangement of resources and about two-thirds of GF100’s complexity—making it a smaller, cooler-running, and believe it or not, more functional (that’s right—this new GPU includes an updated video processor capable of bitstreaming Dolby TrueHD and DTS-HD Master Audio soundtracks from Blu-ray movies). This is a second, much more potent stab at the Radeon HD 5830.

Nvidia’s GF104 Graphics Processor

As you’ll see in the benchmarks, GeForce GTX 460 is faster than GeForce GTX 465 in a great number of tests. But how is this possible from a GPU with fewer shader cores—a GPU designed to be more mainstream than the three billion transistor GF100?

It’s all due to a rearrangement of resources.

On a macro level, GF104 still centers on the concept of Graphics Processing Clusters (GPCs), each containing four Streaming Multiprocessors (SMs).

Things start changing within the SM, though. Instead of the GF100’s 32 CUDA cores per SM, GF104 wields 48 cores per SM. Keeping these more complex SMs fed with information necessitates higher instruction throughput, so we see another enhancement: taking GF100’s two dispatch units per SM to GF104’s four. Similarly, each SM now boasts eight texture units (instead of four).

In the simplest terms possible, this is a wider GPU than GF100. The result is better performance than a scaled-down GF100 in the types of apps that most people play today.

Complete GF104

Now, let’s do some quick math to determine what we’re working with here. A complete GF104 (without any resources disabled), employs two GPCs. Given four SMs per GPC, you’re looking at eight SMs. With the knowledge that there are 48 cores per SM, GF104 can leverage up to 384 CUDA cores, 64 texture units, and eight PolyMorph engines, which Nvidia uses to scale geometry performance.

Complete GF100

The chip’s back-end is a bit different, too. A complete GF100 offers six ROP partition units independent of the GPCs, each capable of outputting eight 32-bit integer pixels per clock (totaling 48). All six partitions are also associated with a 64-bit memory path, yielding an aggregate 384-bit bus. GF104 gets a maximum of four partitions, yielding up to 32 pixels per clock and a 256-bit bus.

Pretty impressive for a chip that’s still manufactured at 40 nm and is comprised of fewer than two billion transistors, right? Though perhaps not as impressive as what AMD has done with its two billion-transistor Cypress—we’ll have to check out the benchmarks for more there.

Chris Angelini
Chris Angelini is an Editor Emeritus at Tom's Hardware US. He edits hardware reviews and covers high-profile CPU and GPU launches.