Nvidia's new mainstream GPU promises an impressive blend of features and performance at an affordable price.
We’ve seen a parade of GeForce 400 series GPUs march out of the TSMC fabs in Taiwan. While they’ve been priced differently and offer differing performance levels, they’ve all consisted of the same chip: the three-billion-transistor monster that was the original GF100. Each one—even the top-of-the-line GTX 480—has had varying numbers of functional units disabled in order to hit power, thermal, and price targets.
Nvidia’s GeForce GTX 460 brings a brand-new GPU, code-named GF104, to the table. This chip is specifically designed for a more mainstream audience, and it delivers an interesting blend of features consisting of DirectX 11-capable shader cores combined with enough texture and render back-end horsepower to easily manhandle previous generations of games. The GF104 will be available in two flavors: one with 24 raster outputs and 768MB of GDDR5 memory, and a slightly beefier model with 32 ROPs and 1GB of GDDR5.
EVGA will be among the manufacturers offering GeForce GTX 460 cards with 1GB frame buffers.
Both the more upscale of the two GeForce GTX 460 cards and the GeForce GTX 465 card have 1GB frame buffers, but the new GTX 460 clocks faster, is outfitted with faster memory, and boasts more texture units, while the older card offers 16 additional shader units.
It’s not quite as simple as having more shaders, though; the GTX 465 has one more computational block, which Nvidia refers to as an SM (streaming multiprocessor). In addition to shader cores, the SM houses L1 cache, texture cache, and the Polymorph Engine (essentially the hardware tessellator.)
* AMD shader cores and Nvidia CUDA cores are not directly comparable.
This block diagram compares the original GeForce GTX 480 to the detuned GeForce GTX 460.
This diagram illustrates the make-up of a streaming processor or SM.
So in theory, the GTX 465 might be better at tessellation, but the GTX 460 1GB should win with games heavy with rendering operations, particularly if you turn up AA. With that thought in mind, let’s look at the performance numbers with a pair of Nvidia reference cards. We ran the same benchmarks on a Radeon HD 5830.
It’s no surprise the 768MB GTX 460 fails to win any of the below benchmarks, because the slower memory comes into play with 4x AA enabled. While many of the games are playable at 1920x1200 with 4x AA, a few clearly aren’t. You wouldn’t want 4x anti-aliasing enabled in any modern first-person shooters, for example. A game like Far Cry 2, which was originally built on DX9 with DirectX 10 features added, fares much better, as do the two simulations.
Best scores are bolded. Our test bed is a 3.33GHz Core i7-975 Extreme Edition in an Asus P6X58D Premium motherboard with 6GB of DDR3/1333 and an 850TX Corsair PSU. The OS is 64-bit Windows Ultimate. All games are run at 1920x1200 with 4x AA.
AMD’s Radeon HD 5830 manages a couple of narrow victories, but it’s largely outclassed. Most of the Radeon HD 5830 cards on the market are street-priced slightly higher than the 1GB GeForce GTX 460, so AMD will need to adjust pricing to remain competitive.
The GeForce GTX 465 is the odd duck out. While it wins or ties several of the benchmarks, it’s also about $50 more expensive. While you could argue it offers more robust raw compute performance, it’s a little lacking in the texturing department relative to the other cards. Nvidia needs to either drop prices for these cards or simply discontinue the GPU.
Meanwhile, Asus, EVGA, Gigabyte, Palit, and others are shipping both versions of the GeForce GTX 460 cards, although SKUs with 768MB frame buffers are easiest to find.
The GeForce GTX 460—especially when paired with a 1GB frame buffer—looks to be a big winner. It could be this generation’s 8800 GT: a relatively affordable videocard that delivers great performance with a 1080p display. You could call it “Fermi Lite,” but it’s definitely a tasty brew.