Date: Monday , July 12, 2010
Today NVIDIA is launching their newest Fermi family GPU, the GeForce GTX 460. The first two Fermi products were released on April 12th of 2010, and included the GeForce GTX 480 flagship and GeForce GTX 470. On May 31st, NVIDIA then launched the GeForce GTX 465 which was met by less enthusiasm and debated merit. Now we have the true gaming sweet spot contender being launched, the GeForce GTX 460 at $199-$229.
During our briefings, NVIDIA repeatedly referred to the GeForce GTX 460 (which bears the alternate moniker "GF104") as "A New Class of Fermi." It seems that NVIDIA is suggesting to us that they are trying to set the GTX 460 apart from the GTX 480/470/465, and give it a life of its own in several key categories. NVIDIA is setting it apart in performance, price, power, noise and heat. We will of course test all of this, and we think you will be pleased with our findings.
NVIDIA likened the GTX 480 to the "Tank" character archetype found in many MMORPGs. The GTX 480 is big and expensive, representing some of the most powerful characters around. The GTX 460, according to NVIDIA, is more like a "Hunter" class. It is smaller, leaner, less complicated, and more approachable, which is a good thing when we’re talking about lower-cost video cards.
Compared to the GTX 480’s 480 CUDA cores and 15 polymorph (tessellation) engines, the GTX 460 packs 336 cores and 7 tessellators. The GTX 480 features 60 texture mapping units to the GTX 460s 56, while the GTX 460 is 2.25" shorter, uses 90W less power at full load, and costs less than half as much.
NVIDIA designed the GTX 460 to meet the needs of those gamers operating in the single largest price domain, right around $200 USD. According to Steam’s May 2010 hardware survey, 31% of its customers use a video card in the $200 range.
NVIDIA recognizes that geometry performance scaling between the GeForce 9800 GT and the GTX 260 was not ideal, so they sought to improve that scaling with the GTX 460, allegedly bringing a performance increase of 4.5X, while scaling shader performance by about 3X. NVIDIA also believes that their performance scaling throughout their DX11 lineup is between 1.5X and 2X that of AMD’s lineup. During our previous testing, we have found performance scaling to be less dramatic than that, with some games running better on the Radeon HD 5800 counterpart to NVIDIA’s GeForce GTX 400 series.
NVIDIA’s reference specification for the GeForce GTX 460 calls for the Graphics cores to be clocked at 675MHz, while the CUDA Processors are bound to that clock rate with a 2:1 ratio, coming out at 1350MHz.
There are two different versions of the GTX 460. One has a 256-bit memory bus with 1024MB of GDDR5 attached, and the other has a 192-bit memory bus with 768MB of memory attached.
NVIDIA did this so that they could tap the $199 market with one part, and the $229 market with a closely related part. The 768MB version is designed to consume a maximum of 150W, while the 1GB version consumes 160W.
The GeForce GTX 460 video cards are short, coming out at 8.25" long, a fill 2.25" shorter than the GeForce GTX 480 and GTX 470. It retains the dual-slot cooling fan and heat-sink and has two Dual-Link DVI outputs and a single mini-HDMI output port. We asked NVIDIA if a mini-HDMI to standard-HDMI dongle was to be included with these video cards, and they said that choice was up to the partner, so we can expect some models to ship with an adaptor and some without.
As for MSRP, NVIDIA suggests a price of $229 for the 1GB model, and $199 for the 768MB model, which positions these video cards squarely against the AMD ATI Radeon HD 5830 video cards.
NVIDIA repeatedly stressed to us that the GTX 460 was made with overclockers in mind, so they are designing these video cards to have significant overclocking headroom. Though the reference clock rate is 675MHz, they expect users to be able to overclock upwards of 800MHz.
They also say that the GTX 460 is extremely quiet. They told us that their testing engineers reported that they were generally unable to hear the GTX 460 over the sound of other fans in their testing systems.
At the beginning of this evaluation, we received two reference cards from NVIDIA. We received one of each version of the GTX 460: a 1GB version and a 768MB version. Since they are both reference cards, they both come with the GPU clocked at 675MHz, the CUDA cores clocked at 1.35GHz, and the memory clocked at 3.6GHz.
From the front, these video cards look identical. In fact, when we took them out of the box in which they were shipped, the only way we could tell them apart was that one of them has an NVIDIA logo sticker on the fan hub, and the other does not. In the first photo above, the GTX 460 768MB is on the bottom, and the 1GB version is on the top. On the back side, the 768MB model had significantly more stickers and looked clean and new. The 1GB version has just a few stickers, and it looks like it has been handled quite thoroughly.
The GeForce GTX 460 looks like a very small version of the GTX 470. It has NVIDIA’s standard high-gloss black plastic heat-sink shroud complete with NVIDIA logos, and a large 80mm fan centered on the shroud. The video card features two 6-pin auxiliary power supply connectors and a single SLI bridge. Tri-SLI is not supported on the GTX 460, so there is some limits in place here with these more budget cards. On the business end, there are two Dual-Link DVI connectors and a single mini-HDMI port.
Looking closely at the back of these video cards, there are a few small differences. First, the 1GB version says "Made in USA", while the 768MB says "PCB made in Hong Kong." This is probably just a difference in these reference cards, as we believe that the retail versions will be made in the same facility. Second the 768MB is missing some surface mount components on the back of the PCB, to the left of the GPU bracket, where the memory would be if it was a 1GB version.
Here are some higher quality product shots provided by NVIDIA.