GeForce GTX 480 architecture features
At the end of 2006, NVIDIA launched the eighth generation of its GeForce family video cards. The flagship of the line was the powerful NVIDIA GeForce 8800 GTX accelerator at that time, which was significantly ahead in performance and technology of the top AMD model of those times - the Radeon X1900 XTX. In order to successfully compete with NVIDIA, AMD needed to create a new top-end accelerator, but due to a series of technological problems, the new product, the Radeon HD 2900 XT, was released about six months later than expected.
A similar situation occurred at the end of 2009 with NVIDIA. With the release of the first accelerator of the AMD Radeon HD 5xxx family - Radeon HD 5870, NVIDIA lost its “leadership title” among single-chip Hi-End solutions, and then, as the line of AMD graphics accelerators expanded, it began to lose ground in almost all segments of the desktop market graphics.
And now, six months later, NVIDIA was finally able to start producing a sufficient number of graphics chips of the new architecture. On March 26, 2010, the latest NVIDIA GeForce GTX 470 and GeForce GTX 480 graphics accelerators were announced.
Contrary to expectations, the senior accelerator of the family - NVIDIA GeForce GTX 480, received not 512 CUDA cores, as previously announced, but all 480. In addition, changes affected texture units, their number is 60, instead of the previously announced 64. But all 48 units ROP and the promised memory bus width of 384 bits remained in place.
16 blocks of the same type (SM) are clearly visible, occupying two-thirds of the core. The input/output interfaces are naturally located along the perimeter of the crystal. And this is what the already packaged GF100 looks like:
The GeForce GTX 480 is significantly superior to the previous top model, the GeForce GTX 285, in almost all characteristics. It is especially worth noting twice the number of CUDA cores, which will undoubtedly have a positive effect on the speed of execution of complex shaders. Also, the new NVIDIA accelerators received the long-awaited support for DirectX 11.
In addition, the GeForce GTX 470/480 supports a technology called NVIDIA Surround, similar to ATI Eyefinity. NVIDIA Surround allows you to use three monitors simultaneously as a single workspace. However, in implementing their version of the technology for complete immersion in the virtual world, the Californians decided to go even further and created a “technological mix” of NVIDIA 3D Vision and NVIDIA Surround, which is called NVIDIA 3D Vision Surround. The essence of this “cocktail” is that you can simultaneously use three monitors and 3D Vision glasses to create the maximum effect of presence.
The GeForce GTX 480 uses a 26,7 cm (10,5") PCB, about a centimeter shorter than the Radeon HD 5870. Additional power (in addition to the PCI Express bus) requires one six-pin and one eight-pin plug. Nvidia says that the board has a TDP of 250 W - significantly less than the Radeon HD 5970, which barely fits under the 300 W ceiling set by the PCI-SIG group - and Nvidia recommends a power supply of 600 W or higher.
But the story doesn't end with the card's additional power supply. Although the GeForce GTX 480 seems less "gluttonous" than AMD's flagship according to its specifications, there were clearly problems with cooling the video card. The card's cooler, in our opinion, is the most aggressive of all the reference design models we've seen before. The radiator removes heat from the surface GPU and memory chips. Four heat pipes transfer heat to an array of aluminum fins, and a special fan blows air through the casing and fins, eventually exhausting it out the back of the graphics card.
The hottest element of the GeForce GTX 480 is the graphics processor. To cool it, a radiator with five heat pipes is used, made using direct contact technology. All five tubes, through a thin layer of thermal paste, come into contact with the metal cover that protects the graphics core.
Supply system GPU uses six phases and is based on the CHL8266 PWM controller. Unlike the power elements produced by Volterra, which are assembled in a single case, the power elements in the GeForce GTX 480 power subsystem are made according to a discrete scheme. Each power phase has a transistor (one in the upper arm and two in the lower). This approach allows for better heat dissipation from the power subsystem elements.
Characteristics of NVIDIA GeForce GTX 480
| Name | GTX 480 |
| Core | GF100 |
| Process technology (µm) | 0.040 |
| Transistors (millions) | 3000 |
| Core frequency | 700 |
| Memory operating frequency (DDR) | 3696 |
| Bus and memory type | GDDR5 384-bit |
| Bandwidth (Gb/s) | 177 |
| Unified shader units | 480 |
| Unified shader unit frequency | 1400 |
| TMU on conveyor | 60 |
| ROP | 48 |
| Shader Model | 5.0 |
| Fill Rate (Mpix/s) | 29500 |
| Fill Rate (Mtex/s) | 45000 |
| DirectX | 11.0 |
| Memory Capacity | 1536 |
| Interface | PCIe 2.0 |
Tests demonstrate the confident superiority of the GeForce GTX 480 over the Radeon HD 5870 (in most benchmarks the card occupies an intermediate position between the HD 5870 and HD 5970), at the same time, the older Cypress is minimally superior to the GeForce GTX 470. Of course, Fermi deserves attention, but about the comfortable using these cards at 60+ dBA noise and at 90+ degrees on the core can be forgotten. Those who definitely want to see NVIDIA GeForce in their system should probably wait for the “colder” revisions of the GeForce GTX 480 and GTX 470. But whatever one may say, the GeForce GTX 480 rightfully receives the performance crown for single-chip solutions and competitors will be able to take it away from it not soon yet.
Dirt 2 DirectX 11
