Today here I am going to provide you the most suitable power supply for GTX 480 and 470 card. So if you are going to buy a Fermi or GF100 series card to build up your own processor then you must also need to concentrate on the power supply you are using. Work is not finished just by adding the faster cpu and high rams. The fuel of the system also plays a major role for proper functioning of the entire system. A low or average can create serious overheating issue on your system destroying your hardware on Overclocking or SLI mode.
The GF100, the first incarnation of the architecture Fermi, has already been the subject of much discussion. We had already proposed two items that are returned in details in this architecture. Manufactured at TSMC in 40 nanometers, the GF100 includes over 3 billion transistors, NVIDIA is not more specific on this figure. With an area of 529 mm ² of the photos published a few days ago that (it is protected by a heat spreader), the GF100 would be a little smaller than the GT200, which was 576 mm ². This is still huge compared to 334 mm ² Cypress AMD and directly influences the production costs for Nvidia. Certainly, it fortunately did not directly affect the final price, but it is clear that commercially, Nvidia will be less comfortable AMD, if he wants to make money with this GPU, or at least not losing too much because the segment of ultra high-end graphics cards to play is primarily intended to promote the brand. Nvidia has also profitable course architecture on other segments with Quadro cards and cards Tesla.
Nvidia announced two versions of GF100. In addition to the GeForce GTX 480 and the GeForce GTX 470, which will be tested shortly, makes its appearance. It uses a GF100 neutered a little more, with 14 SMs and 16 functional memory bus reduced to 320 bits instead of 384 bits. AMD has a large advantage in terms of computing power, but based on the behavior of vector units, less effective. The Radeon HD 5800 also dominate at the fill rate, texturing etc. Note about fill rate that the GeForce GTX 480 and 470 are limited to the level of RIP to 32 pixels per cycle, despite the presence of 48 and 40 ROPs. Finally you will notice the relatively low frequency memories of GDDR5.
GTX 480
The GeForce GTX 480 uses slightly different design from what we saw in the past with Nvidia. Thus an opening in the hull causes the top of the radiator to the air. This upper part is closed, however, so that the hot air passing through it is properly ejected outside the system. The basic system allows for five heat pipes in direct contact with the GPU, rather to his heat spreader.The format of the PCB is identical to that of the GeForce GTX 285 and Radeon HD 5870 reference is 27cm. These however have a somewhat longer view of their hull protruding from the PCB. The card occupies two slots and of course takes a little extra space on its top for heat pipes that stick out. They do, however, no more space than the 8-pin power connectors and 6-pin required.
The card offers two dual-link DVI outputs and a mini HDMI output, which saves a little space to allow the full width of the grid to evacuate hot air. Note that only two of these outputs can be used simultaneously. Although clocked at 924 MHz only (1848 MHz for sending data), the GDDR5 memory, Samsung signed, here is certified to 1250 MHz. While Overclocking this card can earn 15-20% more performance and is very feasible using Afterburner.
The default clock speed for the GeForce GTX 480 is 700 MHz for the GPU , 1400 MHz shader and 3700 MHz memory. With additional tension, it is able to grow to 850 MHz GPU and 1700 MHz shader. The price for this result is not absurd, because the card goes up to 90 C °, the fan is 100% and power consumption to 300 watts.
Geforce GTX 470
Compared to the GeForce GTX 480, the design of the GeForce GTX 470 is more traditional and does not show the top of the radiator. However, it has also two small perforations in the PCB that allow the turbine to be supplied with air by the two sides of the card. Its PCB is also shorter with a length identical to that of the Radeon HD 4800 reference, or about 24.5 cm. It provides a connection similar to that of the GeForce GTX 480: two dual-link DVI outputs and an HDMI 1.3a. Two 6-pin power connectors are required.
The GeForce GTX 470 Super Overclock Edition is a reworked on several points of the GeForce GTX 470. We are entitled to an architecture of power twice as large as the reference model with 12-phase power supply for the GPU and the VRAM 2 and using high quality components (Ultra Durable + VGA). As for technical features, the GTX 470 aboard SOC 448 stream processors, 1280 MB of GDDR5 memory and a 320-bit bus. The operating frequency is 700 MHz for the GPU, 1400 MHz for the shaders and 837 MHz for memory. So we overclock the GPU and original shaders (607 MHz and 1215 MHz for GeForce GTX 470 Classic).
SLI
The SLI system is hot, very hot. At 110 ° C for the GPU hotter, it is even dangerous, yet this time, Nvidia significantly lower frequencies. When we first load test, we are to finish at 1,500 fps to 800 fps! Note that the system was, moreover, not stable. Clearly a case such as the Antec Sonata 3 is not suitable to receive two monsters too greedy. To use the GeForce GTX 480 in SLI will require a more spacious and take care of feeding them specifically with fresh air, or use the water-cooling. Such a system is to be reserved for advanced users. The texturing units of GF100 show here the same performance as the GT200, with the exception of textures that contain the data depth (D32F) whose access gains in performance. You will notice that the GeForce are not able to handle RGB9E5 textures at full speed, unlike the Radeon.
The GeForce GTX 480 and 470 are awesome and here the Radeon level of effectiveness against the rising level of ant aliasing. Despite its 48 ROPs, the GeForce GTX 480 is limited to 32 pixels per cycle or less. With a higher frequency, it is usually a bit strangely behind the GeForce GTX 285, 2 exceptions. The first is a surface with single-channel 32-bit where GF100 more effective. The second concerns the blending of 128 bits, very heavy in memory bandwidth, which is also more efficient.
Bookmarks