From Daily Tech

The GTX 580 is the most powerful single-GPU card solution, according to early testing. It delivers approximately a 10 percent bump from the GTX 480, but costs about 20 percent more. 
Higher retail prices mean that you spend about 20 percent more to get a 10 percent performance boost

Today NVIDIA officially launched the GTX 580, the first card in its Geforce 500 Series.  Like AMD's Radeon 6000 series, the Geforce 500 series isn't a major architecture design and is still produced on the same 40 nm process by Taiwan Semiconductor Manufacturing Company, Ltd.

Despite the fact that the card itself isn't exactly earth-shaking, this launch is clearly a big deal for NVIDIA, as it represents the company closing the gap on release time with AMD.  And the card reportedly picks up right where NVIDIA left off, boosting the company's high end performance even higher.  But how big an impact will it have?  Continue ahead for our thoughts.

The Card

The Geforce GTX 580 is absolutely a performance beast.  In some regards, it is basically an overclocked GTX 480.  But it also bumps the Shader Processor (aka "CUDA core") from 480 (with the GTX 480) to 512.  It also tacks on 4 extra "Special Function Units", taking the total to 64.

The die size has actually shrunk slightly to 520 mm^2, while the transistor count stayed steady at 3 billion.

Clock speeds have been bumped up across the board.  The standard core clock jumps from 700 MHz to 772 MHz.  The shade clock is bumped from 1.401 GHz to 1.544 GHz.  And the GDDR5 memory clock is pushed from 3.696 GHz to 4.008 GHz.

Despite the clock speed increases, the TDP is down from 250 watts to 244 watts.  Impressive.

NVIDIA dubs the new architecture GF110, but essential this appears to be very similar to the GF100 architecture in the GTX 480.

Other than the clocks and number of processor units, pretty much everything else stays the same.  The memory bus is still 384-bits wide and there's still roughly 1.5 GB of it.

The clock increases yield roughly a 4.1 percent increase in pixel fill rate, a 17.6 percent increase in texture fill rate, and an 8.5 percent increase in memory bandwidth.

Clearly things are moving in the right direction for NVIDIA.

[More...] [Comments...]