GeForce GTX 580 in Detail
With a suggested retail price of $499 ($529 at launch via e-tail), its intensions are quite clear, the GeForce GTX 580 was designed for maximum gaming performance. At this price point, the card is around $200 more expensive than the Radeon HD 5870 and about $60 less than some Radeon HD 5970 graphics cards (those are getting harder to get).
This also means that the GeForce GTX 580 is fetching up to $100 more than some GeForce GTX 480 graphics cards. With such a premium, we are expecting big things from the new arrival.
As mentioned before, based on Fermi's third-generation Streaming Multiprocessor (SM) architecture, the GeForce GTX 580 boasts 512 CUDA cores, which is more than twice the shader power of the GT200 core used by the GeForce GTX 280/285. There are also 48 ROP (Raster Operations) units, the same number used by the GeForce GTX 480. The TAU (Texture Addressing Units) have been increased from 60 to 64 which is just 6.6% more than the GeForce GTX 480.
To break it down, the GeForce GTX 580 has 4 Graphic Processing Clusters, 16 Streaming Multiprocessors, 512 CUDA Cores, 64 Texture Units and 48 Raster Operations Units. The graphics clock speed for fixed function units is set at 772MHz, 10% higher than the GeForce GTX 480. Nvidia has also increased the shader frequency by 10% to a more aggressive 1544MHz.
The GeForce GTX 580 is paired with 1536MB of GDDR5 memory clocked at 1002MHz (4008MHz DDR). Combine that with a memory interface of 384-bit and you get a peak theoretical bandwidth of 192.4GB/s, or 8.5% more bandwidth than the GTX 480 and 25% more than the Radeon HD 5870.
Nvidia has slightly reduced the Thermal Design Power (TDP) of the GeForce GTX 580 to 244 watts from the GTX 480's 250 watt rating. Although the TDP has been lowered by a mere 2.5% Nvidia has employed other measures to help keep the GeForce GTX 580 cooler and quieter. The card has a number of power monitoring chips that oversee power draw from the PCIe slot and PCIe power connectors.
By monitoring power consumption, the card determines if it is drawing too much power and throttles down. We saw something similar on the Radeon HD 5970, which would regularly reduce its operating frequency when running load programs such as FurMark and OCCT. Using these programs to measure the load consumption of the GeForce GTX 580 can't really be done accurately anymore.
Other than the PCIe slot, the GeForce GTX 580 pulls power though a pair of external PCIe power connectors. The GTX 580 requires a 6-pin and an 8-pin connector, just like GTX 480. Nvidia recommends using at least a 600-watt power supply with the GTX 580.
The GTX 580 has also received a substantially overhauled cooling design. Notice there are no heatpipes sticking out the top, while the PCB ventilation has also been removed. Whereas the GeForce GTX 480 used a traditional heatpipe heatsink, the GTX 580 utilizes a heatsink with a vapor chamber.
This type of design was first implemented in the Radeon HD 5970 and it seems like a natural progression for the GeForce GTX 580. The fan has also been slightly changed to mimic a design used by AMD. Nvidia now includes a ring around the blower fan fins to reduce vibration and lower the operating volume.
Finally, the I/O end of the GeForce GTX 580 is identical to the GTX 480, featuring a vent across the top of the panel and a pair of DVI ports with a mini-HDMI port below. The display controller now supports HDMI 1.4a, though we should point out that full audio bitstreaming is not supported, which was also the case with the GeForce GTX 480.
This is disappointing because the newer GeForce GTX 460 does carry this feature, so we're not sure why it wasn't ported over the GF110 architecture. Furthermore, the GeForce GTX 580 only supports two displays, so anyone wanting to take advantage of Nvidia's 3D Vision technology will require a second card.
- Ad-free TechSpot experience while supporting our work
- NEW Get a free copy of Superhot: Mind Control Delete, courtesy of GOG
- Our promise: All reader contributions will go toward funding more content
- That means: More tech features, more benchmarks and analysis