A Closer Look

Like on all recent high-end cards from NVIDIA, the cooler is a quite complex piece of technology. The main cooling assembly (first picture) uses a copper base to move the heat away from the GPU as fast as possible. If you look closely you can see the missing memory cooling pad on the front and the back.


The SLI connector is hidden behind a rubber cover. You can combine up to three GeForce GTX 280 cards in SLI to get even more performance.


Power to the card is supplied via 2x 6-pin PCI-Express power connectors. Both are required for operation. Further to the right you see the SPDIF audio input, covered by another rubber cap.


Just like on the G80 (GeForce 8800 GTX/Ultra), the output logic has been moved outside of the GPU and resides in this small chip called NVIO. Compared to the G80's NVIO, this one is a new version called NVIO2. This design approach removes all the analog signals and high frequencies of the TMDS link from the GPU silicon, resulting in (a bit) simplified GPU design.
On the other hand this extra part adds cost to the PCB design. Another advantage becomes apparent if you think "GPGPU/ CUDA" - in pure calculation board designs the display output logic is not needed, which allows a cost advantage when built without.


The GDDR3 chips are made by Hynix and have the model number H5RS5223CFR-N0C. Hynix rates those chips at 1000 MHz (= 1.0 ns cycle time). On the GTX 280 the faster 1200 MHz chips are used, but in our overclocking tests the memory on our sample easily passed the 1200 MHz mark.


This is NVIDIA's massive GT200 GPU, it comes with 1.4 billion transistors in a 65 nm process. Its die size of 576 mm² makes it the biggest and most expensive GPU produced to date. Please note that the metal you see is not the GPU die, but the heatspreader on top of it. As you can see the GTX 260 uses the "-100" variant of the GT200 graphics processor, while the GTX 280 uses the "-300". Now what could "-200" be used for?