NVIDIA GeForce GTX 480 and GTX 470 Review – GF100 and Fermi tackle gaming

The GeForce GTX 480 is the new flagship graphics card from NVIDIA and is based on the full-size GF100 GPU.  The GPU has 480 CUDA cores, 60 texture units and 48 ROPs.  The 480 core count is notable because up until we got product in our hands, we were being told that they would indeed have a 512 core GPU ready for gamers upon launch.  That obviously didn’t happen and because of either power or heat concerns, in conjunction with yield issues, NVIDIA chose to disable one of the Streaming Multiprocessors (SMs) for a total of 15.  I am interested to see how this decision was made as the GPU itself is broken down in to Graphics Processing Clusters (GPCs) that each hold 4 SMs – so now one of those GPCs is an SM short.


Also, because the texture units are directly associated with the SMs at four per unit, the total texture unit count has dropped from 64 to 60.  We have already predicted that the texture performance of the GF100 GPU would be lower than that of the Evergreen GPU from AMD, though NVIDIA claims it is much more efficient than the GT200 implementation, so we’ll have to see how this slight drop in texturing power affects gaming performance.

The ROP count remains the same as we were originally indicated though at 48 so AA performance should not be affected. 

The GTX 480 continues with the same three individualized clock rates that the GT200 had: 700 MHz core clock, 1401 MHz shader clock and a 924 MHz GDDR5 memory clock (effective 3.7 GHz).  The GDDR5 frame buffer sits at 1.5 GB and does use the full 384-bit memory bus for a total memory bandwidth of 177.4 GB/s.  This does best the Radeon HD 5870s 153 GB/s of memory bandwidth that runs on a 256-bit memory bus. 

With the lower texture unit count, the GTX 480 is limited to about 42 GigaTexels/s of texture filtering power while the Radeon HD 5870 gets as much as 68 GigaTexels/s.  Obvious there are some texturing performance differences to be had between the two architectures the question is whether or not that will show itself in gaming environments.

We were never provided estimated clock speeds on the Fermi architecture so I can’t say whether NVIDIA has lowered these or if they are where the company expected them to be.  Many rumor stories are claiming that NVIDIA also had to pull back on them, but we may never know for sure.

Power estimates (from NVIDIA) put the GTX 480 as a 250 watt graphics card with a GPU thermal threshold of 105C – I can vouch for that as our samples were often running at 95C in a typical gaming environment. 

The GeForce GTX 480 graphics card itself has a nice design and is about the same length as the Radeon HD 5870.  The fan on the card is quiet when it can be and even when running at 95C it wasn’t as bad as I expected it to be.  When we did run a pair of them in SLI however, THAT was a different story.  The noise from them was finally drowning out the sound of our Turbo Cool PC Power and Cooling 1200w power supplies!

The silver part you see here with the GeForce branding on it is actually a part of the heatsink design. 

All 1.5GB of the card’s memory resides under the heatsink and the only notable fact from rear of the card is the opening for the fan to attempt to draw in more air for cooling the GPU. 

The GTX 480 requires a single 8-pin and a 6-pin power connection and NVIDIA recommends at least a 600 watt power supply.