Just as the swarm of AMD "Hemlock" Radeon HD 5970 reviews hit the Internet, nVidia decided to answer in a subtle way – on the company’s Facebook page, "somebody" posted a picture showing the desktop version of NV100 GPU – "GF100" in action.
nVidia NV100-Class GPU is action – is this the future GeForce 380?
The card in question looks quite ready for retail, as far as lack of wiring goes, and is running Unigine’s DirectX 11 benchmark. The corner showing framerates is naturally, "cut by accident" but nevertheless, nVidia has desktop parts working, and in just a few weeks, members of the media and partners should start to see these cards live.
The board itself packs an 8-pin plus 6-pin power, meaning the GPU and memory can suck up to 300W, giving a very healthy overclocking headroom. In our discussions with members of nVidia staff, clock expectations and performance are running quite high, and if we compare the thermal characteristics of Tesla 20-Generation, you can see that "GeForce 380" has around 107W for overclocking purposes.
If you are wondering why are we calling the Fermi GPU "NV100", the reason is actually quite simple.
The name for the architecture is Fermi, while the previous architecture [GT200] was called Tesla. But the actual ASIC has several different names, depending on product where it is going to end up. As a generation of silicon, Fermi ASIC carries the name NV100 is the name for GPU itself, GF100 is name for the GeForce part, and we saw GT300GL for Quadro parts and GT300 for Tesla on several Tesla-based documents back in early 2009.
In any case, regardless of how many codenames the company use [AMD’s own HD5970 was codenamed Evergreen as the architecture, Hemlock as the dual GPU part, R800 as the internal name for external companies and finally the board itself was codenamed as Zaphod Beeblebrox].
At the end of the day, it is important what this DirectX 11-class GPUs are doing and are they working properly or not.