At the end of CES 2010, NVIDIA decided to get us up in the early hours of Sunday morning to tell us more about the graphics architecture of its forthcoming high-end GPU, known as Fermi or GF100. The journalist’s life is decidedly hard at times but what wouldn’t we do to learn a bit more on a new GPU?
Let the hype continue……while waiting for the goods. No, NVIDIA hasn’t yet unveiled its forthcoming graphics cards. Sure, they gave us some numbers and demoed a bit but what we got was limited. As you can imagine, their goal is to unveil details of the architecture linked to 3D rendering so as to ensure continued coverage of the GF100 while waiting for the product itself to appear. They’re also aiming to reassure us all by sharing some very targetted information.
Firstly, graphics are important to NVIDIA and the GF100 has been conceived with this as its priority. Some commentators might have been wondering as the company did emphasize the compute side of the architecture first and highlighted it strongly so as to interest a new market. The competition also saw this as an opportunity to attack and some of the press have been using the question to feed some unconsidered debate. NVIDIA isn’t about to ignore its main current market though, whether pro or general consumer, simply for a long shot on the explosion of a new one. That would be suicidal, especially as the GF100 was designed while CUDA was still in incubation. Of course this doesn’t mean that the architecture won’t be successful in a domain other than the priority area but we need to remember that the priority has always been graphics.
To anticipate any criticism that might in spite of everything be levelled at them here, NVIDIA is putting the accent on the fact that the compute aspect is going to become more and more important for graphics, whether directly (post processing) or indirectly (physics, AI and so on).
Next NVIDIA wants to mark the difference between what it claims is the revolutionary new GF100 architecture and the "simple" incremental development represented by the Radeon HD 5000s. The notion of new architecture against the update of an old one is of course purely subjective as well as only representing what a given manufacturer wants to say at a given moment. As their previous architecture had been stagnating for a long time, NVIDIA were late on certain aspects and needed to make more of an architectural jump forward than AMD who have been introducing improvements on a more regular basis. What’s more, this is all cyclical with first one manufacturer then the other having the most evolved architecture. It’s also a dangerous argument as new architectures could be of doubful effectiveness. This was the case with the R600 and the NV30. NVIDIA was successful with the introduction of the G80 but overall a new architecture needs a bit of time to bed down and software as well as physical implementation need to kick in to get the most out of it.
NVIDIA is moreover trying to anticipate any comparison with the NV30 (GeForce FX 5800) by rehabilitating it as an important 3D development. While the NV30 interested the pro sector on certain points, we aren’t convinced. This is no more than an attempt to avoid any negative fallout in terms of the current situation. The NV30 was a poor implementation of DirectX 9 and forced NVIDIA to lie lamentably to try and hide its faults for as long as possible. Rather than an important development, the NV30 and derivatives were a break on development.
Lastly, NVIDIA had something to say on its relationship with developers and the poor image that this sometimes has. Nothing new here however. Of course, NVIDIA’s efforts with developers are important and beneficial. But we aren’t taken in and although NVIDIA denies this vehemently, in some cases and above all when the competition has better products and there’s an important partnerhip with a games maker, it has been responsible for some regrettable practices – the lack of MSAA support in Batman for Radeons being one as far as we’re concerned. Nevertheless we do understand that it may be frustrating for NVIDIA’s development teams to be presented in a bad light when in the enormous majority of cases they simply provide necessary and beneficial help to developers so that PC games can continue to evolve.
If you’ve been following events in the little world of GPUs for the last few years, you’ll probably have a good idea, by reading between the lines, of what the continual hype (photo here, little leak there) means: the GF100 is running late and pure performance in current titles wont be revolutionary. Correspondance with the launch of the GeForce FX here is obvious but that doesn’t mean that results will be the same. The press today is much more attentive than it was then and it would be suicidal to try to pull the wool over our eyes. Moreover, and again in contrast to how things used to be, we’ve been able to speak directly to the architects of the chip to answer numerous technical questions and give us a relatively precise idea of the architecture and its strong and weak points.