During the opening keynote of the GPU Technology Conference, Jen-Hsun Huang unvailled the codenames and the roadmap for the next 2 generation of NVIDIA GPU architectures !So we now publicly know that Johannes Kepler and James Clerk Maxwell are the two next scientists that will succeed to Enrico Fermi !
Sources :
It's interesting to note that communicating so early about unreleased architectures is a major change in NVIDIA marketing strategy, that was used to keep the architecture names and the roadmap secret.
NVIDIA started to communicate publicly about it's architecture names during the release of the first Fermi chip, the GF100, but the previous architecture codenames were never publicly communicated even if the information was available online.
For instance, the name of the Tesla architecture that was the one used in the G80/G9x/GT2xx was unveiled half-publicly through a very good paper from NVIDIA describing the architecture: NVIDIA Tesla: A Unified Graphics and Computing Architecture
All previous NVIDIA architecture names were also discovered a long time ago by the crazy hardcore reverse engineering people working on the NOUVEAU project:
So for the complete family of NVIDIA scientists (according to the NOUVEAU project):
- Celsius : nv10
- Kelvin : nv20
- Rankine : nv30
- Curie : nv40
- Tesla : nv50/G80/G9x/GT2xx
- Fermi : GF100/GF10x
- Kepler (2011)
- Maxwell (~2013)
September 22, 2010 at 1:23 PM
They made a couple of changes to make that slide look nicer.
Tesla was released in 2006 and Fermi in 2010.
This is at least the proof that nVidia wishes to reduce the development cycle time from 3 years to 2 years.
The interesting questions: What would be OpenGL 5 hardware? Kepler or Maxwell?
At least it look like Kepler will be ready for the next console generation and should be the reference next generation games development.
15 DP for Maxwell... humm it looks like it will be 100% of double precision units.
September 22, 2010 at 2:43 PM
I think it mainly depends what is inside OpenGL 5: a simple software API evolution or the support for new hardware class (SM6...).
I don't think we will get a new D3D release by the Kepler timeframe.
If OpenGL 5 is the support for next generation common hardware features, either ATI and NVIDIA are able to reach an agreement together about what are the next common hardware features, and then they make a real OpenGL 5 hardware class by Kepler timeframe, or they wait for MS and then OpenGL 5 may either be simple software enhancements, of be postponed to Maxwell (DX 12 ?) timeframe.
September 22, 2010 at 3:38 PM
If OpenGL 5 was only software API evolution is would be called OpenGL 4.X :p
OpenGL 3 hardware tooks a long time to digest and I think that OpenGL 4 hardware is going to take even longer. I hardly see much innovations for Kepler in the timeframe.
I don't see either an update of D3D for Kepler or maybe something like D3D 11.1.
September 22, 2010 at 3:59 PM
Hum, OpenGL 3 was more a simply software evolution and was still called OpenGL 3 ;-)
But I agree with you that OpenGL 5 should be reserved for next generation hardware. The way I see things is that Kepler innovations will first be exposed as proprietary extensions for GL 4.x, and that common agreed OpenGL 5 hardware features are more likely to be Maxwell timeframe.
September 22, 2010 at 7:45 PM
I am sorry to disagree. The OpenGL everyone expected/wished was Long Peak which was a software evolution. The released OpenGL 3 was a hardware generation release and it's from that time that the ARB decide that major number will represent hardware generations.
A OpenGL 3 hardware is a hardware that supports "geometry" shader, texture arrays, texture buffer, hardware instancing, uniform buffer, 2 channel texture compression formats, conditional rendering, transform feedback, etc...
So no OpenGL 3 is nothing like a software evolution but really stand for a hardware generation just like OpenGL 4 so that an OpenGL renderer could be developed for OpenGL 3.3 and be compatible with all OpenGL 3 hardware.
September 22, 2010 at 7:55 PM
OK good point, my mistake :-)