NVIDIA Fermi performance scaled back?
updated 01:50 pm EST, Thu December 24, 2009
GeForce 300, Tesla may run slower
NVIDIA has quietly scaled back the ambitions of its Fermi-based hardware, a document (PDF) on its own site may have revealed. A specs sheet for the Tesla 20 now shows even the faster hardware using 448 visual processing cores versus the publicly promised 512. It has also ramped down the internal clock for these individual cores to at most 1.4GHz, slightly lower than the current generation.
It's not explicitly stated that the changes made to the Tesla computing cards will apply to future GeForce 300 series or Quadro video chipsets, though the Tesla line often shares much of its underlying feature set with the other elements of NVIDIA's catalog.
Much of the reason for the throttled design is likely to stem from the power consumption, which on a Tesla will draw as much as 225W for a single board. This combined with an already low output of 40 nanometer chips from NVIDIA's manufacturing partner, TSMC, may have pushed NVIDIA to scale back its performance in order to release hardware on time. Fermi is intended to not only provide a major leap in graphics performance but to produce one of the first video architectures explicitly made with general purpose computing in mind.
NVIDIA is due to make announcements at CES but may keep these to mobile designs like the next-generation Tegra, the T20, rather than add significantly to its desktop lineup. [via SemiAccurate]
