Sampler runs at scheduler clock (half the hot clock)
4 samplers per cluster (64 total)
Sampler will do jittered-offset for Gather4 (no idea how, the texture-space offset is constant per call)
4 tris/clock setup and raster
Raster area per unit is now 2x4 rather than 2x16
PolyMorph Engine (heh), effectively pre-PS FF, one per cluster
ROPs now each take 24 coverage samples (up from 8)
Compression is improved, 4x->8x delta drop is less than GT200 clock-for-clock
Display engine improvements
Neviem síce presne čo nV merala, ale GF100 má byť 3,3x rýchlejší ako Cypress s 4offset_gather4. Ten dokáže načítať za takt 80 nefiltrovaných samplov z nezávislých adries - teda 68 GS/s. 3,3x 68 = 224 GS/s pre Fermi. Keď to vydelím 128 Sample-ami, čo je hodnota, ktorú má vedieť Fermi fetchovať za takt, dostanem presne 1,75GHz.
I should pause to explain the asterisk next to the unexpectedly low estimate for the GF100's double-precision performance. By all rights, in this architecture, double-precision math should happen at half the speed of single-precision, clean and simple. However, Nvidia has made the decision to limit DP performance in the GeForce versions of the GF100 to 64 FMA ops per clock—one fourth of what the chip can do. This is presumably a product positioning decision intended to encourage serious compute customers to purchase a Tesla version of the GPU instead. Double-precision support doesn't appear to be of any use for real-time graphics, and I doubt many serious GPU-computing customers will want the peak DP rates without the ECC memory that the Tesla cards will provide. But a few poor hackers in Eastern Europe are going to be seriously bummed, and this does mean the Radeon HD 5870 will be substantially faster than any GeForce card at double-precision math, at least in terms of peak rates.
According to our info, Fermi GTX 480 and GTX 470 names were a last minute change, as even partners were talking about GTX 3xx, but it looks like Nvidia decided to clear up some confusion and went with the GTX 4xx names instead.
The cards should be showcased at Cebit by various Nvidia partners which means that they might get it by the end of February, but as always since we are talking about Fermi here, nothing is certain.
Apparently, the old names for the GF100 were GTX 380 and GTX 360, something that would make sense considering the previous naming history, but as it could create a lot of confusion for the average consumer considering the fact that Nvidia rebranded some old cards as the 300 series, so we guess that this is even better.
On the other hand, we aren't sure why Nvidia decided to go with the GTX 470 and not the GTX 460 name as that would make perfect sense, but this might have something to do with the specification difference between these first cards as, performance wise, they might be quite close to each other.
Of course, the performance numbers are still blurry as Nvidia is keeping these tight under wraps and partners are yet to receive first cards, so things should start to get a bit more clear as we draw closer to the CeBIT show in March.
[..]Launch products will be 448SP parts with 512SP parts being PE parts if shown at all. Given all teh Fermi parts seen todate have been confirmed as 448SP chips, this seems very plausable to me.
ak je to pravda, tak:
zlá správa - 512SPs čipy len ako "Premium Edition" = zle dostupné
dobrá správa - na CES boli všetko len 448SP čipy
podľa nVidie majú predstaviť GF100 karty koncom Q:
This is an incredibly complex processor with 3 billion transistors. It’s the world’s most ambitious GPU design, and it will be the fastest GPU on the planet when it launches… later this quarter..
Volume NVIDIA GF100 products in May 2010
By Hilbert Hagedoorn, February 20, 2010 - 12:22 PM
An interesting development, though I feel it's safe to say we can expect a launch of high-end GF100 based graphics cards somewhere next month Jen-Hsun Huang in his quarterly financial reports mentioned that the Fermi architecture aka GF100 aka NV60 aka G300 aka GT300 aka GeForce Series 400 will start playing a role in Q2 of this year. That means for NVIDIA that volume availability of a series of new products would be released at the end of April or early May, and not March as earlier expected.
It seems that the 40nm Fermi architecture keeps fighting NVIDIA somehow, likely due to yield related issues at the TSMC 40nm node. The good news is that Jen-Hsun Huang implies multiple product, and that covers the mid and high-end range but also professional products like Quadro and Tegra. So in the Q2 timeframe we certainly can expect a lot of new products being pushed out real fast.
