Rys over at Beyond3D has not leaked outright, but teased, a blurred architecture diagram of what appears to be Nvidia’s next-gen GPU. Interestingly, the codename is GF100, which is the third codename we have heard following the popular GT300 and the recently revealed Fermi. We are not sure which of them is the most appropriate, but for this article, we will refer to Nvidia’s next-gen GPU as GF100.

The GT in GT200 stood for “Geforce Tesla”. GF in GF100 could possibly mean “Geforce Fermi”.

While the diagram itself is strongly blurred, a close look reveals significant details, nevertheless.

gf100 [Rumour] Nvidia GT300 Architecture Details revealed

More details next page.


Rys over at Beyond3D has not leaked outright, but teased, a blurred
architecture diagram of what appears to be Nvidia’s next-gen GPU.
Interestingly, the codename is GF100, which is the third codename we
have heard following the popular GT300 and the recently revealed Fermi.
We are not sure which of them is the most appropriate, but for this
article, we will refer to Nvidia’s next-gen GPU as GF100.

The GT in GT200 stood for “Geforce Tesla”. GF in GF100 could possibly mean “Geforce Fermi”.

While the diagram itself is strongly blurred, a close look reveals significant details, nevertheless.

gf100 [Rumour] Nvidia GT300 Architecture Details revealed

First observations are all the details outside the main diagram. The revision date is marked as 27th September 2009 on the top right. Moving to the center, things get illegible, though the last line reads “*x x GT200 stream output buffer”. The x is a number, but can’t be read.

The left bottom corner reads “3.xB transistors, 40nm @ TSMC”. x is hard to pick up, but it looks most like 2. That being a guess, what is certain is that GF100 is a monster – packing in over 3 billion transistors!

At the bottom of the blurred diagram are what looks like 6 MCs, which confirms the 384-bit memory interface we reported earlier. Right above the MC we have a big chunk of ROPs, though it just looks like one big, black blur – which might suggest too many ROPs than can be differentiated by large white spaces.

Things get particularly tricky once we get to the processing units. One of these GF100 SIMD/MIMD units seems quite different from GT200’s SIMD units, as shown here in Beyond3D’s previous GT200 diagram. Replacing the 24 “shader cores” in GT200 are what we can only suggest as “between 16 and 20 grey dots” and “4 blue stars”. Back to the diagram itself, there are 16 of these units. Once the significance of these “grey dots” and “blue stars” are revealed, we will have a major portion of the specifications. What is encouraging, though, is that GF100 seems to be a largely different architecture from the GT200.

That is all we know so far. Perhaps a Photoshop wizard or an eagle-eyed person can shed more light on the missing details.

Reference: Beyond3D