Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: John Walliker who wrote (81616)	3/15/2002 7:31:33 PM
From: SBHX	Respond to of 93625

John W, Vertices/s or triangles/s is a geometry engine characteristic. It is independent of the memory bandwidth. That GeForce4 is 2.5x faster in geometry is mainly a reflection of the GeForce running their core clock > 2.5x faster. (But this is not true, see below) Most of the 3D cards are basically triangle renderers. When you draw the triangles, you are drawing pixels --- basically fillrate. Basically, 3 vertices define the first triangle, but you can use one vertex after that to define a new triangle in the strip etc etc, so vertices/s second can mean triangles/s. If the next triangle shares two common vertices with a preceding one, there is no sense in sending them out again, hence your X3 for triangle is not a valid comparison. IE: the GeForce4 4600 even on geometry is really 7x faster than the sun Majic. GeForce4 basically spent a lot of gates on their Geometry engine or they are able to squeeze 3x as much performance per clock over the majic part. But that's not interesting. Fillrate is where it gets interesting as you want to have very fast texture shaders. To make it complicated, sometimes people use texels in place of pixels (ie: it takes > 1 texture pel to fill in each pel, but that gets complicated). Fillrate is dependent mainly on memory bandwidth. In here, it looks like the GeForce4 is about 4800/163 = 29x faster than this card. But memory is not the culprit here, the Sun chip single render pipe is the problem. I don't think anyone has seen a single render pipe 3D engine since 4 yrs ago. The 4800 is pixels, not texels. See below : nvidia.com I also don't know how the memory on majic chip is organized. If they are using the rdram for their texture memory or frame buffer or Z, they are going to come in far below their theoretical limit in terms of fillrate, unless they sacrifice a lot of area for their texel cache. But their bottleneck is probably the single pipe, not the memory. Most people are trying to do 4 or 8 pipes with 2-4 simultaneous texture engines per pipe. In any event, this is not an impressive graphics part. pheilman is right, Sun doesn't know what they are doing. I would short if this was part of their core business, but it is not. It's a part for 1998, and has no place today. The design team is toast. SbH