User:Ramu50/OLD Index/Junkyard Talkpages A

This hypothesis is TOO OLD, DON'T READ THE CRAP

Physics Processing Unit

(Concept: Overruled, and abandon) - Status: Re-synthesizing Hypothesis

Sorry I am not going to release about MAJC Graphic Card with PPU or the design architecture IBM Cell based, because Intel can copy it to Larrabee for improvements and it is not illegal, since they are both base on the same model.

I have also decide to cancel the release for SLI/Crossfire vs QuadroFX/FireStream and QuadroFX SLI/FireStream Crossfire. --Ramu50 (talk) 06:51, 4 July 2008 (UTC)

(Target finishing dat: Wed, June 25, 2008)--Maybe
Traditional Implanmentation
So a lot of you might ask, well from what we seen in Ageia PhysX, most people would describe it as a processor or a sub unit that is very good at processing algorithm. But the actuall PPU design belongs to Toshiba SpursEngine & IBM Cell processors (used in Playstation 3).

Novodex (physics API)
ALU (NPU + FPU = the variable you input for a physics formula)
SIMD array: ALU (NPU + FPU)
MIMD array: ALU system manager, Vector Processor

Simple forumula like velocity, can be solved by = NPU + FPU
Complex forumula, the NPU and/or FPU will require ALU assistance, things like Square Root,
Trignometry, Statiscal Math, Probability...etc.
e.g. the process that is required to solve, √3, is guided by ALU.
--The ALU would guided by, what range of numbers do you start guessing.
--In Trignometry, the ALU would guide it by, if input is Sin 25 degree, then you either retrieve it from the database in RAM (loaded from HDD) or you try to formulate it from the unit circle

In the graphic card you will have 1 ALU for each vertical SIMD set as the co-processor.
In physics processing the main ALU system manger would systematically manage (or use any type
of method, depending on the instructions implanted) all the formula together, and form it into a flow chart, so an algorithm diagram is created, before the logic processing (or choosing) is performed.

---The logic processing is the actual part of choosing, which follows a very similar scheme in many of web programming script, like if sth....happens....then sth must happen, or for sth....happen...then sth will result.

Vector processor
--(Vector (spatial)), itself is bounded by the 3 laws of physics: Asscoation, Distribution, Commutativity

Association {order of operation doesn't matter, so long you don't change the operand}
Distribution {a way of classifying each operands into a group}
Commutativity {the ability to generate function & using algebra + operators to perform a proof

or a self-check / error-check}
function (multiplication)---branches down into Calculs (analytically)
algebra (addition)----------branches down into Linear Algebra (analytically)

Distribution can be use in matrix, where one vector contain a set of number, and you can sore it in SRAM (caches) or GDDR RAM as a mini-database. Commutativity can perform, error check / self-check, and generate a function. So when an action is generated, it can be stored, if the action appears again but in different transformations (rotation), the ALU can turn it very quickly.

There will be another type of implementation incldued. This is only the basics

This is the message that I post on TigerDirectBlog
GTX 280SLI will be equally the same as 9800GX2, because no additional technology is added.

Overall performance :
(Single Core GPU Graphic Card)
When you have 1 single core GPU, you put them in SLI, you get Parallel computing stability.
Technology is mainly dependant on Driver Management system.

Note: ONLY work for regular SLI (2 graphic card) & Quad SLI (because GPU is compose of 2 sub unit: FPU + NPU) so you HAVE TO have 2:2 or 2:4 ratio.

2:2 (2 GPU on 2 graphic card)
---each GPU usage = 50%
2:4 (4 GPU on 4 graphic card)
---each GPU usage = 25%

Driver Management System
(nForce SLI)---for Nvidia

---method: (SFR, AFR, SLI Antialias)
(GART Crossfire)---for ATI

(Dual GPU Graphic Card)
When you have 2 GPU (or dual core) on a single graphic card, your stability depend on ILP, instruction level parallelism. Which means, "EVERY" single level of code has to be symmetrically parallel.

Code (programming parallelism)
-Nvidia ForceWare (GPU interface)
-ATI Catalyst (GPU interface)
-HLSL (shader)---for DirectX
-GLSL (shader)---for OpenGL
---(WHQL, Window OS certified)

suggested overclock software:
--Nvidia nTune
--ATI OverDrive

Platform / Barebone certifications
(barely any company use it)
Nvidia ESA architecture
AMD GAME

The Driver DOES NOT optimize DirectX and OpenGL. ONLY the shader language of HLSL & GLSL would. THE MESSAGE AFTER THIS IS ADDITIONAL INFO ONLY.

Coming Soon...
　　　　Physics acceleration: SLI, Crossfire, Quadro FX, FireGL and other high-end graphic card

ONLY in Wikipedia