enfrdepluk
Search find 4120  disqus socia  tg2 f2 lin2 in2 X icon 3 y2  p2 tik steam2

GeForce GT 9600

On February 21, 208, the Canadian company officially introduced the 9600 GT video card. It was intended to send the 8600 GTS to its “well-deserved retirement” in the middle-end segment in the price category up to 0. The new G94 core differs from its older “brother” G92 only in quantitative characteristics, and the functional purpose and operating principle of the executive units remained without any or significant changes. In terms of quantitative characteristics, the 9600 GT has 64 universal processors - this is twice as many as the “old man” 8600 GTS, and two times less than the GeForce 8800 GTS (G92) or GeForce 8800 GTX. This time, let's hope that there won't be such a colossal gap in performance compared to older solutions, as there was between the 8800 and 8600.
So, the new G94 core consists of 505 million transistors (for comparison: the G86 had 210 million, the G92 - 754 million, the G80 - 681 million). It operates at a nominal frequency of 650 MHz for NVIDIA GeForce 9600 GTS. The graphics processor uses a unified shader architecture, which has proven itself excellent in the G80/84/92 chips. The idea of ​​​​unifying functional blocks GPU is as follows: previously they were divided into vertex and shader, and now universal blocks are capable of processing any type of instructions without significant losses in performance. This will allow dynamically changing the performance of the core by redistributing resources for the task needed at the moment. As a result, we get a full load of the chip, and as a result, performance increases.
1992621 

In the case of the G94, we see an elementary reduction - 4 shader units, each of which contains 16 streaming processors (Streaming Processor) and 8 texture units (TMU). In total, there are 64 streaming processors and 32 texture units. All G94 processors, as in previous chips, work with increased relatively GPU clock frequencies. In particular, for the 9600 GT it is 1650 MHz. Finally, let's note the frame buffer write blocks (ROP), of which there are 4 in this case (blue blocks next to the L2 cache on the diagram). NVIDIA has not made any adjustments to the operation and concept of building stream processors (SP) since the G84 chips.

For every four stream processors there are two TA texture addressing modules and two TF texture filtering modules. Therefore, now each texture unit, due to the increase in the number of calculated texture addresses, will be able to process twice as many samples as, for example, the G80. Each shader unit has its own L1 cache. It can store not only textures, but also, due to the unification of the blocks themselves, various types of data. All stream processors (SPs) on which the G94 architecture is based are scalar. Why not vector? The reason lies in the fact that, based on studies of shader programs by NVIDIA developers, it was found that the vector architecture uses computing resources quite uneconomically when complex instructions are processed - for example, scalar and vector at the same time (generally speaking, scalar calculations on vector processors are performed very inefficiently ). In light of the recent trend toward a greater transition from vector to scalar computing, the strategy of NVIDIA developers is perhaps becoming clear. Well, what to do with vector program code? Everything is very simple: it is converted into scalar operations directly by the G94 chip itself. As already mentioned, the GeForce 9600 has 4 frame buffer recording blocks (ROP). They have not undergone any changes relative to the G92 architecture and support the following antialiasing methods: multisampling, supersampling and adaptive anti-aliasing.

Characteristics of NVIDIA GeForce 9600 GT

Name GeForce GT 9600
Core G94 (D9P)
Process technology (µm) 0.065
Transistors (millions) 505
Core frequency 650
Memory operating frequency (DDR) 900
Bus and memory type GDDR3 256-bit
Bandwidth (Gb/s) 57.60
Unified shader units 64
Unified shader unit frequency 1625
TMU on conveyor 32
ROP 16
Shader Model 4.0
Fill Rate (Mpix/s) 10400
Fill Rate (Mtex/s) 20800
DirectX 10.0
Memory Capacity 512/1024
Interface PCI-E 2.0