Report an Error

NVIDIA GeForce 8800 GTS 640

G80
Graphics Processor
96
Cores
48
TMUs
20
ROPs
640 MB
Memory Size
GDDR3
Memory Type
320 bit
Bus Width
NVIDIA GeForce 8800 GTS 640 Photo NVIDIA G80 Photo
The GeForce 8800 GTS 640 was a high-end graphics card by NVIDIA, launched in November 2006. Built on the 90 nm process, and based on the G80 graphics processor, in its G80-100-K0-A2 variant, the card supports DirectX 10.0. The G80 graphics processor is a large chip with a die area of 484 mm² and 681 million transistors. Unlike the fully unlocked GeForce 8800 GTX, which uses the same GPU but has all 128 shaders enabled, NVIDIA has disabled some shading units on the GeForce 8800 GTS 640 to reach the product's target shader count. It features 96 shading units, 48 texture mapping units and 20 ROPs. NVIDIA has placed 640 MB GDDR3 memory on the card, which are connected using a 320-bit memory interface. The GPU is operating at a frequency of 513 MHz, memory is running at 792 MHz.
Being a dual-slot card, the NVIDIA GeForce 8800 GTS 640 draws power from 1x 6-pin power connectors, with power draw rated at 143 W maximum. Display outputs include: 2x DVI, 1x S-Video. GeForce 8800 GTS 640 is connected to the rest of the system using a PCI-Express 1.0 x16 interface. The card measures 267 mm in length, and features a dual-slot cooling solution. Its price at launch was 449 US Dollars.

Graphics Processor

GPU Name
G80
GPU Variant
G80-100-K0-A2
Architecture
Tesla
Process Size
90 nm
Transistors
681 million
Die Size
484 mm²

Graphics Card

Release Date
Nov 8th, 2006
Generation
NVIDIA GeForce 8 8800
Production Status
End-of-life
Launch Price
449 USD
Bus Interface
PCIe 1.0 x16

Relative Performance

23%
29%
36%
46%
54%
57%
70%
92%
93%
100%
GeForce 8800 GTS 640
101%
107%
109%
125%
133%
142%
147%
153%
158%
169%
178%
182%
187%
194%
221%
223%
227%
230%
230%
233%
235%
249%
261%
266%
276%
278%
280%
283%
284%
284%
290%
304%
310%
332%
338%
338%
353%
353%
366%
391%
403%
405%
413%
415%
429%
434%
437%
445%
466%
475%
482%
483%
484%
486%
524%
529%
542%
548%
551%
555%
557%
566%
567%
583%
628%
644%
661%
665%
671%
678%
694%
712%
729%
746%
757%
757%
839%
923%
936%
998%
1004%
1017%
1031%
1040%
1044%
1049%
1059%
1064%
1078%
1160%
1175%
1175%
1188%
1299%
1329%
1392%
1437%
1578%
1634%
1782%
1782%
1856%
2172%
2228%
Based on TPU review data: "Performance Summary" at 1920x1080
Performance estimated based on architecture, shader count and clocks.

Clock Speeds

GPU Clock
513 MHz
Shader Clock
1188 MHz
Memory Clock
792 MHz
1584 MHz effective

Memory

Memory Size
640 MB
Memory Type
GDDR3
Memory Bus
320 bit
Bandwidth
63.36 GB/s

Render Config

Shading Units
96
TMUs
48
ROPs
20
SM Count
6

Theoretical Performance

Pixel Rate
10.26 GPixel/s
Texture Rate
49.25 GTexel/s
FP32 (float) performance
228.1 GFLOPS

Board Design

Slot Width
Dual-slot
Length
10.5 inches
267 mm
TDP
143 W
Outputs
2x DVI
1x S-Video
Power Connectors
1x 6-pin
Board Number
P356

Graphics Features

DirectX
10.0
OpenGL
3.3
OpenCL
1.1
Vulkan
N/A
CUDA
1.0
Shader Model
4.0

G80 GPU Notes

PureVideo HD: VP1
VDPAU: Not Supported

Each Streaming Multiprocessor(SM) in the chip architecture contains 8 SPs and 2 SFUs.

Each SP can fulfill up to two single precision operations MAD per clock.

Each SFU can fulfill up to four operations SF per clock (these units can also handle single-precision floating-point multiplications per clock).

The approximate ratio of operations MAD to operations SF is equal 2:1.

The theoretical SP + SFU performance in single-precision floating point operations
[FLOPSsp + sfu, GFLOPS] of the graphics card with shader count [n] and shader frequency [f, GHz], is estimated by the following formula: FLOPSsp+sfu; f × n × 3.

Alternative formula: FLOPSsp+sfu; f × m × ( 8 SPs * 2 (MAD) + 4 * 2 SFUs ). [m] - SM count.

SP - Shader Processor (Unified Shader, CUDA Core)
SFU - Special Function Unit
SM - Streaming Multiprocessor
MAD - ADD+MUL

Retail boards based on this design (1)

Name GPU Clock Memory Clock Other Changes
EVGA 8800 GTS KO ACS³ 580 MHz 850 MHz