NVIDIA GeForce GTX 570 Launch Review

Tue, 2010-12-07 09:40 -- Benjamin Sun

 

                  480 CUDA Cores

                  4 GPC

                  15 SMs

                  15 PolyMorph Engines

                  60 texture units

                  40 ROPs

                  640KB L2 Cache

                  43.9Gigatexels/second texture filtering rate

                  3 Billion transistors

                  Full-speed FP 16 filtering

                  732MHz Core Clock speed

                  1464MHz Shader Clock Speed

                  3800MHz memory clock speed

                  GDDR5 memory

                  320-bit memory interface

                  152GB/second memory bandwidth

                  1280MB memory

                  Vapor Chamber Cooling design

                  2x 6-pin power

                  219W TDP

                  3-way SLI

                  Optimized for SLI cover design

                  External venting

                  New adaptive GPU fan control

                  2-Slots

                  DL DVI x2

                  mini-HDMI

                  3D Vision

                  3D Vision Surround

                  NVIDIA Surround

                  Tile formats that support Z-cull efficiency

                  Full-speed FP16 texture filtering

                  DirectX 11

                  CUDA

                  PhysX

                  OpenGL 4.1

                  32X FSAA

 

Here’s a chart of the GeForce GTX 570 along with the other currently released cards. You will notice that I have chosen to ignore AMD’s products except for the recently released HD 6870 and HD 6850 cards. AMD is set to release their new high-end cards the Cayman series very shortly and it should be an interesting battle.

Chip

NVIDIA GeForce GTX 580

NVIDIA GeForce GTX 570

NVIDIA GeForce GTX 480

NVIDIA GeForce GTX 470

NVIDIA GeForce GTX 460 1GB

NVIDIA GeForce GTX 460

768MB

AMD Radeon HD 6870

AMD Radeon HD 6850

Process

40nm

40nm

40nm

40nm

40nm

40nm

40nm

40nm

Transistors

3 billion

3 billion

3.2 billion

3.2 billion

3.2 billion

3.2 billion

1.7 billion

1.7 billion

Engine Clock

772

732

700

607

675

675

900

775

SPs

512

480

480

448

336

336

1120

960

Shader Clock

1554

1464

1400

1215

1350

1350

900

775

Compute Performance

1581.1

1405.4

1344.96

1088.64

907.2

907.2

2

1.75

Pixel Fillrate

37.06

29.28

33.6

24.28

21.6

16.2

28.8

24.8

Texture fill rate

49.4

43.92

42

34

37.8

37.8

50.4

37.2

Texture Units

64

60

60

56

56

56

56

48

ROPs

48

48

48

40

32

24

32

32

Memory Type

GDDR5

GDDR5

GDDR5

GDDR5

GDDR5

GDDR5

GDDR5

GDDR5

Memory amount

1.536MB

1536MB

1536MB

1280MB

1024MB

768MB

1GB

1GB

Memory data rate

4008

3800

3696

3348

3600

3600

4200

4000

Memory bus

384

320

384

320

256

192

256

256

Memory bandwidth

192.4

152

177.4

133.9

115.2

86.4

134.4

128

TDP

244

215

250

215

160

150

151

127

One could think of the GeForce GF110 chip as a fixed full-speed GTX 480. If you might remember, NVIDIA launched the GTX 480 with 480 CUDA cores enabled, instead of the 512 originally specified when the architecture was announced last year at the GPU Technology Conference. They also released a GTX 470 which had 448 CUDA cores. The GF110 chip is built on 40nm process with 3 Billion transistors.

The GTX 570 has 480 CUDA cores. NVIDIA splits the CUDA cores into four Graphics Processing Clusters each of which has 3 or 4 Shader Multiprocessors. Each SM contains 32 CUDA cores. If you might remember the GeForce GTX 480 had 4 GPCs 3 or 4 SMs per GPC and 32 CUDA cores for a total of 4 GPCS, 15 SMs and 480 CUDA cores meaning the 570 has the same configuration of cores as the 480.

One thing that needs to be talked about is that the GF110 is reengineered at the transistor level meaning that NVIDIA was able to engineer a higher clock speed for the 570 compared to the 470 while also increasing the CUDA core count and improving the power draw at the same time. NVIDIA says the TDP of the 570 is 219W compared to the TDP of the 580 of 244W. Two GTX 570 cards on the same system require 438W of power by themselves.

One of the main features of the GeForce GTX 580 is the ability to run DirectX 11 features like Tessellation and DirectCompute. The 570 has 15 Polymorph engines meaning that it should offer higher Tessellation performance than the 470 across the board in Tessellation performance and slightly higher than the 480 due to the higher clock speed of the 570 versus the 480.

Tessellation is a process where you can increase the geometry of a character or terrain by using a base mesh and subdividing the triangles into a finer mesh of triangles. In AVP, the developer used Tessellation to make the Aliens rib cage stand out instead of flat with it disabled. HAWX2 uses Tessellation to generate terrain with realistic shadows and geometry in the mountains. HAWX 2 uses 1.5 million triangles for the terrain alone.

One of the things NVIDIA did with the GF110 was implement full speed FP16 filtering compared to the half-speed found on the GF100 chip. This is a holdover from the GF104 chip which introduced the feature and should improve performance in filtering operations. The GeForce GTX 570 also has a Vapor Chamber cooler which is slightly different than the one found on the 580. There are also improvements to the Z-Cull on this chip.

NVIDIA released the GeForce GTX 285 with support for 3D Vision. 3D Vision allows you to play games in stereoscopic 3D requiring the use of 120Hz monitors and a set of 3D Vision glasses from NVIDIA to use.  Games like Left 4 Dead 2 and many others support 3D Vision including Lost Planet 2. Playing games in 3D is sort of like watching Avatar in 3D for the first time.  Hand in hand with 3D Vision NVIDIA recently announced 3D Vision Surround which allows their cards to play games on three monitors with 3D Vision on or off. This requires two cards running in SLI mode, but has the advantage of near double the performance of a single card.  Turning on both 3D Vision and Surround has a profound effect on frame rate so two cards are better than one in this regard. It also requires three identical 120Hz monitors which I don’t currently have.

Ageia introduced the PhysX API several years ago and NVIDIA bought the small company after they released a standalone card to do game physics on the dedicated PhysX card instead of the CPU. With the purchase of Ageia, NVIDIA has developed PhysX to be in many games that show off some amazing effects including destructible terrain, smoke, fog, glass effects and so much more. Several recent games including Mafia II, Just Cause 2 and others. With 480 CUDA cores the 570 should offer excellent PhysX performance. Of course as an NVIDIA-exclusive feature PhysX only operates in games that have the feature.

 

NVIDIA Strikes again, this time the new GTX570 finds its home in the top tier of gaming cards available on the market!

Pages