[Tested and Burned] SAPPHIRE Radeon HD 6970 2048MB Review

2010/12/23 JeGX

SAPPHIRE Radeon HD 6970 Review Index

3 – SAPPHIRE Radeon HD 6970 OpenGL performances

Testbed:
– CPU: Core i7 960 @ 3.2GHz
– RAM: 4GB DDR3 Corsair Dominator
– Motherboard: GIGABYTE X58-A UD5
– Windows 7 64-bit
– Graphics drivers: R263.09
– PSU: Corsair AX1200

3.1 FurMark (OpenGL 2)

FurMark 1.8.2 has been used for the test. FurMark homepage is HERE.

Settings: 1920×1080 fullscreen, no AA, no postFX, 60sec, Xtreme mode UNCHECKED.

Rule: The higher the number of points, the faster the card is.

8362 points (140 FPS) – EVGA GeForce GTX 580 SC, ***OCP disabled***, GPU core: 880MHz

8259 points (138 FPS) – ASUS ENGTX580 ***OCP disabled***, GPU Core: 871MHz, VDDC: 1.088V

7769 points (130 FPS) – EVGA GeForce GTX 580 SC, ***OCP disabled***, GPU core: 797MHz

6470 points – EVGA GeForce GTX 480

6341 points (FPS: 106) – SAPPHIRE Radeon HD 6970, GPU core: 880MHz, PowerTune: +20%

5420 points – ATI Radeon HD 5870

5161 points – MSI GeForce GTX 470

4641 points (FPS: 78) – SAPPHIRE Radeon HD 6870, GPU core: 1000MHz

4583 points (FPS: 76) – SAPPHIRE Radeon HD 6870, GPU core: 980MHz

4484 points (FPS: 74) – ASUS EAH6870

4310 points (FPS: 72) – SAPPHIRE Radeon HD 6870

4243 points (FPS: 71) – EVGA GeForce GTX 580 SC, OCP enabled

3912 points (FPS: 65) – SAPPHIRE Radeon HD 6970, GPU core: 880MHz, PowerTune: 0

3884 points – MSI N460GTX Cyclone 768D5 OC

3824 points (FPS: 64) – SAPPHIRE Radeon HD 6970, GPU core: 880MHz, PowerTune: -20%

2772 points – MSI R5770 Hawk

Remark: Radeon HD 6970 real score is the one with PowerTune set to +20%.

3.2 TessMark (OpenGL 4)

Hardware tessellation is one of the big features of Direct3D 11 and OpenGL 4 capable graphics cards. TessMark is a new benchmark focused only on the tessellation engine of DX11 class cards. It’s a pure tessellation benchmark, it does not contain complex shader or other heavy texture fetches. TessMark shows an overview of the tessellation engine raw power, that’s all. DX11 specifies that the tessellation factor can vary from 1.0 up tp 64.0. Of course, for tessellation factors like 32 or 64, most of the tessellated triangles are smaller than… a pixel. In those cases, tessellation is useless and in a real world application such as a game, high tessellation factors won’t be used. But in the case of a synthetic benchmark, it’s always instructive to see how cards can handle the whole range of tessellation level.

TessMark 0.2.2 has been used for the test.

Settings: 1920×1080 fullscreen, no AA, 60sec, map set 1.

TessMark - OpenGL 4 tessellation

Rule: The higher the number of points, the faster the card is.

Tessellation factor 8.0: moderate

53151 (888FPS) – EVGA GTX 580 SC

52188 (872FPS) – ASUS ENGTX580

48084 – EVGA GeForce GTX 480

44090 (735FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat11.6

38191 – MSI GeForce GTX 470

30512 – MSI N460GTX Cyclone 768D5 OC

27718 (462FPS) – Sapphire HD 6870, GPU core: 1000MHz

27469 (458FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat10.12

26223 – ASUS EAH6870

24161 (403FPS) – ATI Radeon HD 5870

23131 (386FPS) – Sapphire HD 6870

20745 – MSI R5770 Hawk

Remark: Cayman GPU implements AMD’s 8th generation tessellator (Barts is the 7th and Cypress is the 6th) and for low tessellation level, Cayman (HD 6970) should process tessellation twice faster than Cypress GPU (HD 5870). This is clearly not the case and this is certainly due to the OpenGL driver that has not been updated to reflect Cayman new architecture.

Tessellation: Cayman vs Cypress

Tessellation factor 16.0: normal

33266 (555FPS) – EVGA GTX 580 SC

32666 (545FPS) – ASUS ENGTX580

29196 – EVGA GeForce GTX 480

23316 – MSI GeForce GTX 470

19398 (323FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat11.6

17452 – MSI N460GTX Cyclone 768D5 OC

9255 (154FPS) – Sapphire HD 6870, GPU core: 1000MHz

8555 – ASUS EAH6870

8229 (137FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat10.12

8177 (136FPS) – Sapphire HD 6870

8018 (134FPS) – ATI Radeon HD 5870

7669 – MSI R5770 Hawk

Tessellation factor 32.0: extreme

15427 (257FPS) – EVGA GTX 580 SC

15128 (252FPS) – ASUS ENGTX580

13008 – EVGA GeForce GTX 480

9997 – MSI GeForce GTX 470

6729 – MSI N460GTX Cyclone 768D5 OC

3397 (57FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat11.6

2299 – ASUS EAH6870

2246 (38FPS) – Sapphire HD 6870

2156 (36FPS) – ATI Radeon HD 5870

2122 (35FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat10.12

2159 – MSI R5770 Hawk

Tessellation factor 64.0: insane

4940 (82FPS) – EVGA GTX 580 SC

4840 (81FPS) – ASUS ENGTX580

3963 – EVGA GeForce GTX 480

3169 – MSI GeForce GTX 470

1959 – MSI N460GTX Cyclone 768D5 OC

594 (10FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat11.6

585 – ASUS EAH6870

574 (10FPS) – Sapphire HD 6870

565 – MSI R5770 Hawk

550 (10FPS) – ATI Radeon HD 5870

539 (9FPS) – Sapphire HD 6970, GPU core: 880MHz – Cat10.12

3.3 ShaderToyMark (OpenGL 2)

ShaderToyMark 0.1.0 is an OpenGL 2 benchmark, developed with GeeXLab,
and focused on pixel shaders only. The pixel shaders are heavily based on math (few texture fetches) and then ShaderToyMark can be seen as a kind of GPU computing benchmark.

ShaderToyMark - OpenGL 2 pixel shader

Settings: 960×540 windowed, no AA, 60sec

316 points (52 FPS) – EVGA GTX 580 SC

306 points (51 FPS) – ASUS ENGTX580

263 points (43 FPS) – GeForce GTX 480

234 (39FPS) – Sapphire HD 6970

189 points (31 FPS) – ATI Radeon HD 5870

184 points (30 FPS) – ASUS EAH6870

179 (29FPS) – Sapphire HD 6870

156 points (26 FPS) – MSI N460GTX Cyclone

104 points (17 FPS) – MSI R5770 Hawk

46 points (7 FPS) – GeForce 9800 GTX

36 points (6 FPS) – EVGA GTX 280

33 points (5 FPS) – GeForce GTX 260

Remark: ShaderToyMark is nice because it simply shows the raw processing power. The peak shader arithmetic of the HD 6970 is 2703 GFLOPS and the one of the HD 6870 is 2016 GFLOPS.
The ratio between GLOPS is 2703 / 2016 = 1.3 while the ratio between scores is: 234 points / 179 points = 1.3.

3.4 Maxon CINEBENCH R11.5 (OpenGL 2)

CINEBENCH is an OpenGL benchmark based on Cinema4D.

Maxon CINEBENCH R11.5

Settings: windowed mode 1920×1080.

61.79 FPS – Sapphire Radeon HD 6870

61.60 FPS – Sapphire Radeon HD 6970

61.22 FPS – Radeon HD 5870

42.84 FPS – GeForce GTX 480

42.71 FPS – EVGA GTX 580 SC

42.53 FPS – MSI N460GTX Cyclone

3.5 OpenGL 4 Mountains demo

Mountains demo is an OpenGL 4 demo that shows hierarchical-Z map based occlusion culling in action.

OpenGL 4 Mountains demo

Settings: default window size: 1024×768, ICR enabled (Instance Cloud Reduction), Hi-Z enabled and dynamic LOD enabled.

684 FPS – EVGA GTX 580 SC

674 FPS – ASUS ENGTX580

568 FPS – EVGA GTX 480

390 FPS – Sapphire Radeon HD 6970

350 FPS – MSI N460GTX Cyclone 768D5

255 FPS – ASUS EAH6870

235 FPS – Sapphire Radeon HD 6870

231 FPS – Radeon HD 5870

220 FPS – MSI R5770 Hawk

3.6 GluxMark2 (OpenGL 2)

GluxMark2 is a purely syntethic OpenGL benchmark
and tries to measure performance from every point of view by using programmable graphics pipeline (vertex, geometry and fragment/pixel shaders).

OpenGL 2 - GluxMark2

Preset: high-end (1920×1080, MSAA: 8X)

11002 points (OpenCL: +4318 = 15320 points) – EVGA GTX 580 SC

9216 points (OpenCL: +2776 = 11992 points) – GeForce GTX 480

8099 points – Radeon HD 5870

8010 points (OpenCL: +709 = 8719 points) – Sapphire Radeon HD 6970

6615 points – Sapphire Radeon HD 6870

5367 points (OpenCL: +2789 = 8156 points) – MSI N460GTX Cyclone

3.7 Unigine Heaven (OpenGL 4)

For this last OpenGL test, I used Ungine Heaven 2.1, one of the standard Direct3D / OpenGL synthetic benchmark.

OpenGL 4 - Unigine Heaven 2.1

Settings: 1920×1080 fullscreen, OpenGL rendering, tessellation: normal, shaders: high, AA: 4X, 16X anisotropic filtering.

48.6 FPS, Scores: 1224 – EVGA GTX 580 SC

46.4 FPS, Scores: 1168 – ASUS ENGTX580

38.7 FPS, Scores: 974 – EVGA GeForce GTX 480

28 FPS, Scores: 704 – SAPPHIRE Radeon HD 6970

24.5 FPS, Scores: 617 – MSI N460GTX Cyclone 768D5 OC

15.9 FPS, Scores: 400 – ATI Radeon HD 5870

13.6 FPS, Scores: 342 – ASUS EAH6870

13.5 FPS, Scores: 339 – SAPPHIRE HD6870

9 FPS, Scores: 227 – MSI R5770 Hawk