The GPU jamboree - how we test
GPU comparisons |
|||||||
|---|---|---|---|---|---|---|---|
| Graphics cards | Current pricing | GPU clock (MHz) | Stream processors | Shader clock (MHz) | Memory clock (MHz) | Memory bus (bits) | Graphics driver |
| HIS Radeon HD 5970 2,048MB | £425 | 725 | 1,600 x 2 | 725 | 4,000 | 256 x 2 | Catalyst 10.10P |
| AMD Radeon HD 6970 2,048MB | £299 | 880 | 1,536 | 775 | 5,500 | 256 | Catalyst 10.12P |
| AMD Radeon HD 6950 2,048MB | £225 | 800 | 1,408 | 775 | 5,000 | 256 | Catalyst 10.12P |
| Sapphire Radeon HD 6850 1,024MB XF | £290 | 775 | 960 x 2 | 775 | 4,000 | 256 x 2 | Catalyst 10.10P |
| Sapphire Radeon HD 6850 1,024MB | £145 | 775 | 960 | 775 | 4,000 | 256 | Catalyst 10.10P |
| Sapphire Radeon HD 5870 1,024MB | £235 | 825 | 1,600 | 825 | 4,800 | 256 | Catalyst 10.10P |
| HIS Radeon HD 6870 1,024MB | £180 | 900 | 1,120 | 900 | 4,200 | 256 | Catalyst 10.10P |
| HIS Radeon HD 5850 1,024MB XF | £360 | 725 | 1,440 x 2 | 725 | 4,000 | 256 x2 | Catalyst 10.10P |
| HIS Radeon HD 5850 1,024MB | £180 | 725 | 1,440 | 725 | 4,000 | 256 | Catalyst 10.10P |
| NVIDIA GeForce GTX 580 1,536MB | £389 | 772 | 512 | 1,544 | 4,008 | 384 | ForceWare 262.99 |
| NVIDIA GeForce GTX 570 1,280MB | £259 | 732 | 480 | 1,464 | 3,800 | 320 | ForceWare 263.09 |
| ASUS GeForce GTX 480 1,536MB | £320 | 700 | 480 | 1,401 | 3,698 | 384 | ForceWare 262.99 |
| ASUS GeForce GTX 470 1,280MB | £189 | 607 | 448 | 1,215 | 3,348 | 320 | ForceWare 260.89 |
| KFA
GeForce GTX 460 1,024MB SLI |
£290 | 675 | 336 x 2 | 1,350 | 3,600 | 256 x 2 | ForceWare 260.89 |
| KFA
GeForce GTX 460 1,024MB |
£145 | 675 | 336 | 1,350 | 3,600 | 256 | ForceWare 260.89 |
| EVGA GeForce GTX 460 768MB | £125 | 675 | 336 | 1,350 | 3,600 | 192 | ForceWare 260.89 |
Test bench |
||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CPU | Intel Core i7 980X Extreme Edition (3.33GHz, 12MB L3 cache, hexa-core, LGA1366 - Turbo Boost enabled) | |||||||||||
| Motherboard | ASUS P6X58D Premium | |||||||||||
| Motherboard BIOS | 1002 | |||||||||||
| Memory | 6GB Corsair DDR3 | |||||||||||
| Memory timings and speed | 9-9-9-24-1T @ DDR3-1,600 | |||||||||||
| PSU | Corsair HX1000W | |||||||||||
| Monitor | Dell 30in 3007WFP - 2,560x1,600px | |||||||||||
| Disk drive(s) | Corsair Force F120 SSD | |||||||||||
| Operating system | Windows 7 Ultimate, 64-bit | |||||||||||
Benchmarks |
||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Aliens vs. Predator | DX11, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 2xAA, 16xAF, very high quality. | |||||||||||
| Just Cause 2 | DX10, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 4xAA, 16xAF, Dark Tower benchmark | |||||||||||
| Mafia II | DX9, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, AA on, 16x AF, built-in benchmark | |||||||||||
| Battlefield: Bad Company 2 | DX11, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 4xAA, 16xAF, ultra quality, FRAPS-recorded benchmark. | |||||||||||
| Call of Duty: Black Ops | DX9, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 4xAA, ultra quality, FRAPS-recorded benchmark. | |||||||||||
| Crysis Warhead | DX10, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 4xAA, gamer quality, Frost map, FRAPS-recorded benchmark. | |||||||||||
| DiRT 2 | DX11, 1,680x1,050, 1,920x1,080 and 2,560x1,600 resolutions, 4xAA, ultra quality, London map. | |||||||||||
| Unigine Heaven v2.1 | DX11, 1,920x1,080, various tessellation modes | |||||||||||
| 3DMark Vantage b1.0.2 | DX10, Run at the 'high' preset and overall, GPU and Feature Test 4 scores noted | |||||||||||
| Temperature | To emulate real-world usage scenarios, we record GPU core temperature both when idle and whilst playing Call of Duty: Black Ops. And, for a worst-case scenario, we throw in numbers from the FurMark stress test. | |||||||||||
| Power consumption | To emulate real-world usage scenarios, we record mains power draw both when idle and whilst playing Call of Duty: Black Ops. And, for a worst-case scenario, we throw in numbers from the FurMark stress test. | |||||||||||
| Noise | A PCE-318 noise level meter is placed at GPU height and 20cm away from the card installed inside a Corsair 700D chassis with side panel on. | |||||||||||
Notes
Seven games, nine benchmarks, 16 GPU setups, of which three are multi-GPU. That's a lot of benchmarking!
We find it interesting that NVIDIA has encouraged partners to drop the pricing of its excellent GeForce GTX 570 GPU. Our analysis of Radeon HD 6970 and HD 6950 suggests that the two new AMD GPUs will sandwich this NVIDIA card. The very latest Cayman pricing has been set accordingly, it seems.
A few numbers to look out for in the following graphs. Radeon HD 6950/70 are shaded in red, while GeForce GTX 580 and GTX 570 are shaded in green. Also look out for how well the two-card Radeon HD 6850, Radeon HD 5850, and GeForce GTX 460 1GB perform.
We like to benchmark with 'out-of-the-box' performance with driver-default settings. But there's been plenty of hoo-hah over AMD reducing driver-default image quality for post Catalyst 10.9 drivers. The situation is documented here. Bearing this in mind, we've manually adjusted the image-quality settings for the Radeon HD 6950 - disabled Catalyst A.I and slid filtering quality to High Quality - and shown it in all the graphs. It's identified by the IQ suffix. This, we hope, provides an apples-to-apples comparison to NVIDIA's default image quality.
However, please bear in mind that we will continue to use driver defaults in all future reviews, unless there's a marked difference in image quality. Further, the out-of-the-box HD 6950's numbers are the ones you should concentrate on.
