Table of Doom
But this wouldn't be a HEXUS review without a huge table to be confused by.
| GPU | NVIDIA GeForce GTX 590 3,072MB | NVIDIA GeForce GTX 580 1,536MB | NVIDIA GeForce GTX 570 1,280MB | NVIDIA GeForce GTX 480 1,536MB | Radeon HD 6990 OC 4,096MB | Radeon HD 6990 4,096MB | Radeon HD 6970 2,048MB | Radeon HD 6950 2,048MB | Radeon HD 5970 2,048MB |
|---|---|---|---|---|---|---|---|---|---|
| Transistors | 6.0bn | 3.0bn | 3.0bn | 3.0bn | 5.28bn | 5.28bn | 2.64bn | 2.64bn | 4.3bn |
| Die size | 2 x 520mm² | 520mm² | 520mm² | 529mm² | 2 x 389mm² | 2 x 389mm² | 389mm² | 389mm² | 2 x 334mm² |
| GPU | Fermi v2 | Fermi v2 | Fermi v2 | Fermi | Cayman | Cayman | Cayman | Cayman | Cypress |
| General clock | 607MHz | 772MHz | 732MHz | 700MHz | 880MHz | 830MHz | 880MHz | 800MHz | 725MHz |
| Shader clock | 1,214MHz | 1,544MHz | 1,464MHz | 1,401MHz | 880MHz | 830MHz | 880MHz | 800MHz | 725MHz |
| Memory clock | 3,414MHz | 4,008MHz | 3,800MHz | 3,696MHz | 5,000MHz | 5,000MHz | 5,500MHz | 5,000MHz | 4,000MHz |
| Memory interface | 384-bit x 2 3,072MB GDDR5 | 384-bit, 1,536MB GDDR5 | 320-bit, 1,280MB GDDR5 | 384-bit, 1,536MB GDDR5 | 256-bit x2, 4,096MB GDDR5 | 256-bit x2, 4,096MB GDDR5 | 256-bit, 2,048MB GDDR5 | 256-bit, 2,048MB GDDR5 | 256-bit x2, 2,048MB GDDR5 |
| Memory bandwidth | 163.9GB/s x 2 | 192.4GB/s | 152GB/s | 177.4GB/s | 2 x 160GB/s | 2 x 160GB/s | 176GB/s | 160GB/s | 2 x 128GB/s |
|
Geometry |
8 | 4 | 4 | 4 | 2 | 2 | 2 | 2 | 1 |
|
DP speed |
1/4 | 1/4 | 1/4 | 1/4 | 1/4 | 1/4 | 1/4 | 1/4 | 1/5 |
| Shaders | 1,024 | 512 | 480 | 480 | 3,072 | 3,072 | 1,536 | 1,408 | 3,200 |
| GFLOPS | 2,486 | 1,581 | 1,405 | 1,345 | 5,407 | 5,099 | 2,703 | 2,253 | 4,640 |
| Texturing | 128ppc
bilinear 128ppc FP16 |
64ppc
bilinear 64ppc FP16 |
60ppc
bilinear 60ppc FP16 |
60ppc
bilinear 30ppc FP16 |
192ppc
bilinear 96ppc FP16 |
192ppc
bilinear 96ppc FP16 |
96ppc
bilinear 48ppc FP16 |
88ppc
bilinear 44ppc FP16 |
160ppc
bilinear 80ppc FP16 |
| ROPs | 96 | 48 | 40 | 48 | 64 | 64 | 32 | 32 | 64 |
| ROP rate | 58.3 | 37.1 | 29.3 | 33.6 | 56.4 | 53.1 | 28.2 | 25.6 | 37.1 |
| GTexel/s INT8 | 77.7 | 49.4 | 35.1 | 42 | 168.96 | 159.36 | 84.48 | 70.4 | 116 |
| FP16 rate | 77.7 | 49.4 | 35.1 | 21 | 84.48 | 79.68 | 42.24 | 35.2 | 58 |
| Board power (TDP) | 365W | 244W | 219W | 250W | 450W | 375W | 250W | 200W | 244W |
| HDMI | 1.4a | 1.4a | 1.4a | 1.4a | 1.4a | 1.4a | 1.4a | 1.4a | 1.3a |
| Retail price | £569 | £385 | £270 | £200 | £530 | £530 | £285 | £205 | £500 |
Who is the real competitor?
Right-o, the numbers are suitably large from this very select band of cards. The lower-than-GTX 580 clocks of the 590 card play out in commensurately lower FLOPS, texturing and ROP rates. We clearly know that it won't best two GTX 580s in SLI, but another intriguing question is whether it can topple two GTX 570s tied together? You see, while the GTX 570 has an inferior architecture - fewer shaders and a narrower memory-bus width, for starters - higher clockspeeds help offset this disadvantage. What's more, the combined price of two second-rung Fermi cards is lower than a single GTX 590. Interesting, indeed.
The spec. table always paints AMD Radeon cards in a better light, for the HD 6990, the obvious cross-company competitor, spanks the GTX 590 on the shading and texturing fronts. But specs do not a card make, as AMD knows all too well from recent history.
Unlike the board-mounted switch of awesomeness on the Radeon HD 6990 card, there's no obvious hardware-based method of increasing the performance farther, unless, of course, you fiddle with overclocking programs and eke out extra speed. Practically all cards will be reference in design, towing the 607MHz/3,414MHz clocks of NVIDIA's card, and it will be a while before we see full-custom jobs on the GTX 590, if at all.
