Scan's TekSpek

Our Aim
To provide you with an overview on New And existing technologies, hopefully helping you understand the changes in the technology. Together with the overviews we hope to bring topical issues to light from a series of independent reviewers saving you the time And hassle of fact finding over the web.

We will over time provide you with quality content which you can browse and subscribe to at your leisure.

TekSpek GPU - Graphics
NVIDIA GeForce GTX 670

NVIDIA GeForce GTX 670

Date issued:

NVIDIA is on something of a roll at the moment; it has arguably the best single- and dual-GPU graphics cards on the market in the form of the GeForce GTX 680 and GeForce GTX 690. But as nice as these ultra-high-end solutions are, the entry price into this lofty club is steep - starting at around £400 and rising to almost £900 for the best of the best.

In an effort to extend the reach of the new Kepler GPU to a wider audience, NVIDIA has now unleashed another enthusiast-class GPU, the GeForce GTX 670.

GPU manufacturers adopt a common tactic of launching with the best-possible single-GPU card first and then, a short while later, bringing in a slightly cheaper, less powerful version. Whether you care for the pragmatics of silicon manufacturing or not, this second-rung GPU usually takes in those GPUs that don't quite make the A-grade, which in this case is the GTX 680. Manufacturing partner TSMC determines which wafers lie on the low side of the yield curve, for whatever reason, and NVIDIA repurposes them as the slightly slower parts. That, in a nutshell, is how the GeForce GTX 670 comes into existence.

NVIDIA, therefore, has to make a trade-off between performance and silicon yield. It wants to make the most money out of costly GPU wafers but doesn't want to hamstring this second-rung GPU too much, considering that it'll go up against AMD's robust Radeon HD 7970 and HD 7950 GPUs, so let's see just what kind of machinations are afoot.

GPU GeForce GTX 680 (2,048MB) GeForce GTX 670 (2,048MB) GeForce GTX 580 (1,536MB) Radeon HD 7970
Radeon HD 7950
DX API 11.1 11.1 11 11.1 11.1
Process 28nm 28nm 40nm 28nm 28nm
Transistors 3.54bn 3.54bn 3.0bn 4.3bn 4.3bn
Die Size 294mm² 294mm² 520mm² 352mm² 352mm²
Processors 1,536 1,344 512 2,048 1,792
Texture Units 128 112 64 128 112
ROP Units 32 32 48 32 32
GPU Clock (MHz) 1,006 (1,058) 915 (980) 772 925 800
Shader Clock (MHz) 1,006 (1,058) 915 (980) 1,544 925 800
GFLOPS 3,090 2,459 1,581 3,789 2,867
Memory Clock (MHz) 6,008 6,008 4,008 5,500 5,000
Memory Bus (bits) 256 256 384 384 384
Max bandwidth (GB/s) 192.3 192.3 192.4 264 240
Power Connectors 6+6 6+6 8+6 8+6 6+6
TDP (watts) 195 170 244 250 200
GFLOPS per watt 15.84 14.46 6.32 15.15 14.34

We have five of the best GPUs in the graphics-card business lined up in the table above: three from NVIDIA and two from AMD. The wonders of a 28nm process mean that while the GTX 670/680 fit in more transistors - read performance - than last-generation GTX 580, their die size is over 40 per cent smaller.

GeForce GTX 670 vs. GTX 680

The real comparison, however, is between the two premium NVIDIA cards. Leading on from what we discussed above, GTX 670 shares the same die as GTX 680, though the performance compromise comes by the way of fewer shaders and lower frequencies. GTX 670 misses out on one of the GTX 680's SMX units - down to seven from eight - and decreases the default core/shader frequency to 915MHz, from 1,006MHz. And that's it; no other obvious performance concessions are apparent.

One SMX unit is home to 192 cores, 32 special-function units, 32 load/stores, 16 texture-units, one setup engine (PolyMorph 2.0), and four warp schedulers that can each dole out two 32-thread instructions per clock. This means, overall, GTX 670 has 1,344 cores and 112 texture units, or about 12 per cent less than GTX 680.

The base clock is set to 915MHz though, on average, it goes up to 980MHz when evaluated across a wide range of games. Fewer cores and a lower core clock translate to approximately 20 per cent less computational and texturing throughput but NVIDIA gives it a helping hand by keeping the memory speed at a GTX 680-matching 6,008MHz.


It's pretty easy to hypothesise GTX 670's performance; it should benchmark at around 90 per cent of a full-fat GTX 680. One obvious benefit of removing a few parts of the architecture is lower power-draw. GTX 680 is hardly a hot, thirsty beastie, granted, but GTX 670 drops the maximum TDP down by 25W, to 170W: nice.

As shown by benchmarks from leading online publications, GeForce GTX 670 performance is roughly within 10 per cent of the full-fat GeForce GTX 680. Putting the huge amount of performance into perspective, the new GeForce GTX 670 is more than 22 per cent faster than the last-generation king, GeForce GTX 580!


NVIDIA's GeForce GTX 670 offers almost all of the Kepler-based GeForce goodness of the GTX 680, but does so at a more agreeable price point. At launch, we can expect stock-clocked models to retail at around £330 (almost £100 less than GeForce GTX 680), rising to £350-plus for partner-overclocked editions from many of the usual suspects.

High-performance NVIDIA graphics cards remain an expensive prospect, and we won't see mainstream GeForce GTX 600-series cards until later in the year, but as it stands the GeForce GTX 670 offers a tremendous level of performance at just over £300.

A full range of NVIDIA GeForce GTX 670 cards are stocked by SCAN. Please head here to peruse our catalogue.