Feature Articles

NVIDIA GeForce GTX 670 - Gaining Momentum, Building Steam

By Kenny Yeo - 10 May 2012

1344 Cores to the Fore

1344 Cores to the Fore

After a lull period, the two graphics giants - AMD and NVIDIA - have awaken. In early January, AMD fired the first salvo by releasing the Radeon HD 7970, the first of their new Southern Islands cards.

In retaliation, NVIDIA released the GeForce GTX 680 in late March. The new GeForce GTX 680 was a winner in our books, offering superb performance and decent power and thermal characteristics. The guys in green then followed up with the massive dual-GPU GeForce GTX 690, reclaiming the title of fastest single graphics card. And now, with no letup, NVIDIA is following up with the GeForce GTX 670.

Although yields for the fully-enabled GF104 chip has been good, a good number of them have failed to make the cut. And this has enabled NVIDIA to quickly release the GeForce GTX 670, which is essentially a cutdown GeForce GTX 680.

Along with the reference GeForce GTX 670 from NVIDIA, we will also be taking a close look at two customized examples from ASUS and Palit.

The GeForce GTX 670 is based on the same GK104 core that powers the flagship and the two actually share many similarities, but with some crucial differences. For those new to NVIDIA’s latest GeForce cards, it’s best to check out our review of the GeForce GTX 680 first to get up to speed on the improvements and latest technologies that NVIDIA’s latest cards offer.

As for the GeForce GTX 670, for starters, it loses one SMX (Streaming Multiprocessor). This means it has lesser CUDA cores - 1344 as compared to 1536; and also lesser texture mapping units - 112 versus 128. However, because it loses an SMX and not an entire GPC (graphics processing cluster), it still retains the same number of raster operating units - 32 - and also has the same 256-bit wide memory bus interface.

Elsewhere, clock speeds have also gone down to differentiate the GeForce GTX 670 from the flagship GeForce GTX 680. The base clock speed of the GeForce GTX 670 stands at 915 MHz, whereas the typical boost clock provided by GPU Boost, a new dynamic overclocking technology not unlike Intel’s Turbo Boost, is 980MHz. Memory clock speed, however, remains unchanged at a blazing 6008MHz DDR. This means overall memory bandwidth remains the same at 192.3GB/s.

So on paper, the GeForce GTX 670 certainly does look formidable and it’ll be interesting to see how it matches up against the flagship GeForce GTX 680, considering it is priced a whole US$100 lesser at US$399, yet only lacks a single SMX and has slightly lower core clock speeds. More importantly, at US$399, we are eager to find out how it'll match up against its targeted rival, the similarly priced AMD Radeon HD 7950.

In this review we will also be looking at two customized GeForce GTX 670 cards from ASUS and Palit. Both cards are factory-overclocked and sport unique customized coolers, so it will be interesting too to see how these two cards perform against the reference card.

Here’s a quick look at how the GeForce GTX 670 stacks up against its competitors.

GeForce GTX 670 and competitive SKUs compared

Model NVIDIA GeForce GTX 670 NVIDIA GeForce GTX 680

AMD Radeon HD 7970

AMD Radeon HD 7950 NVIDIA GeForce GTX 580 AMD Radeon HD 6970
Core Code GK104 GK104  Tahiti XT  Tahiti Pro GF110 Cayman XT
Transistor Count 3540 million 3540 million  4300 million  4300 million 3000 million 2640 million
Manufacturing Process 28nm 28nm 28nm 28nm 40nm 40nm
Core Clock 915MHz 1006MHz 925MHz 800MHz 772MHz 880MHz
Stream Processors 1344 Stream processing units 1536 Stream processing units  2048 Stream processing units 1792 Stream processing units 512 Stream processing units 1536 Stream processing units
Stream Processor Clock 915MHz 1006MHz  925MHz 800MHz 1544MHz 880MHz
Texture Mapping Units (TMU) or Texture Filtering (TF) units 112 128  128 112 64 96
Raster Operator units (ROP) 32 32  32 32 48 32
Memory Clock 6008MHz GDDR5 6008MHz GDDR5  5500MHz GDDR5 5000MHz GDDR5 4000MHz GDDR5 5500MHz GDDR5
DDR Memory Bus 256-bit 256-bit  384-bit 384-bit 384-bit 256-bit
Memory Bandwidth  192.3GB/s 192.3GB/s  264GB/s 240GB/s 192.4GB/s 176GB/s
PCI Express Interface PCIe ver 3.0 x16 PCIe ver 3.0 x16  PCIe ver 3.0 x16 PCIe ver 3.0 x16 PCIe ver 2.0 x16 PCIe ver 2.0 x16
Molex Power Connectors  2 x 6-pin 2 x 6-pin  1 x 6-pin, 1 x 8-pin 2 x 6-pin 1 x 6-pin, 1 x 8-pin 1 x 6-pin, 1 x 8-pin
Multi GPU Technology SLI SLI  CrossFireX CrossFireX SLI CrossFireX
DVI Output Support 2 x Dual-Link 2 x Dual-Link  2 x Dual-Link 2 x Dual-Link 2 x Dual-Link 2 x Dual-Link
HDMI 1 1 1 1  1 (mini-HDMI) 1
DisplayPort 1 (version 1.2) 1 (version 1.2) 2 (version 1.2 HBR2) 2 (version 1.2 HBR2) None 2 (version 1.2)
HDCP Output Support Yes Yes  Yes Yes Yes Yes
Street Price Launch Price: US$399 Launch Price: US$499  ~US$479 ~US$399 ~US$509



Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.