GeForce GT 430 brings Fermi to $79

Say hello to Nvidia's cheapest DirectX 11, Fermi-based graphics processor yet: the GeForce GT 430. While Nvidia stressed during its presentation last week that partners will set their own prices, it did let slip a $79 target figure. At that price, the newcomer is headed for a confrontation with AMD's Radeon HD 5570, which has been delivering bargain DX11 graphics around that same price point since February.

Before getting into any competitive comparisons, let's take a quick look at what makes the GeForce GT 430. This card is Nvidia's first desktop product to feature the GF108 graphics processor, a 585-million-transistor piece of silicon built on the same 40-nm TSMC fab process as the rest of the Fermi series. My measurements peg the GF108's die area at approximately 115.5 mm², a little less than half the size of the GF106 chip inside the GeForce GTS 450

A block diagram of the GF108. Source: Nvidia.

Fittingly, the GF108 has half the number of shader multiprocessor blocks as the GF106—just two, each with 48 stream processors and one texture block. That means the GF108 has 96 SPs and is capable of filtering and sampling 16 texels per clock cycle. Nvidia has also included a single ROP partition, which can output four pixels per clock, as well as two 64-bit memory interfaces, for a total interface width of 128 bits. (For reference, the GeForce GTS 450 can churn out 16 pixels per clock and features a 128-bit memory interface, although it doesn't utilize the GF106 GPU to its full potential.)

Nvidia quotes standard clock speeds of 700MHz core, 1400MHz shader, and 900MHz memory for the GeForce GT 430. The memory speed applies to the card's 1GB of DDR3 RAM, so it makes for a transfer rate of 1.8GT/s and total bandwidth of 28.8GB/s for the whole card. By the way, the card should have a maximum power draw of about 49W—again, roughly half the GTS 450's 106W. The low power consumption might be why Nvidia's reference design has a half-height circuit board.

Here's Asus' take on the GeForce GT 430, the ENGT430, touting the aforementioned half-height design. This isn't a reference card, though; Asus has thrown on a custom cooler, whose fan purportedly boasts a higher-than-average resistance to dust. This more tightly sealed fan should, the firm claims, increase the life span of the card by 25%. Asus also says it used particularly efficient and durable electrical components.

So, how does the GT 430 stack up against its most direct competitor, the Radeon HD 5570, and the next rung up Nvidia's product ladder, the GeForce GTS 450?

  Peak pixel
fill rate
Peak bilinear
INT8 texel
filtering rate*
*FP16 is half rate
Peak shader
GeForce GT 430 2.8 11.2 28.8 269
GeForce GTS 450 12.5 25.1 57.7 601
Radeon HD 5570 (DDR3) 5.2 13.0 28.8 520
Radeon HD 5570 (GDDR5) 5.2 13.0 57.6-64 520

This theoretical comparison isn't all that flattering to the GT 430. The DDR3-based Radeon HD 5570 has higher theoretical numbers pretty much across the board, except for memory bandwidth. (AMD has a GDDR5-powered 5570 that offers a sizeable memory bandwidth increase, although I can't seem to find anyone selling it right now.) The GT 430's pixel fill rate looks particularly anemic next to the AMD offerings. Beyond direct match-ups, one thing to note is that these ~$80 cards are all considerably weaker than the GeForce GTS 450, which retails for as little as $122.99 at Newegg. Anyone even remotely serious about gaming would do well to spend the extra 40 bucks, I think.

Ultimately, as I pointed out earlier this month, the GeForce GT 430 may be part of a dying breed. The next generation of desktop processors will feature much-improved integrated graphics, which may be potent enough to discourage folks from spending $80 on a relatively underpowered discrete GPU. Perhaps we'll see quicker $80 GPUs by then, but I wouldn't be surprised to see an increasing number of cash-strapped gamers gravitate toward slightly more expensive (and much more powerful) solutions in 2011.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.