Nvidia's GeForce GTX 590 graphics card

March has certainly been a month of extremes around here. We kicked it off with a look at the Core i7-990X, a world-beating six-core CPU, and then moved on to the absolutely epic Radeon HD 6990. After that, we investigated a pair of breathtakingly fast SSDs. Now, we're back on the graphics beat with a premium offering from Nvidia, the GeForce GTX 590. Like the Radeon HD 6990, the GeForce GTX 590 is a dual-GPU video card planted firmly at the top of the lineup.

This is Nvidia's first attempt at a dually product in quite some time, at least in the frenetically paced graphics market. The last one, the GeForce GTX 295, debuted over two years ago. As we noted in our 6990 review, cramming two high-end GPUs onto a dual-slot expansion card isn't easy; power and thermal limitations often define these products, more so than most. That's probably one reason we didn't see a dual-GPU entrant in the GeForce GTX 400 series. The first-gen chips based on the Fermi GPU architecture were famously late and thermally constrained, making them iffy candidates for the SLI-on-a-stick treatment.

The GF110 GPU in today's high-end GeForce cards is still a rather enormous chip, but it's a little easier to tame—and is a formidable rival to the Cayman GPU in the Radeon HD 6900 series. Naturally, then, Nvidia has cooked up an answer to the Radeon HD 6990, one that reveals a decidedly different approach to the extreme dually graphics card.

Sizing up Gemini
Code-named "Gemini" during its development, the GTX 590 has a pair of GF110 chips onboard, and those GPUs haven't had any of their onboard hardware disabled. Unit counts therefore mirror those for a pair of GeForce GTX 580 cards in SLI. Yet in order to keep the GTX 590 within a manageable power limit, Nvidia has dialed back the clock speeds to levels well below the GeForce GTX 570's. The GTX 590's core clock is just 607MHz, and the GDDR5 memory ticks along at 854MHz—or about 3.4 GT/s. So, although these are fully-enabled GF110 GPUs, the GTX 590's projected rates for key graphics capabilities look very much like a pair of GeForce GTX 570s, not two full-on GTX 580s.

Here's a quick look at the numbers.

We're assuming perfect scaling from one GPU to two in the figures above, which isn't always how things work out in practice. However, these are simply theoretical peaks, and even the most efficient GPUs don't always maintain these rates in real applications.

On paper, at least, the GTX 590 just beats out the Radeon HD 6990 in ROP throughput and memory bandwidth, two keys to fast operation at high resolutions with edge antialiasing, but it's slightly slower in other areas. We wouldn't sound any alarms about the GTX 590's vastly slower theoretical shader arithmetic rates. Nvidia's shader architecture tends to be more efficient, delivering performance comparable to AMD's in many cases, if not superior. Meanwhile, the GTX 590 absolutely crushes the Radeon HD 6990 in peak triangle rasterization rate, which is but one indication of the GF110's quite real end-to-end superiority in geometry processing and DirectX 11 tessellation throughput. The question there is whether or not Nvidia's geometry processing advantage will matter in real games, and it's a vexing one.

All in all, the GTX 590 looks to be endowed with outrageously high specifications. Yet those specs look very much like those of the primary competition, the Radeon HD 6990. This is gonna be a close one, folks.

The card

Dude. Glow.

Like its competition, the GTX 590 presents dual 8-pin aux power inputs to the user, threatening to require a PSU upgrade. The card's max power rating, or TDP, is 365W, just 10W below the peak power deliverable through the combination of a motherboard's PCIe x16 slot and a couple of those 8-pin auxiliary inputs. Not coincidentally, that's also 10W below the Radeon HD 6990's TDP.