Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

Nvidia's GeForce GTX 780: a Titan for the rest of us

Mark Hachman | May 27, 2013
The same GPU powering the $999 Titan is now available in a $649 cousin, and the GeForce Experience utility comes out of beta.

Peddie said that he was quite impressed with the card's performance.

"Our writeup was very favorable," Peddie said, because of the card's performance and its associated 3GB of memory. "It's definitely going to be a big contender."

Peddie also praised Nvidia's branding. "They could sell soap dishes and people would buy it," he added.

Specs: A chip off the Titan
The numbers carve out a new high-end niche for Nvidia to play in. The GTX 780's interior hides 12 SMX units with 2304 CUDA cores. The memory subsystem consists of six 64-bit memory controllers (with a memory bus width of 384 bits) with 3GB of GDDR5 memory. Nvidia clocks the card's base speed at 863MHz, with the ability to accelerate, via Boost technology, to 900MHz. The memory speed that the GTX 780 uses is 6008MHz. It consumes 250 watts.

What does all this mean? Well, PCWorld hasn't officially tested it, but the GTX 780's performance shouldn't be too far off that of the Titan, for the price: The GTX falls short of the Titan by 384 CUDA cores, though its the memory size is just 3GB, versus 6GB for the Titan. Remember, however, that the Titan is also one of the underpinnings of "Titan," Oak Ridge National Laboratory's supercomputer--and the fastest in the world.

The board itself is 10.5 inches long, and includes two dual-link DVIs, a HDMI connector, and a DisplayPort connector, to boot.

Boost 2.0 and software improvements
Nvidia has improved the GPU Boost capabilities within the new card, so it can run as fast as possible without overheating. Early versions of the Boost technology were essentially the same as Intel's Turbo Boost and other overclocking techniques used across the industry: They ramped up clock speeds as fast as possible, within a given power envelope, which meant that exceeding that power threshold could raise chip temperatures to dangerous levels.

With Boost 2.0, Nvidia now uses on-chip thermal sensors for more accurate measurement. Boost 1.0 and Boost 2.0 are essentially the same, but Nvidia replaced the power target with a specific temperature target: 80 degrees C. If Nvidia's on-chip logic thinks that the temperature will soon exceed that figure, the chip steps down its clock speed until it reaches a safe point.

Boost 2.0 also includes new tools for tweaking the GTX 780's performance. Users can increase the thermal temperature limit to increase performance further. Nvidia said that it would not allow a second type of performance increase, known as overvoltaging, on its own cards. Third-party cards may include this feature, however, enabling users to increase the chip's operating voltage, and thus permitting increased clock multipliers. But using this feature does pose a risk of damaging the GPU, and users must acknowledge the risk before they can unlock the technology.

 

Previous Page  1  2  3  Next Page 

Sign up for CIO Asia eNewsletters.