Kamis, 29 Maret 2012

Kepler Unveiled: Nvidia's GTX 680 Benchmarked In-Depth!

Johannes Kepler once composed, “Nature uses as little as possible of anything.”

Nvidia’s newest GPU, code-named Kepler after the In in german math wizzard, looks to be motivated by that quotation, as much as by the unique Kepler’s statistical ability. The new GPU—the GTX 680— provides fantastic style power, but needs only two 6-pin PCI Communicate energy connections. It’s a big leaving from the last-generation GTX 580, which was quick, but energy starving.

We’ll discuss efficiency soon, but we will first look at Kepler’s actual structure.

Smaller Implies Bigger
Kepler GPUs are designed using a 28nm developing procedure, enabling Nvidia to develop in more tour in less die place. 

Like Fermi, Kepler is a flip structure, enabling Nvidia to range the style up or down by including or subtracting efficient prevents. In Fermi, Loading Multiprocessors, or SMs for brief, are the primary foundations from which the GTX 500 range of GPUs were designed. The CUDA primary variety in the SMs could differ. For example, each SM prevent in the GTX 560 Ti included 48 CUDA cores, while the GTX 580 SM was designed with 32 cores. The GTX 580, however, had a complete of 16 SMs of 32 cores each, for a complete of 512 CUDA cores.

Kepler’s efficient prevent is the SMX. Kepler GPUs are designed on 28nm, which permitted Nvidia’s designers to range elements a bit diversely. So Nvidia improved the variety of cores in the Kepler SMX to a gorgeous 192 CUDA cores each.

The GTX 680 GPU is designed from eight SMX prevents, organized in combined categories known as GPCs (graphics efficiency clusters). This gives the GTX 680 a huge 1,536 CUDA cores.

The SMX does not just home the CUDA cores, however. Designed into each SMX is the new Polymorph website, which contains the hardware-tessellation website, installation, and relevant functions. Also provided are 16 surface systems. This gives the GTX 680 a complete of 128 surface systems (compared with the 64 surface systems provided in the GTX 580). Perhaps surprisingly, the storage cache has modified a bit—each SMX still has 64KB of L1 storage cache, aspect of which can be used as distributed storage for GPU figure out. However, that indicates the complete L1 storage cache have reduced a bit, since there are only eight SMX systems in the GTX 680, not 16 as with GTX 580. The L2 storage cache is also lesser, at 512KB rather than the 768KB of Fermi.

Another exciting modify is that pre-decoding and reliance verifying has been offloaded to application, whereas Fermi managed it in components. What Nvidia got in come back was better instructions efficiency and more die area. Perhaps surprisingly, the transistor depend of the GTX 680 GPU is 3.5 thousand, up only a little from the 3 thousand of the GTX 580. The die dimension have reduced, however, to a much more controllable 294mm2—by comparison, Intel’s Exotic Link 32nm quad-core CPU die is 216mm2.

Textures, Antialiasing, and More
One of the chilly new functions from an real system viewpoint is bindless designs. Just before Kepler, Nvidia GPUs were restricted to 128 several textures; Kepler increases that by enabling designs to be designated as required within the shader system, with up to 1 thousand several designs available. It’s unlikely whether game titles will use that many designs, but certain kinds of structural making might advantage.

Nvidia constantly integrate its exclusive FXAA antialiasing function, but has included a new function that it’s getting in touch with TXAA. The “T” appears for “temporal.” TXAA in its conventional function is actually a version of 2x multisampling AA, but differs the choosing design eventually (i.e., over several supports.) The outcome is better side excellent than even 8x MSAA, but the efficiency hit is more like 2x multisampling.

Another awesome new function that will also gradually be backed in mature Nvidia GPUs is Flexible Vsync. Currently, if you secure straight synchronize to your monitor’s renew amount (typically 60Hz, but as great as 120Hz on some displays), you are going to get easier game play. However, you might see a fall over their words as the shape amount falls to 30fps or below, due to the outcome supports being fixed to vsync. On the other hand, if you run with vsync off, you may see shape carrying, as new supports are sent to the screen before the old one is finish.

Adaptive Vsync hair the shape amount to the straight renew amount, until the car owner finds the shape amount losing below the renew amount. Vsync is then incapable momentarily, until the shape amount improves above the observe renew amount. The overall outcome is much easier efficiency from the customer's viewpoint.

Finally, Nvidia has beefed up the movie website, developing in a devoted scribe website able of development H.264 high-profile movie at 4x – 8x real-time. Energy utilization is low in this function, eating single-digit h, rather than the shader-driven hundreds of h of past GPUs.

The GTX 680 Design Card
Nvidia designed an enhanced routine panel to coordinator the GTX 680 GPU. The panel will send with 2GB of GDDR5, with the standard storage time operating at 6008MHz—the first panel to send with 6GHz GDDR5. The GTX 680 also presents GPU Increase, an idea obtained from the world of x86 CPUs. GPU Increase improves the primary time rate if the inner heat atmosphere allows. This allows video games that offer brighter overall fill to get extra efficiency as required. In another leaving, the GTX 680 provides 1 clock—the shader lamps are now the same as the primary time regularity. Product bins will likely show both the platform and boost lamps on the box. As with lately launched AMD products, the GTX 680 is completely PCI 3.0 certified.

A few significant elements are involved when analyzing the specifications. First, this is a 256-bit large storage program, in contrast to the 384-bit program of AMD’s Radeon HD 7970. Nvidia creates up for this with both enhanced memory-controller performance plus greater clocked GDDR5. The shape shield is “only” 2GB, but that was enough to run our most strenuous standards at 2560x1600 with all details stages maxed and 4x MSAA permitted.

Also value getting in touch with out is Nvidia’s new commitment to performance. The GTX 680 is considerably more energy effective than its forerunner, with a highest possible TDP of just 195W. Lazy energy is about 15W. We saw the energy benefits in our benchmarking.

The GTX 680 is also the first single-GPU card from Nvidia to assist more than two shows. Customers can add up to four shows using all four slots. Nvidia was curiously reticent about talking about its DisplayPort 1.2 execution, which should allow for even more screens once 1.2 able screens and locations appear on the landscape later this season.

The GTX 680 air conditioning is a finish upgrade, using a pointed fin collection, sound dampening, and a high-efficiency warm tube. The card was very silent under fill, though perceptually about the same as the XFX Radeon HD 7970’s twin-cooling-fan style. Of course, having a more energy effective GPU style is a big help. The GTX 680 is no DustBuster.

How Does It Perform?
We rough the GTX 680 against two past GTX 580 styles, the a little bit overclocked EVGA GTX 580 SC and the more intensely overclocked EVGA GTX 580 Categorized. The XFX Radeon HD 7970 Dark-colored Version was also provided. We ran our regular standard package at 2560x1600 with 4x MSAA permitted, along with the FutureMark and Unigine artificial testing.

The GTX 680 clearly requires most of the standards, though the XFX HD 7970 eked out a number of is the winner. Observe that it’s possible some of these standards are actually becoming CPU restricted, even with 4x MSAA, but it’s challenging to say for certain. Which is very likely the situation with HAWX 2, where the mature GTX 580 Classified—albeit a intensely overclocked GTX 580—manages a 1fps benefits.

The GTX 680’s lazy energy scores are amazing, too. The complete program energy at lazy was just 116W, 8W better than the XFX card. However, Nvidia does not integrate anything like AMD’s ZeroCore technological innovation, which decreases energy to a uncovered 3W when the display is converted off (as when Windows seven card blanks the display.) Even better is the energy under load—the GTX 680 is the only GPU to run at under 300W at complete fill.

The GTX 680 we examined is Nvidia’s referrals card, and it’s likely that some companies will send store credit charge playing cards at higher primary time rates of speed. Retail credit charge playing cards will be available upon release (March 22). Nvidia is costs the card at $500, but costs may differ a bit based on producer. That $500 cost tag considerably undercuts AMD’s Radeon HD 7970 costs by as much as $100, which creates the GTX 680 look even better for high-end players.

The GTX 680 looks to restore Nvidia the efficiency title temporarily presented by AMD, and is cost reduced, to start. What exactly is most fascinating, however, is that Kepler likely has some headroom for even higher energy intake, which may allow Nvidia to send an even higher-end GPU when required. The efficiency horserace carries on, and while the top identify now is supposed to be to Nvidia, the organization also needs to produce mid-range GPUs to contest with AMD’s more latest item goes. In the long run, players will advantage from more alternatives and competitors. It’s a win all around.


0 komentar:

Posting Komentar