TweakTown NewsRefine News by Category:
VideoCardz.com is reporting that NVIDIA is set to launch their flagship mobile GPU based on the Kepler architecture during Computex in Taipei next month. The NVIDIA GEFORCE GTX 680M is not a full GK104 Kepler GPU, nor does it even sport half of the CUDA cores of its desktop version.
The GEFORCE GTX 680M features just 744 CUDA cores, with some listings showing the GPU to have 768 cores, so this should be confirmed during Computex itself. Are you wearing socks? You mighr want to take them off in advance so they don't get blown off: the GPU has much higher GDDR5 memory capacity of.. 4096MB! 4GB of RAM on a notebook-based GPU!
It shares the desktop model's memory interface of 256-bit, and rumors swirling around put its power consumption at 100W. The chip is a second revision of N13E-GTX 680M chip - A2 silicon. The card will support SLI (!) and of course, DirectX 11.1. Performance numbers, that's what we all want, right? We're looking at it being 37-percent than the GEFORCE GTX 670M, with the first leaked benchmark coming from a Chinese website. The GPU hits 4,905 points in 3DMark 11's Performance Preset.
Today NVIDIA has announced a new line of graphics cards for professional applications. The new Tesla GPUs are called the K10 and K20 and are designed to perform exceptionally for high performance computing (HPC) scientific and technical applications that use GPU-accelerated computing. These Tesla's were designed to be high performance and extremely power efficient.
Kepler is three times as efficient as the Fermi architecture which had established itself as a new standard for computing when released 2 years ago. The Tesla K10 features two GK104 chips which produce 4.58 teraflops of performance and 320GB/s of memory bandwidth. The K10 is optimized for oil and gas exploration and the defense industry.
The Tesla K20 is where the GK110 makes its first appearance. The K20 is the new flagship product for the Tesla line of GPUs and provides three times the compute performance of any Fermi-based Tesla product and supports the Hyper-Q and dynamic parallelism capabilities. The K20 is expected to be available in the fourth quarter of 2012.
NVIDIA right now own the performance crown for their GEFORCE GTX 680, and GTX 690 GPUs, but AMD aren't just going to lie down and take it. The latest rumor spinning onto the Internet is that we should expect AMD to ramp up the Radeon HD 7970 reference core clock from 925MHz to 1GHz so that they can reclaim the single-GPU performance spot.
We already have cards clocking in at over 1GHz on the core from various partners, but a reference design from AMD would make this much easier. The rumored cards would launch as "GHz Edition" cards, which we already see in the 7800-series range. Why are AMD doing this now, and not at launch?
AtomicMPC had AMD explain that "yields are now better, their average voltage required to hit 925MHz is much lower than it was on early ES revisions, and most chips are happily hitting 1250MHz now". Is this enough for AMD to win back the performance crown? Or would NVIDIA just do the same thing and crank up their clock speeds once AMD do it? The competition is about to heat up, peeps.
The GEFORCE GTX 670 action over the weekend has been entertaining to say the least. I was busy for most of the weekend and only had my smartphone to check the going ons of the Internet, our site, our Facebook page, and to keep up with news.
If you didn't already know, Shane has posted up a glorious preview of the GTX 670's performance, Cameron smashed out an awesomely written piece on why we didn't receive the GTX 690 for review, and now we have news that GALAXY has a single-slot GTX 670 that will launch shortly.
Now we're staring down the barrel of a single-slot GTX 670 which sports the NVIDIA reference design PCB. The VRM area looks to be located near the front of the card, and the PCB appears to be cutting off at two-thirds the length of the card. Since the GTX 670's PCB is shorter, the fan being longer has to mean something, right?
Some ships are more leaky than others, just like some companies are more leaky than others. NVIDIA, at least with the Kepler launch, has been one of the more leaky companies of late. Pictures and benchmarks of the GTX 680 were surfacing weeks before the product launched and it seems like the same is holding true for the GTX 670.
That picture above is claimed to be the upcoming GTX 670. Accurate specifications such as CUDA core count, frequency, etc are still unknown. Based upon the picture, the GTX 670 will still require two 6-pin PCIe power connectors, which seems a bit excessive. It also features two SLI connectors. The board itself is small, but still appears to use a dual-slot cooler.
The new card produces some respectable numbers in both 3DMark Vantage and 3DMark 11. The leaked benchmarks show the GTX 670, GTX 670 OC, GTX 570 OC, and HD 7950. Vantage performance sees the 670 earn 29471 whereas the 7950 earns only a paltry 24035. Moving to 3DMark 11, the 670 earns 7353 and the 7950 gets only 6418.
AnandTech described the beasty and impressive new NVIDIA GEFORCE GTX 690 dual-GPU video card rather well with just three adjectives in its review yesterday - Expensive, Rare and Fast. They along with a bunch of other websites like ours got a review sample directly from NVIDIA. We didn't, but that was no surprise to us at all.
We have attempted for years to work with NVIDIA, but it hasn't worked. Some years ago we started breaking NVIDIA GEFORCE launch dates and posting our reviews early on purpose because NVIDIA would not support us properly. We didn't expect any more or less than the treatment that other media get. They would never send us review samples of new video cards. Our response was simple - you don't play nice with us, we won't play nice with you. We posted many GPU reviews well ahead of the launch time. Some will say you broke the NDA, but let's make it very clear - you have to have an NDA to begin with to break it. We haven't signed an NVIDIA NDA for a very long time. They'd have to firstly communicate with us for that to even happen...
Overclocking guru K|ngp|n has managed another incredible feat. It wasn't too long ago that he managed to take an NVIDIA GTX 680 up to 1957MHz with the help of some extreme cooling. This time he has managed to push an EVGA GTX 680 up to 1442MHz on basic air cooling alone. That is quite the feat and proves just how great the Kepler architecture is.
K|ngp|n used an EVGA GTX 680 SC, which features a default frequency of 1058MHz (1124MHz at Boost). He pushed 1.212V through the core of the GPU to achieve this feat. That voltage is at the very top of the voltage limit set by NVIDIA. The memory was also overclocked to 1812MHz which is also a big improvement.
Of course when you are running a card like that, especially at those clocks, you need a strong system to run with it so that it doesn't bottleneck. In this case, K|ngp|n used i7-3960X CPU (overclocked to 4.98GHz)and memory clocked at 1245.7MHz (2490MHz effective). This achieved a 3DMark 11 scoring of P12745.
NVIDIA are on a roll, nothing can stop them right now it seems and now we have the first pictures of MSI's GEFORCE GTX 670. The GTX 670 will be NVIDIA's third SKU based on the GK104 GPU, and is set to compete directly with Team Red's Radeon HD 7950 GPU.
The card looks to be pure reference design, and only sports two 6-pin PCI-e connectors. I know you want to get into the specifications of the card itself, so lets do that, shall we? The GTX 670 sports 1344 CUDA cores, 112 TMUs, 32 ROPs, a 256-bit wide GDDR5 memory interface with 2GB of RAM and clock speeds on the Core of 900MHz, 1250MHz or 5GHz effective on the memory.
Non-reference cards that come through in the next few months should sport much shorter PCBs thanks to slim VRM requirements, and just 8 memory chips. DIsplay outputs include two dual-link DVIs, one DisplayPort and one HDMI. The box does state "OC Edition", which means it should come out of the box with overclocked speeds. Also, I'm interested to see what this new technology "DispalyPort" can do, you'll know what I mean when you see it.
Word on the street is that NVIDIA is extremely happy with the 28nm yields of the Kepler architecture and so they decided to launch two high-end cards before stripping down the chip and releasing more value-priced cards. Even with the good yield of the 28nm architecture, the GTX 680 is out-of-stock almost everywhere.
Apparently, even with the good yield, it appears that there is quite the selection of chips that aren't performing up to snuff. NVIDIA is, according to sources, preparing to launch two new, cut-down versions of the GK-104. The GTX 670 (670Ti) will be powered by the GK104-335-A2 whereas the GTX 660 (660Ti) could be powered by a different revision.
The 670 is said to feature 1344 CUDA cores with a 256-bit memory bus and 2GB of GDDR5 memory. Clocks for the chip should be somewhere around 915-950MHz for the core and 1.25GHz for the memory. The 660 should feature a fully disabled GPC (Graphics Processing Cluster) disabled. This means it will feature 1152 CUDA codes with a cut-down 192-bit memory bus. This memory bus would force 768MB or 1.5GB of memory.
VR-Zone expects pricing to look like:
- $999 - GTX 690 4GB
- $579 - GTX 680 4GB OC (Preferred AIB Pricing)
- $499 - GTX 680 2GB
- $379 - GTX 670 4GB (Preferred AIB Pricing)
- $399 - GTX 670 2GB
- $249 - GTX 660 (Ti?) 1.5GB
Cards are expected to be announced sometime next week with wide availability by Computex in early June.
NVIDIA GTX 690 shows up in wooden crate, confirms what reviewers were thinking when crowbar showed up
Last week, NVIDIA sent out boxes that contained crowbars to reviewers around the web. The crowbar had the words "for use in case of zombies or..." and the NVIDIA logo. Nothing more and nothing less. LegitReviews hypothesized that it could be used to open a wooden crate and they were right: a wooden crate was delivered to their office this morning.
As you can see in the picture, the crate warns of "weapons grade gaming power" and has more writing on the side. "0b1010110010", which is is binary for "690" is one of the lines on the side, but I can't decipher what the other two lines mean. Of course you can already guess that the GEFORCE GTX 690 was inside.
With the pry bar that was sent out last week employed, the top of the crate was no match. Inside sat the GTX 690 in all of its $999 glory. Drivers, as of yet, are unavailable and the GTX 690 is set to be available in limited supply May 3. It looks like reviewers will have fewer than 3 days to do their magic before the new card goes on sale.
"The GTX 690 is truly a work of art-gorgeous on the outside with amazing performance on the inside," said Brian Kelleher, senior vice president of GPU engineering at NVIDIA. "Gamers will love playing on multiple screens at high resolutions with all the eye candy turned on. And they'll relish showing their friends how beautiful the cards look inside their systems."