NVIDIA GTX 680 ReviewBosco , airman - March 22, 2012
Category: Video Cards
» Discuss this article (308)
For computer enthusiasts around the world, both AMD and NVIDIA fans, today is a very special day. Today, it is finally here. The GTX 680 has arrived! Many rumors have been circulating for months as to what NVIDIA has been up to with Kepler and what we should all expect. Countless questions have come up regarding Kepler's performance, technologies, power consumption, and general comparability to AMD's current flagship card, the HD 7970 (Tahiti). Today, those questions will be put to rest for good. AMD and NVIDIA fans will either laugh or cry, and those who simply don't care will hopefully be educated enough as to which card to buy if they have been holding out for Kepler! Though no "leaked" information has been presented by staff here on OCC, a brisk trot around Google prior to today would have landed you onto plenty of websites with pictures of the card, hardware specifications, and even some benchmarks. As such, some of you reading this may already have an idea what to expect — at least what it looks like and some other physical trait — but benchmarks prior to today don't really have any credibility. For those who have played fair and haven't peaked, I hope you are as excited as I am to see what this card can do.
Two weeks ago, NVIDIA held a press conference in San Francisco, California where hundreds of editors gathered from around the world for the opportunity to witness the unveiling of the company's latest card. At that time, sitting down and waiting for the presentation to start, I'm sure that many of us really didn't know where to set our expectations. We only knew one thing — Kepler is coming. By the end of the presentation, I can say that I was very impressed at what the card had to offer when it came to its hardware. A clear message that NVIDIA conveyed is that it's not always about raw performance numbers and that efficiency now plays a huge role in the overall rating of a product. NVIDIA was proud to introduce the GTX 680 as the fastest, most efficient GPU ever built, and performance per watt is what is important. Though a bold claim, NVIDIA's confidence was very contagious at this event. That being said, I can't wait to put the card on the hot seat to really see if NVIDIA is truly as proud of the GTX 680 as it seems to be!
I cannot express in real words how excited I am to have the opportunity to prepare the launch review for this card. Since I can't tell you how excited, I'm just going to have to sum it up as "Squeee!" Okay, so maybe I might be too excited, but I'm trying to stay in my chair. I really hope that the "Kepler" GTX 680 is everything and more than NVIDIA says it is! So, let's get into it.
The card to be presented in this review is NVIDIA's reference card. As such, I have no packaging to share, so I will be starting right away with pictures of the card. Featuring a black cooler with green accents and a black PCB, the GTX 680 retains the typical NVIDIA reference card style. It is 10.5" in length, just like the GTX 580 and GTX 480. I'll be taking the cooler off soon to get a look at the cooler and lay eyes on the Kepler core. Minding the fact that this card has 1536 CUDA cores, three times that of the GTX 580, I'm interested to see what's under there.
In case you haven't heard, it had previously been confirmed that the Kepler cards would support NVIDIA Surround on a single card. NVIDIA Surround is the equivalent to AMD Eyefinity, with a few added features, such as bezel-peeking and windows that maximize to a single monitor (instead of maximizing to the entire array). To support NVIDIA Surround on one card, the NVIDIA GTX 680 has two Dual-Link DVI ports, an HDMI 1.4a port, and a full-size DisplayPort 1.2 connection. With four connections, four monitors can be simultaneously operated — such as a 3+1 NVIDIA Surround setup for up to an incredible 7680 x 1600 landscape. A small set of holes in this I/O plate allow hot air from the card to be vented out the back of the case. The opposite side of the card is sealed, unlike the GTX 580, which had a small vent whose purpose was to allow additional airflow to enter the card while in an SLI setup. If the cooling for the NVIDIA GTX 680 is half as good as NVIDIA says it is, then the lack of a little vent on the end won't be an issue.
The part that definitely dropped a lot of jaws at the conference was when we learned that the card only uses two 6-pin connectors. Not a 6-pin and an 8-pin, two 6-pin connectors. Not even my old HD 5870 or a GTX 280 used two 6-pin connections. Some how, NVIDIA has managed to fit three times the CUDA cores than a GTX 580 on a board with a TDP of less than 200W, which is 20% less than the 250W TDP of the HD 7970. GPU power delivery is handled by a 4-phase power design, with two additional power phases dedicated for the GDDR5 memory. The secret and the proof is all in the pudding. I may not know the secret, but I can provide the proof. As usual, we still find the SLI bridge connections opposite the power connections, closest to the I/O bracket. The amount of CUDA cores with running just two GTX 680s in SLI would require six GTX 580s and a mind-blowing 1400W of power. Sure, a core-for-core comparison is not proper for assuming performance metrics, but it lends itself to a comparison of performance per watt. NVIDIA measures Kepler as providing two times the performance-per-watt over Fermi.
NVIDIA has told us about power consumption being lower, but what about noise levels and temperatures? Well, we were specifically briefed on the three key features that allow the GTX 680 to deliver a quiet gaming experience. First, acoustic dampening material used in the highly-balanced GPU fan keeps noise and vibration levels low (blade passing frequency tone is minimized). A unique fin stack design is shaped for better airflow, minimizing noise levels further. Finally, a high-efficiency embedded vapor chamber at the heart of the cooler is able to provide the cooling that the Kepler core requires at lower airflow, reducing the maximum fan speed. With the shroud removed, the cooler looks similar to the GTX 580, but lacks the slanted edge of the fin stack. The cooler is constructed from a copper base with aluminum fins so that it remains lightweight and cost-effective.
What's under the cooler and at the heart of the PCB is what is truly groundbreaking. The Kepler core employs 8 SMX units. Each SMX unit contains 192 CUDA cores, totaling the 1536 that we have mentioned. Fermi has 16 SM units, but with only 32 CUDA cores per unit. The Kepler core has 8 geometry units, 4 raster units, 128 texture units, and 32 ROP units. A very exciting figure that we got to hear is that the GTX 680 has the world's first 6Gbps GDDR5 memory. What does 6Gbps compare to in MHz? Well, the math is easy — 6000 MHz. That's right, 2048MB of 6GHz (6008MHz to be exact), 256-bit, 192.26 GB/s memory. It's sickening! In a good way, of course.
Now that the card has physically been explored, I'd like to take the time to share with you its specifications and features, as we always do. This way, numbers and other physical stuff can easily be seen! Let's keep moving.