Nvidia’s Titan arrives to take the performance crown – the Preview
Nvidia is announcing its second “designer card” today, the $999 GeForce GTX Titan, based on its GK110 GPU on 28nm Kepler DX11.1 architecture. This new 7.1 billion transistor single-GPU flagship card is a continuation of Nvidia’s strategy for an exotic card that is aimed at the highest end of PC gamers to capture their hearts and their wallets. The industrial designs of the GTX 690 and the Titan are aimed particularly at gamers who might spend over $3000 on three video cards for Tri-SLI who want the absolute fastest and yet a reasonably quiet PC – at any cost. It is also aimed at the small form factor PC to give them a powerful but small video card, and finally at the smaller cost-conscious programmers to grow CUDA by allowing full double precision for the first time in a “gaming” GPU even though it does nothing for gaming.
The GTX Titan is the culmination of years of Nvidia’s efforts with their new DX11.1 Kepler architecture, their single-GPU flagship video card. Yet this card does not replace the GTX 690 which also retails for $999. Nvidia believes that they both have their place in their product line-up. This time, Nvidia is aiming for just-below dual-GPU GTX 690 performance on a single card. The specifications of the Titan are quite impressive:
Because we received a Titan from Nvidia just yesterday, we will bring you a two-part review. First up today is the introduction to Titan and in the second part on Thursday, we are comparing the performance of the GTX 690 with the new Titan, as well as with AMD’s very fastest offering, the HD 7970 at GHz edition speeds. There are also two very limited editions of the dual-GPU HD 7970-X2 cards made by AMD’s partners which each retail for $1500 but they are halo cards as only 1000 cards of each were made.
We will also compare the GTX 680 as well as the dual-GPU GTX 590 from the last generation and AMD’s last generation, HD 6990 which consisted of two downclocked HD 6970s on a single PCB. Both of these last generation 40nm dual-GPU cards were faster than the GTX 680 and HD 7970 when the new 28nm cards launched, and we shall see if there are any changes due to driver improvements. In particular, AMD’s Catalyst drivers have shown good performance improvents over the past few months, and the HD 7970 GHz edition has taken the single-GPU performance crown from the GTX 680 – and Nvidia wants it back!
Soon you will see us pit the stock and overclocked GTX Titan against the reference GTX 690, GTX 680 and GTX 590 against our PowerColor reference design HD 7970 at GHz edition at boost speeds (1050MHz) and the HD 6990, using 30 modern games and 7 synthetic benchmarks mostly using 1920×1080 and 2560×1600 resolutions, as well as 15 game benchmarks at 5760×1080 using the GTX 690, Titan, GTX 680 and HD 7970 GHz. We are also comparing the performance of our last generation reference dual-GPU video cards, HD 6990 and the GTX 590 (below) as they were – up until today – the fastest video cards of AMD’s and Nvidia’s last 40nm generation.
We shall also compare Nvidia’s 3-panel Surround working now off of a single GTX 690 at 5760×1080 resolution. Lastly, we also bench 3D Vision 2 and PhysX, ‘on’ versus ‘off’ at the very popular 1920×1080 resolution. However, today is not the day. Today is the introduction to the GTX Titan, and on Thursday, we shall reveal our benchmarks.
What’s New with Titan?
Nvidia’s marketing buzzwords for the GTX 600 series launch were, “Faster. Smoother. Richer.” The GTX Titan is also designed for extreme efficiency and high performance.
Faster
The GTX Titan’s Kepler architecture is SMX-based, now with 2688 CUDA cores. It promises better geometry and texture processing than Fermi thanks to its improved instruction throughput and redesign. In addition, Nvidia brings “GPU Boost II” – a more dynamic way to boost clocks speeds and maximize performance for each game, now based on temperature instead of assumed power draw.
Smoother
New kinds of anti-aliasing – FXAA and TXAA – are now said to compete with MSAA in terms of IQ while not sacrificing performance. And there is a new “Adaptive VSync” that helps to reduce tearing and stuttering associated with regular VSync. Great hardware needs great software to support it and Nvidia is also a software company. They claim to give a lot of attention to how the frames are delivered with a special eye on reducing lag and input delay thus minimizing jitter. Nvdia now gives more voltage unlocking options with Titan than with the rest of the Kepler GPUs, and even a way to overclock your display.
Richer
With Titan as with the rest of the GTX 600 series, it is possible to play games spanning 3 displays in Surround or in 3D Surround from a single GeForce GPU, something Fermi could not manage. And this time, the GTX Titan brings two dual-link DVI connectors plus HDMI and a DisplayPort for a 4th accessor display. PhysX continues to be improved.
How does the GTX Titan compare with its rival, AMD’s cards
Part two of this evaluation which is due in two days, attempts to analyze and compare GTX 690, GTX 590 and GTX 680 performance with the new Titan. We also include HD 7970 at GHz edition performance as well as AMD’s fastest card, their dual-GPU HD 6990 and we will announce a performance winner. We will also look at the details to see what this new Nvidia Kepler GK110 GPU brings to the table for a thousand dollars.
First, let’s take a closer look at the new GTX Titan.
holy geeeez
Hi there, first of all thank you for a great site. I love your extended benchmark suite, it’s so great to see a site not use the same few games again and again. I have a few things though, that I’m missing.
1) It would be nice if you could take a screenshot of the image settings in the Catalyst and nvidia control panel. There are just so many settings, and I’m not quite sure how to interpret what you write.
2) Looking at your previous “The war of the WHQL” article, I was missing an easy way to compare the overall performance. Something like a performance index with and without AA would be great. And perhaps highlight the highest score in the table
3) I’m not sure what the most normal/popular clock speed of the 680 and 7970 is, but since few people buy reference cards, I think you should use a clock speed that more closely resembles what people get when they buy an Asus, MSI or Sapphire card and so on. The performance difference is quite large between the slowest and fastest Asus 670 for instance. The TOP model (which I think is the most popular one), has a GPU Boost Clock of 1137 MHz and a GPU Base Clock of 1058 MHz, compared to a GPU Boost Clock of 980 MHz and a GPU Base Clock of 915 MHz. In benchmarks the fastest model is a lot faster, there is really a big difference but many don’t seem to pay any attention to it. Also your 7970 is listed as overclocked? I’m not sure what the most popular speed of a 7970 is again, but it just caught my eye.
4) Don’t stop testing for smoothness
Thank-you for your comments.
1) As to the control panels, they are set to default (“use application settings”) except that Nvidia’s has power limitations removed, High Quality is used instead of Quality and Vsync is off. In the AMD CP, High Quality is used instead of quality, surface and other optimizations including tessellation are OFF (application settings override CP setting)
2) We almost never bench without AA. We do highlight the highest scores when we are only looking at two sets of drivers to see the performance changes. We don’t usually do it when we are comparing 4 sets of drivers (2-AMD and 2-Nvidia)
3) We use the reference clocks for a GTX 680 and the reference clocks for a HD 7970 at GHz speeds (with the boost locked on)
4) We are resuming frame time benching this week
You might consider joining ABT forum. We’d love to have your input there!