More Nvidia Pascal Titan X Details Emerge
Last week, Nvidia CEO Jen-Hsun Huang dropped some bombshell news at a Standford University artificial intelligence meetup. Huang revealed the existence of the Pascal-based Titan X to some of the brightest minds in the AI field in attendance and gave the first units to a few of the lucky researches on hand.
After the reveal, we didn’t really know much about the new Titan X. Nvidia said that it is the most powerful single GPU ever built and revealed that the GPU will deliver 11 TFLOPS of floating point performance from 3,584 CUDA cores that boost at 1.53GHz. We were also told the Titan X would come equipped with 12GB of GDDR5X that delivered 480GB/s of memory bandwidth.
Today, Nvidia answered many of our questions about the upcoming graphics behemoth. We don’t have raw performance numbers for you yet, but we can confirm several details.
Nvidia’s new Titan X is not technically a member of the GeForce family, but it will deliver unheard of gaming performance, nonetheless. Nvidia told us the Titan X is being aimed at professional content creators and deep learning researches, but the company acknowledged that the card will undoubtedly be used by gamers with exceptionally deep pockets.
The Pascal Titan X supports all the same technologies that the rest of the Pascal GPUs are designed for, including simultaneous multi-projection, asynchronous compute and support for Nvidia Ansel. The Titan X will also support SLI, but if you were hoping for more than 2-way SLI, you’ll be disappointed. Nvidia is sticking to its guns with dropping support for 3-way and 4-way SLI configurations, even with its top of the line hardware.
The Titan X features a new GPU called the GP-102, which is built on a 471mm2 die. Nvidia confirmed that the Titan X follows the same structure as the other 10-series cards. In other words, the 3584 Cuda Core GPU will feature 224 texture units, and we expect to see 96 ROPs.
When asked about the name, Nvidia said that Titan X is being considered a brand of its own. The company doesn’t expect there to be much confusion in the marketplace, but it plans to label the Pascal-based GPUs clearly to help avoid such a problem.
The Titan X will go on sale on August 2 for $1,200. Nvidia will sell the cards directly through its own website and won’t be allowing partners to build and sell their own variations.
| Nvidia Titan X (Pascal) | |
|---|---|
| Cuda Cores | 3584 |
| Base Clock | 1471 MHz |
| Boost Clock | 1531 MHz |
| Memory Capacity | 12GB |
| Memory Speed | 10 Gbps |
| Memory Bus | 384-bit |
| Memory Bandwidth | 480 GB/s |
| Technology Support | Simultaneous Multi-Projection |
| VR Ready | |
| Nvidia Ansel | |
| G-Sync | |
| GameStream | |
| GPU Boost 3.0 | |
| Direct X 12 | |
| Vulkan | |
| Open GL 4.5 | |
| Display Support | 7680x4320 @ 60Hz |
| DP 1.4 | |
| HDMI 2.0b | |
| Dual Link DVI | |
| GPU Height | 4.376" |
| GPU Length | 10.5" |
| GPU Width | dual-slot |
| Max GPU Temp | 94-degrees C |
| GPU Power Draw | 250W |
| PSU Requirements | 600W |
| Power Connectors | 1x 6-pin + 1x 8-pin |
Got to be $2500+
$1200 is about AUS $1600
It also follows the logic of what many gamers found out. You could SLI two TI's for a better price/performance than Titans. Nvidia got a lot of grief from Titan owners over that. This totally explains why they are trying to separate it from the crowd.
Similar things have happened with the Quadro series over the years in between different gens. There was a time when some found they could use 970's, 980's, or Titans for a lot less money and get close to the same or better performance. It's just that Nvidia's top consumer cards are so close in the performance tiers that they are trying to find the niche of "affordable" pro cards.
Shuh-nuh-nuh, nuh-nuh-nuh, Bat-Card....
The new Titan X is for people with more money than patience. It could also ease the burden on the GP104 demand, slightly.
Ouch. Maxwell was really a big step up, you know? A single 1080 would probably outrun your 780 Ti SLI setup.
So, you might have a point. Sit out the 10xx generation, since the step up to 11xx might be just as big.
It doesnt matter what consumer market you are in you will pay more for the use of new tech or the best tech.
I for one will be getting 1. But i will limit this to 1 as that will tide me over until Volta and the HMB2 is at a better optomisation out of box.
Remember AMD used HMB1 last year on the fury x 4gb vs 6gb of 980ti or 12gb of titan x 2015. This was an attempt to hit close to 980ti in stats and price with a small amount of the more expensive memory.