Nvidia is the top suppliers of the Graphics cards for every market; their cards cover every aspect of the industry from gaming, servers and for research purposes. Nvidia’s CEO Jensen Huang is particularly known for his interest in AI development, the result of which is the Titan and Tesla series cards. Both are directed towards machine learning purposes, Nvidia specially made Tensor cores for these cards, these cores have more machine learning capabilities than the CUDA cores that are found in their gaming lineup.
At the end of last year, Nvidia revealed the mighty Titan V that succeeds Titan XP which was based on Pascal architecture. The 3000-dollar card was based on the new Volta architecture that according to Nvidia will define the future of the Graphics cards. The retail version was packed with such beefy specifications that got it the crown of being the first card with computing power more than 100 TFLOPs mark at 110 TFLOPs, for reference the graphical computing power of the GPU found in the worlds most powerful console capable of providing 4k games at 60FPS the XBOX One X is only 6.1 TFLOPs.
The basic specifications of that card were 5,120 CUDA cores accompanied by 640 Tensor cores and 96 render output units; the card had 12GB of HBM2 memory stack clocked at 1.7Gbps. The base clock speed is 1200Mhz, and the boost speed was 1455Mhz. The GV100 GPU based on VOLTA and built using the newer 12nm FFN process had a total of 21.1 billion transistors on the 815mm2 die. The card needed 250 Watts from the socket.
Yesterday at an event in the Salt Lake City the Nvidia’s CEO Jensen Huang gave away 20 graphics cards based on the TITAN V to the leading AI researchers at the Computer Vision and Recognition conference. The limited edition and by limited we mean only 20 cards are a better version of the already best TITAN V Graphics card. These are called “Nvidia TITAN V CEO’s edition,” no details on the retail version of these cards are announced if Nvidia wanted to boast about this card we would have gotten details about it earlier, but it seems that Nvidia wants to give back to its customers and to contribute in the AI research. The speech given by Jensen on the stage of the conference was showing his company’s gratitude towards the researchers.
He said, “There are all kinds of research being done here. As someone who benefits from your work, as a person who is going to enjoy the incredible research you guys do — solving some of the world’s grand challenges — and to be able to witness artificial intelligence happen in my lifetime, I want to thank all of you guys for that, Huang said, “You guys bring me so much joy.”
Via Nvidia blog
The specification bit has not been entirely revealed by Nvidia other than the 32GB of HBM2 memory that would ease the struggles faced by the researchers who had to face the bottlenecking problems of the TITAN V, that too had a lot of memory, but sometimes handling the massive amount of data by the neural networks requires more than 12GB of VRAM.
Other than this the card has the same amount of CUDA cores and tensor cores, but with more render output units (ROPs) making a huge leap in the memory bandwidth. The core clock speed of the cores might remain the same as the TITAN V, other than this the base GV100 GPU is the same but one of the disabled SMs (Stream multiprocessors) is now enabled that would increase the computing power of the GPU and will enable it to utilize the massive amount of the VRAM completely.
The architecture remains the same GV100 GPU is made with 12nm FFN processors, the transistor count also remains the same. All these improved specifications do not require extra juice from the socket which means Nvidia has tweaked here and there to keep the TDP 250 Watts.
The most important feature that will highlight the card in the league of heavyweights is its computing power which now is 125 TFLOPs a massive increase that will ease the process of running complex machine learning algorithms.
Now as a PC gamer who is looking for the next-gen gaming graphics cards from Nvidia, and after hearing these cold words from Nvidia’s CEO, “it (Next generation) is a long time from now”, it is my right to remind Nvidia that people are looking for the next generation of GeForce lineup.
On the other hand, the company’s commitment with the researchers is admirable since all of these “very” limited edition cards were given away at the conference. There is still no word about the retail version of this card; we are looking for Nvidia’s response on this. If they decide it to release for the public expect it to be even more expensive than the TITAN V and close to the price of the $10,000 Nvidia Tesla V100 card since it’s the “CEO’s edition.”