vacancies advertise contact news tip The Vault
facebook rss twitter

Nvidia launches Titan V (Volta GV100) graphics card

by Mark Tyson on 8 December 2017, 19:31


Quick Link:

Add to My Vault: x

On Thursday evening at NIPS (the Conference and Workshop on Neural Information Processing Systems) Nvidia launched the Titan V. It refers to this new product as “the most powerful graphics card ever created for the PC” and it features Nvidia’s latest and greatest ‘Volta’ graphics architecture.

The brand new Nvidia Titan V is based upon the same GPU we saw employed in the Tesla V100 HPC accelerator launched back in May. Later that month we noticed an Nvidia intern ‘leak’ a Titan Volta card but for some reason or other it has taken Nvidia six or seven more months to launch such a product.

As an aperitif to the full specs, let’s consider some of the standout figures, as highlighted by Nvidia CEO Jensen Huang at NIPS yesterday. Titan V’s 21.1 billion transistors deliver 110 (deep learning) teraFLOPS of horsepower, 9x that of its predecessor, and extreme energy efficiency, Huang told the crowd at NIPS. The Volta V100 GPU packs 5129 CUDA cores, 640 Tensor cores, has a boost clock of 1455MHz, and features 12GB of HBM2 memory on a 3072-bit wide memory interface.

Nvidia Titan V specs

  • 6 Graphics Processing Clusters
  • 80 Streaming Multiprocessors
  • 5120 CUDA Cores (single precision)
  • 320 Texture Units
  • 1200MHz Base Clock (MHz)
  • 1455MHz Boost Clock (MHz)
  • 850MHz Memory Clock
  • 1.7Gbps Memory Data Rate
  • 4608K L2 Cache Size
  • 12288 MB HBM2 Total Video Memory
  • 3072-bit Memory Interface
  • 652.8GB/s Total Memory Bandwidth
  • 384 GigaTexels/sec Texture Rate (Bilinear)
  • 12nm Fabrication Process
  • 21.1 Billion Transistor Count
  • 3 x DisplayPort, 1 x HDMI Connectors
  • Dual Slot Form Factor
  • One 6-pin, One 8-pin Power Connectors
  • 600 Watts Recommended Power Supply
  • 250 Watts Thermal Design Power (TDP)
  • 91° C Thermal Threshold

Further interesting aspects of this card are that it features a 6- and an 8-pin power connector, it has 16 power phases, and its 250W TDP is attractive considering the performance on offer. As per the leak in May, this card doesn’t work in SLI, instead NVLink will be used for multi-GPU configurations.

Nvidia is targeting researchers and scientists with this latest Titan release, as you might have guessed from the launch venue and presentation. There is free AI software available from Nvidia to use with this card (including the Nvidia TensorRT inferencing optimiser) and help for developers working in deep learning and HPC.

If you are interested in the new Nvidia Titan V, it will be released on 30th Dec (approx) and will cost £2700/$2999 with a limit of two per customer. You can pre-order right now.

HEXUS Forums :: 15 Comments

Login with Forum Account

Don't have an account? Register today!
250w? That's incredibly promising for mainstream Volta GPUs in the future.

From the AT article:

Moving on and diving into the numbers, Titan V features 80 streaming multiprocessors (SMs) and 5120 CUDA cores, the same amount as its Tesla V100 siblings. The differences come with the memory and ROPs. In what's clearly a salvage part for NVIDIA, one of the card's 4 memory partitions has been cut, leaving Titan V with 12GB of HBM2 attached via a 3072-bit memory bus. As each memory controller is associated with a ROP partition and 768 KB of L2 cache, this in turn brings L2 down to 4.5 MB, as well as cutting down the ROP count.

In terms of clockspeeds, the HBM2 has been downclocked slightly to 1.7GHz, while the 1455MHz boost clock actually matches the 300W SXM2 variant of the Tesla V100, though that accelerator is passively cooled. Notably, the number of tensor cores have not been touched, though the official 110 DL TFLOPS rating is lower than the 1370MHz PCIe Tesla V100, as it would appear that NVIDIA is using a clockspeed lower than their boost clock in these calculations.

As mentioned earlier, NVIDIA is unsurprisingly pushing this as a compute accelerator card, especially considering that Titan V features tensor cores and keeps the TITAN branding as opposed to GeForce TITAN. But there are those of us who know better than to assume people won’t drop $3000 to use the latest Titan card for gaming, and while gaming is not the primary (or even secondary) focus of the card, you also won't see NVIDIA denying it. In that sense the Titan V is going to be treated as a jack-of-all-trades card by the company.

To that end, no gaming performance information has been disclosed, but NVIDIA has confirmed that the card uses the standard GeForce driver stack. Now whether those drivers have actually been optimized for the GV100 is another matter entirely; Volta is a new architecture, markedly so at times. Speaklng solely off the cuff here, for graphics workloads the card has more resources than the Titan Xp in almost every meaningful metric, but it's also a smaller difference on paper than you might think.

As for NVIDIA's intended market of compute and AI users, the Titan V will be supported by NVIDIA GPU Cloud, which includes a number of deep learning frameworks and HPC-related tools.

Its made on an improved TSMC 16NM node known as 12NM which has lower leakage:

The GP102 is 471MM2 and the GV100 is 815MM2 in comparison. Now look at the relatively small increase in FP32 numbers(which is most relavant for gaming).

This is more a FP64 and deep learning based computation GPU. It should be in theory be quicker than the GP102 going by the higher FP32 numbers and more bandwidth,but I think Nvidia Ampere is what most gamers might be looking at next year:

I would expect far less transistors dedicated towards FP64 performance,etc and for it to have a much higher FP32 ratio compared to FP64 performance.
250w? That's incredibly promising for mainstream Volta GPUs in the future.

Not really - it's wider than the old titan, but also slower. This isn't looking like the same step-change in performance we got from pascal compared to maxwell so far, I'd want to see something bigger and faster for the same power before it suggests any great improvement
If I had a use for the tensor cores which is basically the main selling point here then it's a bargain but for anything else it's a ‘lol how much’ response.
It is a professional card, price range is good for people that need this.. I wonder if AMD has something up their sleeve for this move, but one can only speculate.