TITAN V launch strengthens machine learning lead for NVIDIA

on December 14, 2017
Reading Time: 3 minutes

Earlier this week, NVIDIA launched the Titan V graphics card at the NIPS (Neural Information Processing Systems) conference in Long Beach, to the surprise of many in the industry. Though it uses the same Volta architecture based GPU that has been shown and discussed and utilized in the Tesla V100 product line for servers, this marks the first time anything based on this GPU design has been directly available to the consumer.

Which consumer though, is an interesting distinction. With its $3000 price tag, NVIDIA positions the Titan V towards developers and engineers working in the machine learning fields, along with other compute workloads like ray tracing, artificial intelligence, and the oil/gas industry. With the ability to integrate a single graphics card into a workstation PC, small and growing businesses or entrepreneurs will be able to develop applications utilizing the power of the Volta architecture and then deploy them easily on cloud-based systems from Microsoft, Amazon, and others that offer NVIDIA GPU hardware.

Giving developers this opportunity at a significantly reduced price and barrier to entry helps NVIDIA cement its position as the leading provider of silicon and solutions for machine learning and neural net computing. NVIDIA often takes the top down approach to new hardware releases, first offering it at the highest cost to the most demanding customers in the enterprise field, then slowly trickling out additional options for those that are more budget conscience.

In previous years, the NVIDIA “Titan” brand has targeted a mixture of high-end enthusiast PC gamers and budget-minded developers and workstation users. The $2999 MSRP of the new Titan V moves it further into the professional space than the enthusiast one, but there are still some important lessons that we can garner about Volta, and any future GPU architecture from NVIDIA, with the Titan V.

I was recently able to get a hold of a Titan V card and run some gaming and compute applications on it to compare to the previous flagship Titan offerings from NVIDIA and the best AMD and its Radeon brand can offer with the Vega 64. The results show amazing performance in nearly all areas, but especially in the double precision workloads that make up the most complex GPU compute work being done today.

It appears that gamers might have a lot to look forward to with the Volta-based consumer GPU that we should see arriving in 2018. The Titan V is running at moderate clock speeds and with unoptimized gaming drivers but was still able offer performance that was 20% faster than the Titan Xp, the previous king-of-the-hill card from NVIDIA. Even more impressive, the Titan V is often 70-80% faster than the best that AMD is putting out, running modern games at 4K resolution much faster than the Vega 64. Even more impressive, the GV100 GPU on the card is doing this while using significantly less power.

Obviously at $3000, the Titan V isn’t on the list of cards that even the most extreme gamer should consider, but if it is indicative of what to expect going into next year, NVIDIA will likely have another winner on its hands for the growing PC gaming community.

The Titan V is more impressive when we look at workloads like OpenCL-based compute, financial analysis, and scientific processing. In key benchmarks like N-body simulation and matrix multiplies, the NVIDIA Titan V is 5.5x faster than the AMD Radeon RX Vega 64.

Common OpenCL based rendering applications use a hybrid of compute capabilities, but the Titan V is often 2x faster than the Vega graphics solutions.

Not every workload utilizes double precision computing, and those show more modest, but noticeable improvements with the Volta GPU. AMD’s architecture is quite capable in these spaces, offering great performance for the cost.

In general, the NVIDIA Titan V proves that the beat marches on for the graphics giant, as it continues to offer solutions and services that every other company is attempting to catch up to. AMD is moving forward with the Instinct brand for enterprise GPU computing and Intel is getting into the battle with its purchase of Nervana and hiring of GPU designer Raja Koduri last month. 2018 looks like it should be another banner year for the move into machine learning, and I expect its impact on the computing landscape to continue to expand, with NVIDIA leading for the foreseeable future.