Not lengthy after Nvidia produced the substantial-effectiveness gaming card – the GTX 1080 Ti – the rumour mill whirred into action for the upcoming-generation GPU, codenamed Volta.
Those people rumours began a few of yrs in the past and due to the fact then Nvidia officially unveiled the to start with card primarily based on the new engineering, the Tesla V100.
Then, at the close of 2017, it released the Titan V, a $3,000 workstation graphics card making use of much the exact same GPU as the V100. Neither card is interesting to gamers, but they do issue the way in direction of customer cards primarily based on Volta. No doubt these will be with us just before as well lengthy.
Considering the fact that AMD’s RX Vega cards unsuccessful to beat even Nvidia’s GTX 1080 (enable by itself the Ti edition), it could be that Nvidia is in no hurry to start Volta cards to consumers.
If you might be searching to acquire new card right now, check out our roundup of the very best graphics cards.
What is Nvidia Volta?
Volta is the title for the microarchitecture – a design and style which will be utilised as the foundation for the upcoming generation of customer graphics cards (as properly as the workstation and enterprise things).
The present-day generation, like the Titan Xp and GTX 1080-1050, are primarily based on Pascal architecture. Prior to that, it was Maxwell.
Initially, the strategy was to go to Volta after Maxwell, but for what ever reason Pascal has arrived in the interim.
So when will customer Volta cards arrive?
That is a problem with no definitive response right now. Some rumours pointed to a 2017 start, but the sensible money was always on a 2018 start, partly due to the fact of responses by memory maker SK Hynix (see below).
The Tesla V100 datacentre card takes advantage of the GV100 GPU, and it is probable the customer graphics cards will have a GV104 chip. Once again, there are rumours that they’ll be referred to as the GTX 2080, 2070, 206 etc. This would be simpler to comprehend than incrementing by one hundred as ranges have completed for the last several yrs.
In any other case the successor to the GTX 1080 would be the GTX 1180. We are going to have to hold out and see, nevertheless.
What about effectiveness and specs?
It is way as well early to say if Volta will deliver the kind of sizeable effectiveness gains that Pascal did over Maxwell.
What is a little clearer is the course of action the GPU will use. Intel is relocating to 10nm for Cannon Lake, and that was the strategy for Volta. However, shrinking the transistors is challenging (component of the reason why Intel is stuck with 14nm for 4 full generations of Core processors) and the present Volta chips ended up making use of a 12nm course of action from TSMC – the business that can make the chips for Nvidia.
The Tesla V100 and Titan V cards equally use HBM – substantial-bandwidth memory – just like AMD’s Vega cards. Earlier customer cards from Nvidia have gone with common GDDR RAM and even though it would be nice if Nvidia kept the technically-greater HBM for Volta-primarily based cards, it could not materialize.
Which is due to the fact memory maker SK Hynix published a press launch in April 2017 which “introduced the world’s swiftest 2Znm 8Gb(Gigabit) GDDR6(Graphics DDR6) DRAM. The solution operates with an I/O facts level of 16Gbps(Gigabits per second) per pin, which is the industry’s swiftest. With a forthcoming substantial-close graphics card of 384-bit I/Os, this DRAM processes up to 768GB(Gigabytes) of graphics facts per second.”
It did not specify which maker that graphics card was coming from, but with AMD’s Vega utilizing HBM2 memory and the press launch also stating that “SK Hynix has been arranging to mass deliver the solution for a client to launch substantial-close graphics card by early 2018”, you’d think it was the upcoming-gen GeForce cards.
The Titan Xp with 12GB of GDDR5X has a memory bandwidth of 547.7GB/s, so it is interesting that SK Hynix’s GDDR6 will run at 16Gb/s and – with a 384-bit interface card – will give up to 768GB/s. That is really a jump: the Tesla V100 – with HBM2 – has a bandwidth of 900GB/s, so a putative Volta-primarily based Titan with GDDR6 wouldn’t be as well significantly driving.
The new Titan V – for context – offers 652.8 GB/s of bandwidth.
We’ll be pursuing the enhancement of Volta intently, and updating this report when any new rumours (and points) about the customer cards floor, so maintain examining back again.