The Nvidia Volta graphics cards are set to house the next generation of GPUs from team GeForce and, if the gaming cards deliver the same performance boost as the monstrous professional GPUs, the new tech is going to be a little bit special. They follow the mighty GTX 1080 Ti, and its Pascal-based brethren, and here's everything we know about them so far.
Check out our pick of the best graphics cards around today.
Nvidia Volta specs
The professional-class Volta – the Tesla V100 – uses TSMC's 12nm FinFET design and the full GV100 GPU has 5,376 CUDA cores. Imagine a GTX 2080 Ti with that.
Nvidia Volta price
Don't shoot the messenger, but it's possible the consumer Volta cards will push
prices up again. $699 for the GTX 2080? Not beyond the realms of possibility...
Nvidia Volta performance
We're too early in the release cycle for there to be any performance figures flying around, but we're hoping for greater efficiency and better with DX12 and Vulkan.
The Nvidia Volta graphics architecture is the next big thing coming out of the green team’s GPU skunkworks and has had its first professional-level, ultra-expensive processor, the V100 officially unveiled. Volta is the silicon successor to the Pascal generation of graphics cards, the generation that brought us the mighty GTX 1080 Ti and Titan Xp, and is gearing up to deliver the GTX 1070 Ti soon. But can it deliver the same generational performance boost offered by Nvidia's impressive last-gen graphics silicon?
If the gaming performance of the Volta GPUs we'll end up seeing in our PCs next year provide anything like the performance delta the first Tesla V100 silicon is displaying against the last-gen professional Pascal cards then we're in for a treat. Volta, in the pro space, is out-performing Pascal by 132%.
As is Nvidia’s wont the new GPU architecture is taking its name from a famous historical scientist. Alessandro Volta gave his name to the Volt having been a pioneer of electrical energy and its storage. He was also the discoverer of butt-gas - fun little science fact for you there.
The existence of the Volta design was first unveiled, at least in theoretical form, at Nvidia’s Graphics Technology Conference way back in 2013. It was originally meant to be the GPU silicon which followed directly on from the Maxwell architecture (which made up the GTX 900-series of graphics cards), but a year later up pops the Pascal design used in the most recent 10-series GeForce parts, pushing the prospective Volta chips further back.
But Volta now seems to be back on track, with devices appearing in drivers, and in vague rumours from Chinese websites, so let’s delve into what it is, what it’s going to mean for our gaming PCs and when you can get one jammed into your rig.
This is probably the most important question of all about the upcoming Nvidia Volta architecture: when is it going to be launched? Nvidia themselves have been relatively tight-lipped about the whole thing, apart from appearing on oft-changing GPU roadmaps, but we’re starting to see more wider mentions of the technology, whether implicit or specific.
Early 2018 is still probably the most realistic release date for the consumer version of the Volta GPU, despite the professional GPUs being delivered this year.
In February the makers of benchmarking tool AIDA64 found reference to the top-end Volta GPU, the GV100, in some new Nvidia drivers. According to a post on Videocardz the Finalwire programmers were trying to unearth some traces of the then-unlaunched GTX 1080 Ti, but instead found reference to the full-fat Volta GPU, the GV100.
We heard a bit more about Volta at Nvidia’s Graphics Technology Conference in May and the pro Volta cards have already started shipping to the lucky few with cash enough to spend on the advanced new accelerator boards. Jen-Hsun took to the stage at the introduction event waving around the Volta silicon, proclaiming the giant GPU a masterpiece. He doesn't look far off the mark to be honest...
But that’s the top-end, professional-level chip being built for supercomputers like the Summit machine likely to gain sentience at Oak Ridge in 2018 and doom us all. The GV104 silicon, however, is likely to be the consumer-facing version of Volta which makes its way into our graphics cards, and is also probably what SK Hynix were referring to when they announced the new GDDR6 memory standard.
When they announced the arrival of GDDR6 Hynix mentioned they were “planning to mass produce the product for a client to release high-end graphics card [sic] by early 2018 equipped with high performance GDDR6 DRAMs.” I’m not the only one to think that is a reference to Nvidia and the release of GDDR6-based Volta GPUs.
Team red are set to be refreshing their AMD RX Vega GPUs in the new year, but the new 12nm design is still likely to be sticking with the same HBM2 memory standard they've been using up to now. It would likely take too much of a redesign to backtrack on the rare, expensive memory tech now.
There have been rumours from Chinese site, MyDrivers.com, claiming Nvidia are planning to release the GeForce 20-series, widely expected to be the nomenclature used for Volta cards, in the third quarter of this year. That would mean consumer-ready Volta GPUs hitting the shelves by September this year, which would be a pretty speedy turnaround as they only started talking about the full-fat professional GV100 core in May with those shipping in Q3 at the earliest.
I'm sticking with 2018 for now, because it's looking increasingly like those rumours linked into the Pascal refresh finding form in the upcoming GTX 1070 Ti.
There have been many figures thrown around as to just what transistor lithography Nvidia will be jamming into their next-gen Volta graphics cards, with the latest being a return to the Taiwan Semiconductor Manufacturing Company (TSMC) for the manufacturing of 12nm Volta GPUs, having previously dallied with Samsung for their 14nm GP107 silicon.
The Tesla V100 Volta chip is sporting the 12nm FinFET design from TSMC so there's no reason to think it'll be any different for the consumer releases.
TSMC have also said their entering volume production of their 12nm FinFET chips in the fourth quarter of this year. That plays into the idea of a full launch of Nvidia's consumer-facing Volta cards relatively early in 2018.
Originally Volta was supposed to be built using TSMC's new 10nm process, but the pace of transistor shrinkage has become rather laggardly in recent years. Other rumours had Nvidia sticking with TSMC’s existing 16nm tech in order to be able to stick to their roadmap and get actual Volta cards on the shelves in 2018. It's now looking like somewhere between the two.
TSMC’s 10nm process, unlike Intel’s Cannonlake design, is expected to be more of a stop-gap measure between their current process and the almost mythical 7nm lithography. To put that into perspective 7nm is about the height of three Tom Cruises standing on each other’s shoulders. But getting down to 10nm seems to be more of a challenge than maybe even they expected. I’m guessing either the price is prohibitive or the yields too low to offset the performance benefits of a newer design. Or a combination of both.
TSMC have though announced this stop-gap measure, which they’re nominally calling 12nm. It’s apparently based on their existing 16nm design, but with density, performance and energy efficiency improvements. Whether this 12nm node will genuinely be packed with 12nm transistors, or whether it’s just going to be clever marketing, is about as clear as thermal paste.
The current thinking then is Nvidia have now taken this updated 12nm design to use in their Volta GPUs to augment the new architecture’s own performance and power improvements with the slight extra performance bump of a shrunken lithography. With it largely being based on the existing 16nm node there shouldn’t be any concerns about launching a brand new GPU architecture with a brand new lithography. Nvidia have been burnt by those technological flames before… anyone remember the super-heated GTX 480?
We also know the top Volta GPUs will be running HBM2 memory, as detailed by Oak Ridge’s Summit specs and subsequently in Jen-Hsun's introduction of the architecture in May. The GV100 is the super-star GPU, sporting both the NVLink interface - instead of PCIe 3.0 - along with the second-gen stacked high-bandwidth memory design. The V100 chip comes with 16GB of HBM2 capable of an incredible 900GB/s of memory bandwidth.
It’s possible the HBM2 memory will feature in the cut-down, PCIe-based GV102 version built for the Quadro crowd and later the Volta Titan cards. The strategy for the current Pascal generation, however, has been to run both HBM2 and the NVLink interconnect on the Tesla accelerators with a switch to GDDR5X and PCIe 3.0 for the less-expensive Quadro and consumer variants. I doubt that will change for Volta, with the GV102 parts probably sporting the new GDDR6 technology.
The first consumer-focused Volta GPUs, however, will be the GV104 chips, potentially set to go into the GTX 2080 and GTX 2070 cards. These would be built using GDDR6, allowing Nvidia to pump large capacity, high-performance frame buffers into their cards without the extra expense of HBM2. GDDR6 isn’t as pacey as HBM2 but is definitely speedier than GDDR5X, offering a data rate of 16Gbps as opposed to the former’s 14Gbps top speed. That also means it comes with a chunk of extra bandwidth too.
Hynix have stated that with a 384-bit memory bus, the sort of design usually favoured by Nvidia’s high-end graphics cards, they are able to offer memory bandwidth of up to 768GB/s, which isn't far off the 900GB/s of the V100's HBM2 design. The new Titan Xp is using GDDR5X and can only manage 548GB/s with its mighty 12GB setup.
With GDDR6 running so quickly it would be unlikely Nvidia would bother with HBM2 for the consumer cards.
In terms of the actual Volta GPU architecture itself there is not a lot to say until we get a little more detail from Nvidia. What their engineers do need to do, however, is to target improved performance from the newer low-level APIs. Compared with AMD’s fourth-gen GCN architecture, used in the Polaris cards, Nvidia’s Pascal GPUs are generally off-the-pace when it comes to Vulkan and DirectX 12 gaming speed.
And now for salty Volta rumour time, people... an image was pulled from Facebook, by an eagle-eyed Redditor, purporting to be of a Volta-based Titan prototype.
The picture was posted to the page of someone claiming to have taken the image from a summer intern for Nvidia (complete with their photo ID card) with them stating in the comments that it was a potential Nvidia Titan X Volta card. Is this a Titan Xv, or an intern misunderstanding what they were actually seeing?
I'm inclined to believe it's the latter because, from everything we've seen recently with the Tesla V100 unveiling, it doesn't look like Volta is anywhere near a consumer release. If Nvidia does decide to lead with a Volta-based Titan as the vanguard for their new generation of consumer graphics cards I'll be seriously surprised.
The card in the picture looks like a Quadro professional card, with the same NVLink connections on the top-side of the board as the Quadro GP100. To me that would suggest this might have actually been a prototype for the original Pascal-based Titan X, used to test a potential HBM2-powered variant. The old Titan X had a modified GPU inside it, which had the same number of CUDA cores as the Quadro card, though Nvidia never released a consumer-facing version sporting the super-expensive HBM2 memory.
There's also a separate GPU sitting in the pictured machine's primary PCIe slot, which might indicate another card was being used for video output rather than this Titan prototype.
Oh hai, welcome to price speculation corner. Obviously we don’t know how Nvidia are going to price their new cards, much of that will probably come down to their relative performance compared with AMD’s high-end Vega parts, but we can still make some educated guesses based on Nvidia passim.
There was, however, a worrisome section in the MyDrivers.com piece about a rumoured accelerated release for Volta. In it they suggest the GeForce 20-series of products will see re-planning of the prices and market positions which “will increase the price of single card to enhance profits.” My understanding of Chinese is pretty woeful (read: nonexistent) so I’m relying on Mr. Google Robot’s comprehension skills, so it’s possible there’s a misunderstanding born of the translation into English.
But if true, and the prices of single cards are being adjusted upwards, that would be the second generation in a row where Nvidia have pushed prices skywards. With the 10-series Pascal-based cards the GTX 1080 was released at an unprecedented level, especially when taking the reference/Founders Edition shenanigans into account.
If $699 is going to become the de facto standard for Nvidia’s high-end graphics cards, as shown with the GTX 1080 Ti cards, then these are worrying times. Unless you’re AMD and confident you can keep undercutting Nvidia’s GPUs with your own refreshed Vega graphics cards. Gotta get them built in volume first guys...
Although we’ve seen reference to the GV100 Volta GPU in Nvidia’s graphics drivers, meaning there must be testing going on right now, there haven’t been a lot of interesting and/or apocryphal benchmarks leaking out to whet our appetites. That’s probably down to the fact the GV100 is going to be tested on applications which likely bare little relevance to any of the performance benchmarks we might recognise.
That said, in a pure generation-on-generation test between the pro-level Pascal P100 and Volta V100 GPUs the newer Nvidia architecture is posting performance that is 132% faster than the last-gen chips. If we get anywhere near that level of performance boost from gaming applications Volta will be stunning. Realistically that's unlikely because these tests are based on Geekbench benchmarks, run in a Linux environment, using the specific CUDA API, and not Shadow of War at 4K.
The professional Volta cards have been tuned especially for AI workloads, with a new Tensor core design at its heart. These new silicon slices won't have anything to do with gaming, at least for the moment, so if those are the only real improvements with Volta it's possible there won't be a huge performance uplift over Pascal at all.
Obviously we’re still expecting some increased gaming performance from Volta, and I’ve already spoken about the necessity for it to better deal with the low-level APIs of Vulkan and DirectX 12, but we should also expect some improved efficiency born both of the slightly shrunken GPU production process as well as the redesigned architecture itself.
Given that Volta is the namesake of the Italian gentleman credited with the invention of the battery you’d certainly hope for some improved efficiency. On the notebook side Pascal made great strides forward for the performance of mobile GPUs, and Volta ought to carry that progress even further.
When he announced the existence of the Volta code-name at GTC 2013 Jen-Hsun Huang explained that "I love that name Volta because it would suggest that it will be even more energy efficient."
Fingers crossed it’s not just a suggestion…
This page is regularly updated with all the latest news, rumours and information about the Nvidia Volta GPU architecture. Bookmark away and let us know your thoughts about it in the comments below.
Click on the quick links to get back up to the section of your choice.