Ampere is the successor to Nvidia's Turing line of RTX 2000 series graphics cards and will power the next generation 3000 series. These new cards will use a new, smaller process node that will bring Nvidia cards into line with the AMD RX 5000 series and improve performance and efficiency across the product stack by significant margins.
Many details about the cards remain under wraps, but the closer we get to the expected release date, the more details become known, which gives us a good idea of what to expect from the next-generation Nvidia GPU line.
Pricing and availability
Ampere was originally supposed to debut in the first half of 2020, but Nvidia later made it clear that it would appear later in the year. We have known since June that pre-production cards were issued to researchers and cloud service providers such as Google Cloud. Playing cards don't have a fixed release date, but since AMD's next-generation RDNA2 cards are due to be released before the new generation of the console is released (somewhere between October and December), it wouldn't be a surprise if Nvidia hits the market earlier, perhaps between August and September.
Pricing remains under wraps too, and will likely do so until Nvidia makes an official announcement about models in an unveiling of the first amp GPU series. The only amp GPU that we know is currently in the wild is the AI-focused DGX A100, priced at $ 200,000 – and gives us no clues as to what to expect from mainstream cards .
However, AMD's next generation RDNA2 GPUs are expected to compete directly with some of Nvidia's most powerful graphics cards for the first time in three years. This could mean that the green team sees more competitive prices than Turing. As is known, prices for the entire product stack have been increased, which has given players a much more expensive upgrade path – especially in the midst of the 2018 cryptocurrency mining madness.
While it is possible that using a new process node for amp card production could offer some savings in manufacturing costs that reduce costs for buyers, amp GPUs are reported to be very large. This reduces the yield figures for each batch produced and may therefore increase costs.
Nvidia's ampere is based on a brand new process node that comes from a 12nm TSMC design for the RTX series. What it has to do, however, remains to be determined. Initial reports said that it would be a 7 nm EUV process from Samsung. Recent rumors point to the 7 nm node from TMSC. However, the latest rumors from June 2020 point to Samsung's 8 nm node instead. Whichever is used, it should improve performance and efficiency, with Nvidia's own design changes adding new features and further improving performance.
Ampere will support a third generation of tensor cores that unlock faster and more powerful deep learning super samples and possibly additional AI-driven functions. It will support the second generation of high bandwidth HBM2 memory (the cost of which could drive up the price of some cards) and GDDR6X, which should offer a larger memory bandwidth with reduced power requirements than GDDR6.
The new architecture will also support PCIExpress 4.0.
Despite its improvements and changes, Ampere does not appear to have reduced the size of any GPUs. The amp-based A100 accelerator uses the largest chip of all newer graphics cards, although it is denser, with much higher transistor numbers than anything we've seen before. Power consumption is also higher, suggesting that Nvidia's new cards may be more power hungry than their predecessors, despite a more efficient design.
Nvidia Super graphics cards have increased performance in 2019, but Ampere should offer a lot more. Dan Baker / Digital Trends
With AMD's rumors of large Navi 20 graphics cards that are expected to deliver world-class gaming performance at the highest level and bring red team-powered ray tracing capabilities to the market, Nvidia is expected to further improve performance, though Ampere will keep its market leader and top performance. Apart from the improvements made with process node reduction, it has been shown that Ampere is able to use a far larger number of streaming multiprocessors (and therefore CUDA cores) than its predecessors. The A100 Amp AI GPU has almost 7,000 of them – a huge increase of almost 2,000 CUDA cores over everything we've seen so far, even for business cards.
With a new generation of memory and supposedly up to 24 GB of it in the top models, this could mean a significant performance improvement over the best Turing cards. In a HardwareLeaks report, an unannounced Nvidia Ampere GPU scored over 18,000 points in a 3DMark Time Spy run. It was more than 30% faster than a standard 2080 Ti and only slightly behind the heavily modified and overclocked Kingpin edition of the EVGA RTX back 2080 Ti.
Below in the product line, WCCFTech has put together some rumors that the new card generation is taking a step up the performance ladder and is likely to maintain similar, if not improved, prices. The new generation RTX 3080 is said to have the same number of CUDA cores as the 2080 Ti and slightly less new generation memory for comparable, if not slightly improved performance.
The 3070 Ti is said to be slightly more powerful than the standard 2080, perhaps more similar to the 2080 Super, with the Standard 3070 replacing the RTX 2080.
All of this is highly speculative until we know more, but it suggests that the Turing generation of amps for overall gaming performance will be more similar in terms of performance compared to its predecessor than the bigger leap in performance we've seen with 10th generation Pascal cards.
Given the larger chips and the potential additional power consumption of the new generation cards, it is not unlikely that Nvidia will have to improve the cooling of the Founders Edition. While new designs are constantly being tested internally, one has surfaced in some leaked images suggesting that Nvidia could radically revise its cooling solution for ampere GPUs.
These coolers, shown in a production plant, are said to be connected to the RTX 3080, which indicates that they are used at least for the high-end amp GPUs. They feature a unique cooling and circuit board design, with a fan attached to the end of the card's PCIe bracket to draw air up and drive over the jacketed heat sink in the middle of the card. The circuit board ends in a concave V shape so that a secondary fan can line up with the circuit board and let the hot air out over the card.
This is certainly unique, especially for an internal Nvidia design, and raises many questions about its potential and whether we will see anything similar from third parties in the future. Given the usual layout of modern slot machines, this would help vent the hot air instead of blowing it out of the (horizontal) "top" and "bottom" of the card, as is the case with many multi-fan designs today.
Although this alleged Nvidia design is not confirmed, the only RTX 3000 series card we've seen is an Asus 3080 Ti model in a Videocardz report that uses a much more traditional three-fan cooling solution.
It is possible that the fans have a higher performance or hide a denser arrangement of heat sink fins underneath to cope with the additional heat emission. However, it appears that a unique cooling solution for amps is not mandatory. However, if a more robust solution is required, this can increase material costs and the price for higher-level GPUs.
What about ray tracing?
As for performance, we don't know exactly what Ampere will do with ray tracing, but it seems almost certain that Nvidia will continue to support the new lighting technology for its new generation cards. This effectively made ray tracing part of the modern gaming conversation with Turing, even if only a few games support it almost two years after the release of these cards.
A rumor surfaced in May 2020 that Ampere will quadruple the ray tracing performance across the GPU range, effectively eliminating any kind of bottlenecks when the feature was activated. This could allow top-tier cards to play games at 4K resolution and reasonable frame rates with ray tracing enabled, and low-end cards can take advantage of this without sacrificing higher frame rates at less intense resolutions.
This would make a big contribution to making ray tracing more widely available to gamers.