Rumor: NVIDIA’s 7nm Ampere coming in the first half of 2020
Nvidia’s 7nm (nanometer) Ampere GPUs have long been in the rumor mill since AMD announced its 7nm Navi GPUs in January 2019. Since then, three different sources have talked about Ampere.
On June 5, DigiTimes cited NVIDIA Korea, which indicated that it would use Samsung’s 7nm node for its new GPUs. Last Friday, German website Igor Lab stated that NVIDIA’s long-rumored next-generation Ampere GPUs should arrive no later than mid-2020.
How reliable are the Ampere rumors?
Let’s take a step back and look at NVIDIA’s GPU roadmap. NVIDIA launches next-generation GPU architecture every two years, typically before AMD.
In the second half of 2018, NVIDIA launched its Turing GPU architecture built on TSMC’s (TSM) 12nm node with a real-time ray-tracing feature. AMD launched its 7nm Navi GPUs in July 2019. These Navi GPUs compete with NVIDIA’s existing Turing GPUs on performance, excluding ray tracing, at a lower price.
AMD’s Navi delivers competitive performance with the help of the 7nm node and RDNA (Radeon DNA) architecture. AMD’s Navi launch forced NVIDIA to launch a Super version of its Turing-based GeForce GPUs at competitive prices. This shows that NVIDIA is bracing itself for a GPU war with AMD in the gaming space.
Now, AMD plans to launch its next-generation RDNA2 architecture in 2020. The Navi 21 and Navi 23 GPUs built on the RDNA2 architecture are expected to have a ray-tracing feature. AMD even dubbed these GPUs “NVIDIA killers.”
With AMD bringing ray tracing to its GPUs, NVIDIA feels the pressure to move quickly. So, the above rumor of NVIDIA Ampere architecture coming in the first half of next year makes logical sense. This is in line with the timeline of NVIDIA and rival AMD’s GPU roadmaps as well.
Can NVIDIA’s Ampere beat AMD’s “NVIDIA Killers”?
According to the Wccftech article, NVIDIA’s Turing GPUs can perform ray tracing at 1080p 30fps (frames per second) for light to moderate path ray-tracing workloads. According to TweakTown, Ampere will have to compete with AMD’s RDNA2-based GPUs, which would deliver 4K 60fps gaming performance as well as some ray tracing. So, we can expect a jump in ray-tracing performance to more than 30fps.
NVIDIA expects to improve performance with the help of the 7nm node and tweaks in new architecture design. If the rumor that Samsung (SSNLF) plans to build NVIDIA’s next GPUs is true, the latter would benefit from the former’s EUV (extreme ultraviolet) process. Currently, there is a lot of ambiguity around this rumor.
Will NVIDIA build Ampere on Samsung’s 7nm node?
In July, the Korea Herald reported that NVIDIA Korea chief Yoo Eung-Joon confirmed rumors that Samsung will build advanced GPUs on its 7nm EUV node. Tom’s Hardware reached out to NVIDIA in July. In response, NVIDIA operations head Debora Shoquist noted, “Recent reports are incorrect – NVIDIA’s next-generation GPU will continue to be produced at TSMC. NVIDIA already uses both TSMC and Samsung for manufacturing, and we plan to use both foundries for our next-generation GPU products.”
This is where the ambiguity begins. The above statement neither confirms nor denies the rumor that the Samsung 7nm node will be used to build Ampere. This means that NVIDIA Korea’s chief is right, as he never stated that all Ampere GPUs will be built on Samsung 7nm.
If NVIDIA plans to use both TSMC and Samsung, it most likely would use TSMC’s 7nm Plus node, as it uses EUV technology just like Samsung. TSMC’s 7nm node uses standard multi-patterning.
In the past, NVIDIA used both foundries for its Pascal GPUs. It built GTX 1050, 1050 TI, and GT 1030 cards on Samsung’s foundry and the rest on TSMC’s foundry. We believe that history could repeat with Ampere. NVIDIA could use TSMC for high-end Ampere GPUs and Samsung for midrange and low-end GPUs.
What does the TSMC 7nm supply shortage mean to NVIDIA?
Recently, there were rumors that TSMC is facing a supply shortage, which resulted in its increased lead time for the 7nm node. This means that any new 7nm products would have to wait for six months instead of the usual two months to be manufactured.
There is a risk that TSMC might give priority to Apple over its other customers. Moreover, AMD switched its complete 7nm production to TSMC this year. In light of these developments, NVIDIA might change its foundry partner. Although NVIDIA can use several permutations and combinations, what matters is that it selects the best combination that also generates high returns.