Nvidia to become TSMC's first customer for A16 technology
TSMC's A16 process is less interesting for Apple. Nvidia is therefore the obvious first major customer.
(Image: Carsten Spille / heise medien)
Nvidia's next but one GPU generation Feynman could be released with the advanced production technology A16 from the world's largest chip contract manufacturer TSMC. Nvidia is said to be considering its use, also to gain a technical advantage over AMD. CTEE reports on this from Taiwan.
TSMC plans to start series production of A16 chips towards the end of 2026. Nvidia has scheduled the Feynman generation for 2028 and could possibly ship samples as early as 2027. Rubin and Rubin Ultra will first appear in series production in 2026 and 2027.
Rear power supply
A16 combines TSMC's 2-nanometer production (N2) with a backside chip power supply, called Backside Power Delivery or Super Power Rails. While the transistor structure remains largely identical to N2, the metal layers for powering the transistors move from the top to the bottom. This untangles the data paths on the top side and, according to TSMC, significantly improves the electrical characteristics, close to a standard generation leap.
Also practical for Nvidia: TSMC is building increasingly complex constructions from several chips to increase performance.
Videos by heise
Economically unexciting for smartphones
Apple is usually the first major customer for chips with TSMC's latest production technology, followed by Mediatek. However, backside power delivery is less interesting for mobile chips due to the higher costs and complex structure. Apple & Co. could therefore jump directly from N2 to A14 without backside power supply and then be faster than Nvidia again.
Nvidia certainly has the money for state-of-the-art production technology. In the AI environment, Nvidia can raise prices far enough to keep the gross margin above 70 percent.
Chip contract manufacturers themselves never make the prices for silicon wafers and therefore chips public. According to rumors, the cost of a single wafer will rise to 30,000 US dollars with the N2 generation. Nvidia's Blackwell AI accelerators contain two GPUs measuring a good 800 mm², of which not even 100 fit on a wafer with a diameter of 300 mm.
For GeForce graphics cards, the switch to A16 seems rather unlikely due to the costs. Nvidia could therefore further split development between gaming and AI chips.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(mma)
