Nvidia Ada Doubles GPU Performance

Nvidia’s newest gaming GPU, code-named Ada Lovelace, builds on the Ampere design with faster clock speeds, far more cores, a massive cache memory, and faster ray tracing.

Linley Gwennap

Tapping a big transistor improvement, Nvidia’s new Ada Lovelace GPU delivers more, more, more for leading-edge gamers. Powering the new RTX 4090 graphics card, the chip has more cores, more ray-tracing units, more clock speed, and a much bigger cache memory. The result is more than twice the peak performance.

Ada relies on the same basic core design as the previous Ampere generation while adding new features to the tensor unit and ray-tracing engine. But the shrink from 8nm to 4nm allowed Nvidia to raise the core count 50% to 128, comparing the 4090 with the earlier 3090, while the boost clock also jumps 50% to 2.5GHz. Together, these two changes yield 2.3x higher peak teraflops per second (teraflop/s). The smaller transistors additionally help the design pack 72MB of cache, 12x more than the previous generation. Power jumps from 350W to 450W, although the older 3090 Ti version draws 450W as well.

In addition, Nvidia announced two RTX 4080 cards that feature smaller configura¬tions of the Ada design. Compared with the older RTX 3080, they offer 30–60% performance gains while raising the list price by similar amounts. Of course, GPUs rarely sell for their list price these days, but the hikes indicate the company is trying to lock in higher prices for mainstream customers. The RTX 4090 went on sale this week, while the RTX 4080 cards are due in November. The company also disclosed an Ada-based RTX 6000 card for the professional market.