What’s in a Name?

Before we get into the details, we must make a note of nomenclature. Ampere is the successor to the Turing microarchitecture, which debuted a processing component that NVIDIA calls an RT core. RT stands for ray tracing, and the headline feature of Turing was its ability to provide real-time ray tracing. That also explains—sort of—the cryptic acronym RTX, used to signify graphics cards with real-time ray tracing features. For more information, read The Turing GPU Architecture and NVIDIA’s RTX Graphics Cards.

Ampere introduced the next generation of RT cores, which, according to NVIDIA, can provide up to 2x the throughput of Turing’s RT cores. Ampere also updated another type of core, called a Tensor Core, purpose-built for machine learning and now on its third generation. Ampere’s Tensor Cores offer 2x the throughput of Turing’s, along with support for two new data formats, TF32 and BFloat16. Finally, Ampere also updated its CUDA cores, NVIDIA’s main graphics processing components, to 2.5x the FP32 throughput of Turing’s CUDA cores.

So in a name like NVIDIA RTX A5000, the RTX signifies that it offers real-time ray tracing, and the A presumably stands for Ampere. But there is one curious omission: the word Quadro, NVIDIA’s long time brand name for professional graphics cards. Make no mistake: these new Ampere cards are the professional version (contrast with the consumer GeForce tier, made for gaming; see What’s the Difference Between GeForce and Quadro Graphics Cards?). So why no más Quadro? Who knows, but I say good riddance.

A Closer Look at the New NVIDIA Ampere Graphics Cards

The desktop GPUs (the NVIDIA RTX A5000 and A4000) double the memory of their predecessors, with the RTX A4000 offering 16GB of GDDR6 memory and the RTX A5000 offering 24GB, both with error-correcting code (ECC) support. It’s interesting to note that these cards are not equipped with the higher bandwidth GDDR6X memory that’s found in the higher-end GeForce RTX cards (for more details, see The GeForce RTX 30 Cards Are Insane. What’s Coming Up for Quadro?). The only pro Ampere card released prior to today’s announcement, the NVIDIA RTX A6000, also uses GDDR6 memory (48GB of it) rather than GDDR6X. Perhaps we’ll see the faster memory in the as-yet-unannounced top end Quadros—er, I mean, NVIDIA RTXs. The new desktop graphics cards are also equipped with PCIe Gen 4 to double the data bandwidth of the previous generation.

The laptop GPUs (confusingly sharing the names of their desktop counterparts) use the latest version of NVIDIA’s Max-Q technology, a special design that accommodates the size and noise targets of thin and light laptops. In addition to the new Ampere GPUs, NVIDIA also announced two new Turing-based laptop GPUs, the NVIDIA T1200 and T600, which will serve as entry-level alternatives to integrated graphics.

Finally, the new data center GPUs promise to boost virtual workstation performance. NVIDIA claims the A10 will offer 2.5x the performance of the previous generation and the A16 will provide 2x the user density for virtual desktop infrastructure (VDI). NVIDIA is clearly excited about these and related solutions; Huang referred to the data center as “the new unit of computing” in his keynote this morning.

 The new Ampere desktop and data center GPUs will be available later this month, and the new laptop GPUs are expected to ship in mobile workstations starting in Q2.

Tech Specs of the New NVIDIA Ampere Graphics Cards

Here are the specs NVIDIA has released for the new desktop and laptop cards:

Specs for the NVIDIA A10 and A16 are less consistent, but here’s what NVIDIA has provided for each:

In its announcement of the new graphics cards, NVIDIA made sure to emphasize how the new cards will facilitate a new way of working.

“Hybrid work is the new normal,” commented Bob Pette, vice president of Professional Visualization at NVIDIA. “RTX GPUs, based on the NVIDIA Ampere architecture, provide the performance for demanding workloads from any device so people can be productive from wherever they need to work.”

Tests of the Ampere-based NVIDIA RTX A6000 by workstation provider Puget Systems showed it to be "the fastest professional graphics card they have ever tested," as reported by Tom's Hardware. Though none of the newly announced desktop cards exceed the specs of the A6000, it seems likely that they'll offer similar improvements over the previous Quadro cards. Early customer testimonials provided by NVIDIA bear this out.

“This new NVIDIA Ampere GPU microarchitecture is a huge leap forward," praised Robert Cervellione of architecture and consulting firm Woods Bagot. "I was shocked by the performance gain I saw with the NVIDIA RTX A5000 versus the equivalent model previous generation RTX when running applications such as Chaos Vantage and NVIDIA Omniverse. It’s clear the NVIDIA RTX A5000 professional GPU was designed for this intensive multi-tasking compared to game level cards.”

Erick Green of all-terrain vehicle manufacturer Polaris added: “Speed is everything when we need to evaluate new concepts for the most adventurous vehicles, and the NVIDIA RTX A5000 really delivers what we need. The basic viewport rendering is incredibly fast in Octane Render—5x faster—and unlocks things we couldn’t have even tried before." 

What are your thoughts on the new Ampere-based graphics cards? We're looking forward to putting the new cards to the test here on engineering.com. Stay tuned for our findings.