The crown for the fastest GPU is constantly changing hands, but right now it belongs to Nvidia’s Ampere A100 GPU, which posted some impressive numbers this week in OctaneRender’s OctaneBench software.
Word comes via Jules Urbach, CEO of OctaneRender’s parent company, OTOY. According to Urbach, who wrote about the achievement via Twitter, The A100 has posted a score of 446 in OctaneBench. This result comes with RTX turned off on the datacenter-focused GPU, so it’s possible that it could climb higher. To put it in context, here are the scores of a few Turing-based cards:
- Nvidia Titan V: 401 points (11%)
- Quadro RTX 8000: 328 points (35%)
- GeForce RTX 2080 Ti: 302 points (48%)
The Ampere A100 is a beast of a card with 6,912 CUDA cores, 432 Tensor cores, 40 GB of HBM2e memory, and a bandwidth of up to 1,555 Gbps, so it’s no surprise that it dwarfs a consumer card like the 2080 Ti. It’s also likely the gains of the 30-series Ampere cards will be smaller than that of the A100, which is meant to be doing much higher workloads for more clients.
Comparing the A100 to other cards
Another big question mark with regard to these numbers comes from Nvidia’s competition, and therein lies a complication for these numbers. As VideoCardz notes, OctaneBench only runs on Nvidia cards. AMD’s upcoming “Big Navi” card will reportedly pull some impressive numbers, and it’s anyone’s guess where Intel’s Xe architecture sits on the performance scale. But OctaneBench won’t be the source to go to for measuring performance on these GPUs. It’s just that Nvidia has been the undisputed champ for so long that these numbers work as comparisons for now.
Benchmarks, of course, are just that–benchmarks. They’re not real-world numbers, and they don’t always reflect what we’ll see when we apply consumer-grade cards to consumer tasks like playing games. Nvidia’s brand new top-shelf card, it turns out, is top-shelf, and this may be indicative of what we can expect once we can see Ampere in real-world situations. But for now, it’s a guess.