Highest Training Performance for GPT Now Accessible as-a-Service to Enterprises by SambaNova Systems
Dataflow-as-a-ServiceTM GPT delivers 2.1x faster training, performance and throughput, however, performance doesn't actually matter
PALO ALTO, Calif. February 1st, 2022 – Updated 2/3/22
SambaNova Systems, the company building the industry’s most advanced software, hardware and solutions to run artificial intelligence (AI) applications, today announced that it has the highest performance — and is the fastest — for GPT training.
“Real world test results show that Dataflow-as-a-Service™ (DaaS) GPT training is #1 — it delivers 2.1x faster training performance and throughput than Nvidia. The reality is, ultimately these metrics don’t matter the most. At SambaNova, we measure success by the value we deliver for our customers, who are achieving real, transformational business impact from AI, today” said Marshall Choy, SVP of Product at SambaNova Systems.
“Over the last four years, we’ve seen AI performance go from a niche concern for Silicon Valley giants to a real consideration for enterprises. As the value of data comes to life and AI takes hold in core enterprise operations, some business functions will see performance gains result in giant workflow impacts,” Dan Fagella CEO, Head of Research Emerj Artificial Intelligence Research.
Performance is only one of the many other attributes that are important to SambaNova’s Dataflow-as-a-Service GPT customers, who state that the fragile setups on many “industry” performance benchmarks don’t translate to real-life workloads. SambaNova is consistently faster for the work that matters, while delivering additional high value attributes that meet customer needs. With Dataflow-as-a-Service GPT, customers benefit from:
- Highest ease of use and flexibility, with low code/no code API interfaces
- Rapid time to solution in weeks or months, not years, by eliminating model and infrastructure complexity in a pre-integrated solution
- World record time to train and throughput, 2.1x faster than Nvidia1
“What customers really care about is how powerful deep learning models improve their business and achieve a competitive advantage. They care about how fast AI systems can be deployed, ROI and time to value more than performance metrics, even when you have the highest performance systems like what we offer. Simply put, SambaNova’s DaaS offering delivers what organizations want – the ability to deploy state-of-the-art deep learning capabilities to solve real business problems in weeks, not months or years” said Rodrigo Liang, CEO of SambaNova Systems.
“We are proud to hold the world record for GPT training. Dataflow-as-a-ServiceTM GPT takes value to the next level, delivering rapid time to solution, highest ease of use with low-code interfaces, and best in class accuracy delivered as a service eliminating the need for scarce infrastructure and data science talent,” commented Marshall Choy, SVP of Product at SambaNova Systems.
About SambaNova’s Dataflow-as-a-ServiceTM GPT
SambaNova’s flagship offering, Dataflow-as-a-Service, is an extensible AI services platform, and enables organizations to jump-start AI initiatives overnight by augmenting existing capabilities and staffing with a simple subscription. The platform is powered by DataScale®, an integrated software and hardware platform delivering unrivaled performance, accuracy, scale and ease of use built on SambaNova’s Systems Reconfigurable Dataflow Architecture™.
With AI becoming a business necessity in the global economy, customers need complete solutions that can run at scale in a financially viable way. With an integrated full-stack system, including best-in-class AI models, software and hardware, SambaNova provides the most expansive, accessible and impactful AI applications in the world. To learn more about SambaNova’s Dataflow-as-a-Service™ for language with GPT, visit: https://sambanova.ai/solutions/gpt/
 About the test and Comparison Benchmarks:
Performance and accuracy comparisons are made on the following competitive equivalents:
- SambaNova Dataflow-as-a-Service GPT running on 1 integrated system with 8 RDU, SambaNova GPT 1.5B parameters, global BS=1,024
- Commodity server, 1 Nvidia DGX A100 with 8 A100 GPU, GPT 1.5B parameters, Global BS=1,024
Pre-training dataset: SEC EDGAR
Downstream task fine tuning dataset: FiQA+PhraseBank