Editorial illustration for Nvidia breaks MLPerf records with 288 GPUs as AMD, Intel pursue other goals
Nvidia Shatters MLPerf Records with 288-GPU Powerhouse
Nvidia breaks MLPerf records with 288 GPUs as AMD, Intel pursue other goals
Nvidia posted its latest MLPerf results, and the numbers are predictably huge. They used 288 of their newest GPUs to set records. The company is playing a game of pure scale, and right now, no one else has enough pieces to sit at that table.
AMD and Intel are in the room, but they brought their own games. AMD's submissions focus on percentage comparisons against Nvidia's top-tier B200 and B300 chips. That's the closest thing we have to a direct fight.
But those comparisons only exist for the specific models AMD chose to run. They didn't try to match Nvidia's 288-GPU cluster. Intel, for its part, is chasing different goals entirely.
Nvidia puts its cumulative MLPerf wins since 2018 at 291 - nine times more than all other submitters combined.
This gets to the core of the problem. Nvidia's 2.7x gain came from better software on the same hardware. AMD's 3.1x jump came from a new chip design.
They are measuring entirely different kinds of progress. So who won? It depends on what you think matters more.
Nvidia is now pushing for a new benchmark called MLPerf Endpoints. It would measure real-world API performance, a step closer to what customers actually experience. The promise is a common ruler.
The awkward detail is that Nvidia itself is leading the charge to define this new ruler inside the industry consortium. The company that dominates the current game is also writing the rules for the next one.
Related Reading
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic NVIDIA NeMo powers telco reasoning model for autonomous network workflows
Industry Applications
Related Topic China offers cheaper electricity to AI firms abandoning NVIDIA chips
Business & Startups
From Archives NVIDIA's 288-GPU Blackwell Ultra Sets New MLPerf Inference Throughput Record
From Archives DeepMind study finds six traps that let a few poisoned docs hijack AI agents
From Archives Nvidia-backed ThinkLabs raises USD 28M, shows AI power‑grid results with SCE
Business & Startups
From Archives Nvidia's 3B Nemotron-Cascade 2 wins math and coding gold; recipe open‑source
Open Source
Common Questions Answered
How many GPUs did Nvidia use to break MLPerf records?
Nvidia used a 288-GPU cluster to set new performance benchmarks in MLPerf Inference v6.0. This massive GPU configuration allowed them to demonstrate unprecedented scaling and performance across multiple models, including multimodal and video models.
What makes the MLPerf Inference v6.0 benchmark unique compared to previous versions?
MLPerf Inference v6.0 is the first benchmark to include multimodal and video models, expanding the scope of performance testing beyond traditional AI workloads. This new inclusion provides a more comprehensive view of AI computational capabilities across different types of AI models and processing requirements.
Why is a direct performance comparison between Nvidia, AMD, and Intel challenging in this MLPerf benchmark?
Each company highlighted different metrics and submitted varying models, making a straightforward performance comparison impossible. For instance, Nvidia's 2.7x software improvement differs fundamentally from AMD's 3.1x generational leap, which represents a new chip architecture rather than pure software optimization.
Further Reading
- NVIDIA Extreme Co-Design Delivers New MLPerf Inference Records — NVIDIA Developer Blog
- NVIDIA Sets MLPerf Inference v6.0 Records with Blackwell Ultra Platform — StorageReview
- NVIDIA Blackwell Ultra GPUs Crush MLPerf Benchmarks with 2.7x Performance Gains — Blockchair
- AMD Delivers Breakthrough MLPerf Inference 6.0 Results — AMD Blogs
