AI Benchmark Battle: NVIDIA vs AMD in MLPerf v5.1 Results

NVIDIA Blackwell Ultra Shatters AI Performance Records in MLPerf v5.1 Benchmarks.NVIDIA’s next-generation Blackwell Ultra AI accelerator has decisively topped the latest industry benchmarks. The GB300 chipset demonstrated staggering performance gains in the newly released MLPerf Inference v5.1 results. AMD’s competing Instinct MI355X also posted strong showings, signaling intensified competition in the high-stakes AI hardware sector.The MLCommons consortium published its latest benchmark results today. These tests provide a critical measure of AI processing power for data centers.

NVIDIA Blackwell Ultra GB300 Delivers Massive Performance Leap

NVIDIA’s Blackwell Ultra GB300 showcased exceptional capabilities in the DeepSeek R1 benchmark. It achieved a 45% performance gain over the previous GB200 in a 72-GPU configuration. An 8-GPU comparison showed a similar 44% uplift, nearly matching NVIDIA’s promised 50% improvement target for the new architecture.In server-oriented DeepSeek R1 testing, the Blackwell Ultra maintained impressive advantages. It recorded a 25% gain in the 72-GPU category and a 21% improvement in 8-GPU submissions. These results validate NVIDIA’s architectural enhancements for real-world AI workloads.According to MLCommons data, the Blackwell Ultra platform set numerous per-accelerator records across multiple benchmarks. It achieved 5,842 tokens/second/GPU in DeepSeek-R1 offline processing and 224 tokens/second/GPU in Llama 3.1 405B offline testing. The chipset demonstrated particular strength in reasoning tasks, outperforming the previous Hopper architecture by 4.7x in offline and 5.2x in server comparisons.

AMD Instinct MI355X and Intel Arc Pro B60 Show Competitive Results

AMD’s Instinct MI355X made a strong showing in the Llama 3.1 405B benchmark. The platform delivered a solid 27% performance increase compared to previous generation hardware. This demonstrates AMD’s growing competitiveness in the AI accelerator market.In Llama 2 70B offline testing, the Instinct MI355X achieved remarkable throughput. A 64-chip configuration generated 648,248 tokens per second. The 32-chip setup produced 350,820 tokens per second, while the 8-chip configuration reached 65,770 tokens per second. This performance represented a 2.09x advantage over NVIDIA’s GB200 in the 8-chip comparison.Intel‘s Arc Pro B60, while targeting different market segments, also participated in the benchmarks. It scored 3,009 tokens per second in Llama 2 70B testing. Industry analysts note its value proposition differs significantly from the data-center-focused chips from NVIDIA and AMD.

Why Players Are Hunting Nod-Krai Legends in Genshin Impact

The MLPerf v5.1 results confirm NVIDIA’s continued dominance in AI acceleration while revealing meaningful competition from AMD. These performance breakthroughs will directly impact how enterprises deploy and scale their AI infrastructure in coming years.

Info at your fingertips

What are MLPerf benchmarks?

MLPerf is an industry standard AI benchmark suite. It provides fair and objective performance comparisons between different AI hardware platforms.

When will Blackwell Ultra chips be available?

NVIDIA has not announced specific availability dates. Industry analysts expect initial deployments in late 2025 for select enterprise customers.

How does AMD’s MI355X compare to NVIDIA’s Blackwell?

AMD’s solution shows competitive performance in specific benchmarks. NVIDIA maintains an overall performance advantage, particularly in reasoning tasks.

Why are AI benchmarks important?

These benchmarks help enterprises evaluate hardware for AI workloads. They provide standardized performance metrics for informed purchasing decisions.

Which companies use MLPerf results?

Major cloud providers, enterprise IT departments, and AI researchers rely on these benchmarks. They guide multi-million-dollar infrastructure investments.