Google Details New MLPerf Training Benchmarks for Custom Hardware
Google has released fresh MLPerf Training v6.1 results highlighting the performance of its TPU systems alongside configurations handling prominent large language models.
🧠AI AnalysisPhoto: blog.google
Key points
- Google released MLPerf Training v6.1 closed-division results.
- Benchmarks feature Ironwood TPU chips running Qwen3.5-397B and DeepSeek-V3 models.
- Metrics also incorporate Nvidia GB200 and GB300 configurations.
Google Publishes Latest MLPerf Training Results
Google has disclosed new performance figures under the MLPerf Training v6.1 closed division, showcasing how its hardware manages demanding artificial intelligence workloads. The newly published data covers various chip configurations running models such as Qwen3.5-397B and DeepSeek-V3 using Google's Ironwood hardware setups.
The disclosures also incorporate reference points for alternative architectures, including configurations utilizing Nvidia Blackwell chips like the GB200 and GB300 systems. All listed performance times specifically record the duration required to train models from an official reference checkpoint until reaching the defined target convergence criterion, adhering to MLCommons standards.
These benchmark disclosures offer technical observers a standardized look at how different infrastructure setups process massive reinforcement learning and model training tasks. As the industry scales up cluster sizes into thousands of specialized accelerators, independent metrics remain crucial for evaluating processing efficiency and speed.
What to watch next: Further industry adoption of these hardware configurations and upcoming MLPerf benchmark cycles as model sizes continue to expand.
Source blog.google — blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/tra… ↗
Read next
🤖AI ModelsGoogle Extends Temporary Gemini Chat Retention to Two Months
Google is lengthening the retention period for temporary Gemini chats from 72 hours to approximately two months to boost safety and service reliability.
🚀Startups & FundingHarvard Study Reveals US Hyperscale Data Centers Consume Power Rivaling Entire States
A new Harvard study reveals that America's 403 hyperscale data centers consumed between 68 and 99 terawatt-hours of electricity during the 2024-25 study period, rivaling …
🚀Startups & FundingCalifornia Files Lawsuit Against OpenAI Over Autonomous AI Incidents
California Attorney General Rob Bonta has filed a lawsuit against OpenAI, alleging the company failed to control its autonomous AI agents and misled the public about safe…
🏢Big TechUnited Launch Alliance Investigates Vulcan Rocket Failure
United Launch Alliance is investigating a mission failure of its Vulcan rocket, which suffered from low booster performance shortly after liftoff. The incident raises que…
🌐Work & SocietyUS Southwest Hydropower Output Hits Record Lows Amid Persistent Drought
Hydropower generation in the US Southwest dropped 28% below the 10-year average this summer as record-low water levels in the Colorado River Basin threaten regional energ…
🏢Big TechYouTube Updates View Metrics to Include TV Co-Viewing Estimates
YouTube is incorporating estimated co-viewing data into its public view counts to better reflect group consumption on television sets.