Xiangyao Yu

Assistant Professor at University of Wisconsin-Madison. Research interest: database systems.

[CMU Database Group @CMUDB]

Today's Postgres vs. World Seminar Speaker: Dr. Xiangyao Yu (@xiangyao_yu) will present the architecture of the @siriusgpudb GPU-native database system. Zoom talk open to public at 4:30pm ET. YouTube video available after: https://t.co/gMBILMnVzv

[NVIDIA AI Developer @NVIDIAAIDev]

Congrats to the team behind Sirius on setting new ClickBench records🏅

Their open-source SQL engine powered by NVIDIA CUDA‑X achieved the lowest relative runtime and 7× cost efficiency on NVIDIA GH200. Using cuDF, Sirius brings GPU‑accelerated analytics to DuckDB.

Learn more: https://t.co/F8nz51ypBA

[Wall St Engine @wallstengine]

$NVDA and University of Wisconsin–Madison has unveiled Sirius, a GPU-native DuckDB engine that hit #1 on ClickBench, delivering at least 7.2x better cost efficiency vs top CPU systems.

Separately, Nvidia’s Nemotron 3 Nano open-source LLM is out, with larger models coming 1H26

[Xiangyao Yu @xiangyao_yu]

🔥Sirius hits #1 on ClickBench!

GPU-native SQL accelerating DuckDB with NVIDIA CUDA-X. GPUs are ready for serious data analytics.

Blog: https://t.co/j0ismOLk93
Github: https://t.co/4P7iFzTHQl

@duckdb @nvidiadeveloper @RAPIDSai

[Disseminate: The Computer Science Research Podcast @DisseminatePod]

🚨 "Disaggregation: A New Architecture for Cloud Databases" with Xiangyao Yu [@xiangyao_yu] is available now!

Watch on YouTube ➡️ https://t.co/Qn1yyyppaP

Listen on Spotify ➡️ https://t.co/oTKs1MTKVt

Listen on Apple ➡️ https://t.co/tgRwSLBfcl

[Murat Demirbas (Distributolog) @muratdemirbas]

[new blog post]

Disaggregation: A New Architecture for Cloud Databases

https://t.co/zTM8BaSIem

[Xiangyao Yu @xiangyao_yu]

@awsTO Thank you Aws!

[Xiangyao Yu @xiangyao_yu]

Honored to receive the #VLDB2025 Early Career Award this year. Grateful to @wiscdb, my students, collaborators, mentors, and family for making this journey possible. Excited to keep pushing the boundaries of database systems!

Paper: https://t.co/GvHpC59hWB

[Wisconsin DB Group (@wiscdb.bsky.social) @wiscdb]

Congrats to @bwyogatama for defending his thesis on GPU data analytics! Recently, he’s been leading @siriusgpudb, which is a GPU-native SQL engine built on top of @duckdb. After graduation, he’s headed to @nvidia!

[PVLDB @pvldb]

Vol:18 No:12 → Disaggregation: A New Architecture for Cloud Databases https://t.co/k1sTbTL512

[Wisconsin DB Group (@wiscdb.bsky.social) @wiscdb]

Yifei Yang's paper "Accelerate Distributed Joins with Predicate Transfer" expands on previous work (CIDR 2024). It shows 1) cost-based adaptive execution strategy minimizing network cost 2) pruning algorithm, eliminating transfers w/out positive contributions to performance.

[Xiangyao Yu @xiangyao_yu]

Check out the papers:

🔗 [CIDR'24] Predicate Transfer: https://t.co/p8IuhmGUTP
🔗 [SIGMOD'25] Distributed PT: https://t.co/32HpgezXX1
🔗 [SIGMOD'25] Robust PT in DuckDB: https://t.co/viSCBfEj3b

[Xiangyao Yu @xiangyao_yu]

Join order doesn't matter—if you pre-filter non-contributing rows.
Predicate Transfer (PT) [CIDR'24] does this using Bloom filters across joins.

At @SIGMODConf'25, we extend PT:
📄#1: Distributed PT (3x speedup over no PT)
📄#2: Robust PT in DuckDB (1.5× speedup + robust plans)

[Wisconsin DB Group (@wiscdb.bsky.social) @wiscdb]

Congrats to Elena Milkai on defending her thesis on real-time transactional analytics! One of her projects, HERMES (VLDB 2025), enables data consolidation across transactional and analytical engines without system changes. After grad, she’s headed to @salesforce’s DB team!

[Xiangyao Yu @xiangyao_yu]

Registration link: https://t.co/bexfyUT0Y6

If you're into GPU infra or high-perf analytics, let’s connect!

[Xiangyao Yu @xiangyao_yu]

🚀Excited to speak at the #STACSummit in Chicago on May 21! @bwyogatama and I will make the case for GPU acceleration beneath the database stack and present SiriusDB, our GPU-native execution engine offering drop-in acceleration across multiple data systems.

[Xiangyao Yu @xiangyao_yu]

Presented our poster "GPU Databases—The New Modality of Database Analytics" at #GTC25 last evening with @bwyogatama. We’ll be around at GTC for the next few days—if you're interested in chatting about GPUs or data analytics, let’s connect! 🔥 @siriusgpudb

[Xiangyao Yu @xiangyao_yu]

🚀Exciting news! Our project "Towards the New Modality of GPU Database Analytics" has been selected for the NVIDIA Academic Grant Program! Huge thanks to NVIDIA for the GPU hours. Looking forward to pushing the boundaries of GPU data analytics together! #NVIDIAGrant @NVIDIAAIDev

[Abigale Kim (@abigalekim.bsky.social) @abigale_kim]

this past Friday, I presented a poster at @WisconsinCS visit days on our research group's cool new GPU-accelerated DBMS, @siriusgpudb !
cc: @xiangyao_yu @bwyogatama @wiscdb

[Huanchen Zhang @huanchenzhang]

Join ordering may not be a critical challenge for future optimizers anymore! Check out our latest paper on robust query processing (to appear in SIGMOD'25): https://t.co/dY9OPe6QMU @xiangyao_yu @andy_pavlo @pateljm @peterabcz @yytian