{"paper":{"title":"ABI: A tightly integrated, unified, sparsity-aware, reconfigurable, compute near-register file/cache GPU architecture with light-weight softmax for deep learning, linear algebra, and Ising compute","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A tightly integrated near-memory GPU architecture called ABI achieves 6-16 times speedup and 6-13 times energy savings on convolutional neural networks, graph networks, linear programming, large language models, and Ising workloads compared","cross_cats":[],"primary_cat":"cs.AR","authors_text":"Jaydeep P. Kulkarni, Siddhartha Raman Sundara Raman","submitted_at":"2026-02-15T18:19:06Z","abstract_excerpt":"We present a tightly integrated and unified near-memory GPU architecture that delivers 6 to 16 times speedup and 6 to 13 times energy savings across Convolutional Neural Networks, Graph Convolutional Networks, Linear Programming, Large Language Models, and Ising workloads compared to MIAOW GPU. The design includes a custom sparsity-aware near-memory circuit providing about 1.5 times energy savings, and a lightweight softmax circuit providing about 1.6 times energy savings. The architecture supports reconfigurable compute up to INT16 with dynamic resolution updates and scales efficiently across"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We present a tightly integrated and unified near-memory GPU architecture that delivers 6 to 16 times speedup and 6 to 13 times energy savings across Convolutional Neural Networks, Graph Convolutional Networks, Linear Programming, Large Language Models, and Ising workloads compared to MIAOW GPU.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The design assumes that the custom sparsity-aware near-memory circuit and lightweight softmax can be integrated with negligible area, latency, and power overheads while maintaining the claimed scalability across problem sizes and reconfigurability up to INT16.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ABI is a unified near-memory GPU architecture delivering 6-16x speedup and 6-13x energy savings for deep learning, linear algebra, and Ising compute via sparsity-aware circuits and lightweight softmax.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A tightly integrated near-memory GPU architecture called ABI achieves 6-16 times speedup and 6-13 times energy savings on convolutional neural networks, graph networks, linear programming, large language models, and Ising workloads compared","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2ff0dfe7f94f83f9fbe69c3c813d3a9f0a9b1f0f70c6baefedea557d0fa03e3f"},"source":{"id":"2602.14262","kind":"arxiv","version":3},"verdict":{"id":"43eaabfd-83e3-4975-a808-80fd554e7da2","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T21:52:26.277525Z","strongest_claim":"We present a tightly integrated and unified near-memory GPU architecture that delivers 6 to 16 times speedup and 6 to 13 times energy savings across Convolutional Neural Networks, Graph Convolutional Networks, Linear Programming, Large Language Models, and Ising workloads compared to MIAOW GPU.","one_line_summary":"ABI is a unified near-memory GPU architecture delivering 6-16x speedup and 6-13x energy savings for deep learning, linear algebra, and Ising compute via sparsity-aware circuits and lightweight softmax.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The design assumes that the custom sparsity-aware near-memory circuit and lightweight softmax can be integrated with negligible area, latency, and power overheads while maintaining the claimed scalability across problem sizes and reconfigurability up to INT16.","pith_extraction_headline":"A tightly integrated near-memory GPU architecture called ABI achieves 6-16 times speedup and 6-13 times energy savings on convolutional neural networks, graph networks, linear programming, large language models, and Ising workloads compared"},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.14262/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"780f5f8f79ef0d0e0b238ba6f3fa281feeaf6d6c703b63fa649d076e9d7bd337"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}