{"article":{"annotation":{"article_id":"slackswap-activation-memory-gpu-training-2026","available_at":"2026-10-13T01:40:32.731817Z","geography_revision":"ef6ee0010dda5fe7b240fd514749b43e214c57da73222d2e40ec83b276dd0977","id":"9b5ec3e857ab9c22f27a1798abbf946b759d8104d9ffc99345490e5604b94daf","input_sha256":"cf86f452065734efc0dda8dda6e009e69735a1fbf30a82f2d1e9f0ea3870744f","limitations":["Rule-derived subjects are candidates, not verified editorial labels or calibrated model probabilities.","Coverage is distinct from event location. Unknown country/city/coordinates remain absent; no automatic geocoding.","Article geometries keep their stated precision and methodology; reproducing them is not independent coordinate verification."],"method":"controlled-rules-v1","places":[{"admin1":null,"city":null,"coordinate_note":null,"coordinate_reference_system":"OGC:CRS84","coordinate_source":null,"country":"US","evidence":[{"field":"country","language":"en","note":"Explicit preserved editorial country/region; no city or national centroid inferred","source_urls":["https://arxiv.org/abs/2610.13003"]}],"geometry":null,"id":"iso3166-1:US","kind":"country-or-area","labels":{"ar":"الولايات المتحدة الأمريكية","en":"United States of America","es":"Estados Unidos de América","fr":"États-Unis d’Amérique"},"match_status":"candidate","precision":"regional","role":"coverage","uncertainty_m":null}],"rules_sha256":"fb5ad3c27c91cb840f7a0a05f7f64a5cb26ad0374409ff9457f277c7092e22b4","schema_version":1,"subjects":[{"concept_id":"medtop:13000000","confidence":null,"confidence_method":null,"evidence":[{"field":"category+topics","language":"en","note":"Preserved editorial category/topic; broad candidate only","source_urls":["https://arxiv.org/abs/2610.13003"]}],"method":"controlled-rules-v1","rule_id":"technology-broad","status":"candidate"}],"taxonomy_revision":"1f2db31d811fca07e99b16d402c10e2cc2965c9df1c7b41d9be348b40a24e7fb","taxonomy_version":"IPTC Media Topics 2026-Q2"},"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=en","category":"technology","content":{"key_points":[],"paragraphs":["Intermediate activations saved for backpropagation can exceed GPU memory during neural-network training. SlackSwap profiles a training step and treats the transfer time that can be hidden during computation as a budget: it moves activations that are costly to rebuild within that budget and recomputes the rest.","In a discrete-event simulation parameterized for an A100 and PCIe 4.0, with the activation-memory budget set to half the no-memory-limit GPU’s peak, SlackSwap retained 96.3% of ideal throughput for BERT-Large, 97.6% for GPT-2 XL and 99.6% for a Llama-7B-style model. The authors say all results are simulated and that effects such as allocator fragmentation are not modeled; these are not measurements from physical GPU training."],"sections":[]},"corrections":[],"country":"US","data":{"blocks":[{"columns":[{"key":"model","label":{"ar":"النموذج","de":"Modell","en":"Model","es":"Modelo","fr":"Modèle","pt":"Modelo","tr":"Model"},"language":"en","type":"text"},{"key":"ideal_throughput","label":{"ar":"معدل الأداء مقارنة بالمثالي","de":"Durchsatz im Verhältnis zum Ideal","en":"Throughput relative to ideal","es":"Rendimiento respecto al ideal","fr":"Débit par rapport à l’idéal","pt":"Débito face ao ideal","tr":"İdeale göre verim"},"type":"number","unit":"percent"}],"description":{"ar":"النتائج ناتجة عن محاكاة وليست تشغيلًا على GPU فعلي.","de":"Die Ergebnisse stammen aus einer Simulation, nicht von einem physischen GPU-Lauf.","en":"Results are from simulation, not a run on a physical GPU.","es":"Los resultados proceden de una simulación, no de una GPU física.","fr":"Les résultats proviennent d’une simulation, pas d’un GPU physique.","pt":"Os resultados são de simulação, não de uma execução numa GPU física.","tr":"Sonuçlar gerçek GPU çalıştırması değil, benzetimdir."},"id":"simulated-training-throughput","methodology":{"ar":"تقيّم الدراسة ثلاثة أعباء عمل للمحوّلات عند ميزانية ذاكرة تنشيط قدرها 50% في محاكاة أحداث منفصلة مضبوطة وفق A100 وPCIe 4.0.","de":"Die Studie bewertet drei Transformer-Arbeitslasten bei einem Aktivierungsspeicherbudget von 50 % in einer ereignisdiskreten Simulation mit Parametern für A100 und PCIe 4.0.","en":"The study evaluates three transformer workloads at a 50% activation-memory budget in a discrete-event simulation parameterized for an A100 and PCIe 4.0.","es":"El estudio evalúa tres cargas Transformer con un presupuesto de memoria de activaciones del 50 % en una simulación de eventos discretos parametrizada para una A100 y PCIe 4.0.","fr":"L’étude évalue trois charges Transformer avec un budget mémoire d’activations de 50 % dans une simulation à événements discrets paramétrée pour un A100 et PCIe 4.0.","pt":"O estudo avalia três cargas Transformer com um orçamento de memória de ativações de 50% numa simulação de eventos discretos parametrizada para A100 e PCIe 4.0.","tr":"Çalışma, A100 ve PCIe 4.0 özelliklerini temel alan ayrık olay benzetiminde 50% etkinleştirme belleği bütçesindeki üç Transformer iş yükünü değerlendiriyor."},"rowTranslations":[{"model":{"ar":"BERT-Large","de":"BERT-Large","en":"BERT-Large","es":"BERT-Large","fr":"BERT-Large","pt":"BERT-Large","tr":"BERT-Large"}},{"model":{"ar":"GPT-2 XL","de":"GPT-2 XL","en":"GPT-2 XL","es":"GPT-2 XL","fr":"GPT-2 XL","pt":"GPT-2 XL","tr":"GPT-2 XL"}},{"model":{"ar":"نموذج على نمط Llama-7B","de":"Llama-7B-ähnlich","en":"Llama-7B-style","es":"modelo tipo Llama-7B","fr":"modèle de type Llama-7B","pt":"modelo do tipo Llama-7B","tr":"Llama-7B benzeri"}}],"rows":[{"ideal_throughput":96.3,"model":"BERT-Large"},{"ideal_throughput":97.6,"model":"GPT-2 XL"},{"ideal_throughput":99.6,"model":"Llama-7B-style"}],"sourceUrls":["https://arxiv.org/abs/2610.13003"],"title":{"ar":"معدل التدريب المحاكى","de":"Simulierter Trainingsdurchsatz","en":"Simulated training throughput","es":"Rendimiento de entrenamiento simulado","fr":"Débit d’entraînement simulé","pt":"Débito de treino simulado","tr":"Benzetilmiş eğitim verimi"},"type":"table","updatedAt":"2026-10-13T01:31:44Z"}],"version":1},"data_downloads":[{"formats":{"csv":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026/data/simulated-training-throughput?lang=en&format=csv","json":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026/data/simulated-training-throughput?lang=en&format=json"},"id":"simulated-training-throughput"}],"data_types":["table"],"editorial_illustration":{"alt":"Conceptual illustration of technology, processors and connected information.","asset_id":"newsverum-v1-technology","avif_variants":[{"height":320,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-480.avif","width":480},{"height":480,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-720.avif","width":720},{"height":640,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-960.avif","width":960},{"height":800,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1200.avif","width":1200},{"height":1024,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.avif","width":1536}],"credit":"Newsverum","documentary":false,"height":1024,"kind":"symbolic-editorial-illustration","language":"en","scope":"topic","url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.webp","variants":[{"height":320,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-480.webp","width":480},{"height":640,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-960.webp","width":960},{"height":1024,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.webp","width":1536}],"width":1536},"event_id":"event-50c6776f-cabc-5f36-8bd2-f715b760aea9","event_period":"2026-10-09","event_period_basis":"source-review","id":"slackswap-activation-memory-gpu-training-2026","image":null,"importance":0.6,"language":"en","modified_at":"2026-10-13T01:40:12+00:00","publication_basis":"publication-record","published_at":"2026-10-13T01:40:12+00:00","reading_minutes":1,"region":"world","section_illustration":{"alt":"Conceptual illustration of technology, processors and connected information.","asset_id":"newsverum-v1-technology","avif_variants":[{"height":320,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-480.avif","width":480},{"height":480,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-720.avif","width":720},{"height":640,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-960.avif","width":960},{"height":800,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1200.avif","width":1200},{"height":1024,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.avif","width":1536}],"credit":"Newsverum","documentary":false,"height":1024,"kind":"symbolic-editorial-illustration","language":"en","scope":"section","url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.webp","variants":[{"height":320,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-480.webp","width":480},{"height":640,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-960.webp","width":960},{"height":1024,"url":"https://newsverum.com/assets/newsverum/illustrations/v1/technology-1536.webp","width":1536}],"width":1536},"slug":"slackswap-activation-memory-gpu-training-2026","sources":[{"checkedAt":"2026-10-13","name":"arXiv","publishedAt":"2026-10-09","url":"https://arxiv.org/abs/2610.13003"}],"summary":"Researchers at UC Merced and the University of Texas at Austin introduced SlackSwap, a planner for deep-learning training. It recomputes some activations when the transfer budget between a GPU and host memory is exceeded; reported results come from simulation, not hardware measurements.","text":"Researchers at UC Merced and the University of Texas at Austin introduced SlackSwap, a planner for deep-learning training. It recomputes some activations when the transfer budget between a GPU and host memory is exceeded; reported results come from simulation, not hardware measurements.\n\nIntermediate activations saved for backpropagation can exceed GPU memory during neural-network training. SlackSwap profiles a training step and treats the transfer time that can be hidden during computation as a budget: it moves activations that are costly to rebuild within that budget and recomputes the rest.\n\nIn a discrete-event simulation parameterized for an A100 and PCIe 4.0, with the activation-memory budget set to half the no-memory-limit GPU’s peak, SlackSwap retained 96.3% of ideal throughput for BERT-Large, 97.6% for GPT-2 XL and 99.6% for a Llama-7B-style model. The authors say all results are simulated and that effects such as allocator fragmentation are not modeled; these are not measurements from physical GPU training.","title":"SlackSwap combines data transfers and recomputation in GPU-limited training","topics":[],"translations":[{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=tr","language":"tr","url":"https://newsverum.com/tr/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=en","language":"en","url":"https://newsverum.com/en/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=ar","language":"ar","url":"https://newsverum.com/ar/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=fr","language":"fr","url":"https://newsverum.com/fr/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=es","language":"es","url":"https://newsverum.com/es/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=de","language":"de","url":"https://newsverum.com/de/news/slackswap-activation-memory-gpu-training-2026"},{"api_url":"https://newsverum.com/api/v1/articles/slackswap-activation-memory-gpu-training-2026?lang=pt","language":"pt","url":"https://newsverum.com/pt/news/slackswap-activation-memory-gpu-training-2026"}],"url":"https://newsverum.com/en/news/slackswap-activation-memory-gpu-training-2026"},"version":"1"}