Pith. sign in

Paper Citation Record · LEDGER

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure

As of 22 August 2026, this Paper Citation Record lists 100 of 296 outbound references and 1 inbound Pith citation observation for arXiv:2507.07223.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2507.07223 v2

Coverage vector

measured 100 of 296 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-06T18:50:52.423393Z

measured 101 of 101 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-22T06:32:14.747728+00:00

measured 1 of 1 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-15T15:21:29.510318Z

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: pith, observed 2026-08-15T15:21:30.520212Z

Reference resolution

100 of 296 outbound references displayed

  • verified exact0
  • verified fuzzy0
  • unresolved97
  • parse uncertain0
  • malformed identifier0
  • metadata mismatch3

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 959de874-0ed6-4b10-a622-7ab39d29eb88 · outbound

This paper cites Avoiding another AI winter,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Avoiding another AI winter,

Reference 1

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:44.627502Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:44.627502Z digest=sha256:c11728f03d1341819f2c67280f1bf9b236afa7e7129cda646252e32f80e59ef4

Observation ce54bb8e-08e7-4648-8652-de5c02771f8e · outbound

This paper cites A Brief History of AI: How to Prevent Another Winter,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure A Brief History of AI: How to Prevent Another Winter,

Reference 2

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:44.717214Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:44.717214Z digest=sha256:94a325e8df2f019bdbfb523c3b2d86588c0032db35d6cd3b9e792684790a3f3f

Observation e8c772b2-e5cd-4620-90ae-7a392cdf2523 · outbound

This paper cites Attention is all you need,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Attention is all you need,

Reference 3

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:44.810710Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:44.810710Z digest=sha256:e3c34dc19cc9e4d10427fe3964b80f3127dea9442821330feea4f4261d6da3fe

Observation 207cb565-a9de-4eff-a1aa-95b64c6b5ad5 · outbound

This paper cites Annotated History of Modern AI and Deep Learning,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Annotated History of Modern AI and Deep Learning,

Reference 4

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:44.888240Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:44.888240Z digest=sha256:633bd5570982c2b0b7e2f0a294ff4b95d5e88a67171365f72eb88a432cc6cc32

Observation 5c7a5b91-6ee6-44d7-8990-4428913cba25 · outbound

This paper cites ImageNet Classification with Deep Convolutional Neural Networks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure ImageNet Classification with Deep Convolutional Neural Networks,

Reference 5

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:44.990434Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:44.990434Z digest=sha256:a87906f89bf2a3a531ec6a4c1549a36e91926436837de99ab5ff008779ad56c0

Observation efc7b7dd-0b0d-40f5-824c-5c1d099a5531 · outbound

This paper cites BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,

Reference 6

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.083350Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.083350Z digest=sha256:ea3d49b8c324831befd10d8a128a72e676b9e54be03b005dde63410c992e56d5

Observation b6e0c890-a0e1-48ac-a03e-279326fb2505 · outbound

This paper cites The Llama 3 Herd of Models.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure The Llama 3 Herd of Models

Reference 7

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.175041Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.175041Z digest=sha256:a8eb65b6d54c503051741006612064c02014f5dc51d3a56420d8bb5df9bb22bc

Observation 13bf8b61-7b83-4743-8764-ec3ce4cbf352 · outbound

This paper cites Generative Adversarial Nets,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Generative Adversarial Nets,

Reference 8

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.188306Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.188306Z digest=sha256:e515e108ba062c385e8912ae96329a71e4124741d4f29ab891ada67e29d3247d

Observation 8d03c592-3f66-4338-9dae-a5080df60f22 · outbound

This paper cites Learning to learn by gradient descent by gradient descent,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Learning to learn by gradient descent by gradient descent,

Reference 9

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.249588Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.249588Z digest=sha256:e028c05a4016b78a6e45f41a16b9c233dff758df49e2e1da4d7917c28e8806bb

Observation c643b260-79ef-4c73-9b22-1d47c1d1100d · outbound

This paper cites Gradient Descent: The Ultimate Optimizer,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Gradient Descent: The Ultimate Optimizer,

Reference 10

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.360307Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.360307Z digest=sha256:7c4b678a57bd18c23606d0911c2e7be125e5561d5eb5492641c8803aefb94599

Observation 4b53a767-b11b-4031-8f05-726cf381178f · outbound

This paper cites An overview of gradient descent optimization algorithms.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure An overview of gradient descent optimization algorithms

Reference 11

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.468946Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.468946Z digest=sha256:a633dcdc69d60a33398d859d365c096f4c50d2bd6945fd61778bff298e47a9c0

Observation 230dcd83-dac4-4422-82a6-054fa3ad0dd1 · outbound

This paper cites Optimization methods for large-scale machine learning,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Optimization methods for large-scale machine learning,

Reference 12

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.594404Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.594404Z digest=sha256:4c7b9a95cb770bfb63dc3217e403d44830f3f5eefe067892d788fd1f918d3040

Observation 36e9eb5a-9b22-4de1-a768-9b076e810478 · outbound

This paper cites Practical recommendations for gradient-based training of deep architectures,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Practical recommendations for gradient-based training of deep architectures,

Reference 13

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:45.774798Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:45.774798Z digest=sha256:7ae5c5523d9c4c5cf06e638271de1493912a69829ed3c72342f0b7783b18dbfe

Observation 86498f9e-16f8-4fd0-9321-efb53d7b24b0 · outbound

This paper cites On the Number of Linear Regions of Deep Neural Networks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure On the Number of Linear Regions of Deep Neural Networks,

Reference 14

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.005976Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.005976Z digest=sha256:6af4c401ba27819f68cf5164e8d724488913d9b3315845cf54ad6710728cfa05

Observation 2a6e391b-8884-42e6-a895-4e20279cfca5 · outbound

This paper cites Representation learning: A review and new perspectives,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Representation learning: A review and new perspectives,

Reference 15

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.243461Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.243461Z digest=sha256:67a6b377b6c81e1ba3f46791f645a8b07e5e847d6411ce47d4f98d6807c32aa3

Observation 817ff0a6-3d06-4c31-ada0-b89333443442 · outbound

This paper cites Intrinsic dimension of data representations in deep neural networks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Intrinsic dimension of data representations in deep neural networks,

Reference 16

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.411121Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.411121Z digest=sha256:472d5dbf12f438e47e3448c48433c388dc617cd4d34142121d7bd1b4bfef8255

Observation 8bd1f63e-ac89-4826-b89c-8f43c349539c · outbound

This paper cites High dimensional, tabular deep learning with an auxiliary knowledge graph,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure High dimensional, tabular deep learning with an auxiliary knowledge graph,

Reference 17

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.553457Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.553457Z digest=sha256:84ec759546769965781d73e7474ad14e84b0b3bfdee22e0904a1f5e387bd02d6

Observation cb93f79f-2a97-42c1-8597-a5156d2935ab · outbound

This paper cites General-purpose, long-context autoregressive modeling with Perceiver AR,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure General-purpose, long-context autoregressive modeling with Perceiver AR,

Reference 18

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.684131Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.684131Z digest=sha256:bc18501f3f78fe74ec01c45c0593ca8abebef71b30c981af981f11f51c099325

Observation 2cf42575-8cc0-4ea4-9900-a535202646f8 · outbound

This paper cites A survey on hardware accelerators: Taxonomy, trends, challenges, and perspectives,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure A survey on hardware accelerators: Taxonomy, trends, challenges, and perspectives,

Reference 19

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.788043Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.788043Z digest=sha256:8bbe42b0332db9d754828f9b06e13bb97920071c247d04a00e5385a82135ecc3

Observation 31a8081c-30cd-4d7f-9c56-80b53a4dc450 · outbound

This paper cites The transformational role of GPU computing and deep learning in drug discovery,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure The transformational role of GPU computing and deep learning in drug discovery,

Reference 20

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:46.954360Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:46.954360Z digest=sha256:9095cd70eef3ac63f58d5468c108efd6816bdc26cfc4081a8504b518daeb4788

Observation 6f31e56b-8ae3-4121-8ef2-2e28ea43073b · outbound

This paper cites ZeRO: Memory optimizations Toward Training Trillion Parameter Models,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure ZeRO: Memory optimizations Toward Training Trillion Parameter Models,

Reference 21

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:47.093029Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:47.093029Z digest=sha256:170d66fce58bbd33034e20c34c119f0962b5f51388add6f8483a3097a72a8377

Observation 6e48f871-6030-4f03-9662-04c96aa0b323 · outbound

This paper cites DeepSpeed: System Optimizations Enable Training Deep Learning Models with Over 100 Billion Parameters,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure DeepSpeed: System Optimizations Enable Training Deep Learning Models with Over 100 Billion Parameters,

Reference 22

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:47.300565Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:47.300565Z digest=sha256:e769393f14ddec2bbc764d22c2dd9ffd4c12957e51389c118437e796231067cd

Observation b9639a43-3504-44fe-8d70-49190af0d5a7 · outbound

This paper cites Gpu graph processing on cxl-based microsecond-latency external memory,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Gpu graph processing on cxl-based microsecond-latency external memory,

Reference 23

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:47.455266Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:47.455266Z digest=sha256:815850eb303dcacfc6c4ce0977f8e0a3d58c526445df0d99c1851690311183b3

Observation 4532eeee-b5d6-4a51-b978-070f45b6e814 · outbound

This paper cites EMOGI: Efficient Memory-access for Out-of-memory Graph-traversal In GPUs.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure EMOGI: Efficient Memory-access for Out-of-memory Graph-traversal In GPUs

Reference 24

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:47.659632Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:47.659632Z digest=sha256:120c9645d2c5457ae02f7d153f89908f9e271df6f69f76381b676ce564743623

Observation 6ff19642-e4de-4110-9ead-c0d8813baeab · outbound

This paper cites Efficient memory management for large language model serving with pagedattention,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Efficient memory management for large language model serving with pagedattention,

Reference 26

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.041272Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.041272Z digest=sha256:b419dd71ff1e3136fbce308c8f4cfb42c375d432d7bea4f8c166408934d54762

Observation 65a9e8be-924a-4f32-a4cd-654c6ff95f35 · outbound

This paper cites H2O: Heavy-hitter oracle for efficient generative inference of large language models,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure H2O: Heavy-hitter oracle for efficient generative inference of large language models,

Reference 27

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.209373Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.209373Z digest=sha256:90e55ded374f537632fa2786340ee5d5a2fff2710ca84030ed14de359deab086

Observation ec03af62-3cd0-4b87-82fb-c947d50952e7 · outbound

This paper cites Keyformer: Kv cache reduction through key tokens selection for efficient generative inference,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Keyformer: Kv cache reduction through key tokens selection for efficient generative inference,

Reference 28

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.378999Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.378999Z digest=sha256:c3a56a081d85bb504fd9a4b81750a664b7b086f937422be412eb7d7b9f27549a

Observation dd52f4e3-87c9-48cb-bd38-916e95856bfe · outbound

This paper cites TCCL: Co-optimizing Collective Communication and Traffic Routing for GPU-centric Clusters,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure TCCL: Co-optimizing Collective Communication and Traffic Routing for GPU-centric Clusters,

Reference 29

Resolution
metadata mismatch
raw_fallback, observed 2026-08-06T18:50:56.859861Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-22T06:32:14.747728+00:00.

source=pdf_text observed=2026-08-06T18:50:48.493360Z digest=sha256:45fbf6ec4aabef2ffe119865465772338732054adb99e11021270d75ef158ae3

Observation 4df1ebe9-bdec-4b05-9be5-cb2efc8b23d6 · outbound

This paper cites Sˆ3: Increasing GPU Utilization during Generative Inference for Higher Throughput,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Sˆ3: Increasing GPU Utilization during Generative Inference for Higher Throughput,

Reference 30

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.717974Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.717974Z digest=sha256:06f1633bc29546a7f309f723b9b8054a87be411ed992b42f8e4f966ecf8f3a63

Observation 5873df17-6e2b-4c7e-9730-45ba117901b1 · outbound

This paper cites NVIDIA Grace GPU.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA Grace GPU

Reference 31

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.869405Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.869405Z digest=sha256:3fddf6276334b7297beb0391f5105f2af1d14e2e009617dac41ae1884647f789

Observation a7302004-5471-4415-bbfc-672d8c95ee24 · outbound

This paper cites Characterization and prediction of deep learning workloads in large-scale GPU datacenters,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Characterization and prediction of deep learning workloads in large-scale GPU datacenters,

Reference 32

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:48.929610Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:48.929610Z digest=sha256:39454111e3a722e433288c9a7bfeb0fe59d26cbadc4ace574b44ac3c8386dbc1

Observation 52ad905f-5962-4bdf-8664-8935afb33c2f · outbound

This paper cites Towards Memory Disaggregation via NVLink C2C: Benchmarking CPU-Requested GPU Memory Access,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Towards Memory Disaggregation via NVLink C2C: Benchmarking CPU-Requested GPU Memory Access,

Reference 33

Resolution
metadata mismatch
raw_fallback, observed 2026-08-06T18:50:56.774562Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-22T06:32:14.747728+00:00.

source=pdf_text observed=2026-08-06T18:50:49.011951Z digest=sha256:5a38f43ec72688e4d0995c8994492b3361d993b9e4c214a80f98a08382669afb

Observation c3db7e59-9f25-477b-a3bf-d9e8af32f857 · outbound

This paper cites Poseidon: An Efficient Communication Architecture for Distributed Deep Learning on GPU Clusters,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Poseidon: An Efficient Communication Architecture for Distributed Deep Learning on GPU Clusters,

Reference 34

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.068169Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.068169Z digest=sha256:ab8901b7b4c9fbf0ff2e2c21f9850b8f757324444643966bfba08678cdea2470

Observation f178814c-9fc0-44bd-b2a1-dfed9ba92ff0 · outbound

This paper cites A memory perspective: The effects of finetuning llms with highbandwidth memory,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure A memory perspective: The effects of finetuning llms with highbandwidth memory,

Reference 35

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.161709Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.161709Z digest=sha256:4643909b40464941b2131e3bc40517268995068fe408754cae8f0e152d010946

Observation 41c58c71-2b96-4316-9dda-f8b511a7f751 · outbound

This paper cites Whitepaper: NVDIA GH200 Grace Hopper Superchip Architecture,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Whitepaper: NVDIA GH200 Grace Hopper Superchip Architecture,

Reference 36

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.244384Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.244384Z digest=sha256:ff4c73bafeda67d06bd938c981d923a0806a201325a00767e1ede403945ef173

Observation f158507a-488c-4d1c-9ce8-2a1e5656de15 · outbound

This paper cites Streambox-hbm: Stream analytics on high bandwidth hybrid memory,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Streambox-hbm: Stream analytics on high bandwidth hybrid memory,

Reference 37

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.329840Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.329840Z digest=sha256:91d3d3eb2a00fb2dbb3a0be43b9cc90a5abd92bc99d24c24ef8f274205a3fc5b

Observation a5aaf8ac-80e0-4118-84af-6ce1e7d42c60 · outbound

This paper cites Performance evaluation and optimization of HBM- enabled GPU for data-intensive applications,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Performance evaluation and optimization of HBM- enabled GPU for data-intensive applications,

Reference 38

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.400186Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.400186Z digest=sha256:f4e354574bbe6190a60e00ce7099534629af751b65dc1190c34ea03ce99866c5

Observation d77be27f-0c10-4952-b990-5bcfd6dedef1 · outbound

This paper cites NVIDIA GB200 NVL72 Delivers Trillion-Parameter LLM Train- ing and Real-Time Inference.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA GB200 NVL72 Delivers Trillion-Parameter LLM Train- ing and Real-Time Inference

Reference 39

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.473655Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.473655Z digest=sha256:5ad265cda31b173bcfeca9f7c388335db15a62a7c7881b767ba46a84093e6679

Observation fcc8e50f-8c18-44fa-974a-e35db944713d · outbound

This paper cites Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity,

Reference 40

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.560002Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.560002Z digest=sha256:fc4dbf2e40320cb1c827443408134a3d21bde16165ca48c1b2d0cb7db3ec40d5

Observation f91d6942-3f99-4749-8581-fa9414ec7d1a · outbound

This paper cites Retrieval augmented language model pre-training,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Retrieval augmented language model pre-training,

Reference 41

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.633004Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.633004Z digest=sha256:6672eb8ad7f607377a82a838c69801e312a682a7685f5cb3faaa730476cc3528

Observation 6abaf47c-5112-4a2f-9d2d-4bc41a12d153 · outbound

This paper cites Improving language models by retrieving from trillions of tokens,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Improving language models by retrieving from trillions of tokens,

Reference 42

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.697847Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.697847Z digest=sha256:b61a7046f3942c81fecad1f87123394905035eaf7f350ae372ed807695425d9c

Observation ad1cc1f3-8f4d-4aa9-b791-18906dfe0490 · outbound

This paper cites Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism

Reference 43

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.785560Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.785560Z digest=sha256:f564909d9f72209300a402fb6bf55f17e88afbbb68d71caedacf4e3bb216c3b3

Observation 5fd623e4-e2a4-498f-8dd7-f679d22a3529 · outbound

This paper cites How Much Knowledge Can You Pack Into the Parameters of a Language Model?.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure How Much Knowledge Can You Pack Into the Parameters of a Language Model?

Reference 44

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.872748Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.872748Z digest=sha256:0a7d9cff9be5fd25b57209d8f10199c4c19bb7fbffa464cda68c7cdb5e87dcea

Observation cc938e01-2d74-4ec3-84ff-7ed3d3851811 · outbound

This paper cites Pipedream: generalized pipeline parallelism for dnn training,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Pipedream: generalized pipeline parallelism for dnn training,

Reference 45

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:49.938398Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:49.938398Z digest=sha256:7f4fa7ef06d3ca785cc1a22575a14d10aed26e6ba270d17d0417e45d60ef0596

Observation 72163969-a342-4f63-bccd-49060ae35bb6 · outbound

This paper cites Number of parameters in gpt-4 (latest data),.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Number of parameters in gpt-4 (latest data),

Reference 46

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.005542Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.005542Z digest=sha256:591f47400d18a0555778d8442e0dd9deae38be557451c280c12ac2d04079c864

Observation fa00673f-bea8-4376-b300-31c63625aabf · outbound

This paper cites NVIDIA GB200 NVL72.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA GB200 NVL72

Reference 47

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.070127Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.070127Z digest=sha256:7d069bdae07a8a15181bd336a3a4ce731d97e909fec78b702e66f54838610feb

Observation b22a1f94-3158-4dcb-a281-edc8b83a7dc3 · outbound

This paper cites NVIDIA GB300 NVL72.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA GB300 NVL72

Reference 48

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.156769Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.156769Z digest=sha256:7ab22bc3ac1ffbc5c25ed35213b4dafe341143d0f5e44f7348981a10f7c4f910

Observation 3818b815-baf6-40d4-a591-367491c0162a · outbound

This paper cites NVIDIA DGX GB300.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA DGX GB300

Reference 49

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.224150Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.224150Z digest=sha256:de227a6149ca3fb5b7d3937736ae844ef3c8223b726325cc123669edf2ea50f0

Observation c3f6e128-acb6-4590-929f-faf76ad6794f · outbound

This paper cites Understanding communication characteristics of distributed training,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Understanding communication characteristics of distributed training,

Reference 50

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.324135Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.324135Z digest=sha256:ffbd8afcc11a1e1e86bd4601b022e6931841789afc26011a5164906fe65955ca

Observation cd589e9a-f548-497f-b789-fbd9755e7b40 · outbound

This paper cites MAD-Max Beyond Single-Node: Enabling Large Machine Learning Model Acceleration on Distributed Systems,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure MAD-Max Beyond Single-Node: Enabling Large Machine Learning Model Acceleration on Distributed Systems,

Reference 51

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.407491Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.407491Z digest=sha256:df61056643d251fa42e9a420c74cf8c68dbc6d4aad8b3a832d4709d8473b6b3a

Observation 169b84dd-eb35-4d65-8765-b7364f8470bc · outbound

This paper cites FLUX: Fast Software-based Communication Overlap On GPUs Through Kernel Fusion.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure FLUX: Fast Software-based Communication Overlap On GPUs Through Kernel Fusion

Reference 52

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.463847Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.463847Z digest=sha256:3c351dbe2b3d7ad88f9dd7e33c5b46eba6827a33b43e64fccd8f07262a17de71

Observation 46abc68b-1e72-4eaa-98c6-089714962a1d · outbound

This paper cites Collective communication library NCCL,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Collective communication library NCCL,

Reference 53

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.531679Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.531679Z digest=sha256:2e05c10adbfa473df38718078ea453842b14aa55ae79f3064453fe9e96fef22c

Observation aca4cb32-e627-4872-86f0-1456b4214eb4 · outbound

This paper cites Lancet: Accelerating mixture-of-experts training via whole graph computation-communication overlapping,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Lancet: Accelerating mixture-of-experts training via whole graph computation-communication overlapping,

Reference 54

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.586469Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.586469Z digest=sha256:8720dfa7916cb78e3f4dc222209285ef53e3dc484d20ea78e03d9313ca32d01a

Observation 1c4d3d5b-b99a-44a5-b246-645605945842 · outbound

This paper cites Traci: Network acceleration of input-dynamic communication for large-scale deep learning recommendation model,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Traci: Network acceleration of input-dynamic communication for large-scale deep learning recommendation model,

Reference 55

Resolution
metadata mismatch
raw_fallback, observed 2026-08-06T18:50:56.469937Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-22T06:32:14.747728+00:00.

source=pdf_text observed=2026-08-06T18:50:50.686463Z digest=sha256:30c62fe489c72862deaa1b931fad939f4a545b251756a7f3d899b972fed97c01

Observation 471bbb5c-5a05-4a5c-9cf5-fda77f7e335a · outbound

This paper cites GPU graph processing on cxl-based microsecond-latency external memory,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure GPU graph processing on cxl-based microsecond-latency external memory,

Reference 56

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.771412Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.771412Z digest=sha256:f6f860b5986626e51b0d42c152b3eb63b6ede89ab8877d27e09b4f9247bc2ea9

Observation 384a5e5c-0585-4f59-816a-35cb297168eb · outbound

This paper cites Understanding routable{PCIe} performance for composable infrastructures,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Understanding routable{PCIe} performance for composable infrastructures,

Reference 57

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.839408Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.839408Z digest=sha256:5a511ee39d7438cc2af5d1c6c70a02089190e24888c9a022531456922a0db1bc

Observation 30621a73-9791-4094-b55c-09cf1298a79e · outbound

This paper cites Understanding PCIe performance for end host networking,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Understanding PCIe performance for end host networking,

Reference 58

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.922505Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.922505Z digest=sha256:af0850186c573c44bc324f21bdeca0d24b71f875340371e2be255266ad798806

Observation 841d44a9-30e2-43be-8d86-9e77e01648f1 · outbound

This paper cites CXL 1.0 specification,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure CXL 1.0 specification,

Reference 59

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:50.962356Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:50.962356Z digest=sha256:c88fe1408026bd388295674eda80870bdb164bf4f460245e1bf33eb77747bb17

Observation bb516739-9b5e-49d5-b7bd-c50bda4ccffd · outbound

This paper cites CXL 2.0 specification,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure CXL 2.0 specification,

Reference 60

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.007942Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.007942Z digest=sha256:4e01f89a8b4ac90ea824241c327b5f39c73f186c15999753399b15a3acd6a490

Observation b9ed063f-1f98-4bb9-a208-cfc37af202b0 · outbound

This paper cites CXL 3.2 specification,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure CXL 3.2 specification,

Reference 61

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.145890Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.145890Z digest=sha256:7096a22ab19f18d0707bff8750a3f02239bb25ab3b6d4d92d9628172f0a37869

Observation a26f8c62-0d99-410a-b902-9805103c873e · outbound

This paper cites Overcoming the memory wall with CXL-Enabled SSDs,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Overcoming the memory wall with CXL-Enabled SSDs,

Reference 62

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.321615Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.321615Z digest=sha256:7dcc29cf17dd540586f59a98dfedc1cf769cafc4c427d612cbf1961b78fce69b

Observation 24c01a0a-fe55-4b2d-83d5-28555188efb3 · outbound

This paper cites An introduction to the compute express link (cxl) interconnect,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure An introduction to the compute express link (cxl) interconnect,

Reference 63

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.497367Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.497367Z digest=sha256:9db67a493aa707aeec00e57a5b724a0325010ca1f10fc9733c60dd11b97358a6

Observation f547036c-fa1a-48f2-9ade-afaed31ea7c9 · outbound

This paper cites Hello bytes, bye blocks: Pcie storage meets compute express link for memory expansion (cxl- ssd),.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Hello bytes, bye blocks: Pcie storage meets compute express link for memory expansion (cxl- ssd),

Reference 64

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.592359Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.592359Z digest=sha256:0187d150996cb0f71bb6ae77058c25e31486a8d2892371c058dc0df0079e475f

Observation 82da996e-dfcc-425a-924f-599f1447b1f9 · outbound

This paper cites Pond: CXL-Based Memory Pooling Systems for Cloud Platforms,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Pond: CXL-Based Memory Pooling Systems for Cloud Platforms,

Reference 65

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.698491Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.698491Z digest=sha256:27a0eb9e5e39ef5944118a2f7432aa9b1500de80067db0f243297579d25fb255

Observation ff480925-c66a-48f1-90df-47112c03486f · outbound

This paper cites Breaking the Memory Wall with Compute Express Link (CXL),.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Breaking the Memory Wall with Compute Express Link (CXL),

Reference 66

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.855162Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.855162Z digest=sha256:a9dfcf7acb26c895f9e6989028c61db03da1e8a4092c738ab85c46286cee6b41

Observation f68a0c7c-ec19-412b-be8f-7ec0cdc963ca · outbound

This paper cites Direct access,{High-Performance} memory disaggregation with {DirectCXL},.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Direct access,{High-Performance} memory disaggregation with {DirectCXL},

Reference 67

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:51.978517Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:51.978517Z digest=sha256:9b950725d2deb4452a8f3b77810e7470eddb413e5c8e207bb937deabb4f9b245

Observation c7687572-0c16-4271-9ff6-d708d5201e6b · outbound

This paper cites Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,

Reference 68

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.098820Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.098820Z digest=sha256:d35c864ffb8898dd305f2a2198f2ec6aa5f7812fbe4fa68ec8e40a2927945578

Observation a0e41e77-acd9-42ee-9fb5-3b0c00fbbe77 · outbound

This paper cites Active retrieval augmented generation,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Active retrieval augmented generation,

Reference 69

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.250936Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.250936Z digest=sha256:053c33697f4e6d7d868ad6a52bb298eb88c125a3a33d70785a889d1e96413dda

Observation bc36a144-68c5-4bea-a581-9335f12d6625 · outbound

This paper cites Benchmarking large language models in retrieval-augmented generation,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Benchmarking large language models in retrieval-augmented generation,

Reference 70

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.273606Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.273606Z digest=sha256:026059b080dfedf7fa5090b32e4bb92b59c8b120ca7a6673ddbc8a5d36dd2521

Observation d9633924-eb71-482b-afd5-f45472ed02e3 · outbound

This paper cites From Local to Global: A Graph RAG Approach to Query-Focused Summarization.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure From Local to Global: A Graph RAG Approach to Query-Focused Summarization

Reference 71

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.278520Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.278520Z digest=sha256:354f01d1e3973cd9334bb6eb640f78e8fc904b650954793a20f9a51306fbc193

Observation 06914274-d58b-457b-8d19-abd0f881ab54 · outbound

This paper cites KV Cache is 1 Bit Per Channel: Efficient Large Language Model Inference with Coupled Quantization,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure KV Cache is 1 Bit Per Channel: Efficient Large Language Model Inference with Coupled Quantization,

Reference 72

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.282773Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.282773Z digest=sha256:8cfe66686bc353e81959b4b6e4611cfa5e0bb1945adec6ad163aea5ef3e0faf7

Observation 2b512ef9-215c-46e6-8276-0a7e88ee8044 · outbound

This paper cites InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache Management,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache Management,

Reference 73

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.286774Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.286774Z digest=sha256:ca2f1d8c184e1e44aaf12057eeae06c5eab90f9174ba4762f8bdced9c1a65fe0

Observation cd5008a8-7a74-49ce-b623-cf7ef4fcb953 · outbound

This paper cites an unresolved cited work.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Unresolved cited work

Reference 75

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.295326Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.295326Z digest=sha256:0fcd43c6a26133dbec47b631af8485d2a21a393ea133c3996b7465332d726fc6

Observation 4e619249-3789-4d1a-b210-d866683c063d · outbound

This paper cites UALink 200G 1.0 Specification,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure UALink 200G 1.0 Specification,

Reference 76

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.299507Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.299507Z digest=sha256:fbf231362ad75221ac7d44e1f0890c3edf8a204086e820fae894dfb1de289bee

Observation 0648a33f-7136-4ac6-b069-f1d56f84b3dc · outbound

This paper cites [Online].

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure [Online]

Reference 77

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.304004Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.304004Z digest=sha256:1f4a56837c8761a2e1bdf866ed7edf5c3548f315b7ee10669c819b9657e4b544

Observation eb1fa165-8f59-43ed-8ad9-299a250c5d4a · outbound

This paper cites How Ultra Ethernet and UALink Enable High-Performance, Scalable AI Net- works,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure How Ultra Ethernet and UALink Enable High-Performance, Scalable AI Net- works,

Reference 78

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.307870Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.307870Z digest=sha256:d3dcdc235aefaf9e58be216e88a73f82b00e56be49fc1f65e50166d10bb4e1ed

Observation 56a82bf3-06c3-4ccb-b828-cd3aef461c0b · outbound

This paper cites What is NVLink?.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure What is NVLink?

Reference 79

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.312269Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.312269Z digest=sha256:1f961220c9153be1ddd969495efe0f588de6a4dc9b332a69c34afd53dc5e6721

Observation 7cf12210-d465-4efd-bb1e-0ff26e842dde · outbound

This paper cites NVLink NVIDIA.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVLink NVIDIA

Reference 80

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.316486Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.316486Z digest=sha256:322966aa8e2a239ed04884e592562f9b150bd0e11ff31955271193821687f40b

Observation 46b9b429-6b4d-4d19-ad94-3b5522fb34dd · outbound

This paper cites WNVIDIA NVLink and NVLink Switch,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure WNVIDIA NVLink and NVLink Switch,

Reference 81

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.320597Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.320597Z digest=sha256:af28d60e25708e4ed68f024a194eccf9b612218e9997b0482eb264dd4b3beada

Observation 934bb20d-f772-4c9b-92df-c6c03597161e · outbound

This paper cites NVIDIA Fusion,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA Fusion,

Reference 82

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.325885Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.325885Z digest=sha256:a5a75f488c2f503807041e7177d8ed4b3685b850fba5a8d406d1eaed788f5095

Observation abe2fbb5-ab79-4646-84d3-f4a186d36bf0 · outbound

This paper cites NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure With NVIDIA Partner Ecosystem,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure With NVIDIA Partner Ecosystem,

Reference 83

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.330598Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.330598Z digest=sha256:31d46bb29073487c504937715e831fc729e9a7ae096ca8cb8de0d3815f9e8f5c

Observation 2c3d6217-7bca-42b3-b630-3322f6115450 · outbound

This paper cites Design guidelines for high performance RDMA systems,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Design guidelines for high performance RDMA systems,

Reference 84

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.335714Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.335714Z digest=sha256:08c0a9fd3f174c3a6018b8be1998b697689a18750a21a674211e1858aae68eed

Observation 00107cb5-d202-45bc-9945-85443b3620ad · outbound

This paper cites An optimized rdma qp communication mechanism for hyperscale ai infrastructure,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure An optimized rdma qp communication mechanism for hyperscale ai infrastructure,

Reference 85

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.340336Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.340336Z digest=sha256:8b0e5bf4df188ac9369c22691c76bf8988a4a5bd879221fb9df08ed6829929b3

Observation 8930f5e3-5903-4a69-9dae-15bcfe274ece · outbound

This paper cites Rdma over commodity ethernet at scale,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Rdma over commodity ethernet at scale,

Reference 86

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.344945Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.344945Z digest=sha256:64abc2947698e37fd6d3a4b1c86fd7977606743846b973c9cf9d5f8dddad8366

Observation 73cf4e6c-14cb-4bf8-b9f0-d9ba13039319 · outbound

This paper cites Rdma over ethernet for distributed training at meta scale,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Rdma over ethernet for distributed training at meta scale,

Reference 87

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.349414Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.349414Z digest=sha256:4578b33359c306bd6e59367c45814d11fceaa5f004473af520becc5618206d11

Observation c6752674-f8d7-42d1-8955-2a70fdc79b8a · outbound

This paper cites Graph 500 benchmarks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Graph 500 benchmarks,

Reference 88

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.354129Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.354129Z digest=sha256:8bef74983b6ac50736a2ef395984820c543b8b0cb0a0ef5f74a166f5f127dce4

Observation 0cbfd1c4-2aa3-4710-97ab-01553aa7803d · outbound

This paper cites NASA Parallel Benchmarks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure NASA Parallel Benchmarks,

Reference 89

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.358617Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.358617Z digest=sha256:f42ef188bc27b89fb36979616357e2d7e723940637d4791cecec001f1072f308

Observation b079bcc4-4f69-4538-88a1-67e2a3d2a180 · outbound

This paper cites Lattice Boltzmann Method Benchmarks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Lattice Boltzmann Method Benchmarks,

Reference 90

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.362917Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.362917Z digest=sha256:9dbd104d5161a8a84fa3c5e747f3ba78e41f191a7d6fa281747d0047cfde4211

Observation 9eadc4c9-ba07-4553-a645-e7379d2df155 · outbound

This paper cites Tealeaf: A mini-application to enable design-space explorations for iter- ative sparse linear solvers,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Tealeaf: A mini-application to enable design-space explorations for iter- ative sparse linear solvers,

Reference 91

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.368008Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.368008Z digest=sha256:da870a62d1e9a0b1ce973ac8da1d3b83f18cbef3b6f17643d479b81d7b901cc9

Observation a3c2b24c-ff96-4394-96e0-379e07187362 · outbound

This paper cites Advances in Numerical Modeling for Heat Transfer and Thermal Management: A Review of Computational Approaches and Environmental Impacts,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Advances in Numerical Modeling for Heat Transfer and Thermal Management: A Review of Computational Approaches and Environmental Impacts,

Reference 92

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.373255Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.373255Z digest=sha256:8f52acebc1df17fe7c2e933d4094d47cdd34e0289ccaed188e1472e6986de45d

Observation 28ac1eaa-66f3-4df4-ac04-895eedf55958 · outbound

This paper cites Sequence to sequence learning with neural networks,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Sequence to sequence learning with neural networks,

Reference 93

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.377684Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.377684Z digest=sha256:a85b9e7efe92d96b85278e39692efefa642ace298f5bc1100b85daabc5ce084b

Observation 5a373215-d156-48c6-b4a4-efc312586920 · outbound

This paper cites State-of-the-Art Speech Recognition with Sequence-to-Sequence Models,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure State-of-the-Art Speech Recognition with Sequence-to-Sequence Models,

Reference 94

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.382589Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.382589Z digest=sha256:6fbaa36d97b20b2e6a6e8328710522c9969a09e9af2ccf05f74b983ef1745049

Observation 96a0f40e-6fc5-4f28-9edc-39dac6ccafc8 · outbound

This paper cites Multi-task Sequence to Sequence Learning.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Multi-task Sequence to Sequence Learning

Reference 95

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.387160Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.387160Z digest=sha256:c1a901e619e737160de6e40379cd0f32e5b5f4856b33ca8fc6e54caa5dfc3d82

Observation 52159462-23da-4ded-b9ea-d1eb5d68801b · outbound

This paper cites Improving language understanding by generative pre-training,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Improving language understanding by generative pre-training,

Reference 96

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.391709Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.391709Z digest=sha256:7f99535d86105817869f5705319c2d450d73005a993609bdb746dc47bc3af743

Observation 3ad91b25-060f-4df5-b3be-a912d60c31ca · outbound

This paper cites New models and developer products announced at devday,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure New models and developer products announced at devday,

Reference 97

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.396528Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.396528Z digest=sha256:12bcf39e69dc86eb079ea3c88cb2e9880b12cbbdf440c4d106a08aa1571e6393

Observation 90d55f46-fc11-435e-b221-471e0fc7b53e · outbound

This paper cites Models. Llama3.1, 3.2, 3.3, Llama4.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Models. Llama3.1, 3.2, 3.3, Llama4

Reference 98

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.401383Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.401383Z digest=sha256:992b97dbbadddac61f24c364e4e06a972e4afada27e73292ffb8cb6e5b10323f

Observation b0939831-f130-44e4-8f90-42609091dea7 · outbound

This paper cites GPT-4 Turbo,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure GPT-4 Turbo,

Reference 99

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.406098Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.406098Z digest=sha256:6115d8391f0166cc3b272460c650067f690aeed61b573d445c8307ceff08a529

Observation 034394ad-decb-4ee1-b18a-097dd6a12cf2 · outbound

This paper cites Gemini 2.5: Our most intelligent AI model,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Gemini 2.5: Our most intelligent AI model,

Reference 100

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.410359Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.410359Z digest=sha256:e83b52b38e177b64110bb6809c72f1a2a39ebf14c6c467c6b062de791f2691c1

Observation a1b23261-f358-4822-8701-3df548b49dcf · outbound

This paper cites Report: Google DeepMind Gemini,.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Report: Google DeepMind Gemini,

Reference 101

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.414527Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.414527Z digest=sha256:7e59051fe4557505e5c9d1be01af10eb0c309777330f362e5cd0450cc0659580

Observation e3ebedf8-4822-4a8d-965b-a541e76b50f9 · outbound

This paper cites Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation.

Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

Reference 103

Resolution
unresolved
no resolver link, observed 2026-08-06T18:50:52.423393Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-06T18:50:52.423393Z digest=sha256:9b4cb7a921334060e306623bfcf49ba7f1f240f5662710ddd8fee0f7bb05d156

Pith citing papers

Observation 4d5fb4b4-e3ab-4423-91e0-b0b78e115360 · inbound

NUNA: Characterizing and Mitigating Non-Uniform Network Access in Multi-Die GPU Scale-Up Systems cites this paper.

NUNA: Characterizing and Mitigating Non-Uniform Network Access in Multi-Die GPU Scale-Up Systems Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure

Reference 45

Resolution
verified exact
local_arxiv, observed 2026-08-15T15:21:30.524084Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-22T06:32:14.747728+00:00.

source=pdf_text observed=2026-08-15T15:21:29.510318Z digest=sha256:9b695bc8723c120fce10747d411a674c7095b30cec4c2ad4084a1ec26fe3b0f1