As the number of cores and threads in throughput accelerators such as Graphics Processing Units (GPU) increases, so does the importance of on-chip interconnection network design. This article explores throughput-effective Network-on-Chips (NoC) for future compute accelerators that employ Bulk-Synchronous Parallel (BSP) programming models such as CUDA and OpenCL. A hardware optimization is \u201cthroughput effective\u201d if it improves parallel application-level performance per unit chip area. We evaluate performance of future looking workloads using detailed closed-loop simulations modeling compute nodes, NoC, and the DRAM memory system. We start from a mesh design with bisection bandwidth balanced to off-chip demand. Accelerator workloads tend to demand high off-chip memory bandwidth which results in a many-to-few traffic pattern when coupled with expected technology constraints of slow growth in pins-per-chip. Leveraging these observations we reduce NoC area by proposing a \u201ccheckerboard\u201d NoC which alternates between conventional\u00a0full<\/i>\u00a0routers and\u00a0half<\/i>\u00a0routers with limited connectivity. Next, we show that increasing network terminal bandwidth at the nodes connected to DRAM controllers alleviates a significant fraction of the remaining imbalance resulting from the many-to-few traffic pattern. Furthermore, we propose a \u201cdouble checkerboard inverted\u201d NoC organization which takes advantage of channel slicing to reduce area while maintaining the performance improvements of the aforementioned techniques. This organization also has a simpler routing mechanism and improves average application throughput per unit area by 24.3%.<\/p>\n","protected":false},"excerpt":{"rendered":"
As the number of cores and threads in throughput accelerators such as Graphics Processing Units (GPU) increases, so does the importance of on-chip interconnection network design. This article explores throughput-effective Network-on-Chips (NoC) for future compute accelerators that employ Bulk-Synchronous Parallel (BSP) programming models such as CUDA and OpenCL. A hardware optimization is \u201cthroughput effective\u201d if […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":null,"msr_publishername":"ACM","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"ACM Transactions on Architecture and Code Optimization (TACO)","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"ACM Transactions on Architecture and Code Optimization (TACO)","msr_number":"","msr_organization":"","msr_pages_string":"35","msr_page_range_start":"35","msr_page_range_end":"","msr_series":"","msr_volume":"10","msr_copyright":"","msr_conference_name":"","msr_doi":"https:\/\/doi.org\/10.1145\/2512429","msr_arxiv_id":"","msr_s2_paper_id":"","msr_mag_id":"","msr_pubmed_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_original_fields_of_study":"","msr_release_tracker_id":"","msr_s2_match_type":"","msr_citation_count_updated":"","msr_published_date":"2013-09-01","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"http:\/\/dl.acm.org\/citation.cfm?doid=2509420.2512429","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_s2_pdf_url":"","msr_year":0,"msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":0,"msr_s2_match_confidence":0,"msr_microsoftintellectualproperty":true,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13552,13547],"msr-publication-type":[193715],"msr-publisher":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-361865","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-hardware-devices","msr-research-area-systems-and-networking","msr-locale-en_us"],"msr_publishername":"ACM","msr_edition":"ACM Transactions on Architecture and Code Optimization (TACO)","msr_affiliation":"","msr_published_date":"2013-09-01","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"35","msr_chapter":"","msr_isbn":"","msr_journal":"ACM Transactions on Architecture and Code Optimization (TACO)","msr_volume":"10","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"http:\/\/dl.acm.org\/citation.cfm?doid=2509420.2512429","msr_doi":"https:\/\/doi.org\/10.1145\/2512429","msr_publication_uploader":[{"type":"url","title":"http:\/\/dl.acm.org\/citation.cfm?doid=2509420.2512429","viewUrl":false,"id":false,"label_id":0},{"type":"doi","title":"https:\/\/doi.org\/10.1145\/2512429","viewUrl":false,"id":false,"label_id":0}],"msr_related_uploader":"","msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"","msr_influential_citations":0,"msr_reference_count":0,"msr_arxiv_id":"","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[{"id":0,"url":"http:\/\/dl.acm.org\/citation.cfm?doid=2509420.2512429"}],"msr-author-ordering":[{"type":"user_nicename","value":"aliba","user_id":30943,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=aliba"},{"type":"text","value":"John Kim","user_id":0,"rest_url":false},{"type":"text","value":"Tor M. Aamodt","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"article","related_content":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/361865","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/361865\/revisions"}],"predecessor-version":[{"id":403121,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/361865\/revisions\/403121"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=361865"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=361865"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=361865"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=361865"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=361865"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=361865"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=361865"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=361865"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=361865"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=361865"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=361865"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=361865"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=361865"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}