{"id":138938,"date":"2025-07-14T11:11:58","date_gmt":"2025-07-14T09:11:59","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/gpuhammer-variant-ataki-rowhammer-na-pamyat-gpu"},"modified":"2025-07-14T11:11:58","modified_gmt":"2025-07-14T09:11:59","slug":"gpuhammer-variant-ataki-rowhammer-na-pamyat-gpu","status":"publish","type":"post","link":"https:\/\/prohoster.info\/ro\/blog\/news\/gpuhammer-variant-ataki-rowhammer-na-pamyat-gpu","title":{"rendered":"GPUHammer \u2014 o variant\u0103 a atacului Rowhammer asupra memoriei GPU","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Cercet\u0103torii de la Universitatea din Toronto au prezentat prima attack de tip Rowhammer aplicat\u0103 pentru a corupe con\u021binutul memoriei video. Capacitatea de a efectua atacul, care duce la substituirea a p\u00e2n\u0103 la 8 bi\u021bi de date, a fost demonstrat\u0103 pe GPU-ul discret NVIDIA A6000 cu memorie video GDDR6. Ca exemplu practic, este ar\u0103tat cum poate fi folosit\u0103 o astfel de corup\u021bie pentru a interveni \u00een execu\u021bia modelurilor de \u00eenv\u0103\u021bare automat\u0103, reduc\u00e2nd semnificativ (de la 80% la 0.1%) acurate\u021bea rezultatelor emise prin schimbarea valorii unui singur bit.

\"GPUHammer<\/center> <\/p>\n

P\u00e2n\u0103 acum, crearea atacurilor de tip Rowhammer pentru memoria video a fost \u00eengreunat\u0103 din cauza dificult\u0103\u021bii de a determina aranjamentul fizic al memoriei \u00een cipurile GDDR, \u00eent\u00e2rzierilor mari la accesul la memorie (de 4 ori mai lent) \u0219i a frecven\u021bei mai mari de actualizare a memoriei. \u00cen plus, realizarea cercet\u0103rilor a fost complicat\u0103 de utilizarea \u00een cipurile GDDR a mecanismelor proprietare de protec\u021bie \u00eempotriva manipul\u0103rilor care duc la pierderea prematur\u0103 de sarcin\u0103, pentru analiza c\u0103rora era necesar\u0103 crearea de bancuri de teste hardware speciale bazate pe FPGA. <\/p>\n

Pentru a studia memoria video, cercet\u0103torii au creat o nou\u0103 tehnic\u0103 de inginerie invers\u0103 a DRAM GDDR, iar \u00een atac a fost folosit\u0103 optimiz\u0103ri pentru accesul la memorie utilizate \u00een organizarea calculului paralel, care au fost folosite ca amplificatoare ale intensit\u0103\u021bii accesului la celulele individuale. Pentru a efectua atacul pe GPU-ul NVIDIA a fost folosit cod low-level CUDA. <\/p>\n

\u00cen codul utilizatorului CUDA nu se efectueaz\u0103 acces\u0103ri la adresele fizice ale memoriei, dar \u00een driverul NVIDIA s-a realizat reflectarea memoriei virtuale \u00eentr-o aceea\u0219i memorie fizic\u0103, de care cercet\u0103torii au profitat pentru a calcula ofsetul memoriei virtuale \u0219i a determina aranjamentul bancurilor de memorie. Recunoa\u0219terea acces\u0103rilor la diferitele bancuri de memorie a fost realizat\u0103 prin analiza \u00eent\u00e2rzierilor, care difer\u0103 \u00een func\u021bie de accesul la unul sau la mai multe bancuri de memorie. <\/p>\n

\u00cen sistemele care folosesc \u00eempreun\u0103 GPU-uri, cum ar fi servere<\/a> Pentru implementarea modelelor de \u00eenv\u0103\u021bare automat\u0103, atacul permite unui utilizator s\u0103 modifice con\u021binutul memoriei video cu datele altui utilizator. Ca m\u0103sur\u0103 de protec\u021bie, compania NVIDIA recomand\u0103 activarea codurilor de corec\u021bie a erorilor (ECC, Error Correction Codes) cu comanda «nvidia-smi -e 1» sau utilizarea seriei de pl\u0103ci grafice care suport\u0103 OD-ECC (On-Die ECC), cum ar fi GeForce RTX 50, RTX PRO, GB200, B200, B100, H100, H200, H20 \u0219i GH200. <\/p>\n

Conform cercet\u0103torilor, activarea ECC pe sistemele cu GPU A6000 duce la o sc\u0103dere a performan\u021bei \u00een execu\u021bia modelului de \u00eenv\u0103\u021bare automat\u0103 de aproximativ 10% \u0219i la o reducere a volumului de memorie de 6,25%. Unelte pentru realizarea atacului \u0219i pentru efectuarea ingineriei inverse a layout-ului de baz\u0103 al memoriei video pe sistemele cu GPU NVIDIA au fost publicate pe GitHub. <\/p>\n

Atacul RowHammer permite distorsionarea con\u021binutului anumitor bi\u021bi din memoria DRAM prin citirea ciclic\u0103 a datelor din celulele de memorie adiacente. Deoarece memoria DRAM reprezint\u0103 un aranjament bidimensional de celule, fiecare av\u00e2nd un condensator \u0219i un tranzistor, citirea constant\u0103 a acelea\u0219i zone de memorie duce la fluctua\u021bii de tensiune \u0219i anomalii, provoc\u00e2nd o pierdere mic\u0103 de sarcin\u0103 \u00een celulele adiacente. Dac\u0103 intensitatea citirii este mare, atunci celula adiacenta poate pierde o cantitate suficient de mare de sarcin\u0103, iar ciclul de regenerare nu va reu\u0219i s\u0103 restabileasc\u0103 starea ini\u021bial\u0103, ceea ce va duce la o modificare a valorii datelor stocate \u00een celul\u0103. <\/p>\n

Metoda de atac Rowhammer a fost propus\u0103 \u00een 2014, dup\u0103 care a \u00eenceput o \u201ejoac\u0103 de-a \u0219oarecele \u0219i pisica\u201d \u00eentre cercet\u0103torii \u00een domeniul securit\u0103\u021bii \u0219i produc\u0103torii de hardware \u2014 produc\u0103torii de cipuri de memorie au \u00eencercat s\u0103 blocheze vulnerabilitatea, iar cercet\u0103torii au g\u0103sit noi modalit\u0103\u021bi de a o ocoli. De exemplu, pentru a se proteja \u00eempotriva RowHammer, produc\u0103torii de cipuri au ad\u0103ugat mecanismul TRR (Target Row Refresh), dar s-a dovedit c\u0103 acesta blocheaz\u0103 distorsionarea celulelor doar \u00een cazuri particulare, f\u0103r\u0103 a proteja \u00eempotriva tuturor variantelor posibile de atac. Metode de atac au fost dezvoltate pentru cipuri DDR3, DDR4 \u0219i DDR5 pe sisteme cu procesoare Intel, AMD \u0219i ARM. De asemenea, au fost g\u0103site modalit\u0103\u021bi de ocolire a corec\u021biei erorilor ECC \u0219i s-au propus variante de atac prin re\u021bea \u0219i prin executarea codului JavaScript \u00een browser.
\n
Sursa: opennet.ro<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

\u0418\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0438 \u0438\u0437 \u0423\u043d\u0438\u0432\u0435\u0440\u0441\u0438\u0442\u0435\u0442\u0430 \u0422\u043e\u0440\u043e\u043d\u0442\u043e \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u0438\u043b\u0438 \u043f\u0435\u0440\u0432\u0443\u044e \u0430\u0442\u0430\u043a\u0443 \u043a\u043b\u0430\u0441\u0441\u0430 Rowhammer, \u043f\u0440\u0438\u043c\u0435\u043d\u044f\u0435\u043c\u0443\u044e \u0434\u043b\u044f \u0438\u0441\u043a\u0430\u0436\u0435\u043d\u0438\u044f \u0441\u043e\u0434\u0435\u0440\u0436\u0438\u043c\u043e\u0433\u043e \u0432\u0438\u0434\u0435\u043e\u043f\u0430\u043c\u044f\u0442\u0438. \u0412\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u044c \u0441\u043e\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044f \u0430\u0442\u0430\u043a\u0438, \u043f\u0440\u0438\u0432\u043e\u0434\u044f\u0449\u0435\u0439 \u043a \u043f\u043e\u0434\u0441\u0442\u0430\u043d\u043e\u0432\u043a\u0435 \u0434\u043e 8 \u0431\u0438\u0442 \u0434\u0430\u043d\u043d\u044b\u0445, \u043f\u0440\u043e\u0434\u0435\u043c\u043e\u043d\u0441\u0442\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0430 \u043d\u0430 \u0434\u0438\u0441\u043a\u0440\u0435\u0442\u043d\u043e\u043c GPU NVIDIA A6000 \u0441 \u0432\u0438\u0434\u0435\u043e\u043f\u0430\u043c\u044f\u0442\u044c\u044e GDDR6. \u0412 \u043a\u0430\u0447\u0435\u0441\u0442\u0432\u0435 \u043f\u0440\u0430\u043a\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0433\u043e \u043f\u0440\u0438\u043c\u0435\u0440\u0430 \u043f\u043e\u043a\u0430\u0437\u0430\u043d\u043e \u043a\u0430\u043a \u043c\u043e\u0436\u043d\u043e \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u0442\u044c \u043f\u043e\u0434\u043e\u0431\u043d\u043e\u0435 \u0438\u0441\u043a\u0430\u0436\u0435\u043d\u0438\u0435 \u0434\u043b\u044f \u0432\u043c\u0435\u0448\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u0432 \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u0435 \u043c\u043e\u0434\u0435\u043b\u0435\u0439 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0438 \u0441\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e (\u0441 80% \u0434\u043e […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":138939,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-138938","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd47GPUHammer \u2014 o variant\u0103 a atacului Rowhammer asupra memoriei GPU | ProHoster","description":"Cercet\u0103torii de la Universitatea din Toronto au prezentat primul atac de tip Rowhammer, aplicat pentru distorsionarea con\u021binutului memoriei video.","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/news\/gpuhammer-variant-ataki-rowhammer-na-pamyat-gpu","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"ro_RO","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47GPUHammer \u2014 \u0432\u0430\u0440\u0438\u0430\u043d\u0442 \u0430\u0442\u0430\u043a\u0438 Rowhammer \u043d\u0430 \u043f\u0430\u043c\u044f\u0442\u044c GPU | ProHoster","og:description":"\u0418\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0438 \u0438\u0437 \u0423\u043d\u0438\u0432\u0435\u0440\u0441\u0438\u0442\u0435\u0442\u0430 \u0422\u043e\u0440\u043e\u043d\u0442\u043e \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u0438\u043b\u0438 \u043f\u0435\u0440\u0432\u0443\u044e \u0430\u0442\u0430\u043a\u0443 \u043a\u043b\u0430\u0441\u0441\u0430 Rowhammer, \u043f\u0440\u0438\u043c\u0435\u043d\u044f\u0435\u043c\u0443\u044e \u0434\u043b\u044f \u0438\u0441\u043a\u0430\u0436\u0435\u043d\u0438\u044f \u0441\u043e\u0434\u0435\u0440\u0436\u0438\u043c\u043e\u0433\u043e \u0432\u0438\u0434\u0435\u043e\u043f\u0430\u043c\u044f\u0442\u0438.","og:url":"https:\/\/prohoster.info\/ro\/blog\/news\/gpuhammer-variant-ataki-rowhammer-na-pamyat-gpu","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2025-07-14T09:11:59+00:00","article:modified_time":"2025-07-14T09:11:59+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"138938","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":"default","schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-02-09 17:49:20","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2026-01-23 13:50:21","updated":"2026-02-09 17:49:20","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/138938","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/comments?post=138938"}],"version-history":[{"count":1,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/138938\/revisions"}],"predecessor-version":[{"id":159226,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/138938\/revisions\/159226"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media\/138939"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=138938"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=138938"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=138938"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}