- Imemori ye-CPU cache (L1, L2, L3 ngisho ne-L4) inciphisa kakhulu ukubambezeleka uma kuqhathaniswa ne-RAM futhi iyisihluthulelo sokusebenza kwangempela.
- Uhlu lwe-cache lulinganisa umthamo kanye nesivinini: i-L1 kanye ne-L2 nge-core ngayinye, i-L3 eyabiwe, futhi kwezinye izimo i-L4 njengokusekelwa kwe-GPU.
- Ukubambezeleka, i-bandwidth, kanye nesilinganiso sokushaya kwe-cache kunquma ukusebenza kwemidlalo kanye nomthwalo womsebenzi onzima, kanye nesivinini se-RAM.
- Ubuchwepheshe obufana ne-3D V-Cache ye-AMD bandisa i-L3 cache ngokusebenzisa i-3D stacking, okuthuthukisa kakhulu ukusebenza kwemidlalo.
Uma kuxoxwa ngama-processor, ukugxila cishe njalo kugxile kuma-cores, imvamisa, kanye nenqubo yokukhiqiza, kodwa inkumbulo ye-cache kanye nokubambezeleka kwayo kusalokhu kunganakwa kakhulu naphezu kokuba yisihluthulelo ekusebenzeni kwangempela kwesistimu. Ukuqonda okwenzekayo phakathi kwe-CPU, i-cache, kanye ne-RAM kusenza sibone ukuthi kungani ama-processor amabili anemvamisa efanayo engenza ngendlela ehluke kakhulu.
Eminyakeni yamuva nje, imiqondo efana ne -cache latency, i-bandwidth ye-L1/L2/L3 , kanye nobuchwepheshe obufana ne-3D V-Cache ye-AMD sekube ngabadlali ababalulekile kuma-benchmarks, imidlalo, kanye nokuhlolwa kokusebenza kwe-C++, lapho ukukala ngisho nama-nanosecond ambalwa omehluko kungashintsha ngokuphelele umphumela. Ake sihlukanise konke okungemuva kwe-CPU cache latency, ukuthi kungani ibaluleke kangaka, nokuthi ilinganiswa kanjani ekusebenzeni.
Kusukela kusikhala se-CPU-RAM kuya ekuzalweni kwememori ye-cache
Ngawo-80, isivinini seprosesa sakhula ngokushesha kakhulu kunesivinini sememori . Ama-CPU aqala ukusebenzisa imiyalelo ngesivinini esisheshayo, kuyilapho izikhathi zokufinyelela kwi-RAM zahlala zinde. Umphumela waba yisithiyo esingaguquki: iprosesa yachitha isikhathi esiningi ilinde ukuthi idatha ifike.
Ukuze kuxazululwe lesi sikhala sokusebenza, onjiniyela bethule inkumbulo yesikhashana njengesendlalelo esiphakathi phakathi kwe-CPU ne-RAM. Lo mbono ulula kodwa unamandla: ukugcina, eduze kakhulu neprosesa kanye nenkumbulo esheshayo kakhulu, idatha nemiyalelo okungenzeka ukuthi i-CPU iyidinga kakhulu esikhathini esifushane. Lokhu kunciphisa ukubambezeleka okusebenzayo futhi kufihla ukuhamba kancane kwe-RAM.
Ku-PC yesimanje singahlukanisa ngokucacile amazinga amathathu okugcina: isitoreji esikhulu (i-HDD, i-SSD), esikhulu kodwa esihamba kancane; i-RAM , esheshayo kakhulu kodwa isalokhu ine-latency ebalulekile; kanye nememori ye-cache ehlanganiswe ku-CPU , encane ngomthamo, kodwa eshesha kakhulu.
I-cache ayiyona iprosesa kuphela: ama-hard drive, ama-SSD, ama-GPU, amaphrinta, namanye amadivayisi nawo anezinqolobane zawo zangaphakathi zokusheshisa ukufinyelela kwedatha. Kodwa-ke, i-cache ye-CPU inomthelela oqondile kakhulu kokuhlangenwe nakho kwansuku zonke komsebenzisi.
Iyini i-CPU cache memory futhi isebenza kanjani?
I-CPU cache iyimemori ye-SRAM ehlanganiswe ne-chip yeprosesa , ekwazi ukusebenza ngesivinini esikhulu kakhulu futhi ine-latency yezingxenyana ze-nanosecond. Ngokungafani ne-RAM (DRAM), ayidingi ukuvuselelwa njalo, ibiza kakhulu ukuyikhiqiza, futhi inomthamo omncane kakhulu; ngakho-ke, igcinelwe idatha yesikhashana nje, ebalulekile.
Umsebenzi wayo oyinhloko ukusebenza njenge- buffer esheshayo kakhulu phakathi kwama-core e-CPU ne-RAM . Uma uhlelo luqaliswa, ikhodi yalo nedatha kuqala kulayishwa kusuka endaweni yokugcina kuya esikhaleni sekheli le-RAM. Ukusuka lapho, isilawuli sememori esihlanganisiwe se-CPU siletha amabhlogo okungenzeka kakhulu ukuthi asetshenziswe ku-cache, silandela amaphethini okufinyelela kanye nama-algorithms okubikezela.
Inqubo evamile imi kanje: i-CPU icela ikheli lememori, kuqala ihlole i -cache ye-L1 , bese kuba yi-L2, bese kuthi ekugcineni kube yi-L3. Uma idatha itholakala kunoma yiliphi lala mazinga, kuvela "ukushaya" kwe-cache, isicelo sinikezwa ngokushesha, futhi imijikelezo eminingi yewashi iyalondolozwa. Uma ingatholakali kunoma yiliphi lala mazinga, kuvela "ukwehluleka" kwe-cache, futhi idatha kumele ilandwe ku- RAM, ene-latency ephezulu kanye ne-bandwidth engasebenzi kahle emisebenzini emincane.
Umklamo walolu hlobo lwenkumbulo uhlose ibhalansi: ama-cache amancane, ashesha kakhulu eduze kwenhliziyo , kanye namazinga amakhulu kodwa ahamba kancane njengoba eqhubekela phambili kude namayunithi okusebenza. Wonke umlingo wokusebenza we-CPU yesimanje uncike ekwandiseni inani lama-hits nokunciphisa izindleko zama-misses.
Amazinga e-cache: L1, L2, L3 ngisho ne-L4
Amaprosesa anamuhla ngokuvamile anamazinga amathathu e-cache ahlanganiswe ne-chip uqobo: i-L1, i-L2, ne-L3 . Amanye amamodeli athile afaka nezinga le-L4, elivame ukusetshenziswa njenge-chip ehlukile efana ne-eDRAM yemisebenzi ethile kakhulu, njengokuhlinzeka ngezinsiza ezengeziwe ku-GPU ehlanganisiwe.
Izinga ngalinye le-cache libonakala ngonxantathu wezici: ibanga elibonakalayo ukusuka enkabeni, ukubambezeleka, kanye nomthamo. Uma lisondele kumayunithi okusebenza, ukubambezeleka kwehla, kodwa futhi nosayizi walo mncane, ngoba izindleko kuma-transistors zanda kakhulu.
I-Cache L1: umugqa wokuqala wokudubula
I-cache ye-L1 iseduze kakhulu ne-kernel futhi iyona eshesha kunazo zonke izi-cache . Ivame ukuhlukaniswa ngamabhulokhi amabili ahlukene: i-cache yedatha ye-L1 (L1D) kanye ne-cache yemiyalelo ye-L1 (L1I). Eyokuqala igcina ama-operands okufanele acutshungulwe, kanti eyesibili igcina imiyalelo echazwe kahle ezosetshenziswa yi-kernel.
I-core ngayinye ine-cache yayo ye-L1, engabiwanga nabanye, okusho ukuthi akukho ukuvumelana okuqondile kwe-L1 phakathi kwama-core ahlukene. Ngokuvamile, sikhuluma ngama- 32 KB e-L1D kanye nama-32 KB e-L1I nge-core kuzakhiwo eziningi zesimanje, yize eminye imiklamo ikhulisa la manani. Inani elikhulu kancane lingatholakala kwamanye amaprosesa noma amaseva asezingeni eliphezulu.
Ngokusebenza, i-L1 inikeza ukubambezeleka kwe-sub-nanosecond kanye nesivinini sokufunda esiphezulu sezinkulungwane ze-GB/s kumathuluzi okulinganisa afana ne-AIDA64. Isibonelo, iprosesa yesimanje ye-Ryzen ingafinyelela ukubambezeleka kwe-L1 cishe okungu-0,7 ns ngama-bandwidth adlula u-2.700 GB/s, ngentengo yosayizi omncane, njenge-cache ye-L1 engu-512 KB iyonke kuwo wonke ama-cores.
I-L2 Cache: ibhalansi phakathi kosayizi nesivinini
I-cache ye-L2 ingemuva kwe-L1 njengesendlalelo sesibili sokusekelayo . Ukubambezeleka kwayo kuphakeme kancane, kodwa kusephansi kakhulu uma kuqhathaniswa ne-RAM. Umsebenzi wayo ukugcina idatha engalingani ne-L1 kodwa esasetshenziswa njalo.
Ezindaweni eziningi zokwakha zabathengi, i-core ngayinye ine-cache yayo yangasese ye-Layer 2 , engahlukaniswanga ngedatha nemiyalelo futhi ngokuvamile isukela ku-256 KB kuya ku-1 MB nge-core, yize esizukulwaneni sakamuva samaseva, osayizi abangu-1 MB nge-core noma ngaphezulu bavamile. Ngokuphathelene nokusebenza, ama-bandwidth angaphezu kuka-1.300 GB/s ane-latencies engaba ngu-2,7 ns ajwayelekile.
Njengoba inkulu, ungqimba lwe-L2 lusiza ekunciphiseni amaphutha abengafinyelela ku-L3 noma i-RAM ukube bekukhona i-L1 kuphela, kugcinwa ibhalansi enhle phakathi komthamo kanye nokubambezeleka . Kwezinye izindawo, ama-core amaningi aqoqwa ngamaqoqo abelana ngengqimba efanayo ye-L2, ethula izici ezengeziwe zokubumbana kanye namabhasi angaphakathi.
I-L3 noma i-LLC (I-Last Level Cache)
I-cache ye-L3 ivame ukubizwa ngokuthi i-LLC (Last Level Cache) ngoba, kuma-processor amaningi edeskithophu, iyizinga lokugcina le-cache ngaphambi kwe-RAM. Iyi-cache enkulu kunazo zonke futhi ehamba kancane ngaphakathi kwephakheji eyinhloko ye-CPU.
Ngokungafani nama-cache e-L1 ne-L2, i-cache ye-L3 ivame ukwabelwana ngayo phakathi kwawo wonke ama-core noma amaqembu amakhulu ama-core (isibonelo, amabhlogo ayisishiyagalombili kwamanye amadizayini e-AMD). Lokhu kuvumela noma yimuphi umgogodla ukuthi uphinde usebenzise idatha elethwe ngomunye, okwandisa izinga lokushaya lilonke, yize kuyinkimbinkimbi kakhulu i-logic yokuvumelana.
Kuprosesa yedeskithophu ephakathi nendawo, akuvamile ukubona i-L3 cache isukela ku-4 MB kuya ku-32 MB, kuyilapho ama-CPU eseva, njengochungechunge oluthile lwe-AMD EPYC, efinyelela kalula kumakhulu e-MB . Intengo okufanele ikhokhwe ukubambezeleka okuphezulu, okungaba cishe yi-10 ns, kanye ne-bandwidth ephansi kune-L1 kanye ne-L2 cache, yize isabalulekile impela: cishe i-900 GB/s ezinhlelweni ezinamandla.
I-L4 cache: amacala akhethekile
I-L4 cache iyisimo esikhethekile esingaveli kuma-CPU amaningi abathengi . Ngokuvamile isetshenziswa njenge-eDRAM engaphandle kwephakheji ye-CPU eyinhloko, kodwa isondelene kakhulu nebhodi lomama, futhi isetshenziswa kuma-processor anama-GPU ahlanganisiwe ukuhlinzeka ngokuthuthukiswa okwengeziwe ku-bandwidth yehluzo.
Isibonelo esidala kwakuyi-Intel Core i5-5775C, eyayihlanganisa i-6 MB ye-cache ye-L3 ne- 128 MB ye-eDRAM esetshenziswa njenge-cache ye-L4 ye-Iris Pro 6200 GPU yayo ehlanganisiwe. Le memori isebenze njenge-buffer yedatha yehluzo, inciphisa ukucindezeleka ku-RAM yesistimu futhi ithuthukisa ukusebenza kwemidlalo uma kuqhathaniswa namanye ama-iGPU ngaphandle kwalolu sekelo.
Ukubambezeleka kwe-cache kanye nomthelela wayo ekusebenzeni
Igama eliyisihluthulelo uma sikhuluma nge-CPU cache yi -latency: isikhathi esithathwa yi-CPU ukufinyelela idatha egcinwe ezingeni elithile lememori. Lokhu kubambezeleka kukalwa ngama-nanoseconds (ns), futhi yize lezi zikhathi zingase zibonakale zincane kakhulu, zifakwe ezigidini zokufinyelela ngomzuzwana, zenza umehluko omkhulu.
Kuhlelo olujwayelekile lwe-cache, i-latency ye-L1 iyona ephansi kakhulu (ngaphansi kuka-1 ns), i-latency ye-L2 iphindaphindwa yizici eziningana (isibonelo, ama-2,7 ns), futhi i-latency ye-L3 iyanda futhi (ingafinyelela ku-10 ns noma ngaphezulu, ikakhulukazi uma iku-chip ehlukile). Uma kungekho hit kunoma iyiphi yama-cache, iprosesa kumele ifinyelele i-RAM, lapho ama-latency angakhuphuka khona abe amasekhondi ambalwa e-nano, ngisho nangememori ye-DDR5 esheshayo.
Isibonelo sangempela: ngememori ye-Ryzen 7 7700X kanye ne-DDR5 ku-6.000 MT/s kanye ne-CL30, singabona ama-RAM latency ajwayelekile angaba ngu-70 ns, uma kuqhathaniswa no-0,7 ns ku-L1, 2,7 ns ku-L2, kanye no-10 ns ku-L3. Lo mehluko uchaza ukuthi kungani kubaluleke kangaka ukuthi i-CPU ithole idatha ku-cache: ukwehluleka kwe-cache ngakunye kuphoqa ukulinda isikhathi eside kakhulu.
Ukubambezeleka akuxhomekile kuphela kubuchwepheshe bememori obusetshenziswayo, kodwa futhi nasebangeni elibonakalayo kanye ne-topology . Emiklamo yakudala lapho ama-cache e-L2 ne-L3 ayefakwe khona kubhodi lomama, kude kakhulu ne-CPU, ukubambezeleka kwakuphezulu kakhulu futhi ukusebenza kwaphazamiseka. Ukuhlanganisa wonke ama-cache kuphakheji ye-CPU kunciphisa kakhulu lokhu kulibaziseka.
Ngaphezu kokubambezeleka, i -bandwidth ye-cache esebenzayo (elinganiswa ngama-GB/s) ikhombisa ukuthi ingakanani idatha engadluliselwa ngeyunithi yesikhathi. Lapha, i-L1 iphinde ibusa kakhulu ngesivinini sokufunda esidlula u-2.000 GB/s kuma-benchmark okwenziwa, kulandelwe yi-L2 ne-L3 enamanani aphansi kakhulu, kodwa isese ngaphezu kwe-RAM.
Ama-hits, ama-cache miss, kanye nokuvumelana kwe-cross-core
Uma i-CPU ifuna idatha bese iyithola kwelinye lamazinga e-cache, siyibiza ngokuthi i -cache hit . Uma ingayitholanga, kwenzeka iphutha le-cache , bese i-CPU kufanele isebenzise izinga elilandelayo noma, esimweni esibi kakhulu, iye kwimemori eyinhloko. Uma kufinyelelwa ama-hits amaningi emazingeni aphezulu, kulapho ukusebenza kwesistimu iyonke kuba ngcono khona.
I-cache ayigcini nje ngokungeza ukubambezeleka; ingabangela nokuthi i-CPU idingeke iphindaphinde imijikelezo yomsebenzi noma ihlele kabusha imiyalelo ngoba idatha ayikatholakali, okuphula "ipayipi langaphakathi leprosesa futhi kunciphisa ukusetshenziswa kwamayunithi ayo asebenzayo.
Kumaprosesa anezinsika eziningi, ukuhambisana kwe-cache nakho kuyadlala indima ; okungukuthi, ukugcina umbono ohambisanayo wedatha eyabiwe phakathi kwama-cores anezinsika zawo ze-L1 ne-L2. Amaphrothokholi okuhambisana okuyinkimbinkimbi aphatha ukungasebenzi nokubuyekeza imigqa ye-cache lapho omunye umgogodla uguqula idatha, okwengeza ithrafikhi yangaphakathi futhi kungathinta ukubambezeleka okusebenzayo.
Ukwakhiwa kwangaphakathi nakho kudlala indima. Kumiklamo ye-monolithic njengabaprosesa abaningi be-Intel Core, wonke ama-core afinyelela i-cache eyodwa ye-L3 ene-latency efanayo. Ku-chiplet noma i-MCM architectures, njengezizukulwane zokuqala ze-AMD Ryzen, i-cache ye-L3 ihlelwe ngamabhulokhi (CCX, CCD), futhi amanye ama-core angafinyelela ngqo ingxenye ye-cache ye-L3, ethula i-latency eyengeziwe lapho isuka kwelinye ibhulokhi iye kwelinye.
Ukubambezeleka kwe-cache uma kuqhathaniswa ne-RAM: imvamisa, izikhathi kanye nebhasi
Uma ulwazi lungekho ku-cache, i-RAM iyaqala ukusebenza . Lapha, imiqondo emibili ivame ukudideka: i-latency kanye ne-frequency. I-frequency (evame ukuvezwa ku-MT/s noma ku-MHz) ikhombisa ukuthi idatha ingadluliselwa kangakanani ngomzuzwana, kuyilapho i-latency (ku-ns noma ngezikhathi ezifana ne-CL16, CL30, njll.) ikhombisa ukuthi kuthatha isikhathi esingakanani ukuthi idatha yokuqala ilethwe.
I-RAM enemvamisa ephezulu kodwa ukubambezeleka okuphezulu kakhulu inganikeza i-bandwidth enhle kodwa isikhathi sokuphendula esibi kakhulu, okujezisa izinqubo ezibucayi njengemidlalo noma imisebenzi ethile enokufinyelela okungalingani. Ngokuphambene nalokho, amamojula anezikhathi eziphansi athuthukisa isivinini sokufinyelela kokuqala, noma ngabe i-bandwidth yawo eluhlaza ayimangalisi kangako.
Ububanzi bebhasi phakathi kwe-CPU ne-RAM nabo bubalulekile. Kumapulatifomu amaningi edeskithophu, isiteshi ngasinye sememori sinebhasi lama-bit angu-64, futhi ngokucushwa kweziteshi ezimbili, lokhu kukhuphuka kube ama-bits asebenzayo angu-128, okuphinda kabili inani ledatha elingahamba ngesikhathi esisodwa phakathi kweprosesa ne-RAM.
Uma inhlanganisela yemvamisa, ukubambezeleka, kanye nobubanzi bebhasi isebenza kahle kakhulu , kulapho ukuphuthelwa kwe-cache kuzoba buhlungu khona. Noma kunjalo, ayikho i-RAM yamanje engasondela kuzibalo ze-latency kanye ne-bandwidth ze-cache ze-L1, L2, noma ze-L3, ngakho-ke okubalulekile kuhlala kukhulisa ukushaya kwe-cache.
I-RAM ye-Scratchpad vs i-Auto Cache
Ngaphakathi kweprosesa, asitholi nje kuphela ama-cache azenzakalelayo alawulwa yihadiwe. Amanye amadizayini afaka phakathi i-Scratchpad RAM , uhlobo lwememori yangaphakathi eshesha kakhulu, ngokungafani ne-cache, engalawulwa ngokwayo.
Umehluko ucacile: imemori ye-cache iphinda ngokusobala imigqa yedatha eseduze namakheli asetshenziswayo, ilandela ama-algorithms angaphakathi, futhi umhleli wezinhlelo akanalo ukulawula okuqondile phezu kwayo. Ngokuphambene nalokho, i-RAM ye-Scratchpad isebenza njenge-RAM encane yendawo, lapho isofthiwe ngokwayo inquma ukuthi iyiphi idatha okufanele igcinwe, ukuthi ihlelwe kanjani, nokuthi isuswe nini.
Le ndlela ivame kakhulu kuma-processor afakiwe, ama-DSP, kanye namanye ama-GPU, lapho kudingeka khona amaphethini okufinyelela abikezelwa kakhulu futhi ukungaqiniseki kokugcinwa kwesikhashana okuzenzakalelayo kufanele kugwenywe. Kuma-PC namaseva edeskithophu, umsebenzisi wokugcina akavamile ukusebenzisana ngokusobala ne-Scratchpad RAM.
Ukubambezeleka kwe-cache, ukuhlolwa kokusebenza, kanye nemiphumela exakile
Ezweni lama-benchmarks, kuvamile ukukala i-cache latency kuma-nanoseconds kanye nama-bandwidth ku-GB/s ezingeni ngalinye (L1, L2, L3) usebenzisa amathuluzi anjenge-AIDA64, noma ngokuthuthukisa izivivinyo zakho ku-C++ ezenza ukuskena okuthile kwama-array amakhulu.
Lezi zivivinyo zingaveza ukuziphatha okuxakile. Isibonelo, uma ulinganisa i-bandwidth yokufunda ye-L2 ne-L3 ohlelweni olubonakala lungasebenzi, kungenzeka ukubona ukuthi amanani e-L2 ngezinye izikhathi ehla kusuka cishe ku-80 GB/s kuya cishe ku-60 GB/s , kanti amanani e-L3 angawa kusuka cishe ku-45 GB/s kuya ngaphezu kuka-35 GB/s ngaphandle kwesizathu esibonakalayo.
Incazelo evamile ukugoqa, noma izinguquko eziguquguqukayo ku-frequency yangaphakathi ye-CPU kanye nezimo zamandla. Kodwa-ke, kunezimo lapho, lapho kuvulwa ithuluzi lokuqapha njenge-HWINFO, amaphuzu abuyela ngokumangalisayo kumanani "amahle". Uma ithuluzi livaliwe, amaphuzu ayashintshashintsha futhi.
Okuvame ukwenzeka ukuthi izinhlelo ezifana ne-HWINFO zigcina i-CPU isezimweni zokusebenza okuphezulu , ziphoqa amaza azinzile futhi zivimbele izingxenye zendandatho yangaphakathi noma i-cache ukuthi ingangeni kumamodi okulala okujulile. Uma i-benchmark isebenza ngaphandle kokuthile okunye okusebenzayo, iprosesa ivame ukonga amandla, okunganciphisa i-bandwidth esebenzayo elinganisiwe ngezikhathi ezithile, yize ukubambezeleka okumaphakathi kungashintshi kakhulu.
Ezimweni ezinjalo, kuye kwaphakanyiswa izixazululo ezifana nokugcina intambo "yokugcina iphila" eyenza umsebenzi olula ukuvimbela i-CPU ukuthi ilale, kodwa akulula ngaso sonke isikhathi ukuphinda umphumela wethuluzi lokuqapha, ngoba ukwakheka kwe-microarchitecture ngayinye kunesethi yayo yezimo zokulala kanye nezinqubomgomo zangaphakathi.
Imemori ye-cache kanye nokusebenza kwemidlalo
Emidlalweni yevidiyo, i-L3 cache idlala indima ebalulekile. Imisebenzi eminingi yokudlala izuza kakhulu ngokukwazi ukugcina idatha eyengeziwe kanye nemiyalelo eduze kwe-CPU , kusukela ezakhiweni zesigcawu kuya kudatha yefiziksi kanye ne-logic yomdlalo. Uma ukufinyelela kwe-RAM kuncane, i-FPS izinzile kakhulu kanye nokuguquguquka kwesilinganiso sohlaka okuncane.
Iprosesa ene- cache ye-L3 eningi futhi esetshenziswa kahle inganikeza ukuthuthuka okuphawulekayo ekuzinzeni kwesilinganiso sohlaka, ikakhulukazi emidlalweni yanamuhla ephatha inani elikhulu lezinto futhi yenza imibuzo eminingi ephindaphindwayo ezakhiweni zedatha. Esikhundleni sokufinyelela njalo i-RAM (ehamba kancane kakhulu futhi ine-latency ephezulu), i-CPU ingakhonza izicelo eziningi ngqo kusuka ku-cache yayo ye-L3.
Kodwa-ke, i-cache ayisebenzi ngendlela emangalisayo. I-CPU enama-cores ambalwa, i-IPC ephansi, kanye nama-frequency aphansi izoqhubeka nomkhawulo, ngisho noma kune-cache enkulu ye-L3. Inhlanganisela ephumelelayo yokudlala imidlalo ivame ukuba yi -IPC ephezulu, imvamisa enhle, i-cache eyanele, kanye ne-RAM esheshayo ene-latency ephansi , ukuze lonke iphakheji linciphise izithiyo.
Uma umdlalo odinga amandla uqaliswa, uma ulinganiswa ngokuhambisana ne-benchmark efana ne-AIDA64, ungabona ukuthi i-cache ifinyelela kanjani ku-spike . Ama-micro-stutter amaningi abonakala kwezinye izihloko abangelwa ukuphutha kwe-cache okuphoqa uhlelo ukuthi lufinyelele i-RAM, noma ukushintshashintsha okuphula indawo yedatha emisiwe.
I-3D V-Cache ye-AMD kanye nomjaho we-L3 eyengeziwe
Ukuze kwephulwe umkhawulo ongokoqobo walokho okungafakwa ku-cache ye-L3 engxenyeni eyodwa eyisicaba ye-silicon, i-AMD yethule ubuchwepheshe bayo be -3D V-Cache , obuhilela ukufaka i-chip eyengeziwe ye-L3 cache phezu kwe-processor CCD (chiplet). Lokhu kwandisa kakhulu umthamo we-cache ye-L3 ngaphandle kokwandisa usayizi wesisekelo se-die.
Amaprosesa e-Ryzen ane-suffix ye-X3D, njenge- Ryzen 7 7800X3D noma i-7950X3D, athatha lo mbono ngendlela eyeqisayo. I-7800X3D, isibonelo, ihlanganisa i-32 MB ye-cache ye-L3 ku-CCD eyisisekelo kanye nenye i-64 MB ehlanganisiwe, ehlanganisa i-96 MB ye-cache ye-L3 efinyeleleka yilelo sethi lama-cores. I-7950X3D iqhubekela phambili nge-128 MB ye-cache ye-L3, ngaphezu kwe-1 MB ye-cache ye-L1 kanye ne-16 MB ye-cache ye-L2 esakazwa phakathi kwawo wonke ama-cores ayo.
Umthelela emidlalweni ye-inthanethi mkhulu kakhulu ngoba amazinga e-cache akhuphuke kakhulu , okunciphisa isidingo sokufinyelela i-RAM yezakhiwo eziningi ezazingafaneleki ngokuphelele ku-cache ye-L3. Lokhu kuchaza ukuthi kungani la mamodeli e-X3D eseyisibonelo sokusebenza kahle kwemidlalo ye-inthanethi, ngisho noma kuqhathaniswa nama-CPU anama-core amaningi noma isivinini sewashi esiphakeme kancane.
Okwamanje, imbangi yayo eyinhloko, i-Intel, ayikaze iphindaphinde ikhambi elifanayo. Le nkampani ibonise ukuthi, okwamanje, ikhetha ukugxila ekusebenzeni okulinganiselayo okuphelele kunendlela egxile emidlalweni efana namabhulokhi amakhulu e-L3 cache ehlanganisiwe. Lokhu akuvimbeli ukuthi ikhulise kancane kancane usayizi we-cache yayo ezizukulwaneni ezizayo, kodwa kubonakala sengathi ayinawo amandla okusebenzisa uhlelo olufanayo lwe-V-Cache esikhathini esifushane.
Ungabona kanjani ukuthi iprosesa yakho ine-cache engakanani
Uma ufuna ukwazi ukuthi iyiphi i-cache ye-L1, L2, kanye ne-L3 i-CPU yakho ngaphandle kokufuna amadokhumenti obuchwepheshe ngokuhlanya, enye yezindlela ezisheshayo ukusebenzisa amathuluzi afana ne -CPU-Z ku-Windows. Lolu hlelo lwamahhala lubonisa, kumathebhu akhethekile, osayizi bezinga ngalinye le-cache, kanye nenani lama-cores, imicu, kanye nolunye ulwazi olufanele.
Enye inketho ukuvakashela iwebhusayithi esemthethweni yomkhiqizi (i-Intel ARK, amashidi omkhiqizo we-AMD, njll.), yize ngokuvamile i-cache ye-L3 kuphela enemininingwane ecacile, futhi kufanele ubheke amadokhumenti obuchwepheshe ukuze uthole ama-cache e-L1 ne-L2. Ukubuyekezwa nokuhlaziywa okuvela ezikhungweni zemidiya ezikhethekile kuvame ukuhlanekezela lolu lwazi, kanye nezilinganiso ze-latency kanye ne-bandwidth.
Ukukhetha nokusebenzisa kahle i-RAM kanye nememori ye-cache
I-cache ingafihla amaphutha amaningi e-RAM, kodwa hhayi wonke. Uma wakha ikhompyutha, kubalulekile ukukhetha i-RAM enemvamisa efanele, ukubambezeleka, kanye nomthamo ofanele wokusetshenziswa kwayo okuhlosiwe. Ikhompyutha yomsebenzi wasehhovisi noma ukuphequlula okulula ingasebenza kahle ngamamojula angu-8-16 GB kanye nama-module aphakathi nendawo, kuyilapho kumageyimu nokuhlela amavidiyo, i-16 GB noma i-32 GB ivame ukunconywa.
Kuhlale kunconywa ukuhlola ukuhambisana kwe-RAM nebhodi lomama kanye ne-CPU (uhlobo lwe-DDR3, i-DDR4, i-DDR5, isivinini esiphezulu esisekelwayo, amaphrofayili e-XMP/EXPO, njll.). Kungumqondo omuhle futhi ukuthola ibhalansi phakathi kwemvamisa ephezulu kanye nesikhathi esifanele, kunokugxila kuphela enombolweni ye-MHz ekhanyayo.
Ukuze uthole okuningi ngememori kanye ne-cache, kusiza kakhulu ukugcina uhlelo lwakho lungenazo izinhlelo zangemuva ezingadingekile , ubuyekeze i-BIOS yakho kanye nabashayeli, futhi, kumakhompyutha asetshenziswa kakhulu, susa amafayela esikhashana kanye nezinqolobane zesiphequluli ezingasasebenzi. Noma yini enciphisa ukucindezeleka ku-RAM kanye ne-CPU isiza ukukhulula lezo zinsiza ezisheshayo zalokho okubaluleke ngempela.
Ezingeni le-BIOS, ukuvula amaphrofayili ememori azenzakalelayo (i-XMP, i-DOCP, i-EXPO) ngokuvamile kuyindlela elula yokuqinisekisa ukuthi i-RAM isebenza ngesivinini esithenjiswe umenzi, ukugwema ukunamathela kumanani alondolozayo anciphisa i-bandwidth.
Ukuqonda ukuthi i-cache latency isebenza kanjani, i-hierarchy yayo ye-L1/L2/L3, ubudlelwano bayo ne-RAM, kanye nobuchwepheshe obufana ne-3D V-Cache kukuvumela ukuthi ufunde imininingwane ye-CPU kanye nemiphumela yokulinganisa ngokukhanya okusha: ngale kwe-GHz kanye nenani lama-cores, ukugeleza okuningi kwesistimu kunqunywa kulawo ma-nanosecond ambalwa iprosesa ethathayo ukuthola noma ukungatholi idatha kwimemori yayo eseduze.