Uyilo lwereferensi ye-3.2kW ye-GaN yeziko ledatha amandla e-AI

I-CRPS185 3 Titanium Plus uyilo lwereferensi yeseva evela eNavitas yogqitha iimfuno ezingqongqo ze-80Plus Titanium zokuhlangabezana neemfuno zamandla ezikhulayo zamandla eziko ledatha ye-AI.
Iiprosesa ze-AI ezilambele umbane ezifana ne-Nvidia's DGX GH200 'Grace Hopper' ifuna ukuya kuthi ga kwi-1,600 W nganye, iqhuba iinkcukacha zamandla-nge-rack ukusuka kwi-30-40 kW ukuya kwi-100 kW kwikhabhathi nganye.Ngeli xesha, ngokugxilwa kwehlabathi kugcino lwamandla kunye nokunciphisa ukukhutshwa, kunye nemimiselo yamva nje yaseYurophu, unikezelo lwamandla omncedisi kufuneka lugqithise i-80Plus 'Titanium' echazwe ngokufanelekileyo.

● Ibhulorho yesiqingatha se-GaN edityaniswe kwiphakheji enye
● Isizukulwana sesithathu seGaN amandla IC

Uyilo lwereferensi ye-Navitas lunciphisa ixesha lophuhliso kwaye luvumela ukusebenza kakuhle kwamandla, ubuninzi bamandla kunye neendleko zenkqubo usebenzisa i-ICs yamandla e-GaNFast.La maqonga enkqubo abandakanya isibambiso esipheleleyo soyilo kunye ne-hardware evavanywe ngokupheleleyo, isoftware ebethelelweyo, i-schematics, i-bill-of-materials, uyilo, ukulinganisa kunye neziphumo zovavanyo lwe-hardware.

I-CRPS185 isebenzisa uyilo lwesekethe yamva nje ebandakanya i-totem-pole ye-CCM edibeneyo ye-PFC enebhulorho epheleleyo LLC.Amacandelo abalulekileyo yi-Navitas 'entsha ye-650V ye-GaNFast yamandla e-ICs, ene-robust, i-high-speed edibeneyo ye-GaN drive ukujongana novakalelo kunye nemiba ebuthathaka ehambelana ne-discrete GaN chips.
I-ICs yamandla e-GaNFast nayo inikezela ngelahleko ephantsi kakhulu yokutshintsha, kunye ne-transient-voltage ikhono ukuya kwi-800 V, kunye nezinye iinzuzo zesantya esiphezulu ezifana nentlawulo yesango eliphantsi (Qg), i-output capacitance (COSS) kwaye akukho lahleko yokubuyisela umva (Qrr ).Njengoko ukutshintshwa kwesantya esiphezulu kunciphisa ubungakanani, ubunzima kunye neendleko zamacandelo angenayo kumbane, i-Navitas iqikelela ukuba i-ICs yamandla e-GaNFast igcina i-5% yeendleko ze-LLC-stage system material cost, kunye ne-$ 64 ngombane ngamnye kumbane kwiminyaka eyi-3.

Uyilo lusebenzisa i-'Common Redundant Power Supply' (CRPS) ifomu-factor specification echazwe yi-hyperscale Open Compute Project, kuquka i-Facebook, Intel, Google, Microsoft, kunye noDell.

● Iziko loyilo lwaseTshayina kwiziko ledatha yeGaN
● I-2400W CPRS Unikezelo lwe-AC-DC lune-96% efanelekileyo

Ukusebenzisa i-CPRS, i-platform ye-CRPS185 ihambisa i-3,200 W epheleleyo yamandla kwi-1U kuphela (40 mm) x 73.5mm x 185 mm (544 cc), ukufikelela kwi-5.9 W / cc, okanye phantse i-100 W / in3 ubuninzi bamandla.Oku kukunciphisa ubungakanani be-40% ngokuthelekisa, indlela ye-silicon yelifa elilinganayo kwaye idlula ngokulula umgangatho osebenzayo we-Titanium, ifikelela ngaphezu kwe-96.5% kwi-30% yomthwalo, kwaye ngaphezu kwe-96% yolula ukusuka kwi-20% ukuya kwi-60% yomthwalo.

Xa kuthelekiswa nezisombululo zemveli 'zeTitanium', iNavitas CRPS185 3,200 W 'Titanium Plus' uyilo olusebenza kumthwalo oqhelekileyo we-30% unokunciphisa ukusetyenziswa kombane ngama-757 kWh, kunye nokunciphisa ukukhutshwa kwekharbon diokside nge-755 kg kwiminyaka emi-3.Oku kuncipha kulingana nokonga i-303 kg yamalahle.Ayincedi kuphela abathengi beziko ledatha ukuba bafezekise ukongiwa kweendleko kunye nophuculo olusebenzayo, kodwa ikwanegalelo kwiinjongo zokusingqongileyo zogcino lwamandla kunye nokunciphisa ukukhutshwa.

Ukongeza kwiiseva zamaziko edatha, uyilo lwereferensi lunokusetyenziswa kwizicelo ezifana nokutshintsha / ukuhambisa amandla ombane, unxibelelwano, kunye nezinye izicelo zekhompyutha.

“Ukuduma kwezicelo ze-AI njengeChatGPT sisiqalo nje.Njengoko i-data data rack power power inyuka nge-2x-3x, ukuya kwi-100 kW, ukuhambisa amandla amaninzi kwindawo encinci kubalulekile, "kusho uCharles Zha, i-VP kunye ne-GM ye-Navitas China.

"Simema abaqulunqi bamandla kunye nabayili benkqubo ukuba basebenzisane ne-Navitas kwaye bafumanise ukuba indlela egqibeleleyo yendlela yokusebenza kakuhle, uyilo loxinaniso lwamandla aphezulu inokubiza imali eninzi, kwaye ikhawulezise ngokuzinzileyo ukuphuculwa kweseva ye-AI."