I-ProHoster > Блог > izindaba ze-inthanethi > I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

Ngisho nangesikhathi sokumenyezelwa kwamakhadi evidiyo ochungechunge lokuqala lwe-GeForce RTX 20, abaningi babekholelwa ukuthi ama-Turing GPU awakweleti nhlobo ubukhulu obuncane bokuba khona kwamayunithi engeziwe: ama-RT cores nama-tensor cores. Manje, umsebenzisi oyedwa we-Reddit uhlaziye izithombe ze-infrared ze-Turing TU106 kanye ne-TU116 GPUs futhi waphetha ngokuthi amayunithi amasha ekhompyutha awathathi isikhala esiningi njengoba bekucatshangwa ekuqaleni.

I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

Okokuqala, masikhumbule ukuthi i-Turing TU106 GPU iyi-chip ye-NVIDIA encane kunazo zonke futhi ehlangene kakhulu enama-RT cores akhethekile okulandelela ama-ray kanye nama-tensor cores okusheshisa imisebenzi yobuhlakani bokwenziwa. Ngokulandelayo, iphrosesa yehluzo ye-Turing TU116, ehlobene nayo, iphucwa lawa mayunithi akhethekile wekhompiyutha futhi yingakho kwanqunywa ukuwaqhathanisa.

I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs
I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

I-NVIDIA Turing GPUs ihlukaniswe yaba amayunithi e-TPC, afaka i-pair of multiprocessors yokusakaza (Streaming Multiprocessors), esevele ihlanganisa wonke ama-cores wekhompuyutha. Futhi njengoba kuvela, i-Turing TU106 GPU inendawo ye-TPC engu-1,95 mm² ngaphezulu kune-Turing TU116, noma u-22%. Kule ndawo, i-1,25 mm² ingeye-tensor cores, futhi ngu-0,7 mm² kuphela eye-RT cores.

I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs
I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

Kuvele ukuthi ngaphandle kwe-tensor entsha ne-RT cores, iphrosesa yehluzo ye-flagship ye-Turing TU102, engaphansi kwe-GeForce RTX 2080 Ti, ngeke ithathe u-754 mm², kodwa i-684 mm² (36 TPC). Ngokulandelayo, i-Turing TU104, eyisisekelo se-GeForce RTX 2080, ingase ithathe u-498 mm² esikhundleni sika-545 mm² (24 TPC). Njengoba ubona, ngisho nangaphandle kwama-tensor nama-RT cores, ama-Turing GPU amadala angaba ama-chips amakhulu kakhulu. Okuphawulekayo okwengeziwe Pascal GPUs.


I-Tensor ne-RT cores ayithathi isikhala esiningi kuma-NVIDIA Turing GPUs

Ngakho siyini isizathu sobukhulu obungaka? Okokuqala, ama-Turing GPU abe nosayizi abakhulu benqolobane. Usayizi wama-shader nawo unyusiwe, futhi ama-Turing chips anesethi enkulu yemiyalelo namarejista amakhulu. Konke lokhu kwenze kwaba nokwenzeka ukwandisa kakhulu hhayi kuphela indawo, kodwa nokusebenza kweTuring GPUs. Isibonelo, i-GeForce RTX 2060 efanayo esekelwe ku-TU106 inikeza cishe izinga elifanayo lokusebenza njenge-GeForce GTX 1080 esekelwe ku-GP104. Eyokugcina, ngendlela, inenani elikhulu lama-25% we-CUDA cores, nakuba ithatha indawo engu-314 mm2 uma iqhathaniswa ne-410 mm2 ye-TU106 entsha. 




Source: 3dnews.ru

Engeza amazwana