Kuleli sonto, i-NVIDIA ihlwithe ngokusemthethweni abamisiwe ukumemezela okulandelayo ochungechungeni lwabo lwama-GPU adlala i-graphics-crackin' Ampere microarchitecture yakamuva - i I-RTX A6000 I-Pro Viz GPU kanye ne- A40 Isikhungo Sedatha GPU. Balandela ukutholakala kwe-GPU evamile kokungalindelekile, okufunwa kakhulu, okunzima ukuza nge-RTX 3080 ne-RTX 3090 futhi, kamuva kule nyanga, i-RTX 3070.
Imininingwane ye-NVIDIA Ampere GPU
Uma uthanda i-RTX GPU ye-1st-gen, i-next-gen izokwenzela zonke izinhlobo ze-giddy. Amakhadi e-Ampere akhela phezu kwamakhono e-Quadro RTX futhi ethula ubuchwepheshe be-3rd-Generation of Tensor Cores ukuze kusheshiswe ukusebenza kwe-AI, i-2nd-Generation RT Cores ukusheshisa ukulandelwa kwe-ray, kanye ne-NVLink yesizukulwane sesithathu ukusheshisa ukukala kwe-GPU eminingi.
Yize idatha entsha yokusebenza iphuma nsuku zonke, nansi i-mish-mash ye-specs, okumangazayo ukuthi, isabalale kumasayithi ahlukahlukene, eqoqwe lapha ukuze kube lula kuwe.
| CUDA Cores | I-Ampere GPU Ukuqhathanisa | ||||
| A40 | A6000 | RTX 3070 | RTX 3080 | RTX 3090 | |
| GPU | GA102 | GA102 | GA104 | GA102 | GA102 |
| I-VRAM | 48GB | 48GB | 8GB | 10GB | 24GB |
| Iwashi Lenkumbulo | I-14.5 GB GDDR6 | I-16 GB GDDR6 | I-16 GB GDDR6 | I-19 GB GDDR6 | I-19.5 GB GDDR6 |
| I-Bandwidth yememori | I-696 Gbps | I-768 Gbps | I-448 Gbps | I-760 Gbps | I-936 Gbps |
| Ububanzi Bokuxhumana Kwememori | I-384-bit | I-384-bit | I-256-bit | I-320-bit | I-384-bit |
| CUDA Cores | 10752 | 10752 | 5888 | 8704 | 10496 |
| Ama-Tensor Cores (3rd-Gen) | 336 | 336 | 384 | 576 | 576 |
| I-RT Cores (2nd-Gen) | 84 | 84 | 46 | 68 | 82 |
| I-VRAM | 48GB | 48GB | 8GB | 10GB | 24GB |
| Ukusebenza kweTensor | ? | ? | I-82 TFLOPS | I-119 TFLOPS | I-143 TFLOPS |
| Ukusekelwa kwe-NVLinks | Yebo | Yebo | Cha | Cha | Yebo |
| Uhlu lwamanani | TBA | TBA | $499 | $699 | $1499 |
Uzoqaphela i-A6000 ne-A40 ihlezi cishe phakathi kwe-Geforce 30 Series RTX 3080 ne-RTX 3090 kodwa ngokuzuza ku-VRAM okufana nenkathi ye-Turing Quadro RTX 8000 kodwa ngaphezu kokuphinda kabili isibalo sayo esiyinhloko se-CUDA. Eqinisweni, kungase kusize kakhulu ukuyiqhathanisa ne-Turing GPU esezingeni eliphezulu, ngakho masenze lokho.
| CUDA Cores | I-RTX GPU Ukuqhathanisa | ||
| A40 | I-RTX A6000 | RTX 8000 | |
| GPU | GA102 | GA102 | TU102 |
| Iphrosesa Sizr | I-8 nm | I-8 nm | I-12 nm |
| I-VRAM | 48GB | 48GB | 48GB |
| Iwashi Lenkumbulo | I-14.5 GB GDDR6 | I-16 GB GDDR6 | I-14 GB GDDR6 |
| I-Bandwidth yememori | I-696 Gbps | I-768 Gbps | I-672 Gbps |
| Ububanzi Bokuxhumana Kwememori | I-384-bit | I-384-bit | I-384-bit |
| CUDA Cores | 10752 | 10752 | 4608 |
| Ama-Tensor Cores | 336 | 336 | 576 |
| Ama-RT Cores | 84 | 84 | 72 |
| Ukusebenza kweTensor | ? | ? | I-130.5 TFLOPS |
| Ukusekelwa kwe-NVLinks | Yebo | Yebo | Yebo |
| Uhlu lwamanani | TBA | TBA | $10000 |
I-NVIDIA Ampere GPU Design
Mhlawumbe njenge-sexy njengoba zonke lezi zinombolo ziwumklamo omusha we-GPU. Ngesitayela esicwebezelayo esinamathoni amabili amnyama/egolide, manje sekuwubuqili obuqinisekisa ukufihla lawa ma-GPU ngaphakathi kombhoshongo ogcwele uthuli noma irack. Bheka nje laba buhle:
Ukusebenza kwe-NVIDIA Ampere GPU
Njengoba sekushiwo, idatha entsha yokusebenza kwe-Ampere ibonakala nsuku zonke. I-Techgage ishicilele izindatshana ezimbalwa ochungechungeni lwe-RTX 30 kakade ngokubheka Ukusebenza kwe-RTX 3080 futhi Ukusebenza kwe-RTX 3090. Yize idatha ye-RTX A6000 isazofika, i-NVIDIA iveze ukuqonda okuthile kokusebenza. Uma usebenzisa i-KeyShot ne-SOLIDWORKS, i-Fusion 360, i-PTC Creo noma enye isofthiwe yokumodela, amabhentshimakhi angaphambi kwesikhathi nge-RTX A6000 abonisa ukukhuphuka kokusebenza okungu-275% ngaphezu kwe-RTX 6000. Nakhu okushiwo abanye mayelana nekhadi elisha le-Ampere.
Ikhono lokuphinda kabili noma kathathu isixazululo futhi sisheshise kakhulu ukubona kwethu kwangempela ngesikhathi sangempela samamodeli wokwakha amakhulu, ayinkimbinkimbi ezindaweni zedolobha nge-NVIDIA RTX A6000 ahlaba umxhwele kakhulu,"
U-Paul Renner, umphathi we-visualization kwa-KPF
I-NVIDIA RTX A6000 entsha isivumela ukuthi sichaze kabusha ukuthi yini engenzeka ngokulandela umkhondo wesikhathi sangempela nokufunda ngomshini.”
UDarren Hendler, umqondisi weqembu labantu elidijithali ku-Digital Domain
I-NVIDIA RTX A6000 entsha idlule ebesikulindele futhi yabonisa izinzuzo zokusebenza ezingaphezu kuka-2x ngezigcawu zangaphandle ezilandelwa yi-ray. Kuyamangaza impela.”
U-Guillaume Shan, I-Computer Graphics kanye ne-Visualization Solutions kwa-Group Renault
Bheka ukutholakala kwe-RTX A6000 maphakathi no-December futhi uqale ukubonakala ezindaweni zokusebenza ezivela e-Lenovo, Dell, HP, nezinye ekuqaleni konyaka ozayo. Uma unentshisekelo ezicini zobuchwepheshe zesakhiwo esisha se-Ampere, shaya okuthi Ukubuka konke kwe-NVIDIA Ampere futhi uhlale namabhisikidi kanye ne-cup o' gravy ozoyifunda phezu kwephepha elimhlophe elinyakazayo elingezansi.
Enye futhi ngesilinganiso esihle…
