TSMC, Nvidia & Microsoft: Tarihin Kwamfutar Zamani & Juyin Juya Halin AI

Gabatarwa: Ta yaya aka isa daga bakin fatauci zuwa juyin juya halin AI?

Tafiyar fasaha daga rikicin fatauci na shekarun 1990 zuwa babban guntu RTX Spark N1X na 2026

Abubuwan da ke ciki: Ceto & jagorancin Nvidia, hadin gwiwa & rikici da Microsoft, CUDA, Tensor Cores & NVLink, 2nm & CoWoS na TSMC, da kwamfutocin N1X na 2026.

Babi na 1: Rikicin fatauci na shekarun 1990 & matsalar da Microsoft DirectX ta haifar

A tsakiyar shekarun 1990, an tura Nvidia zuwa bakin fatauci saboda ta yi kokarin kirkirar hotunan 3D ta hanya dabam gaba daya da sauran kamfanonin kwamfuta. Wannan babban rikici ne ya zama mafi muhimmiyar juyawa a tarihin kamfanin.

1. Guntun NV1 & fasahar Quadratic Texture Mapping (QTM) (Mayu 1995)

• Fasaha: Nvidia, wadda Jensen Huang, Chris Malachowsky da Curtis Priem suka kafa a 1993, ta fito da guntunta na farko NV1 a watan Mayu 1995.
• Yadda take aiki: Maimakon gina abubuwan 3D da kananan sassa kamar sauran kamfanoni, guntun NV1 ya yi amfani da hanyar Quadratic Texture Mapping (QTM) wajen zana hotuna da fuskoki masu karkatattu da manyan kusurwa hudu (quads).
• Kasawancin kasuwanci: Ko da yake wannan fasahar da take kirkirar siffofi masu karkatattu da karancin bayanai abin mamaki ce, tana da wuyar shirye-shirye ga masu kirkirar wasanni. Sai kadan daga wasannin Sega, ba ta samun goyon bayan wani software ba.

2. Babbar matsalar da tsara DirectX 1.0 na Microsoft ya haifar (Satumba 1995)

A watan Satumba 1995, Microsoft ta fitar da software DirectX 1.0 (Direct3D) domin kawo wasannin kwamfuta a Windows 95 a karkashin ka'ida daya. A ciki, Microsoft ta ayyana cewa kawai fasahar polygon da take dogaro da kananan sassa (Triangle-based Polygon Rendering) ce za ta zama ka'idar gama gari (Standard). Da yake Direct3D ba ta goyon bayan fuskoki masu karkatattu ba, sababbin wasannin Windows ba su iya aiki a kan guntun NV1, kuma fassarar software (Emulation) ta rage sauri sosai. Da masu kera kwamfutoci suka soke odarsu, fasahar Nvidia ta zama tsohuwa cikin dare daya.

3. Sokewar guntun NV2 & taimakon dala miliyan 5 na Sega (1996)

• Wahalar shawara: Jensen Huang ne ya yanke shawarar soke guntun NV2 a tsakiyar tsara shi, wanda ake kirkirar da fasahar fuskoki masu karkatattu don na'urar wasa Dreamcast ta Sega.
• Kudi na kwanaki 30 kawai: Ko bayan an sallami rabin ma'aikata, Nvidia ta rage da kudi na kimanin kwanaki 30 kawai.
• Karamcin Shugaban Sega: Jensen Huang ya tafi Japan, ya hadu da Shugaban Sega Shoichiro Irimajiri, ya amince fili cewa NV2 kuskure ce. Ya roka ko da an soke kwangilar, Sega ta biya sauran dala miliyan 5 domin Nvidia ta rayu. Irimajiri ya amince ya ceci Nvidia.

4. Komawa da guntun RIVA 128 (1997)

• Da dala miliyan 5 daga Sega, Nvidia ta tsara guntun RIVA 128 (NV3) a hardware daidai da fasahar kananan sassa ta Direct3D ta Microsoft.
• RIVA 128 da ta fito a 1997 ta sayar da fiye da miliyan daya (1 Million) cikin watanni hudu kacal ta ceci Nvidia daga fatauci.

Babi na 2: Faduwar 3dfx & halayen jagorancin Jensen Huang

1. Yadda Nvidia ta wuce 3dfx

• Tsarin duka a daya (RIVA 128): Da yake katunan 3dfx Voodoo suna aiki ne kawai don 3D, masu amfani dole ne su sayi katin 2D dabam. Guntun RIVA 128 ya ba da 2D da 3D a guntu daya.
• Yaƙin Glide vs DirectX: Yayin da 3dfx ta dogara kawai akan API nata Glide, Nvidia ta goyi bayan ka'idojin gama gari: DirectX na Microsoft da OpenGL.
• GPU ta farko a duniya (GeForce 256, 1999): A watan Oktoba 1999, Nvidia ta fito da guntun GeForce 256 mai Hardware Transform and Lighting (T&L) ta kira shi "GPU" (Graphics Processing Unit). Wannan ya rage nauyin lissafin CPU ya kara saurin wasanni sau da yawa.
• Saurin fitarwa na watanni shida: Ta fitar da sabon guntu kowane watanni shida, Nvidia ta tura 3dfx zuwa fatauci ta sayi kadarorinta a shekarar 2000.

2. Salon jagorancin Jensen Huang da aka haifa daga rikici

• Yanayin "za mu fatauci cikin kwanaki 30": Domin girman kai na nasara kada ya zo, ya kafa al'adar yin aiki a faɗake koyaushe.
• Amincewar amincewa da kurakurai (Intellectual Honesty): Barin tsare-tsaren kuskure nan take da koya daga gazawa ba tare da ɓoye ba ya zama muhimmiyar ka'idar Nvidia.
• Kwafin aiki da farko (Simulation First): Ya tabbatar da hanyar gwada guntu da daidaito 100% a cikin kwafin software (emulator) kafin kera shi.
• Kungiya maras matakai (Flat Organization): Ya rage tawagar gudanarwa ta matakai da yawa ya gina tsarin da shugabanni sama da 50 ke kai rahoto kai tsaye gare shi.

Babi na 3: Tarihin hadin gwiwa & rikicin Nvidia & Microsoft (1995–2026)

Babi na 4: Tushen juyin juya halin AI na zamani (CUDA, Tensor Cores & NVLink)

• Software CUDA (2006): CUDA ta buɗe hanyar amfani da dubban cores na GPU a cikin lissafin kimiyya da lissafi ta hanyar harsunan C/C++, ba tare da sanin harshen hotuna ba. A shekarar 2012, tsarin AI mai suna AlexNet ya yi amfani da CUDA a kan katunan wasa Nvidia biyu ya lashe gasar ImageNet — an shuka iri na juyin juya halin AI na yau.
• Tensor Cores: Maimakon ninka lambobi biyu kawai kamar cores na yau da kullum, suna ninka cikakken teburin lambobi (Matrix) a lokaci guda, suna kammala lissafin AI da saurin walƙiya a daidaito karancin bit (FP8/FP4).
• NVLink & NVLink-C2C: A cikin cibiyoyin bayanai, NVLink ce ke haɗa ɗaruruwan GPU kai tsaye su yi aiki kamar babban guntu guda. A kwamfutocin N1X, haɗin NVLink-C2C yana haɗa CPU ta ARM da GPU ta Blackwell a cikin guntu daya yana sa ƙwaƙwalwar ajiya ta haɗin gwiwa 128GB (Unified Memory) ta zama ta gama gari ga dukansu biyu. Don haka ana iya gudanar da manyan tsare-tsaren AI (Local LLMs) ba tare da intanet ba.

Babi na 5: Tarihi & fasahar TSMC (Taiwan Semiconductor Manufacturing Company)

Taiwan, babban birnin guntu na duniya: TSMC ce ke kera fiye da kashi 90% na guntuwan semiconductor mafi wayewa a duniya. Bugu da kari, a Taiwan akwai UMC (kerar guntuwa), MediaTek (kamfanin da ya tsara CPU ta ARM ta guntun N1X tare da Nvidia), Realtek, Novatek, da kamfanin shiryawa na guntuwa ASE Technology su ma suna taka muhimmiyar rawa.

1. Matakodin tarihi na TSMC (1987–2026)

• 1987 – Haihuwar Pure-Play Foundry: Morris Chang ne ya kafa TSMC ranar 21 ga Fabrairu 1987. Ta kirkirar ka'idar "ba tare da tsara guntu da sunanta ba, sai kera guntuwan da sauran kamfanoni suka tsara kawai", ta kafa tushen ci gaban kamfanoni kamar Nvidia, Apple, AMD da Qualcomm.
• 1997–1998 – Hadin gwiwa da Nvidia: Bayan nasarar RIVA 128, TSMC ta amince da wasikar da Jensen Huang ya rubuta ta fara kerar guntuwan Nvidia.
• 2014–2018 – Kwangilar Apple & wucewa Intel: Bayan ta samu kerar guntun Apple A8 a 2014, TSMC ta yi amfani da na'urorin EUV lithography na kamfanin ASML na Holland ta wuce Intel a 2018 ta zama kamfanin kerar guntuwa lamba daya a duniya.
• 2020–2026 – Faɗaɗa zuwa duniya ("Silicon Shield"): Tana aiki a matsayin "garkuwar silicon" ta Taiwan, TSMC tana gina sababbin masana'antu a Amurka (Arizona), Japan (Kumamoto) da Jamus (Dresden) domin rage hadurran siyasar yankuna.

2. Fasahohin 2nm (N2) & CoWoS na TSMC

• Transistors na nanosheet na 2nm (N2): Maimakon tsohon tsarin FinFET, TSMC tana amfani da fasahar Gate-All-Around (Nanosheet) da ke kewaye hanyar wutar lantarki daga sassa huɗu. Wannan yana rage zubewar wuta kuma yana ba da sauri fiye da 10–15% da rage amfani da wuta 25–30% idan aka kwatanta da guntuwan 3nm.
• CoWoS (Chip-on-Wafer-on-Substrate): Fasahar shiryawa ta 3D da ke jera guntuwa dabam-dabam kamar CPU ta ARM, GPU ta Blackwell da ƙwaƙwalwar ajiya mai sauri kusa a kan tushen silicon marar kauri da ake kira silicon interposer su haɗu cikin guntu daya. Wannan ne dalilin saurin guntun Nvidia N1X da ba a taɓa gani ba.

Babi na 6: Kwamfutocin RTX Spark N1X Windows 11 na 2026

Guntun Nvidia N1X ya iso kasuwa ranar 16 ga Oktoba 2026, na musamman ga kwamfutocin Windows 11 kawai (ba ya aiki a kwamfutocin Mac; ana ba da shawarar WSL ga masu amfani da Linux). Ga bayanan manyan samfura masu CPU ta ARM mai cores 20, GPU ta Blackwell mai cores 6,144 da RAM har zuwa 128GB (da farashin kasuwar Australia):

Read more articles