---
title: "O‘zbekistonda H100 ijarasi: dastlabki kirish | UZ Dynamics"
description: "NVIDIA H100: ariza orqali dastlabki kirish, mavjudlik va form-faktor aniqlashtirilmoqda. 80 GB HBM3 · 3,35 TB/s · NVLink."
url: "https://uzdynamics.ai/uz/gpus/h100"
lang: "uz"
image: "https://uzdynamics.ai/uz/gpus/h100/opengraph-image/card?f08c4f789845417d"
alternates:
  ru: "https://uzdynamics.ai/ru/gpus/h100"
  uz: "https://uzdynamics.ai/uz/gpus/h100"
  en: "https://uzdynamics.ai/en/gpus/h100"
---

> For an index of this site's pages, see [llms.txt](https://uzdynamics.ai/llms.txt). A markdown copy of each page is at its address with `.md` appended.

# NVIDIA H100 ijarasi: dastlabki kirish

Hopper · SXM (aniqlanmoqda)

Dastlabki kirish uchun arizalar qabul qilinmoqda. Quvvatlarni O‘zbekistonda joylashtirmoqdamiz. Narxlarni hozircha e’lon qilmaymiz. Mavjudligini javobda aniqlab beramiz.

H100

[H100 uchun ariza](https://uzdynamics.ai/uz/request?source=gpu-page%3Ah100&gpu=h100)

Vazifangizga mos ijara shartlarini yuboramiz. Shartnoma tuzilguncha to‘lov talab qilinmaydi.

- **Har bir GPU xotirasi:** 80 GB HBM3
- **Xotira o‘tkazuvchanligi:** 3,35 TB/s
- **Kartalar orasidagi aloqa:** NVLink 900 GB/s

H100 sxemasi: 5 ta HBM3 steki, jami 80 GB; To‘liq GH100 chipi 6 ta stekka mo‘ljallangan; NVLink 900 GB/s

Soddalashtirilgan sxema, masshtabga rioya qilinmagan

## NVIDIA H100 xususiyatlari

- **Arxitektura:** Hopper
- **E’lon qilingan yil:** 2022
- **Xotira:** 80 GB
- **Xotira turi:** HBM3
- **Xotira o‘tkazuvchanligi:** 3,35 TB/s
- **FP32:** 67 TFLOPS
- **FP16/BF16 siyrakliksiz:** 989,4 TFLOPS
- **FP8 siyrakliksiz:** 1 978,9 TFLOPS
- **FP4 siyrakliksiz:** yo‘q
- **FP64:** 34 TFLOPS
- **Tenzor yadrolarida FP64:** 67 TFLOPS
- **RT yadrolari:** yo‘q
- **NVENC/NVDEC:** NVENC 0, NVDEC 7
- **MIG:** ko‘pi bilan 7 ta MIG bo‘linmasi
- **Kartalar orasidagi aloqa:** NVLink 900 GB/s
- **Quvvat (TDP/TGP):** 700 Vt gacha
- **Form-faktor:** SXM (aniqlanmoqda)

Xotira har bir GPU uchun ko‘rsatilgan. Bir nechta GPU xotirasi alohida: katta model uchun mos dasturiy ta’minot va modelni kartalar orasida taqsimlash kerak. NVLink ularni bitta GPU qilmaydi.

Boshqa variantlar: [H100 NVL](https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/h100/PB-11773-001_v01.pdf), [H100 PCIe](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf)

H100 form-faktori aniqlashtirilmoqda: xotira hajmi, o‘tkazuvchanligi va boshqa xususiyatlar unga bog‘liq. Xususiyatlar NVIDIA rasmiy materiallari bo‘yicha, 26-sentabr, 2026 sanasida tekshirilgan.

## H100 vazifalar bo‘yicha

- Inferens: 32 mlrd parametrgacha LLM (BF16), bitta kartada
- Qo‘shimcha o‘qitish: QLoRA 70 mlrd parametrgacha, LoRA 14 mlrd parametrgacha, bitta kartada
- Noldan o‘qitish: 3 mlrd parametrgacha model, bitta kartada
- Render: RT yadrolari yo‘q
- FP64 hisob-kitoblari: 34 TFLOPS, tenzor yadrolarida 67 TFLOPS

Ochiq formulalar asosidagi taxminiy hisob: bu amaliy sinov natijasi yoki hisob-faktura emas.

## NVIDIA H100 o‘rnatilgan server

Server konfiguratsiyasi (protsessor, xotira, disk, tarmoq) aniqlanmoqda.

Hisob farazi: bitta serverda 8 tagacha GPU. Haqiqiy konfiguratsiyani javobda tasdiqlaymiz.

[8 tadan ortiq GPU? Klasterni muhokama qilish](https://uzdynamics.ai/uz/enterprise?source=gpu-page-cluster%3Ah100&gpu=h100&count=more)

## Manbalar

- [NVIDIA H100](https://www.nvidia.com/en-us/data-center/h100/)
- [NVIDIA H100 Tensor Core GPU Architecture (whitepaper V1.04)](https://dam-cdn.nvd.orangelogic.com/AssetLink/705n6ur546g0uk43w0117r17n8042d73.pdf)
- [NVIDIA Video Encode and Decode Support Matrix](https://developer.nvidia.com/video-encode-decode-support-matrix)
- [NVIDIA Newsroom, March 22, 2022](https://nvidianews.nvidia.com/news/nvidia-announces-hopper-architecture-the-next-generation-of-accelerated-computing)

## O‘xshash GPU modellari

- [NVIDIA A100](https://uzdynamics.ai/uz/gpus/a100)
  80 GB HBM2e · 2,04 TB/s · NVLink
  H100 6 ta ko‘rsatkichdan 5 tasida oldinda. H100 modelida FP32 (67 va 19,5 TFLOPS) hamda tenzor yadrolaridagi FP64 unumdorligi (67 va 19,5 TFLOPS) yuqoriroq. Bir xil: xotira hajmi.
- [NVIDIA H200](https://uzdynamics.ai/uz/gpus/h200)
  141 GB HBM3e · 4,8 TB/s · NVLink
  H200 7 ta ko‘rsatkichdan 2 tasida oldinda. H200 modelida xotira hajmi (141 va 80 GB) hamda xotira o‘tkazuvchanligi (4,8 va 3,35 TB/s) yuqoriroq. Bir xil: FP32, FP16/BF16, FP8, tenzor yadrolarida FP64, NVLink.

Siyrakliksiz FP16/BF16 va FP8 — NVIDIA texnik hujjatidagi H100 SXM5 ustunidan: H200 uchun NVIDIA faqat siyraklik bilan olingan qiymatlarni beradi, ular H100 SXM qiymatlari bilan bir xil.

## Katalogdagi barcha GPU modellari

- [B300](https://uzdynamics.ai/uz/gpus/b300)
- [B200](https://uzdynamics.ai/uz/gpus/b200)
- [H200](https://uzdynamics.ai/uz/gpus/h200)
- [A100](https://uzdynamics.ai/uz/gpus/a100)
- [RTX PRO 6000](https://uzdynamics.ai/uz/gpus/rtx-pro-6000)
- [L40S](https://uzdynamics.ai/uz/gpus/l40s)
- [L4](https://uzdynamics.ai/uz/gpus/l4)
- [RTX 6000 Ada](https://uzdynamics.ai/uz/gpus/rtx-6000-ada)
- [RTX A6000](https://uzdynamics.ai/uz/gpus/rtx-a6000)
- [RTX A5000](https://uzdynamics.ai/uz/gpus/rtx-a5000)

[H100 uchun ariza](https://uzdynamics.ai/uz/request?source=gpu-page-end%3Ah100&gpu=h100)

Vazifangizga mos ijara shartlarini yuboramiz. Shartnoma tuzilguncha to‘lov talab qilinmaydi.

```json
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"UZ Dynamics","item":"https://uzdynamics.ai/uz"},{"@type":"ListItem","position":2,"name":"GPU katalogi","item":"https://uzdynamics.ai/uz/gpus"},{"@type":"ListItem","position":3,"name":"H100","item":"https://uzdynamics.ai/uz/gpus/h100"}]}
```