NVIDIA TITAN V 不專業開箱 - 3C

By Gary
at 2018-03-18T20:46
at 2018-03-18T20:46
Table of Contents
一大早被郵差門鈴吵醒,趕緊開門收貨開箱,拿出一幾乎全白的盒子放在桌上:
https://i.imgur.com/eXLDkXA.jpg
打開後,除了 TITAN V 本體,還有兩本手冊與一條 DP 轉 DVI 的轉接線:
https://i.imgur.com/6vKbiLq.jpg
背板有 3 個 DisplayPort 加 1 個 HDMI 接頭:
https://i.imgur.com/U9K85dq.jpg
將本體拿出靜電袋,準備裝進系統:
https://i.imgur.com/ECRr1Ut.jpg
現有一張 ASUS ROG STRIX-GTX1080TI-O11G-GAMING:
https://i.imgur.com/nXXIVmM.jpg
插好插滿:
https://i.imgur.com/EX7rItr.jpg
開機測試:
https://i.imgur.com/ldY5EDD.jpg
利馬跑個之前寫的tensorflow程式,用小量資料組訓練個一期,結果跟預想差不多,
在不為 GV100 寫客製最佳化程式下其實只比 1080Ti 快 30% 左右,要完整用到那
640 個 Tensor Core 的運算能力,還有很多文件要看程式要寫:
https://i.imgur.com/YGDy8Oi.png
不過如果你的程式有用到雙精度浮點運算(FP64),倒是不用改程式就可以直接獲得
10 倍左右的加速,TITAN V 是 NVIDIA 首次在消費型顯卡下放全速 FP64 計算能力:
https://i.imgur.com/lIgIfMm.png
NVIDIA 的專業卡才會有的 ECC 記憶體,在 TITAN V 上還是沒有開放:
https://i.imgur.com/JjbUC5k.png
最後附上一個簡單的規格比較表:
╒═══════╤═══════╤══════╤══════╤═══════╕
│ │ Titan V │ Titan Xp │GTX 1080 Ti │ V100 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ CUDA Cores │ 5120 │ 3840 │ 3584 │ 5120 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ FP64 Cores │ 2560 │ 120 │ 112 │ 2560 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ Boost Clock │ 1455 MHz │ 1582 MHz │ 1582 MHz │ 1370 MHz │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體大小 │ 12GB │ 12GB │ 11GB │ 16GB │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體介面 │3072-bit HBM2 │384-bit G5X │352-bit G5X │4096-bit HBM2 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體頻寬 │ 653 GB/s │ 547 GB/s │ 484 GB/s │ 900 GB/s │
╞═══════╪═══════╪══════╪══════╪═══════╡
│半精度(Tensor)│ 110 TFLOPS │0.19 TFLOPS │0.17 TFLOPS │ 112 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│單精度 (FP32) │ 13.8 TFLOPS │12.1 TFLOPS │11.3 TFLOPS │ 14 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│雙精度 (FP64) │ 6.9 TFLOPS │0.38 TFLOPS │0.35 TFLOPS │ 7 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ GPU 晶片 │ GV100 │ GP102 │ GP102 │ GV100 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ Die Size │ 815 mm2 │ 471 mm2 │ 471 mm2 │ 815 mm2 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 電晶體數量 │ 21.1B │ 12B │ 12B │ 21.1B │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 製程技術 │TSMC 12nm FFN │TSMC 16nm FF│TSMC 16nm FF│TSMC 12nm FFN │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 售價 │ 10萬 │ 4萬 │ 2.8萬 │ 30萬 │
╘═══════╧═══════╧══════╧══════╧═══════╛
--
https://i.imgur.com/eXLDkXA.jpg
打開後,除了 TITAN V 本體,還有兩本手冊與一條 DP 轉 DVI 的轉接線:
https://i.imgur.com/6vKbiLq.jpg
背板有 3 個 DisplayPort 加 1 個 HDMI 接頭:
https://i.imgur.com/U9K85dq.jpg
將本體拿出靜電袋,準備裝進系統:
https://i.imgur.com/ECRr1Ut.jpg
現有一張 ASUS ROG STRIX-GTX1080TI-O11G-GAMING:
https://i.imgur.com/nXXIVmM.jpg
插好插滿:
https://i.imgur.com/EX7rItr.jpg
開機測試:
https://i.imgur.com/ldY5EDD.jpg
利馬跑個之前寫的tensorflow程式,用小量資料組訓練個一期,結果跟預想差不多,
在不為 GV100 寫客製最佳化程式下其實只比 1080Ti 快 30% 左右,要完整用到那
640 個 Tensor Core 的運算能力,還有很多文件要看程式要寫:
https://i.imgur.com/YGDy8Oi.png
不過如果你的程式有用到雙精度浮點運算(FP64),倒是不用改程式就可以直接獲得
10 倍左右的加速,TITAN V 是 NVIDIA 首次在消費型顯卡下放全速 FP64 計算能力:
https://i.imgur.com/lIgIfMm.png
NVIDIA 的專業卡才會有的 ECC 記憶體,在 TITAN V 上還是沒有開放:
https://i.imgur.com/JjbUC5k.png
最後附上一個簡單的規格比較表:
╒═══════╤═══════╤══════╤══════╤═══════╕
│ │ Titan V │ Titan Xp │GTX 1080 Ti │ V100 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ CUDA Cores │ 5120 │ 3840 │ 3584 │ 5120 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ FP64 Cores │ 2560 │ 120 │ 112 │ 2560 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ Boost Clock │ 1455 MHz │ 1582 MHz │ 1582 MHz │ 1370 MHz │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體大小 │ 12GB │ 12GB │ 11GB │ 16GB │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體介面 │3072-bit HBM2 │384-bit G5X │352-bit G5X │4096-bit HBM2 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 記憶體頻寬 │ 653 GB/s │ 547 GB/s │ 484 GB/s │ 900 GB/s │
╞═══════╪═══════╪══════╪══════╪═══════╡
│半精度(Tensor)│ 110 TFLOPS │0.19 TFLOPS │0.17 TFLOPS │ 112 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│單精度 (FP32) │ 13.8 TFLOPS │12.1 TFLOPS │11.3 TFLOPS │ 14 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│雙精度 (FP64) │ 6.9 TFLOPS │0.38 TFLOPS │0.35 TFLOPS │ 7 TFLOPS │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ GPU 晶片 │ GV100 │ GP102 │ GP102 │ GV100 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ Die Size │ 815 mm2 │ 471 mm2 │ 471 mm2 │ 815 mm2 │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 電晶體數量 │ 21.1B │ 12B │ 12B │ 21.1B │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 製程技術 │TSMC 12nm FFN │TSMC 16nm FF│TSMC 16nm FF│TSMC 12nm FFN │
╞═══════╪═══════╪══════╪══════╪═══════╡
│ 售價 │ 10萬 │ 4萬 │ 2.8萬 │ 30萬 │
╘═══════╧═══════╧══════╧══════╧═══════╛
--
Tags:
3C
All Comments

By Anonymous
at 2018-03-21T09:21
at 2018-03-21T09:21

By Cara
at 2018-03-24T08:10
at 2018-03-24T08:10

By Rachel
at 2018-03-26T15:13
at 2018-03-26T15:13

By Valerie
at 2018-03-31T09:50
at 2018-03-31T09:50

By Heather
at 2018-04-05T07:56
at 2018-04-05T07:56

By Selena
at 2018-04-06T05:57
at 2018-04-06T05:57

By Ina
at 2018-04-09T07:11
at 2018-04-09T07:11

By Audriana
at 2018-04-13T03:22
at 2018-04-13T03:22

By Sarah
at 2018-04-17T08:19
at 2018-04-17T08:19

By Necoo
at 2018-04-21T10:29
at 2018-04-21T10:29

By Skylar DavisLinda
at 2018-04-23T07:43
at 2018-04-23T07:43

By Kyle
at 2018-04-25T15:17
at 2018-04-25T15:17

By Noah
at 2018-04-29T09:59
at 2018-04-29T09:59

By Ingrid
at 2018-05-03T01:50
at 2018-05-03T01:50

By Hedy
at 2018-05-04T14:32
at 2018-05-04T14:32

By Megan
at 2018-05-06T04:39
at 2018-05-06T04:39

By Zenobia
at 2018-05-06T10:22
at 2018-05-06T10:22

By Sarah
at 2018-05-08T02:15
at 2018-05-08T02:15

By Caroline
at 2018-05-12T19:11
at 2018-05-12T19:11

By Ethan
at 2018-05-13T01:29
at 2018-05-13T01:29

By Damian
at 2018-05-13T20:08
at 2018-05-13T20:08

By Agatha
at 2018-05-14T17:51
at 2018-05-14T17:51

By Emily
at 2018-05-18T20:40
at 2018-05-18T20:40

By Eden
at 2018-05-19T17:31
at 2018-05-19T17:31

By Megan
at 2018-05-24T14:44
at 2018-05-24T14:44

By Puput
at 2018-05-27T12:39
at 2018-05-27T12:39

By Ivy
at 2018-05-31T02:02
at 2018-05-31T02:02

By Kelly
at 2018-06-01T19:28
at 2018-06-01T19:28

By Charlie
at 2018-06-02T02:34
at 2018-06-02T02:34

By Ina
at 2018-06-02T17:43
at 2018-06-02T17:43

By Andy
at 2018-06-05T02:16
at 2018-06-05T02:16

By Oscar
at 2018-06-07T18:45
at 2018-06-07T18:45

By Bennie
at 2018-06-10T07:33
at 2018-06-10T07:33

By Callum
at 2018-06-10T21:13
at 2018-06-10T21:13

By Edwina
at 2018-06-11T12:38
at 2018-06-11T12:38

By Frederica
at 2018-06-15T08:27
at 2018-06-15T08:27

By Frederica
at 2018-06-19T08:15
at 2018-06-19T08:15

By Sarah
at 2018-06-20T12:42
at 2018-06-20T12:42

By Odelette
at 2018-06-21T04:48
at 2018-06-21T04:48

By Candice
at 2018-06-25T18:10
at 2018-06-25T18:10

By Anthony
at 2018-06-28T01:55
at 2018-06-28T01:55

By Mason
at 2018-07-01T18:20
at 2018-07-01T18:20

By Franklin
at 2018-07-04T11:27
at 2018-07-04T11:27

By Isla
at 2018-07-07T04:13
at 2018-07-07T04:13

By Steve
at 2018-07-10T20:13
at 2018-07-10T20:13

By Joe
at 2018-07-12T04:37
at 2018-07-12T04:37

By Enid
at 2018-07-15T12:52
at 2018-07-15T12:52

By Kristin
at 2018-07-20T03:21
at 2018-07-20T03:21

By Caroline
at 2018-07-24T15:45
at 2018-07-24T15:45

By Ivy
at 2018-07-29T13:57
at 2018-07-29T13:57

By Iris
at 2018-07-29T22:59
at 2018-07-29T22:59

By Yedda
at 2018-08-02T02:30
at 2018-08-02T02:30

By Puput
at 2018-08-06T04:21
at 2018-08-06T04:21

By Dorothy
at 2018-08-06T20:54
at 2018-08-06T20:54

By Jack
at 2018-08-09T22:35
at 2018-08-09T22:35

By Irma
at 2018-08-11T06:25
at 2018-08-11T06:25

By Ivy
at 2018-08-12T10:44
at 2018-08-12T10:44

By Daniel
at 2018-08-16T15:20
at 2018-08-16T15:20

By Charlotte
at 2018-08-19T07:29
at 2018-08-19T07:29

By Elma
at 2018-08-23T03:36
at 2018-08-23T03:36

By Ina
at 2018-08-23T12:43
at 2018-08-23T12:43

By Joe
at 2018-08-25T00:49
at 2018-08-25T00:49

By Annie
at 2018-08-28T21:04
at 2018-08-28T21:04

By Yedda
at 2018-08-29T08:26
at 2018-08-29T08:26

By Suhail Hany
at 2018-09-01T13:23
at 2018-09-01T13:23

By Gilbert
at 2018-09-04T14:08
at 2018-09-04T14:08

By Rebecca
at 2018-09-07T07:06
at 2018-09-07T07:06

By Anonymous
at 2018-09-10T21:49
at 2018-09-10T21:49

By Daph Bay
at 2018-09-14T21:56
at 2018-09-14T21:56

By Heather
at 2018-09-19T15:08
at 2018-09-19T15:08

By Daph Bay
at 2018-09-22T12:28
at 2018-09-22T12:28

By James
at 2018-09-27T05:39
at 2018-09-27T05:39

By Madame
at 2018-10-01T22:12
at 2018-10-01T22:12

By Delia
at 2018-10-03T17:56
at 2018-10-03T17:56

By Jacob
at 2018-10-04T05:41
at 2018-10-04T05:41

By Callum
at 2018-10-07T00:10
at 2018-10-07T00:10

By Odelette
at 2018-10-09T02:45
at 2018-10-09T02:45

By Elizabeth
at 2018-10-10T08:24
at 2018-10-10T08:24

By Doris
at 2018-10-13T06:49
at 2018-10-13T06:49

By Zenobia
at 2018-10-14T01:09
at 2018-10-14T01:09

By Agnes
at 2018-10-15T22:50
at 2018-10-15T22:50

By Hardy
at 2018-10-17T04:36
at 2018-10-17T04:36

By Xanthe
at 2018-10-17T17:22
at 2018-10-17T17:22

By Eden
at 2018-10-19T10:44
at 2018-10-19T10:44

By Rae
at 2018-10-23T11:13
at 2018-10-23T11:13

By Delia
at 2018-10-23T22:57
at 2018-10-23T22:57

By Brianna
at 2018-10-28T04:01
at 2018-10-28T04:01

By Joe
at 2018-11-01T13:32
at 2018-11-01T13:32

By Elvira
at 2018-11-06T04:44
at 2018-11-06T04:44

By Zora
at 2018-11-07T22:47
at 2018-11-07T22:47

By Linda
at 2018-11-12T11:42
at 2018-11-12T11:42

By Rosalind
at 2018-11-13T15:13
at 2018-11-13T15:13

By Carolina Franco
at 2018-11-18T11:53
at 2018-11-18T11:53

By Zora
at 2018-11-23T02:51
at 2018-11-23T02:51

By Christine
at 2018-11-23T11:50
at 2018-11-23T11:50

By Olga
at 2018-11-24T11:54
at 2018-11-24T11:54
Related Posts
機殼MB500問題

By Margaret
at 2018-03-18T20:12
at 2018-03-18T20:12
保證AMD AM4能用的Ram

By Audriana
at 2018-03-18T17:53
at 2018-03-18T17:53
AMD玩大了,展會用大海報公開笑話Intel

By Enid
at 2018-03-18T16:30
at 2018-03-18T16:30
電競來說目前台灣哪家牌子做的最紅

By Mason
at 2018-03-18T16:04
at 2018-03-18T16:04
40K 遊戲用主機

By Catherine
at 2018-03-18T15:43
at 2018-03-18T15:43