[情報] VEGA 20 可能會給台積電 7nm 已回收

看板PC_Shopping作者 (Lovelive!)時間9年前 (2017/09/09 00:25), 9年前編輯推噓24(25推 1噓 36→)
留言62則, 30人參與, 最新討論串1/1
http://www.digitimes.com/news/a20170908PD210.html AMD will continue to contract Globalfoundries to manufacture its next-generation Vega GPUs, and will have Siliconware Precision Industries (SPIL) as its major packager for the chips, according to industry sources. The supply of AMD's Vega 10 GPUs has fallen short of demand, thanks to robust demand for artificial intelligence (AI), Bitcoin mining and gaming applications since 2017, the sources said. Globalfoundries uses its 14nm FinFET process technology to fabricate AMD's Vega 10-series GPUs, and is expected to land orders for the next-generation Vega 11 series, the sources indicated. Packaging specialist SPIL, which has already obtained orders for AMD's Vega 10-series chips, will continue to hold the majority of backend orders for the Vega 11 series, the sources noted. The popularity of AMD's and Nvidia's high-end GPUs for applications such as AI has drawn much attention to the importance of packaging technologies for 2.5D/3D solutions, the sources identified. Heterogeneous integration for use in high-bandwidth memory (HBM) devices also requires such advanced packaging techniques. Taiwan Semiconductor Manufacturing Company (TSMC) with its CoWoS (chip-on-wafer-on-substrate) technology has reportedly secured orders for AI chips from Nvidia and Google. TSMC has further enhanced its advanced packaging capability eyeing a bigger presence in the supercomputer field, the sources said. AMD's rumored Vega 20 series will be applied to supercomputers, according to the sources. AMD is likely to switch orders to TSMC, which will use its 7nm FinFET process and CoWoS packaging to fabricate the Vega 20-series GPUs, the sources noted. -- 多虧了AI和挖礦 VEGA 10 系列供不應求賣得很好 VEGA 10 和 VEGA 11都是給 GF 14nm 做的 最後一段提到 VEGA 20 可以能直接用 GG 7nm 製程 不過 VEGA 20 是給超級電腦用的，會不會賣給一般消費市場沒有提到 -- ※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 1.165.189.59 ※ 文章網址: https://www.ptt.cc/bbs/PC_Shopping/M.1504887944.A.ED5.html ※ 編輯: KotoriCute (1.165.189.59), 09/09/2017 00:26:24
推
09/09 00:28, , 1F
GG VEGA
09/09 00:28, 1F
推
09/09 00:29, , 2F
礦工到底是怎麼搶vega的啊？量不是很少嗎？
09/09 00:29, 2F
推
09/09 00:29, , 3F
熱身中
09/09 00:29, 3F
推
09/09 00:31, , 4F
VEGA在挖礦的表現不是不佳嗎?會熱賣喔?
09/09 00:31, 4F
推
09/09 00:32, , 5F
礦工直接從管道拉貨，有些根本沒機會出現在通路上
09/09 00:32, 5F
推
09/09 00:34, , 6F
Bitcoin mining 各國記者水準差不多阿
09/09 00:34, 6F
推
09/09 00:35, , 7F
遊戲用GPU會這樣沒落嗎...缺貨缺的好誇張
09/09 00:35, 7F
→
09/09 00:35, , 8F
Digitimes 也是台灣記者啊XD
09/09 00:35, 8F
推
09/09 00:36, , 9F
喔喔XD 原來
09/09 00:36, 9F
→
09/09 00:36, , 10F
digitimes 就電子時報的英文版
09/09 00:36, 10F
推
09/09 00:37, , 11F
只要GG有負面新聞馬上就會又類似的新聞出來，可信度
09/09 00:37, 11F
→
09/09 00:37, , 12F
還是要打個問號
09/09 00:37, 12F
→
09/09 00:38, , 13F
最新路線圖vega不是只有14/14+nm嗎，為啥又有7nm了?
09/09 00:38, 13F
→
09/09 00:39, , 14F
要上也是navi上吧....
09/09 00:39, 14F
推
09/09 00:40, , 15F
GG八月營收比去年衰退呀，需要有其他新聞來消毒
09/09 00:40, 15F
推
09/09 00:40, , 16F
我看這幾天有人放實測,像挖乙太之類的只有40,跟先前
09/09 00:40, 16F
→
09/09 00:41, , 17F
在網上傳的可以破百實在差太大,這樣礦工也搶喔?
09/09 00:41, 17F
→
09/09 00:42, , 18F
而且40還是超頻才有的數字,沒超只有3X
09/09 00:42, 18F
推
09/09 01:00, , 19F
好像是降壓調整參數後有40 還更省電的樣子
09/09 01:00, 19F
→
09/09 01:00, , 20F
單卡算力高還是有吸引力 建置的主機板也是錢
09/09 01:00, 20F
推
09/09 01:08, , 21F
不是有哀鳳8嗎 QQ?
09/09 01:08, 21F
推
09/09 01:13, , 22F
感覺客戶指定要GG的吧
09/09 01:13, 22F
→
09/09 01:18, , 23F
VEGA很有吸引力啦,ETH最強,其他幣種也贏1080TI
09/09 01:18, 23F
→
09/09 01:18, , 24F
挖礦只會越來越難挖,與其省電還不如挖快一點
09/09 01:18, 24F
推
09/09 01:50, , 25F
怎麼7nm的cpu,gpu都給GG啦
09/09 01:50, 25F
→
09/09 02:14, , 26F
都擺明要用膠水了 特地丟回GG實在蠻假的
09/09 02:14, 26F
→
09/09 02:41, , 27F
用7nm的就是Navi了吧 之前roadmap上就有了
09/09 02:41, 27F
→
09/09 02:42, , 28F
到時候不知道還有沒有必要做Vega20這種東西
09/09 02:42, 28F
→
09/09 02:43, , 29F
GPU很難用膠水解決 kuma大之前說過了
09/09 02:43, 29F
→
09/09 02:43, , 30F
Vega上的infinity fabric是給APU用的
09/09 02:43, 30F
推
09/09 02:46, , 31F
你回去翻Raja講的 IF是給ML那塊用的
09/09 02:46, 31F
→
09/09 02:47, , 32F
為了APU才不會沒事塞那堆遊戲沒用的東西進去....
09/09 02:47, 32F
噓
09/09 04:10, , 33F
買不到買不起fuckoff
09/09 04:10, 33F
→
09/09 04:41, , 34F
infinity fabric有用在目前Vega內部單位互連
09/09 04:41, 34F
→
09/09 04:41, , 35F
如果是AI ML運算或挖礦,其實多晶片卡用
09/09 04:41, 35F
→
09/09 04:41, , 36F
IF頻寬連結就很夠了。
09/09 04:41, 36F
→
09/09 04:44, , 37F
Vega上的IF作用跟NVLink類似，都不是給一般人用的
09/09 04:44, 37F
→
09/09 05:38, , 38F
IF不只連外部，內連也可取代Ring bus
09/09 05:38, 38F
→
09/09 05:38, , 39F
但元件內連與跨晶片外連的用途，要分開看待
09/09 05:38, 39F
→
09/09 05:39, , 40F
外連就像Nv link，是連結CPU或其他GPU
09/09 05:39, 40F
→
09/09 05:40, , 41F
那在專業ML市場較可行。工作類型適合分散
09/09 05:40, 41F
推
09/09 07:13, , 42F
假外媒炒新聞是可以賺多少呢？
09/09 07:13, 42F
推
09/09 08:28, , 43F
22K
09/09 08:28, 43F
推
09/09 09:03, , 44F
之前也有類似的消息，7nm部分的單會給TSMC
09/09 09:03, 44F
→
09/09 09:05, , 45F
很合理，下一代xbox ps5的APU，可能還是TSMC生產的
09/09 09:05, 45F
推
09/09 09:32, , 46F
台灣媒體跟作手有特殊關係也不是一天兩天的事情
09/09 09:32, 46F
→
09/09 13:31, , 47F
這看起來只是整理網路八卦就能寫一篇
09/09 13:31, 47F
→
09/09 13:46, , 48F
gf都很緊張發個聲明稿說跟AMD合作良好目前產品都用
09/09 13:46, 48F
→
09/09 13:46, , 49F
他們製程了 感覺就有鬼
09/09 13:46, 49F
→
09/09 13:47, , 50F
7nm又難產惹八
09/09 13:47, 50F
→
09/09 14:04, , 51F
呵呵...GF..對手太強大..
09/09 14:04, 51F
推
09/09 15:08, , 52F
是GF 太爛
09/09 15:08, 52F
推
09/09 15:25, , 53F
裝機速度超緩慢，應該沒有
09/09 15:25, 53F
推
09/09 17:29, , 54F
拜託你別跳台積 沒錢狗跟人家搶產能沒人理你
09/09 17:29, 54F
→
09/09 17:29, , 55F
而且你跳我就不能嘴了操
09/09 17:29, 55F
→
09/09 17:56, , 56F
不能嘴就安靜不要嘴阿，很難嗎？
09/09 17:56, 56F
推
09/09 17:56, , 57F
礦工要去台gg下片嗎
09/09 17:56, 57F
推
09/09 18:25, , 58F
之前的傳言應該是明年一些單會投到GG7
09/09 18:25, 58F
→
09/09 18:31, , 59F
infinity fabric用在內部互聯的看不出來跟傳統做法
09/09 18:31, 59F
→
09/09 18:31, , 60F
有什麼差別 統一互聯方式方便堆疊多核心運作?
09/09 18:31, 60F
→
09/10 18:19, , 61F
也許是利於未來擴大晶片規模。
09/10 18:19, 61F
→
09/10 18:20, , 62F
Intel換掉ring架構就是單位越來越多。
09/10 18:20, 62F
‣ 更多 KotoriCute 的文章
文章代碼(AID): #1PiiI8xL (PC_Shopping)