[情報] 4080規格洩漏更新 cuda降到9728

[NanaMizuki (水樹奈々)]

https://videocardz.com/newz/
nvidia-geforce-rtx-4080-rumored-specs-downgraded-to-9728-cuda-cores

The embargoed RTX 40 specifications for board partners certainly do change a
lot. Hardware leaker Kopite is reporting on any change to the SKUs being
considered for the upcoming models.

Following an update to RTX 4090 and RTX 4070, the RTX 4080 is now in line for
an update. It should be noted that RTX 40 specs did change a few times
before, the most recent update for RTX 4080 was around June. Kopite has since
update the RTX 4090 and RTX 4070 specs, with the only remaining part being
4080.

At least in this case the specs did not change dramatically, there is only an
update to CUDA Core specs, now rumored at 9728. This means a downgrade from
80 active Streaming Multiprocessors to 76. It appears that the memory, TDP
and performance estimates are all the same.

The RTX 4080 is so to require 420W of power by default, and it should ship
with 16GB 256-bit GDDR6X memory. Judging from new specs, all thus far
confirmed high-end RTX 40 models are to feature 21 Gbps modules.

Kopite did not update the 3DMark TimeSpy Extreme score (as he said before
those are not estimates but an actual test result). This means that with
fewer cores, the clock speed might have gone up, although we do not know the
original and current frequency.

The RTX 4080 is expected to be announced at the same time as RTX 4090 which
is probably early October.

4080的規格自從6月被洩漏之後就沒任何新情報了

後來的規格變動都是集中在4090跟4070

根據爆料者@kopite7kimi的說法

4080的cuda從原先洩漏的10240更新成9728

不過@kopite7kimi沒有改變4080 3dmark time spy的分數

這可能意味著4080的cuda是被誤判而不是下修

--
本文作者為水樹奈奈飯

長期關注 im@s 與 LL

--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 36.226.5.248 (臺灣)
※ 文章網址: https://www.ptt.cc/bbs/PC_Shopping/M.1659768527.A.61A.html
→ a85139138: 四位數跟五位數，奇摩子有差
推 bunjie: 少1000
推 novicc: ucda？
推 helba: 沒事 還有很多時間能縮水
推 yafx4200p: 刀
推 blackway0226: 蘇媽 爭氣點
推 henry46277: cuda 才對吧？
→ otosaka: UCDA降低會變成UCLA嗎
推 oppoR20: UCDA是UCLA姐妹校嗎
推 JAY7: UCDA是加拿大學校專用的嗎
推 justicebb: uccu
推 kaj1983: 老黃：倚天切！
推 RC8377: UCC
推 iqafo: 重點是價格合理嗎
推 Zenryaku: 歪樓 笑死
推 mscp: 中間是要塞一個80TI吧，這刀法夠細
推 destroyss: 80Ti再分12G 16G 20GCUDA數也不同 神刀
→ LiNcUtT: 10240直接變80ti
推 wasitora: 單一CUDA效率是3000系列的多少倍呢?
推 AreLies: 這代DT會上AD103嗎
推 hannah5269: UCLA厲害歐，不便宜
→ oiukjyhntgb: 爆料已經麻木了，每次都爆料 然後後面又再下修 河河
推 greg7575: 丸子
推 qweertyui891: 總會有適合你的刀法
→ hong414: 我猜是今年就會出了 期待
推 monguly00: 等好久，要買新電腦，但新顯示卡沒上市…
→ dknyzzz: 能耗比不要太爆衝再考慮買來逛PTT
推 Severine: 老黃真貼心 懂得照顧各個族群
推 Bf109G6: UCDAvis (x)
→ Dcpp2015: 下修就效能也下修吧？
→ fishworm: 原標題是 cuda 數改變為 9728，怎麼翻譯後變 9278 了
→ fishworm: 從 10240 下降 512 變 9728
→ linyoyo: 因為怕30滯銷，所以下修40性能
※ 編輯: NanaMizuki (36.226.5.248 臺灣), 08/06/2022 18:22:52
推 da2286: 7000若夠強不可能下修，反之，下修拉近30系列協助銷庫存
→ erisiss0: 那還不簡單 按這道理我老黃就發一個和30性能完全一樣
→ erisiss0: 價格也一樣 叫做4080的東西 用這理論是不是就沒有庫存了
→ erisiss0: 然後5080性能也和3080一樣，可以幫4080清庫
→ erisiss0: 真是個機靈鬼，這樣就永遠都沒有困擾了呢
→ erisiss0: 據說過去的牙膏廠就好像這樣玩呢
→ erisiss0: 不知道啥原因現在就不這樣玩了呢，CC
推 tint: 4080應該就是用AD103
推 nrxadsl: 好刀工
→ jhbgybj123: 殺小 怎麼被次元刀切了
噓 binco: 哈哈。以老黃的尿性哪會讓你相同價位買到超規的東西
推 kkcity59: 老黃名菜-文思豆腐
推 Transposon: 4080s ready??
→ RaiGend0519: 哈哈
→ RaiGend0519: 看起來可以跳過等下一代了
推 tint: 砍512個CUDA其實影響還好 這一代時脈都會拉很高(對比30系)
推 aegis43210: 跳過就要等兩年了，而且兩年後製程也不會有什麼明顯進
→ aegis43210: 步
→ aegis43210: 雖然之後最高階會用MCM，效能會突飛猛進，但功耗也是
推 kimula01: 再切下去要變成YMCA了
推 luvstarrysky: amd可以加油點嗎 n牌這樣亂搞都沒事的
推 luvstarrysky: 只要rdna3能威脅到4000系列 n牌就不會這樣亂搞了
推 game636610: 再出一個4080s就補上了拉
推 spoon2013: 推奈飯(跑錯版)