我有兩張GPU卡,分別是:
- GTS450,微星 N450GTS-MD1GD3 顯示卡
- GTX580,技嘉 N580D5-15I-B 1536M DDR5/DVI2+miniHDMI1/384-bit
我有兩張GPU卡,分別是:
TECHPOWER 上的頻寬未定義測量方式,故比較起來意義似乎不大…
CUDA中則測量三向速率,比較重要的應該是D2D的速度?
另CUDA中H2D+D2D的數值和,似乎略等於TECHPOWER的數值…?
最後,國外比較GTS450及GTX580
http://www.hwcompare.com/9143/geforce-gts-450-vs-geforce-gtx-580/
提供參考
謝謝caure大的回文。
Bandwidth is the max amount of information (counted in megabytes per
second) that can be transferred past the external memory interface in a
second. The number is worked out by multiplying the card’s bus width by
its memory speed. If the card has DDR type memory, the result should be
multiplied by 2 once again. If DDR5, multiply by ANOTHER 2x.
The better the memory bandwidth, the faster the card will be in general.
唯一有看到的影響如上,
其餘看了說明,都僅說一般的BW都是指D2D最大速率…
另TECHPOWERUP裡,BW確實是算出來的,其改版資訊中:
Fillrate Pixel, Fillrate Texel, Memory Bandwidth will now correctly recalculate when clocks change
只是不知道如何算出來的=w=
總之是篇沒結論的回文XD
Host to Device = M/B main Memory to Graphic Card Memory (PCIe bus bandwidth)
D大的意思是,H to D跟D to H的頻寬取決於:PXIe頻寬、北橋頻寬。
D to H, H to D 是主記憶體加上PCIe bus的綜合頻寬:
至於像GTX580的情況, 程式若寫成一次便從主記憶體拉進大筆待處裡資料 (H2D),