qcmadness

http://www.anandtech.com/show/73 ... -smaller-than-gk110

http://www.xbitlabs.com/news/gra ... s_Next_Quarter.html

引用:

More interesting perhaps is that Matt also gave Forbes a ballpark number on the die size of their new GPU: GK110 is still 30% bigger than the new GPU, or inverted the new GPU is 23% smaller than GK110. While AMD’s small die strategy has been dead for some time, the company has still shied away from large GPUs for various reasons, their largest GPU since the ill-fated R600 (HD 2900 XT) being the 389mm2 Cayman GPU at the heart of the HD 6900 series. 23% smaller than GK110 would put the die size of AMD’s future GPU at around 425mm2, making it slightly larger than Cayman, or roughly the same size as R600.

引用:

“[Next generation AMD Radeon graphics cards] are coming in Q4. I cannot reveal a price-point but we are looking at more traditional enthusiast GPU price-points. We are not targeting a $999 single GPU solution like our competition because we believe not a lot of people have that $999. We normally address what we call the ultra-enthusiast segment with a dual-GPU offering like the Radeon HD 7990. So this next-generation line is targeting more of the enthusiast market versus the ultra-enthusiast one,” said Matt Skynner, corporate vice president and general manager of AMD’s graphics business unit, in an interview with Forbes web-site.

即係又玩efficiency

ccw

so latttttttte

Puff

http://bbs.hk-spot.com/viewthrea ... &extra=page%3D1

不過無論點都會有人彈出黎話「AMD 咁廢架大芯片都造唔出」，或者「卡皇依然係 Titan，訓啦 AMD 」

dom

well, they should be fine as long as driver ready at launch

Puff

引用:

原帖由 dom 於 2013-9-19 13:17 發表
well, they should be fine as long as driver ready at launch

單卡 driver 唔係太大問題掛，係 CFX 會出事多。
3DCentre 估 40-44CU + 4 prim/clk + 384 bit + 32/48 ROP
傻瓜估算法: Bonaire 160mm2 三粒 - (UVD/PCIe/VCE/Display 20mm2 x2) = 440 mm2


[ 本帖最後由 Puff 於 2013-9-19 16:00 編輯 ]

qcmadness

引用:

原帖由 Puff 於 2013-9-19 15:56 發表

單卡 driver 唔係太大問題掛，係 CFX 會出事多。
3DCentre 估 40-44CU + 4 prim/clk + 384 bit + 32/48 ROP
傻瓜估算法: Bonaire 160mm2 三粒 - (UVD/PCIe/VCE/Display 20mm2 x2) = 440 mm2
...

40CU, 384-bit

Puff

引用:

原帖由 qcmadness 於 2013-9-19 17:06 發表

40CU, 384-bit

40 CU 384b 走唔甩，所以依家係歡樂唔收錢開口中遊戲既時間

stupid math super densely packed 440-450mm2 有 42CU/384b 6Gbps/48 ROP/6 prim-per-clock
當然影響因素就首先就 1/4 DP，6 prim/clk 會唔會多過頭，同埋額外 16 ROP 有無需要 (benefit ultra high res)。
砍到 4 prim/clk 可能換得到三至四粒 CU，16 ROP 抵得上 3 粒 CU。另外縮皮用 384b 5Gbps (pitcairn) controller 又可以換 ~6 組 CU



[ 本帖最後由 Puff 於 2013-9-19 18:46 編輯 ]

qcmadness

引用:

原帖由 Puff 於 2013-9-19 18:36 發表

40 CU 384b 走唔甩，所以依家係歡樂唔收錢開口中遊戲既時間

stupid math super densely packed 440-450mm2 有 42CU/384b 6Gbps/48 ROP/6 prim-per-clock
當然影響因素就首先就 1/4 DP，6 prim/clk 會唔會多過頭，同埋額 ...

應該係40
因為講明係high-clock design

dom

引用:

原帖由 Puff 於 2013/9/19 15:56 發表

單卡 driver 唔係太大問題掛，係 CFX 會出事多。
3DCentre 估 40-44CU + 4 prim/clk + 384 bit + 32/48 ROP
傻瓜估算法: Bonaire 160mm2 三粒 - (UVD/PCIe/VCE/Display 20mm2 x2) = 440 mm2
...

老老實實我 想佢 Performance / Watt 再升

dom

A 記依家最大問題係 DRIVER .....GCN 出左咁耐 , 新出既 "Kabini" 到依家都未有正常既 Driver (WHQL)
重未計 Micro-Shuttlering 到依家都未完全FIX
我都唔知講咩好

dom

話口未完 A 記就出左  Catalyst 13.9 WHQL

Puff

引用:

原帖由 dom 於 2013-9-19 21:10 發表
話口未完 A 記就出左  Catalyst 13.9 WHQL

聽講 Dual Graphics 大改善