p*u
发帖数: 2454 | 1 http://www.theinquirer.net/inquirer/feature/1730197/itll-sandy-bridge-
bulldozer-2011
When I first saw this I couldn't believe it; but it makes some sense after
thinking about it. It works like co-processor, for floating points it
dispatches instructions to different cores for more throughput. in another
word it tries to do multi-threading for software developers at hardware
level. so if you write a single threaded rendering engine, it could
finish much faster on Bulldozer than on Sandy Bridge. |
d*****0
发帖数: 68029 | |
s****c
发帖数: 11300 | 3 这是早就在搞的东西了 intel也有类似的项目 但是前景如何未知
我个人持保守，悲观的态度 起码在推土机构架上不要指望这个太多
after
another
【在 p*u 的大作中提到】
: http://www.theinquirer.net/inquirer/feature/1730197/itll-sandy-bridge-
: bulldozer-2011
: When I first saw this I couldn't believe it; but it makes some sense after
: thinking about it. It works like co-processor, for floating points it
: dispatches instructions to different cores for more throughput. in another
: word it tries to do multi-threading for software developers at hardware
: level. so if you write a single threaded rendering engine, it could
: finish much faster on Bulldozer than on Sandy Bridge.
|
i*c
发帖数: 1132 | 4 BD is rumored to run at 4-5 GHz as a result of deep pipeline and reduced
gate delay.
【在 p*u 的大作中提到】
: http://www.theinquirer.net/inquirer/feature/1730197/itll-sandy-bridge-
: bulldozer-2011
: When I first saw this I couldn't believe it; but it makes some sense after
: thinking about it. It works like co-processor, for floating points it
: dispatches instructions to different cores for more throughput. in another
: word it tries to do multi-threading for software developers at hardware
: level. so if you write a single threaded rendering engine, it could
: finish much faster on Bulldozer than on Sandy Bridge.
|
l***j
发帖数: 300 | 5 Intel 的 HyperThreading 是一个 Core 上有两个指令解码单元，所以看上去像两个
Threads. 文章中说的是两个 Core 共享一个指令解码单元，所以一个Thread的指令可
以分散到两个Core上去执行。这有点GPU的概念，如果是16 wide SIMD的话，相当于每
个Core处理8 wide。但是这两个core会有不同的register，怎么把结果最后汇总起来呢
。这个Overhead 会有多大？对于非并行的应用是不是也有提高？ |
