v*6
2 楼
已经gg,记得的一些题目分享给大家,同时也看看大家有什么解法,自己现在还不是很
确定:) 因为是一个architect的职位,所以不是纯软件的问题,有的可能需要一些
power的背景知识。
1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
One is working at 1GHz with 2 operations executed in parallel at the same
time. The other one is at 500GHz with 4 operations. Suppose we have
transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
and here leaky means leakage power of the transistor. Which multiplier
should we use considering the 3 types of transistor?
2.In the new mobile phone, we can either choose to use a 1GHz solo core or
500MHz duo core processor. What tradeoffs should we consider?
3. We run two video game benchmarks on our new designed SOC. The two
benchmarks have the same instruction set. The benchmark with higher power
consumption always work well while the other one always get stuck. What can
be the problems?
4.In Linux, we use virtual address.So each process will think it has 4 GB
memory space even if the real memory is only 2GB. Now suppose we do not have
MMU and programmer use real physical address in their program. We only have
small size of physical memory. How can we design the system?
确定:) 因为是一个architect的职位,所以不是纯软件的问题,有的可能需要一些
power的背景知识。
1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
One is working at 1GHz with 2 operations executed in parallel at the same
time. The other one is at 500GHz with 4 operations. Suppose we have
transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
and here leaky means leakage power of the transistor. Which multiplier
should we use considering the 3 types of transistor?
2.In the new mobile phone, we can either choose to use a 1GHz solo core or
500MHz duo core processor. What tradeoffs should we consider?
3. We run two video game benchmarks on our new designed SOC. The two
benchmarks have the same instruction set. The benchmark with higher power
consumption always work well while the other one always get stuck. What can
be the problems?
4.In Linux, we use virtual address.So each process will think it has 4 GB
memory space even if the real memory is only 2GB. Now suppose we do not have
MMU and programmer use real physical address in their program. We only have
small size of physical memory. How can we design the system?
c*r
3 楼
?
d*y
5 楼
赞一个。
f*5
8 楼
).
))
Can we use different Vt transistors for the two pipelines?
【在 v**********6 的大作中提到】
: 已经gg,记得的一些题目分享给大家,同时也看看大家有什么解法,自己现在还不是很
: 确定:) 因为是一个architect的职位,所以不是纯软件的问题,有的可能需要一些
: power的背景知识。
: 1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
: One is working at 1GHz with 2 operations executed in parallel at the same
: time. The other one is at 500GHz with 4 operations. Suppose we have
: transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
: and here leaky means leakage power of the transistor. Which multiplier
: should we use considering the 3 types of transistor?
: 2.In the new mobile phone, we can either choose to use a 1GHz solo core or
d*k
9 楼
有人说citigold的点数即使转到prestige也不能转航空里程?
不知道是不是真的,求助一下
不知道是不是真的,求助一下
c*p
11 楼
1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
One is working at 1GHz with 2 operations executed in parallel at the same
time. The other one is at 500MHz with 4 operations. Suppose we have
transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
and here leaky means leakage power of the transistor. Which multiplier
should we use considering the 3 types of transistor?
P_dynamic = alpha * f(^2?) * V * C
P_static = V * C * 1/f
第一个f是两倍,C是一倍。第二个f低,C大于两倍(f高的流水线可以多级,硬件复用
效率高)。
然后是看P_D和P_S的trade-off
2.In the new mobile phone, we can either choose to use a 1GHz solo core or
500MHz duo core processor. What tradeoffs should we consider?
a.功耗和频率实际上呈平方关系,两倍的频率就是四倍的功耗。
b.应用场景,强调单线程还是多线程,涉及performance
c.芯片面积,和功耗也间接相关
3. We run two video game benchmarks on our new designed SOC. The two
benchmarks have the same instruction set. The benchmark with higher power
consumption always work well while the other one always get stuck. What can
be the problems?
跑高能耗应用的时候温度上升,降频;降频之后切到了低能耗应用,卡顿但是温度降下
来了,升频,结果又切到了高能耗应用。
4.In Linux, we use virtual address.So each process will think it has 4 GB
memory space even if the real memory is only 2GB. Now suppose we do not have
MMU and programmer use real physical address in their program. We only have
small size of physical memory. How can we design the system?
不知道。把程序的PA映射到内存的address上?这不还是个MMU么……
).
))
【在 v**********6 的大作中提到】
: 已经gg,记得的一些题目分享给大家,同时也看看大家有什么解法,自己现在还不是很
: 确定:) 因为是一个architect的职位,所以不是纯软件的问题,有的可能需要一些
: power的背景知识。
: 1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
: One is working at 1GHz with 2 operations executed in parallel at the same
: time. The other one is at 500GHz with 4 operations. Suppose we have
: transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
: and here leaky means leakage power of the transistor. Which multiplier
: should we use considering the 3 types of transistor?
: 2.In the new mobile phone, we can either choose to use a 1GHz solo core or
One is working at 1GHz with 2 operations executed in parallel at the same
time. The other one is at 500MHz with 4 operations. Suppose we have
transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
and here leaky means leakage power of the transistor. Which multiplier
should we use considering the 3 types of transistor?
P_dynamic = alpha * f(^2?) * V * C
P_static = V * C * 1/f
第一个f是两倍,C是一倍。第二个f低,C大于两倍(f高的流水线可以多级,硬件复用
效率高)。
然后是看P_D和P_S的trade-off
2.In the new mobile phone, we can either choose to use a 1GHz solo core or
500MHz duo core processor. What tradeoffs should we consider?
a.功耗和频率实际上呈平方关系,两倍的频率就是四倍的功耗。
b.应用场景,强调单线程还是多线程,涉及performance
c.芯片面积,和功耗也间接相关
3. We run two video game benchmarks on our new designed SOC. The two
benchmarks have the same instruction set. The benchmark with higher power
consumption always work well while the other one always get stuck. What can
be the problems?
跑高能耗应用的时候温度上升,降频;降频之后切到了低能耗应用,卡顿但是温度降下
来了,升频,结果又切到了高能耗应用。
4.In Linux, we use virtual address.So each process will think it has 4 GB
memory space even if the real memory is only 2GB. Now suppose we do not have
MMU and programmer use real physical address in their program. We only have
small size of physical memory. How can we design the system?
不知道。把程序的PA映射到内存的address上?这不还是个MMU么……
).
))
【在 v**********6 的大作中提到】
: 已经gg,记得的一些题目分享给大家,同时也看看大家有什么解法,自己现在还不是很
: 确定:) 因为是一个architect的职位,所以不是纯软件的问题,有的可能需要一些
: power的背景知识。
: 1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
: One is working at 1GHz with 2 operations executed in parallel at the same
: time. The other one is at 500GHz with 4 operations. Suppose we have
: transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
: and here leaky means leakage power of the transistor. Which multiplier
: should we use considering the 3 types of transistor?
: 2.In the new mobile phone, we can either choose to use a 1GHz solo core or
A*h
12 楼
降级之后还能拿到citigold每个月的typ吗,看到有人说降级后还可以保持以前的好处
,比如prestige减年费。
,比如prestige减年费。
x*1
14 楼
).
))
请给我解释一下 p leakage你怎么得到 vc/f的
http://blaauw.eecs.umich.edu/getFile.php?id=38&sid=nnxdqhqhnosz
还有 p dynamic = alpha c v^2 f
【在 c****p 的大作中提到】
: 1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
: One is working at 1GHz with 2 operations executed in parallel at the same
: time. The other one is at 500MHz with 4 operations. Suppose we have
: transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
: and here leaky means leakage power of the transistor. Which multiplier
: should we use considering the 3 types of transistor?
: P_dynamic = alpha * f(^2?) * V * C
: P_static = V * C * 1/f
: 第一个f是两倍,C是一倍。第二个f低,C大于两倍(f高的流水线可以多级,硬件复用
: 效率高)。
v*6
17 楼
).
))
第一个题,根据我知道的,leakage是和电压工艺成正比,和频率关系好像不大。我也
知道是这样一个tradeoff,但是接下去就不知道怎么分析了,感觉满难的,不知道这个
pipeline关键字能不能起到什么作用。。。。
第三题,我没说清楚,这两个benchmakr是在不同的时候跑的,而且可以想成在同样的
电压和频率下跑的
【在 c****p 的大作中提到】
: 1. Suppose we have 2 pipelined hardware multipliers(or some other hardware).
: One is working at 1GHz with 2 operations executed in parallel at the same
: time. The other one is at 500MHz with 4 operations. Suppose we have
: transistors of 3 types(low leaky(30%),middle leaky(50%) and high leaky(70%))
: and here leaky means leakage power of the transistor. Which multiplier
: should we use considering the 3 types of transistor?
: P_dynamic = alpha * f(^2?) * V * C
: P_static = V * C * 1/f
: 第一个f是两倍,C是一倍。第二个f低,C大于两倍(f高的流水线可以多级,硬件复用
: 效率高)。
j*2
20 楼
真的,因为它们是应税TYP。
http://stirfrypoints.blogspot.com/2015/07/citithank-youtyp_13.h
【在 d***k 的大作中提到】
: 有人说citigold的点数即使转到prestige也不能转航空里程?
: 不知道是不是真的,求助一下
http://stirfrypoints.blogspot.com/2015/07/citithank-youtyp_13.h
【在 d***k 的大作中提到】
: 有人说citigold的点数即使转到prestige也不能转航空里程?
: 不知道是不是真的,求助一下
o*w
23 楼
操,我跟那SB没关系,鄙视那厮
d*k
26 楼
感谢!
那这点数最好的方式就是prestige换1:1.6的aa机票了吧
还有个问题
那信用卡的typ点数可以二次转吗?
比如A转B,B再转C或者C的航空里程?
【在 j****2 的大作中提到】
: 真的,因为它们是应税TYP。
: http://stirfrypoints.blogspot.com/2015/07/citithank-youtyp_13.h
那这点数最好的方式就是prestige换1:1.6的aa机票了吧
还有个问题
那信用卡的typ点数可以二次转吗?
比如A转B,B再转C或者C的航空里程?
【在 j****2 的大作中提到】
: 真的,因为它们是应税TYP。
: http://stirfrypoints.blogspot.com/2015/07/citithank-youtyp_13.h
相关阅读
请问什么意思Amount of credit you've recently obtained or applied for.申请 CHASE SAPPHIRE PREFERRED CARD的问题今年到现在Amex还没有转delta的promotion,是不是就没有了往国内汇款2500刀,求大虾科普!苹果店回国退税问题AA的points多少能订一张单程的从北京到纽约的票包子请教-Amex Delta Plat怎么样SPG 关卡offer 3kwells fargo的bonus可以在online bank查到吗?spg应该添加到已有amex账户上还是开一个新的SPG 30k bonus 值得申吗?今年应该就这样了大衰之后有大福walmart网上买gc总是payment error,怎么回事?卡多了有点力不从心啊关business卡会影响信用记录吗?被barclays关卡了刚申请了carlson club,没有秒过请教pnc开户Kohl's 有别家的gift card买吗