-
Since I feel like injecting more data.
P4:
note the integer dispatch limitation. [Answer 2]

Yohan:
note the integer dispatch limitation. [Answer 2]

Conroe: [Answer 3]
note the integer dispatch limitation.

Nehalem:
note the integer dispatch limitation. [Answer 3]

note that Barcelona has the EXACT SAME integer dispatch limitation. [Answer 3]
While I was gone, I wrote a couple basic micro-benchmarks:
1) A Integer compute limited micro-benchmark [Makes OoO and prediction useless; since it is impossible to mispredict and no hazards exist]
Which quickly found several interesting things:
a) performance was determined entirely by the following formula: time = 100,000 / (clock speed * number of integer instructions per clock executed)
b) When work shifted to more complex integer instruction mix [less adds and more XNORs] AMD began to have a drastic performance advantage
c) When work shifted to more complex floating point instruction mix [less adds and more shuffles] Intel began to have a drastic performance advantage
2) In a compute low, data rich program [b-trees and triple indirection vs arrays and zippers]
Several interesting things were found:
a) Performance was determined ENTIRELY by average latency in nano-seconds for data fetch [CPU clock speed, integer unit count, and etc didn't matter]
b) Caches showed improved performance until reaching 1 MB [Then showed rapidly dropping diminishing returns]
c) prediction seemed to be extremely closely tied to less than a half dozen assumptions [BOTH for AMD and Intel] Ironically both AMD and Intel, share only 3 of them. Any code outside of those assumptions had an average latency of main memory. [Extreme cache thrashing]
3) Finally in the real world; I took the following common applications and modified them to properly monitor them.
They generated the following average IPC under Ideal conditions:
a) Firefox [0.3 - 0.5 basic browsing; 2.7 when rendering video]
b) Open Office [0.01 - 0.7]
c) Apache [0.9 - 1.2]
d) XFREE86 [0.02 - 0.04]
e) GNOME [0.0001 - 0.0002]
f) KDE [0.0001 - 0.0002]
Now, I may be incorrect but what I have drawn from the preceding experiential evidence, is that with the exception of video rendering; The user will see absolutely no reduction in performance by AMD reducing the number of Integer Units.
Fast computers breed slow, lazy programmers
The price of reliability is the pursuit of the utmost simplicity. It is a price which the very rich find most hard to pay.
http://www.lighterra.com/papers/modernmicroprocessors/
Modern Ram, makes an old overclocker miss BH-5 and the fun it was
Posting Permissions
- You may not post new threads
- You may not post replies
- You may not post attachments
- You may not edit your posts
-
Forum Rules
Bookmarks