Overclock.net banner

1 - 3 of 3 Posts

·
Registered
Joined
·
74 Posts
Discussion Starter · #1 ·
http://www.xtremesystems.org/forums/showthread.php?275873-AMD-FX-quot-Bulldozer-quot-Review-(4)-!exclusive!-Excuse-for-1-Threaded-Perf

I've seen lots of methods rumored to boost the Bulldozer's performance in lightly-threaded applications. One of the most popular ones was disabling one core per module, and push clocks further with the surplus voltage and heat capacity.

Please forgive me if this test has been posted before. What was the reasoning behind the theory that the bulldozer could benefit from higher clocks and less cores for lightly threaded applications? I dont know about you guys but all I see is decreased performance.

User Particle said,
Quote:
I appreciate your efforts. However, it's likely that you're largely seeing the performance hit of the cache thrashing issue recently discovered since disabling a core in each module would prevent that contention.
Is he referring to the L1 cache bug? I still do not see the improvement. It looks like all 8 cores running seems to have the best performance, as stated by Majord:
Quote:
The answer from AMD was it was better off in the '2cu 4c' arrangment, and turboing higher.. however your results seem to disagree with that quite a bit.

any chance you can test single thread without affinity set between the two configs? I'd like to see the implications of threads bouncing between modules vs cores
 

·
Registered
Joined
·
323 Posts
Something I read was that Bulldozers L2 and L3 cache's have horrible latency and the cores run out of data to process. Can't say if this has any truth to it but it makes sense.

Quote:
Originally Posted by Grlzzly;15323804
Please forgive me if this test has been posted before. What was the reasoning behind the theory that the bulldozer could benefit from higher clocks and less cores for lightly threaded applications? I dont know about you guys but all I see is decreased performance.
If you only use 1 core per module you raise the TDP ceiling of the 4 remaining cores allowing the higher clocks. Higher clocks = faster processing any day of the week. This also allows the single cores to receive the entire available resources of the module. If your working on lightly threaded applications it wont matter if you have 8 cores or 80, it still all comes down to the IPC of core doing the work. Higher clock, more total instructions available.
 
1 - 3 of 3 Posts
Top