Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

LOSTCIRCUITS	SHORTCUTS: A Process Shift Brisbane by Numbers Test Configurations Memory Subsystem Power Consumption 3D Rendering and Power per Renderpass POV-Ray- Cinebench AV-Encoding 3Dmark'06 F.E.A.R. DOOM3 FarCry L2 Cache Latencies The Plot Thickens Final Thoughts Give Us Some Feedback on this Review
AMD's Brisbane Core - the Transition to 65 nm And the cache latency
(Author: MS, January 5, 2007)

Final Thoughts

We started out doing just a simple performance and power analysis of the latest AMD processor and, all of a sudden, found ourselves having more fun than with any other CPU review in a long time. That is, as long as fun translates into headaches, head scratching, pulling hair and teeth - well not quite. Suffice it to say that we went through a large syringe of Arctic Silver within 2 days, changing CPUs and changing CPUs and changing CPUs. We spent about 1 week just trying to untangle the cache mystery and if we would write up all the results we have, this article would get published by the end of Q3 2007.

In the end, we believe we have enough data to argue that the 65 nm AMD core has a level 2 cache that is marginally slower than the 90 nm counterpart. The question, of course is how to define "marginally"? In this case, we argue that there is an approximately 2 cycles increase in latency across the bench, which translates into roughly 10% higher access latencies for the L2 cache. Of course, different methods of querying the cache will show different results, depending on the algorithms used and the amount of pipelining that the controller can do. The modular cache approach offers a more flexible design to increase cache size without extensive redesign, to put it simple, by adding extra cache blocks similar to adding more memory on a system level. Another possibility of course is that the current architecture is a step towards Barcelona with an added L3 cache. The exclusive nature of the cache levels might require some snooping that translate directly into 2 cycles of extra latency.

How does extra cache latency affect system performance? One thing - or rather a trend - we noticed is that heaviliy multithreaded applications showed the smallest performance hit of the 65 nm part compared to the 90nm version, whereas especially older games showed the greatest performance hit. We have seen 3rd party benchmark results using newer gaming applications like "SplinterCell Chaos Theory" that actually show the Brisbane core as much as 8% ahead of Windsor. All of that translates into the notion that with a smarter cache controoller, the extra arbitration latency of 2 cycles (on average) can be hidden behind the mentioned pipelining of multithreaded applications or of multiple applications in a multithreaded operating system and the performance difference in real life will be more or less in the noise.

Cache and memory benchmark (SiSoft Sandra): Higher is better. 90 nm (brown) vs, 65 nm (red)

The Final Verdict

AMD's 65 nm CPUs are essentially what AMD promised, that is, a process shift without any significant differences in performance. There is that issue with the half-multiplier, though. It is really not clear why this was done and what was the rationale behind this move. We have enough benchmarks that show the "underclocked" processor delivering higher performance, especially in gaming applications but then, who are we to judge.

next page: => More =>