Guide: AM3 CPU's Which Ram Speed is Faster and Which CPU-NB Clock is Best - Page 11

You can give it moar joose! I set mine to 1.37 straightaway..raise it if need be..depends on how the chip reacts.

I might do that with the Sabertooth I have coming. I don't trust this board much for that. At 1.26 and the heatsinks at extremely warm, I'm now blowing a GT 2150 right on them.

I have a question

The 'encoding' test... does that mean like streaming video kind of stuff?

So if you have a stream... you should go with a higher memory clock instead of tighter timings, am I getting that correctly?

And then raise cpu-nb as high as possible, and ht link stay at 2ghz/stock (so if you raise fsb, you should LOWER your ht link multi...).

Hi Belial,

Encoding performance is primarily CPU bound. Increasing the NB frequency by a full 50% (2000 to 3000MHZ) only improves the result in the charts shown by about 4%. What this means is that the task is not being bottle-necked by memory bandwidth. The 4% gain is the result of achieving a slightly better average latency from the CPU through the IMC to the memory and back. (quite literally, the "timing" of communication events has more chances to "hit" on a cycle to the IMC since it's spinning faster, so nanoseconds are shaved here and there).

In each case, the performance difference demonstrated between 1333 and 1600 speeds is less than 1% (with some differences as low as 0.2%). The small difference again, is related to event timing and not bandwidth. The Athlon II/Phenom II architecture is bandwidth bottle-necked at the IMC, not the memory itself. For all practical intents and purposes, choosing between 1333 and 1600 speed on an Athlon II memory controller is primarily an exercise in academics. Such a change will not make or break the ability to run a particular game or encode at a particular quality setting in real time. Keep in mind that the comparisons shown are at specific timings. If you deviate from the settings shown, your results may be different. It's possible that with your particular memory, you may get the best encode performance with the lower memory speed if that speed allows you to run tight enough timings. Again, I must emphasize, it's academic, it's "interesting," it isn't "useful."

Remember, this CPU architecture started on DDR2. As it turns out, no matter what you do with the DDR3 speeds/timings, you won't see a significant improvement in memory bandwidth over what these chips were doing on DDR2. FYI, DDR2 667MT/s in dual channel has over 10GB/s of potential bandwidth. The typical AM3 system running DDR3 has twice the theoretical memory bandwidth, but only musters up about a 10% improvement over that DDR2 performance. Best I've achieved is around 11,500MB/s.

Best of luck,
Eric

Well from what I understood, AMD favors timings and Intel favors speed. Hence on my current system, I went for tighter timings than speed (cut timings by 2, but speed is barely higher than 1333mhz, as you can see in my profile).

But I thought the encode stats were showing that speed>timings for encoding, whatever that was.

im not really sure what to take from your response... that speed isn't that useful? But the graphics show that speed>timings in the encode tests:

I just dont know if this encoding is like the type of encoding you get with streaming xsplit, etc.



Shows that speed>timings for encoding.

Now I stream starcraft 2, that's what i want performance in. And streaming takes up a ton of my pc. so thats my interest here. obviously im aware im not getting big changes anywhere here, but im curious if i should be going for mhz instead of timing now.

Also, here's a source showing that CPU-NB is limited, that higher isn't always better, but that there is actually a sweet spot:


http://www.overclockers.com/forums/showthread.php?t=596023

By the way, just got a phenom ii.