Okay MODS, before I start rambling on again and continue beating this dead horse, what started out as a simple interest in seeing TTL take an unskewed unbiased no BS real world look at RAM freq. vs timings, and not canned mem tests, has now become more of a thread discussing said topic and should probably be moved from TTLC to the Memory forum.
Now to continue: Yes, it is true that cpu intensive apps, especially those operating on very large blocks of data, will be affected more by latency since the cache is being bled dry much quicker and more often than a typical app. Let's face it, the L3 cache of an i7 990x, for example, is still only 12MB, and even running at stock, it still has to feed 6 cores at 3.46GHz. The processor still has to do operations on the data, though, which takes cpu clocks, so it will not be constantly sucking the cache dry every nanosecond. If we look at the number of times the cpu will likely hit that RAM access wall, even with preemptive caching, it could still be many tens of thousands of times per second. The typical tight vs loose timing advantage is 2 clocks and that will only be seen at the beginning of each burst. For the sake of argument, lets say we have 2000MHz RAM and 100,000 times per second execution is delayed due to data block transfers. We take the additional 2 clock hit with cheaper RAM, that's 200,000 clocks at RAM speed which totals 100 microseconds delay every second. That means, even in this intensive scenario, you would only loose 1 second per 1000 seconds (16 minutes and 40 seconds). That's 86.4 seconds per day, and even if it were 4 times that, it would still only be 5 minutes and 45.6 seconds per day. Not the massive benefit one might be led to believe, but in the long run, over weeks and months of 24/7 computing, it could become more significant. Granted, this is a rather simplistic look at the issue and there is a little more to it, but I think, to state the obvious, we can all agree that higher frequency at the same timing would show far more benefit than tighter timing at the same frequency. Of course, getting the lowest latency RAM at the highest frequency your processor can handle is the best of both worlds. No surprise there.
All that being said, even I still opted for CL7 RAM instead of 8 or 9. Why, you ask? Just because I can and that's what I wanted. LOL And so there you have it, even if it meant only a very tiny performance improvement, that's what I chose.
Well, it's been fun kicking this back and forth; that's what forums are for, right? So don't anyone go getting their panties in a bunch over it; it's all good.
So yeah, Tom, if you're even looking at this thing (probably not), just thought it would be interesting to see your take on it, that's all. If ya-all want to move the thread, lock it, delete it, whatever... no biggie. Enjoy.
