compared to a higher frequency. I have done various benchmarks in everest, aida64 and maxxmem. all of these benchmark programs have favoured situations where i have had the frequency higher with the sacrifice of looser timings, be it only by a small amount. are these benchmarks not a good representation of real life situations? do low timings make more of a different than frequency?
To start with, you won't notice any difference at all in real life situations by tightening the timings on your memory. Considering that a small improvement in memory bandwidth accounts for milliseconds in Super Pi etc., there's no way this is a perceptible difference. Regarding your first question, it all depends on the chipset and the architecture. Some DDR2 boards (the EVGA 680i in particular) favoured tight timings over raw MHz, so 1000MHz 4-4-4-8 would produce higher bandwidth than 1066MHz 5-5-5-15. But the same is not true of DDR3 because the architecture has changed and latency doesn't make as much of a difference as it used to. Yes, lowering timings improves bandwidth... but increasing frequency makes more of an impact on performance. If I have my memory set to 1600MHz 6-7-6-18 which sounds very nice and snippy, it gets absolutely stomped on by 2000MHz 8-9-8-24.
And here you should have said again - it gets stomped in benchmarks. Unless your application is perfectly multithreaded and requires extreme memory bandwidth and uses nothing else but memory, then you will notice the difference. Otherwise there is no point even going above 1333CL7/1600CL9.
It eventually becomes subjective - some people will be more than happy with 1333MHz CL9 whereas others might find it sluggish. I had 1333MHz CL7 for ages, and now I use 2000MHz CL8 and don't notice any difference at all...unless I bench. The bandwidth is vastly improved, but not to the extent that I notice it. What's the point of higher frequency kits? Choice. Simple as that. If you think that buying a 2000MHz kit will make your system grow wings, you're buying into marketing propaganda.
Sometimes when you overclock a little too much, performance can be less as it needs to double check errors making the calculation happen twice. If you are getting lower performance it's best to figure out what is causing it and set it so it isn't causing errors in the first place. Memory bandwidth is like a waterpipe, If most loads will go through it no problem then widening the pipe won't make a difference. It's only when the pipe isn't wide enough to get all the water through efficiently that you start to notice a difference. Crysis can be quite demanding of memory.
Tighter timings seem to shine better on AMD-based systems. For the most part, lower timings are hardly noticeable, or you could even say it's simply marketing that they try to convince buyers that they improve system performance. Synthetics-wise and benchmarks, there will be a difference, but it's not as major as upping the frequency speeds of the RAM. You would be better off getting 1866mhz ram at 9-9-9-24 for example instead of getting 1600mhz RAM at 7-8-7-24.
If you're comparing tight timings to 2200MHz, you can forget it lol - 2200MHz very, very fast for system RAM.
Games in general will appreciate tighter timings over bandwidth. 1600 cas 7, 1866 cas 8 are generally the best all-round configs. .....but on a different note if your building new, MEMORY IS SILLY CHEAP THESE DAYS.
beg to differ. Changing timings has very little if any impact on general applications and games. Bit-tech did a memory test just recently showing anything above 1600mhz provided hardly any benefit to FPS in games and programs and timings do nothing other than Cheesecake benchmarks. http://www.bit-tech.net/hardware/memory/2011/01/11/the-best-memory-for-sandy-bridge/1
I think it may be more complicated than that, there are like 20+ different DDR3 timings which are all going to be interlinked somehow, compared to 3 or 4 with DDR2 and below. I think we need a more in depth investigation into timings that take the newer developments and architectures into account. It's also not quite as simple as all games will benefit, even in the days of DDR it was proven that some benefit from having tighter timings, others benefit from raw MHz.
Absolutely. I'm not even really sure what the primary ones do, let alone the many others. To the OP: Are you sure your RAM's stable at these low latencies? I could run my old 1600MHz RAM at 7-7-7-24, which would run things but was unstable and so slowed everything down presumably for the need to recheck stuff (as mentioned by Deders). Not that it matters as if you have it at 2200MHz it'll be laughing. As for benchmarking, if the difference is less than say, 2%, you're probably detecting anomalous differences between runs of a benchmark than the difference of memory settings.
Yes, it is significantly more complicated than people think. Latency and frequency numbers are always used for marketing (and always have been), whereas bandwidth (sustained bandwidth, not peak bandwidth) is rarely quoted... because, by and large, sustained bandwidth is very similar between memory kits even if they have different frequencies and latencies. Let's not forget that memory performance is determined by the system that the memory goes into; memory bus (or unclock, whatever you want to call it) and CPU play a big part in how fast your memory operates.
well i did 3 runs of my current setting and then 3 runs of 2133mhz with slightly tighter timings and the higher frequency came out about 80-100 mb/s faster in aida64 on write and write. yeah im pretty sure they are stable
Very true, this was proven when 1156 came out with a more optimised memory controller (located in the Uncore part of the CPU) that partially made up for lack of triple channel by being better optimised than 1366. A test on an AMD controller would give different results from an Intel controller.
That was on Sandy Bridge. That is a minority of people at this point in time. I do not think DDR3 has any more or less timings than say DDR2, it is simply that all/nearly all sub-timings are exposed to the user in the BIOS. 'Back in the day' you didn't see them, thats not to say they weren't there. All true. However your on a triple-channel system and that is a different argument altogether. Namely for you guys DDR3 1333 9-9-9-24 in each channel is a better/less naff setup than say on a dual-channel system. Again, 8 gigs of fast ram is soo cheap these days it seems worth it to pug up and extra 30 quid over some generic crap.
granted it was on SB but timings havent mattered since Core2 and the release of phenom. timings made an impact on old AMD skt 754 and 939 systems running DDR but since then they make very little impact on REAL performance
