i7 memory comparator?

[bilbat]

LSD,
after a lengthy discussion re i7 memory here:
EX58-UD5 and (possible) RAM problems
I gave some more thought to the matter, and realized the whole 'memory game' has changed; with the MCH, I always recommended fast memory not so much for throughput speed, but as a 'tool' to get your FSB above 1600, where DDR2-800 'runs out of steam' at the minimum 2.0 (1:1) multiplier; now, with the advent of the 15/17, and the 'IOH', it seems to boil down to one thing - latency; and, with all the speeds proliferating out there, comparing latencies, apples to apples, is high unto impossible. I posted this:
X58 memory comparisons
hoping for comment, and I thought to show it to you, as you're 'playing with' the new goodies daily! One problem:
The posted picture 'compressed', making the print iffy to try to read, and 'click to zoom' doesn't work on any of my browsers - I don't know if it works for anyone - if you can't read it, the original excel file (so far) is here:
Socket1366MemoryList.xlsx
I was hoping to see if anyone else could think of something I'm missing - any other criteria that would make it easier to compare i7 (and, if this works out, I will do one for socket 1156 stuff as well) RAM, with an eye toward value. I wanted to get other people's ideas now, as I have a list of about eighty-five parts, and I don't want to have to look 'em all up twice, or three times, as features get added...
I used NewEgg, as everyone (at least, in NA) has access to 'em, and I 'threw out' and part whose manufacturer didn't bother to provide any other timings than CAS (which nixed Kingston, PNY, and some Crucial and Patriot) as what kind of support can you hope for, if they don't even give the standard four-item latency list? It still leaves me with more than six dozen pieces!
Bill

[Psycho101]

Re: i7 memory comparator?

For a full latency timing list in the usual format, isn't there a link to the manufacturers site on the newegg details page? I'm from thew UK but have looked at newegg when advising here in the past.

From a purely personal view, latency balanced with speed can be more complicated than one may think. After all, true latency is measured currnetly in ns. It's therefore entirely possible that DDR-2000 at 9-9-9-27 can have much lower real term latency than DDR-1333 at 7-7-7-21.

When advising anyone on the type of RAM to buy for i5/i7 I always state that more than ~1333 isn't going to give much better results. After all, even the most intense, 8 core aware apps care more about RAM physical size and CPU speed than RAM performance. At the moment, DDR-1333 even at 9-9-9-27 is, for me, the sweet spot in terms of performance and versatility with setting RAM multi's on i5/i7 etc. Any more and any tighter timings just makes for pretty numbers when benchmarking IMO.

[bilbat]

Re: i7 memory comparator?

From a purely personal view, latency balanced with speed can be more complicated than one may think. After all, true latency is measured currnetly in ns.

That's the entire point of the discussion here:
EX58-UD5 and (possible) RAM problems
that latency, for a given DIMM, is a actually a matter of nanoseconds turnaround time, not cycle counts...

For a full latency timing list in the usual format, isn't there a link to the manufacturers site on the newegg details page?

Unfortunately, the simple answer is no! Some of the 'pointers' on NE for one of the RAM mfg's point to pages in Japanese! Few give much at all in real specs, and the ones that are given, are given at 'rated' speed, which, for i7 products, vary from the 'official 1033, all the way up to 2133...

It's therefore entirely possible that DDR-2000 at 9-9-9-27 can have much lower real term latency than DDR-1333 at 7-7-7-21.

Therein lies the rub: the actual calculation is: ((1/2000)[cycle time @2000] x 9 [latency at 2000]) / (1/1333)[cycle time @1333]) rounded up to the nearest integer; so your 2000 9-9-9-21 is likely to run 1333 at 6-6-6-18; but how is the 'generally innumerate' public to know that? - thus, the comparison tool... My 'weighting for value' kind of works like this: if two DIMMs have the same latency (at 1066) but differing prices, the lower priced one is 'weighted' to give a higher 'value' number; if two DIMMs have the same same price, the one with the lower latency (again, 'normalized' to 1066) gets the higher 'value'.

[Psycho101]

Re: i7 memory comparator?

I agree with you and acknowledge the usefulness of your comparison guide for those that do not know that slower Cas, RAS to CAS, tRP and tRAS at higher speeds can even actual latencies out. I just hope that said paople can take the time to think about things and actually read the guide.

Normalising to 1066 is also a good idea.

I have spoken to many a person who thought that "warp speed" RAM was going to "rock their computing world" and then couldn't understand why my S939 2.7GHz (DDR-500) X2 3800+ was out performing their stock clocked version with DDR2 at 667/800. For me, value and bang for my buck (pound) is much more significant. I like pretty flashing lights as much as the next person, but price/performance is crucial to me.

Maybe Lsdmeasap would be able to include the comparison in a sticky such as the "Memory timings explained..." one. 99.9% of people I meet here are pretty alert people and many do read the stickys before posting etc.

[bilbat]

Re: i7 memory comparator?

For me, value and bang for my buck (pound) is much more significant

We agree exactly! I mostly advise over at Tom's Gigabyte forum; I can't think of an inoffensive way to put it, but the majority of our readers are - shall we say 'less astute', than here; they're kind of across a spectrum from NewEggs' EggXpert forum, to a shadow of this. I hate to see these guys blowing fortunes, getting taken by lofty looking specs, that they are ill equipped to analyze, and then being disatisfied whith a system that was a two-month dream for them - until they tried to integrate it!