Slashdot Mirror
Dual Channel Memory Shootout
MDT48 writes "Ever wondered if that expensive low latency memory was worth the cash? These guys have rounded up almost every memory module out there and hammered them. Must have taken them ages, and takes almost as long to read, well worth the effort though."
The Winners
GEIL Not only do they have GEIL placing well in the middle of the lineup both in price and performance, but I have noted that it offers the best of both worlds. It doesn't cost nearly as much as the high high end stuff, and while it doesn't perform *quite* as well, it's still better than the low end stuff. GEIL presents a good opporutinity to chose between Value and High End
Well since it is a 33 page read, how about the link to the end where the conclusions are.
Conclusion
The Winner
and for those too lazy to click
Conclusion
So there we have it, 18 different types of memory benchmarked to within an inch of their life and to prove what? Well, one thing we've proved is that while even value memory may offer sustained levels of high speed operation, when you want to make it to the scary end of the spectrum above 250MHz you generally need to pay the price premium associated with "enthusiast" modules. Cheaper stuff may get you close, but at the very high-end we're afraid you really do get what you pay for.
If you're happy to settle for fast rather than fastest, it seems that reputations and price tags count for little in this game. Good chips on a poorly designed PCB and poor chips on a great PCB will both leave you wanting. And even if you have the best and fastest memory on the planet you still need a motherboard that can do it justice.
Evaluating performance at more reasonable frequencies is slightly easier, but also slightly fuzzier. Some of the benchmark results varied so widely that it was hard to understand what was really going on, and with various tests favouring different attributes and the surprisingly similar stock performance from many of the modules on test, it was almost an exercise in identifying the "Top Dogs", the "Turkeys" and "The Rest".
We hope you enjoyed this roundup and gleaned at least some information from the effort that went into it. As we hope you've seen, or will see when you glance through the benchmark results, memory is a very complex subject and pinning it down in performance terms isn't as simple as you might imagine. We won't let that stop us trying though.
Finally we'll pick out some winners. Remember that for our purposes, overclocking is more a pleasant bonus than an essential feature, and though we have made allowance for overclocking performance it hasn't been given the same weight that it might have been given were we writing this roundup purely for enthusiasts.
Finally, benchmarks do a great job of highlighting strengths and weaknesses in products but it's important to keep things in perspective. Most of us would not notice any tangible performance differences between even the fastest and slowest modules on test here when run at the processor's default speed unless we were specifically looking for them. That's not to say that faster isn't better, it's just not as critical as some would have us believe.
The Winners
Value Editor's Choice - TwinMOS PC4000 CL2.5
Concerns over availability make us a little nervous about handing the value crown to TwinMOS, but we can't argue with the numbers and this is quite simply the fastest memory for the money that we had the pleasure of testing. We only hope the poor availability isn't due to production constraints and look forward to seeing more retailers pick up on what appears to be a very nice product.
Value Recommended - Buffalo PC3200 CL3
We may have had little experience with Buffalo before this roundup but both of the Buffalo modules we were sent acquitted themselves well in testing. For the most part, performance levels in the value group contenders were quite evenly matched but Buffalo did enough to scrape onto our leader board. A very impressive all-round showing earns Buffalo a well-deserved Recommended award.
High-End Editor's Choice - Corsair TwinX1024-3200XL Pro
Perhaps the toughest choice of all, we eventually opted for Corsair's 3200XL Pro for its shear arrogance. While far from cheap, it consistently set the standards for stock performance while flashing a knowing LED at the competition as it did so. For a change we see a product with a bite every bit as bad as its bark.
Hig
30% Troll, 50% Underrated, 10% Interesting
Score:5, Troll
This is for the crowd that puts neon lights in their PC. Not the people who buy ECC memory for their desktops.
Looking at the benchmarks, in most cases I would say..... no
(just in case someone has the urge to actually RTFA...)
And if you would just like a short introduction to what the heck RAM speed means, check out this excellent Arstechnica.com article:
FSB and memory speeds
---------------------
Freedom or Evil: Freevil.net
G. W. Bush says, "You decide!"
The better the cas timing the better the ram... Also more sticks in the banks cases more latency... if you can do one stick 512 vs 2 256 it will give less sytem latency... I wonder what dual channel ram does to system latency... if there is more because 2 sticks have to be addressed...hmmm?
||| I still can't believe Parkay's not butter.
Always nice to have a personal verification on the products you purchase.
This about TwinMos from overclockers NZ:
"Up on closer inspection, we found out that the TwinMOS's PCB is made by Tripod Technology. I've been to the Tripod plant in Taiwan and was pleasantly surprised by their workers self-discipline and how organised the plant was."
In post Patriot Act America, the library books scan you.
"I know faster memory would help, but does it help that much when you follow the advice above?"
It matters MORE not less. When you have swapping or vm going on your bottlenecked by the harddrive. When everything is in ram your bottlenecked by the Ram.
That's why if you increase the memory in a linux or bsd system you tend to see a much more dramatic performance increase than a processor upgrade usually yields (unless it's a several generation jump, like from a p1 200 to a p4).
The processor is much faster than the memory, since you've eliminated the hdd as a bottleneck (and everything else, the memory has a direct channel to the cpu), every increase in memory speed = actual performance increase.
Why do people make wrong claims in such a matter-of-fact manner? Is it ignorance or malice?
Whatever, more modules in general leads to slower access time. More sticks (modules) of RAM adds to the capacitance driven by the memory controller, and therefore increases the propagation delay on those traces. The more connected modules the longer total trace length driven by each IO buffer of the memeory controller. The difference is probably negligible for all but the edgiest overclocker, but there's always a speed advantage for fewer modules.
I did a quick google and found this example to back me up, but it's not really needed as you clearly have no fucking idea what you're talking about and could not cite one single reputable argument to back up your insane claim. One of your friends spewed that nonsense and you're parroting it because it sounded reasonable. Of course, that's only because you have no fucking idea what you're talking about.
everything in moderation
Could it be that parent misunderstood grandparent's extremely brief comment? Could grandparent be referring to the fact that with multiple, independently addressed modules, the memory controller can overlap accesses, leading to vastly improved memory bandwidth (and, given prefetch, latency) in the common case of accessing sequential memory locations? Could it be that grandparent is a system architect, and parent is a hardware engineer?
Nahh. Clearly grandparent is insane, and spewing friends' nonsense. And don't even ask me where the sexual intercourse comes in...
It may not be slower, and may well be faster, if you have a multi-channel memory controller. YMMV. BMSMA.
No, multiple modules will never be faster -- a mulit-channel controller will be the same with one or two modules; slower with more than two (which is possible.)
More modules is never faster from a memory access timing standpoint which is, again, what this discussion is (or should be) all about. You can, of course, increase overall system memory bandwidth by adding one each memory controller and SDRAM module, but that's irrelevant here (even in that case, you'd want faster RAM so that the CAS, RAS, and other latencies can be minimized and bandwidth increased a bit more.)
everything in moderation
Maybe, but that'd be missing the point entirely, since this article is about RAM access timing -- comparing different RAM brands and types in the same system.
Everyone who is paying attention (or thinks for a second) knows that adding memory controllers to the CPU can increase memory bandwidth (it's 2 pipes instead of 1, after all), and this increase is more than the gain found in overclocking your FSB (front-side bus, or memory controller) from 200MHz to 210MHz or going from 3-3-7-11 access timing to 2-2-2-2.
This article is all about how fast you can run your FSB and how tight you can set the timings for various brands of RAM. Not how much more memory bandwidth your system can get by adding memory controllers/channels. That's a CPU upgrade, not a RAM swap. Wrong topic.
everything in moderation
It's one of the older memes that are floating around out there (shows the slashdotter belongs to a different era) :-). This might help!
My Favourite Meme
> Next time, read the question before blasting away in a mad frenzy.
/. breaks thread integrity when showing anything with a threshold. Everything under the threshold is lumped under "X more replies".
I see where your reply comes from, but you're barking at the wrong tree. Here's why:
randyest is NOT replying to the OP question, he's doing so to a -1 modded answer to it that says more modules are faster because some bullshit he pulled out of his ass.
Of course, there are several things that have failed in this communication channel:
1. You didn't check who randy was replying to (by clicking the "parent" link at the bottom of his post).
2. The -1 modded post didn't change the subject line (left it with Re: whatever).
3. Neither did randy.
4.
5. randy should have quoted who he was answering to. Quoting is good.
Anyway, chill a bit guys.
---- Take the Space Quiz!