Date: November 10, 2008 - Author: Rob Williams
Can't decide on the right memory kit to pick up with your brand-new Core i7 PC upon release? This article was designed for you. We aren't comparing brands here, but are rather comparing densities and frequencies against each other to see if there's any point at all in purchasing a higher-end kit. You might just be surprised at our results.
With Intel's official Core i7 launch coming up within the next few weeks, the time to choose which parts you'll need for your new build is now. Whether or not you "need" an upgrade is going to be something only you can answer, but our in-depth preview from last week should be a good starting point. By the end, you should feel a lot more confident about your decision.
In that article, we mentioned that there would be a few considerations you would have to bear in mind if considering an upgrade, or a brand-new build. One of the more important is memory. In the preview, I mentioned that it's best to upgrade to a memory kit that's designed for X58/Core i7, because previous DDR3 kits may have an issue with the new motherboards.
It didn't take too long before I experienced this first-hand, but I did have a workaround. That was to boot up the machine with a kit I knew would work, manually change the voltages and timings, then swap back. After that, the PC booted fine. It goes without saying that this is not an ideal solution for most people, especially if you only have one DDR3 kit on hand. Nor is it much fun having to tweak every-single timing setting in order to have absolute stability.
If money isn't a huge issue (and assuming you're upgrading to i7, it probably isn't), you may be better off opting for a special kit designed exclusively for the platform. That way, you can avoid any potential headaches, and know that you'll be installing a kit designed with this specific platform in mind.
In this brief article, we hope to answer that exact question. The answer will of course depend on personal needs, but I think for most people, what will be "required" is going to be about the same. If you run a workstation PC or a server, then you obviously want more memory than the average consumer, while the average consumer would likely prefer overall speed than density.
Because Core i7 brings a triple-channel memory controller with it, purchasing a kit of RAM for the platform is going to be a very different experience compared to previous builds. The reason is simple... kits will include three sticks of memory, not two. This might seem like an odd-ball way of changing things, but as evidenced in last weeks article, triple-channel can mean huge bandwidth.
Because recent PCs utilize a dual-channel memory controller, common practice was to purchase either a 2GB or 4GB kit, as odd numbers aren't possible with an even number of sticks (at least if you are dealing with even densities and are not mixing and matching), but that changes with i7. Instead of 2GB, 4GB or even 8GB kits, we'll be dealing with 3GB, 6GB and 12GB.
The question comes down to "how much do I need?", and for the most part, this is actually quite easy to answer. The reason is because for most enthusiasts, the choice of 3GB is going to be eliminated right away. At this point in time, 3GB isn't a substantial amount of RAM, and most people have been using 4GB in their machines for the past year at least... so to actually downgrade would be an odd step to take. The major increase in bandwidth doesn't exactly counteract the lack of density, sadly.
With today's games and high-end machines, 3GB can become very limiting, especially if you want to run Windows Vista alongside games at a resolution of above 1680x1050. Today, 4GB is almost a minimum, and even 3GB isn't going to be good enough. On i7 though, it's either 3GB or 6GB if you want your memory to be optimized... there's no in-between.
The other option is to take the outrageous route and pick up 12GB, but that's going to be overkill for the vast majority of people. If you need that much RAM, then you'd likely know it without reading this article. Even intensive render jobs won't usually take advantage of more than 6GB, so I think that will be the most common configuration for most people.
Memory frequencies also come into play, and this article will aim to figure out whether or not faster RAM is actually needed. After all, if Core i7 enables great latencies and insane bandwidth on lower-end kits, is there any reason to go with a "high-end" model? In previous generations, the benefits of having faster RAM was to both increase bandwidth and decrease latencies, and really, things don't change much with i7.
Even before jumping into our results, it's safe to say that the lower latencies will make a far more important difference than increased bandwidth. Under no real circumstance can I see application performance differences on a PC with 20,000MB/s bandwidth compared to another with 30,000MB/s. Lower latencies equal faster transactions, and that's what's going to be important in real-world situations. Things could change in the server market, but this article is designed to take care of our number one type of visitor, the enthusiast.
On the next page, we'll give a brief overview of our configurations and testing methodology, then jump right into benchmarking to see if that high-end kit you've had your eye on is really worth the money.
At Techgage, we strive to make sure our results are as accurate as possible. Our testing is rigorous and time-consuming, but we feel the effort is worth it. In an attempt to leave no question unanswered, this page contains not only our testbed specifications, but also a fully-detailed look at how we conduct our testing.
If there is a bit of information that we've omitted, or you wish to offer thoughts or suggest changes, please feel free to shoot us an e-mail or post in our forums.
The table below lists the hardware for our two current machines, which remains unchanged throughout all testing, with the exception of the processor. Each CPU used for the sake of comparison is also listed here, along with the BIOS version of the motherboard used. In addition, each one of the URLs in this table can be clicked to view the respective review of that product, or if a review doesn't exist, you will be led to the product on the manufacturer's website.
Component |
Core i7 Test System |
| Processor | Intel Core i7 Extreme 965 - Quad-Core, 3.20GHz, 1.30v |
| Motherboard | ASUS P6T Deluxe - X58-based, 2624 BIOS (10/23/08) |
| Memory | DDR3: Qimonda 3x1GB - DDR3-1066 7-7-7-20-1T, 1.56v DDR3: OCZ 3x2GB - Gold DDR3-1600 8-8-8-24-1T, 1.65v |
| Graphics | Palit Radeon HD 4870 512MB (Catalyst 8.9) |
| Audio | On-Board Audio |
| Storage | |
| Power Supply | |
| Chassis | |
| Display | |
| Cooling | |
| Et cetera |
When preparing our testbeds for any type of performance testing, we follow these guidelines:
Because the ultimate goal of this article is to see how much of a real-world difference faster memory avails, we are running a total of five different configurations, four being with our OCZ 6GB kit. We vary the speeds from the very low-end speed of DDR3-800 to the high-end speed of DDR3-1600. We should note that DDR3-800 speeds are not going to be common amongst triple-channel kits, but we've included results to showcase the differences nonetheless.
Most of the same benchmarks we used in our Core i7 preview article return here, with the exception of our two games. To take the place of the games, we use 3DMark Vantage, which is more than intense enough to show us where a bottleneck may lie. For detailed explanations on the tests themselves, please refer to our Core i7 preview.
With that all said, let's jump right into a few of our real-world tests.
One crowd that can appreciate a faster processor is without question, the 3D designers. But given their rendering projects can be so large, can faster memory shave minutes or even hours off of the task? Although our render jobs are nowhere near as complex as what you'd find in a professional studio, any benefits seen here should carry over even more so into larger projects. Sadly, from what we can see below, the gains are very, very minor.
Our fast DDR3-1600 kit managed to take the top spot in all three of our rendering benchmarks, but the overall improvements are minimal at best. All of the increases seen were between 1% - 1.5%, which isn't going to add up to much in the real-world, unless you are dealing with jobs that can take a few days. Even then, the differences are going to be an minutes, not hours.
We know now that 3D designers aren't going to see major improvements with faster memory, but how about the video gurus? Our three tests below had the job of finding out, and again, the differences we found were absolutely minute, although they were a bit better than what we saw above.
The test that showed off the biggest increase was ProShow Gold, which actually scales with RAM speed... something I wasn't expecting. With small projects, like our DVD encode, the increases are incredibly small, but with larger projects, like our HD Video Encode, the real differences begin to be seen. The real gain in a scenario like this would be small, but things would change if you are dealing with much more elaborate jobs. I think it's safe to say that overall, there are no notable performance gains so far.
On the next page, we'll continue our run of benchmarks to see if we can see real gains anywhere.
It's time to take a look at things from a synthetic standpoint, where differences might be seen, even if they aren't reflected in most real-world situations. Thanks to Sandra's offering of every system-related test we could ever need, we spent some time in there, especially in the memory tests, which the results are shown for below.
One thing is clear... faster memory is going to a) have more bandwidth and b) have lower latency. That goes without saying, and the graphs below prove this. In fact, the RAM scales well enough that the bandwidth almost doubles as the frequency doubles. What kills an absolute 100% increase is likely the timings, which have to be higher on faster kit, understandably so.
Our DDR3-1600 shows incredible performance all around. Not only is the bandwidth close to 30,000, but the latencies are far better than what we see with our DDR3-1066 kit. That of course hasn't meant much so far though, as the performance gains have been minor, to say the least. So let's take a look at the remainder of our benchmarks and see what happens.
Whew, alright. Strangely enough, where math is concerned, tighter timings or faster frequencies seem to make no difference whatsoever. The reason I consider this strange is that when Super Pi was the "main" benchmark of choice, tighter timings did improve the overall calculation speed. It seems with today's processors, the extra cores make up for the slower memory of yesterday in some cases.
While RAM speed hasn't made much of a difference in any of our tests so far, things are a little different with Adobe Lightroom. Here, the overall time-to-export decreased as our RAM speed increased, which gives undeniable proof that this is one application that will perform better with faster memory. The differences again are small, but they're some of the best we've seen so far.
Finally, we ran 3DMark Vantage in lieu of real games as it's synthetic nature is going to be better capable in stressing our system to its boiling point, which will help us see if memory is likely to affect gaming performance.
Though 3DMark does a good job of stressing the system, the outputted scores flip-flop so often, that it's impossible to see proper scaling when comparing CPUs or memory speeds. It's more useful in GPU reviews where there actually will be differences between the 3DMark and GPU scores. But what we cared about here was to see if our faster memory increased our gaming performance, and we found that out to not be the case at all. Our DDR3-800 delivered almost identical scores as the DDR3-1600.
So, what's the verdict? Let's talk about it next.
The idea for this article came to me quite a while ago, far before we even knew what Nehalem's real name was going to become. The reason I never got around to taking care of it is simple. In order to realistically see the benefits of faster RAM, the CPU speed has to be kept the same. Until Core i7, that wasn't really that possible, as the only way to be able to change the RAM speed while keeping the CPU the same would be to use a FSB speed of 600MHz. Given how rare that is to hit stable, using it wasn't really a possibility.
Core i7 changed everything, because no matter the CPU speed, you're able to choose a variety of RAM frequencies without touching anything else. While on Core 2, the FSB had to be increased, the Base Clock on Core i7 allows RAM speeds between DDR3-800 and DDR3-2133... giving us a great amount of breathing room to conduct our testing.
Taking a look at the results here, it seems pretty obvious that faster RAM, for the most part, isn't going to make a noticeable difference to anyone. As much as memory companies would love to sell you their fastest part, it's just not going to make a noticeable difference in most of what you do. Even our 3DMark Vantage scores were close to being identical, and that's much more strenuous than any current game on the market.
I'm not going to jump to conclusions quite yet and say that RAM won't make a noticeable difference somewhere, but I haven't seen any evidence of it yet. In this article alone, we tackled 3D rendering, image manipulation, video encoding, mathematics and even gaming, but the only place we saw a real difference was with the synthetic scores of the memory-specific tests.
The question of which scenario would benefit from faster memory at a static CPU clock speed is one that's been hovering around my mind for quite a while. I've asked numerous memory vendors over the past few years that exact question and have never received a straight answer, sadly. You can reach your own conclusions with that one.
That's not to say that bigger differences wouldn't be seen with previous architectures, because I'm confident that we would see more notable differences there. That's mainly because the bandwidth to begin with doesn't get near as high as on i7. But even then, most of the bandwidth there isn't really touched to begin with, and the latencies between Core and Core i7 are roughly the same.
Bottom-line? If you are building a new machine for light work and don't have huge gaming in mind, then a 3GB kit should suit you. If you are a hardcore gamer or heavy multi-tasker, a 6GB kit is going to be the right fit. As for the overall frequency, I think our article speaks about that enough. Personally, given our results here, I'd recommend a DDR3-1066 7-7-7-20. Or if it's possible, find the same frequency with latencies of 6-6-6. I believe that a kit with those specs could even outperform the DDR3-1600 with 8-8-8, except with raw bandwidth.
I'll also open the discussion and debate on whether faster memory has real benefits to most consumers in our discussion thread. If you yourself know of certain scenarios where faster RAM can make a greater than 2% improvement, we'd love to hear about it, and even consider implementing some of those tests into our future memory-related content. Likewise, if any memory vendors out there have comments, feel free to join in and let us know what we're missing. If there's any demand for a follow-up article that takes a more thorough look at things, please feel free to post about that also.
Have a comment you wish to make on this article? Recommendations? Criticism? Feel free to head over to our related thread and put your words to our virtual paper! There is no requirement to register in order to respond to these threads, but it sure doesn't hurt!
Copyright © 2005-2009 Techgage Networks Inc. - All Rights Reserved.