See www.gskill.com. This will be the fifth set of G.Skill memory I installed in a computer without a problem. G.Skill is slightly more expensive that the unbranded generic memory I normally use and offers a better range of speeds.
At the bottom end of the memory market is a mixture of rubbish and reliable older technology. An example of reliable cheap memory is generic memory I used a few years ago that was fitted with Hitachi chips. You do not get the latest fastest chips on that type of memory.
When you want the latest memory speeds, you have to go to to brands. In Sydney, G.Skill have the fastest memory at standard CAS settings. CAS is a measurement of internal delays. There is about a one percent difference between standard and fast CAS settings. Fast CAS requires the latest premium chips and costs about double the price of standard CAS memory. If you use standard CAS speed memory, you save enough money to buy a faster processor or SSD instead of magnetic disk.
The G.Skill memory is available in both CAS 11 and CAS 9 with a price difference of only three dollars per 8 GB, compared to a hundred dollars difference in some other brands. The difference will be only a few seconds per day or a few minutes per month. The price difference is not cost effective when it is hundreds of dollars. The tiny g.Skill difference, only six dollars for my 16 GB, will pay for itself in the first month.
For reference, the standard CAS memory for DDR3-2133 memory, CAS 11, is defined as CL 11-11-11-30-2n for the G.Skill model F3-17000CL11D-8GBXL. The faster CAS, CAS 9, is CL9-11-10-28-2N from model F3-17000CL9D-8GBXM. The first number in the full CL list is the most important with the subsequent numbers providing less and less effect. DDr
DDR3-1333 memory has CAS timings typically of 9, 8, or 7. For DDR3-1600, the CAS options can be 11, 10, 9, or 8. DDR3-2133 can have CAS timings of 14, 13, 12, or 11 but manufacturers are already delivering faster options. CAS numbers are based on clock cycles, not time. DDR3-1333 CAS 6 is about the same speed as DDR3-1600 CAS 7 and DDR3-2133 CAS 9. Overall, CAS has little effect because it is only the delay to the arrival of the first data from the memory. The clock speed decides how fast the data will arrive and is far more important.
The Ripjaws range of memory from G.Skill appears to be standard memory with metal heatsinks clamped on. If you are not overclocking memory, you do not need the metal heatsinks. In fact the heatsinks restrict air flow between the memory sticks and create heat problems. The heatsinks have to be really good to provide a benefit greater than the problems created by the restricted air flow. The Ripjaws design seems to work effectively when not overclocking or using only a low level of overclocking.
When you run your memory hot, either through overclocking or through jamming to much hardware in a poorly ventilated case, the heatsinks can work only if there is airflow over their top edge. Some processor fans will blow the air over the memory. Some do not.
Most motherboards have the memory at the front near the disks. A good case will have, or let you add, a big quiet PWM fan blowing air over the disks. The air should then flow over the memory then be sucked into the processor fan. If the front fan is too low, you might need a second fan. My new Antec Solo II case lets you fit two 120 mm fans to the front. The higher fan will blow straight over the memory.
DDR3 is the fastest standard memory you can buy. There are computers with faster memory but it is proprietary, expensive, and you end up with really cheap shortcuts in other parts of the computer slowing down everything else. As a comparison, on one project a $2000 PC with standard memory performed faster than a $12000 proprietary system despite the proprietary system having fancy fast memory.
The fastest standard DDR3 memory is DDR3-2133. The most common economy DDR3 options are DDR3-1333 and DDR3-1600. The generic and cheap brands are DDR3-1333. You will see a significant difference at DDR3-2133. The extra speed is worth the difference, especially when the price difference is as low as the price difference between the G.Skill memory and everything else.
DDR3 lets you have up to 8 GB per memory slot. For some reason, memory manufacturers, motherboards, and processors are one step behind and handle only 4 GB per slot. This is not a problem because there are 4 slots, giving you a total of 16 GB. When I have lots of applications and files open, neither Linux nor Windows uses more than 2.5 GB.
Why have so much memory? Windows will cache all the file and directory information if there is enough memory. Linux does something similar and appears to require more memory before it will cache everything. For Windows, the file and directory cache appears to be about half the available memory. When you have several terabytes of disk and perform repeated searches, a few extra ggabytes of memory makes subsequent searches and browsing faster.
The memory also helps when I use MySQL and equivalents in local applications. If you have a couple of gigabytes of memory, you can run MySQL at the medium setting. When you have more memory, you can run MySQL at the maximum memory usage settings. Frequently repeated queries are faster.
8 GB of DDR3-1333 is available for AU$44 from a reliable brand in Sydney. 8 GB of DDR3-1600 is available for AU$55 from a reliable brand in Sydney. The cost difference is only $11 and you will save that much time in one day.
8 GB of DDR3-2133 is available for AU$100 from a reliable brand, in this case G.Skill, in Sydney. The price difference from DDR3-1600 is only $45 or an hour of work for the average worker in Sydney. The extra speed will save you more than an hour of work in the first month or week.
In some countries Crucial is a cheaper brand and Kingston a dearer brand. Around here, Kingston is at the bottom end of the brand price range and Crucial further up. G.Skill is not at the bottom of the market anywhere and in some countries it is a premium brand. In every country where I can find a direct comparison, G.Skill is cheaper than Crucial and Corsair for the top speed from standards compliant memory.
Benchmarks are almost meaningless unless they test the exact set of software you want to use. I read a few memory benchmarks where they ran the memory at non standard CAS settings. Other tests use different amounts of memory for each test. Some tests list memory speed results in the wrong order because they are not testing memory speed.
If you use faster memory and the processor chipset can handle the extra speed, some things will be faster while others will run at the speed of the processor, or the disk, instead of the memory. Jumping from 1600 to 2133 will make memory operations 10% faster and not have an effect on anything else.
CAS speed increases have such a small effect that the difference is less than the margin of error in some benchmarks.
Doubling memory size will not alter anything until you run a benchmark that fills the memory. Sometimes you can just run the same benchmark several times and see a difference on the later benchmarks because memory is filling up. In real life you get the same result from opening many applications at the same time.
I used fast memory in a conventional computer using magnetic disks. The applications started and processed at the speed of the magnetic disks. The applications displayed information at the speed of the graphics chip. I then moved to a fast SSD and a fast graphics chip. Everything was faster but the memory speed became a bottleneck. I upgraded to faster memory and achieved the maximum speed for the processing. The difference was noticeable and satisfying.
Memory is currently cheap compared to the other components of a computer. The small price difference for faster memory is an invitation to use the fastest memory in a range. If you are using DDR3, the fastest memory is DDR3-2133 and is worth the small extra cost. Your processor can only use the extra speed if the processor is fast and you use the right chipset. For Intel i5 processors and DDR3-2133, you need the Z68 chipset. Several people using cheaper motherboards with chipsets slower than the Z68 have reported no difference with memory faster than DDR3-1600 and, in some cases, no faster than DDR3-1333.
Your processor can only load applications and data at the maximum speed if the storage device is fast. When you are using a processor that can make use of the maximum speed of DDR3-2133 memory, put your operating system and applications on a good SSD. Most SSDs read data faster than a magnetic disk and will speed up your application load time. Many SSDs have slow write times and will not make data updates faster. When you save some dollars by using g.Skill instead of Corsair memory, use the extra dollars to buy a better SSD with a fast write time to keep up with the rest of your system.
Jumping from my current 1333 memory to 2133 will save more than the extra cost of the memory in less than a month, most likely less than a day on the first busy day. Jumping a standard CAS speed to the higher speed CAS will save the slight extra cost in a month or two. Jumping up from 8 GB to 16 GB will save me something when I run one of my large management processes. G.Skill memory is currently the cheapest way to achieve the desired speed from a reliable brand.