How to Choose the Right RAM for Your Desktop or Laptop PC in 2021 – PCMag
It’s been a good, long run, but the end may be sight for the supremacy of DDR4 RAM, the kind of system memory that powers most consumer PCs these days. That said, we’re certainly not holding back on our PC builds and upgrades to wait for the impending DDR5. (The only thing that might make us do that this year is the dire cost of PC components.) Every new generation of DDR memory, historically, has started off with some stumbles and setbacks. DDR4, meanwhile, has a seven-year track record and endured a long, long shakeout in the consumer-PC market. That means it will have a leg up on both compatibility and developmental maturity for some time to come.
But that long familiarity we have with DDR4 doesn’t make memory upgrades today any less complex. How do you go about buying the best memory configuration for your PC? Whether you are building a PC new or upgrading the one you have, the answers come easily—but only once you know the right questions to ask.
For Starters: How Much Memory Do I Really Need?
We’ve been pushing 8GB of memory as the bare minimum for use under Windows 10 for a while now. It’s easy to find mainstream-priced and even budget Windows 10 laptops and desktops shipping with just 8GB of DRAM. Indeed, that’s the norm, and that’s fine for workaday productivity tasks, light gaming, and minimal multitasking. But that’s not a great starting point for an experienced user.
Even browsing the internet can push memory usage over 50% on such systems, leaving little capacity for any remaining programs, such as PC games or photo editors. And while modern web browsers typically lighten the load when other programs demand more memory capacity, that process can make the machine feel sluggish.
That leads to our general capacity guidelines. We recommend 16GB of main system memory for most users who want to multitask without incurring the above-mentioned slowdown, and 32GB for heavy multi-taskers or those running memory-hogging programs such as video editors.
Beyond that is overkill for most folks. Programs that run optimally with 64GB or more are generally designed for experienced or professional users who already know how much they need—or at least, who know that they need all they can get.
How Much Memory Do I Have?
One of the easiest ways to see how much memory you already have is to open any folder in Windows 10 (in the default view), find the This PC icon on the left, right-click it, and go to Properties. This report below from an older, previously upgraded Dell notebook shows that it has an older Core i5 CPU and 8GB total memory, 7.8GB of which can be used for programs. (The rest is reserved for use by the processor’s integrated graphics.)
A basic memory report in Windows 10
The above report may also list the manufacturer name and model, but it wasn’t available on this system after upgrading from the factory installation to Windows 10. If you need greater detail, you can download a third-party utility that will surface much more system info. Our go-to is the freeware program CPU-Z(Opens in a new window)…
CPU-Z, on the other hand, tells all.
In this older example system, the program’s memory tab shows that this PC has a total of 8GB installed in dual channel mode, running at around a 533MHz clock rate, which works out to the “DDR3-1066” memory type, since all generations of desktop DDR have a doubled data rate.
We also see that the actual memory installed is a pair of 4GB DDR3-1600 modules, which are backward compatible to several slower settings. Note that the slot number of the “SPD” tab has a pulldown menu, which is particularly useful when the default “Slot 1” is empty. Additional tabs show things like the motherboard model, which can be useful if you are having a difficult time figuring out what the chipset and platform are at the core of your PC. (For more about checking out the RAM in your current system, see this feature.)
Okay, So I Know What I Have. What’s the Easiest ‘Next Step’?
The search ends here for some PC upgraders, as some machines simply can’t be upgraded. The above screenshots, for example, come from an old, DDR3-equipped notebook that doesn’t support modules of 8GB each and already has its two slots filled with 4GB memory modules. A key thing for starters: DDR3 is a sign of an older PC, and you can’t simply swap in DDR4 modules in their place. On both laptops and desktops, DDR3 and DDR4 memory are keyed differently from one another and are incompatible. But if you’re not getting this information from the manufacturer, there are other ways to figure it out.
Memory sellers that specialize in end-user sales (notably, Crucial(Opens in a new window) and Kingston(Opens in a new window)) offer online “memory configurators” to help potential customers find an array of compatible memory-module options from their enormous product stacks. Unlike the oft-outdated memory-module “compatibility lists” that system and desktop motherboard manufacturers maintain on a board-by-board level, memory manufacturers’ lists are constantly updated to represent real-time availability. Buyers can simply select the fastest kit of the desired capacity recommended for their system, but with the understanding that these lists typically lean toward the safest parts, rather than, necessarily, the fastest or best-value ones. (Crucial’s is called Crucial System Advisor(Opens in a new window), while Kingston’s is Kingston Memory Finder(Opens in a new window).)
Tracking RAM upgrade possibilities with Kingston’s Memory Finder
Now, if all you want is a memory-capacity boost, and you’re not concerned about eking out every last droplet of performance or overclocking, your search can end there. Using a memory configurator is a safe bet, and it is often the best idea for upgraders of laptops, whose memory-upgrade options are usually pretty limited, anyway.
If you’re a PC enthusiast, though, and are looking at a high-performance desktop, a memory maker’s configurator may not go deep enough. We like picking our own memory, which is where the next parts come in.
What Basic Type of Memory Do I Need?
If you’re not already looking inside your PC, an easy way to figure out the memory format is to look at the system manufacturer’s product page, user manual, or service manual. Most desktop PCs make use of unregistered memory (UDIMMs, commonly just called DIMMs). DIMM stands for “dual inline memory module.”
A lineup of two desktop DDR4 DIMMs (top two rows) and two SO-DIMMs (bottom row)
Notebooks, meanwhile, almost all make use of shorter, “small-outline” DIMMs (SO-DIMMs, also called SODIMMs and pronounced “sew-dims”). Compact desktops will use one or the other of these, depending upon what the designer found to be the best fit for the system mainboard and chassis. The smaller the system, the more likely it is to rely on SO-DIMMs versus regular DIMMs, simply because the former are much smaller in surface area.
A laptop-style DDR4 SO-DIMM
The wider spacing of the components on the printed circuit boards that make up desktop DIMMs allows for additional parts to be installed, such as heat sinks and or even RGB lighting strips for PC modders. Laptop-style SO-DIMMs, on the other hand, are designed to be installed in tight, stacked or overlapping slots, and to be invisible, and thus skip such excess. A pair of each is shown above.
As mentioned, DDR4 is the norm in almost all current laptops and desktops. The basic data rate standard for DDR4 memory is 2,133MT/s (that is, million transfers per second), which transfers at double the clock frequency of 1,066MHz. The basic data rate for DDR3 was 1,066MT/s, which transferred at twice its 533MHz clock frequency. Note that it’s not wrong to label a data rate with “MHz,” since a data cycle is still a cycle: Many industry writers simply choose “MT/s” nomenclature to avoid confusion between it and the clock frequency.
DDR4 was launched primarily at per-module capacities of 4GB to 16GB each, while DDR3 modules favored 1GB to 4GB capacities per module. The upper limits of these specifications were twice as high, but it took several years for 32GB DDR4 and 8GB DDR3 modules to reach the consumer market after the introduction of the memory type. Because of those delays, many older motherboards required a firmware update to support the bigger, later capacity. (As seen in the earlier “old Dell notebook” example, many platforms never got those updates.)
To summarize, at a basic level, most systems should support at least 16GB per module of DDR4-2133 memory, or 4GB per module of DDR3-1066, without overclocking. And again, if you’re simply looking for a memory bump to boost your multitasking and browsing, you can stop there and go with this basic speed of module according to whether you need DDR3 or DDR4. But we like to go past that—when we can!—and fortunately most DIY-minded desktops are designed with the performance credentials to get us there.
So, About Memory Specs: Is High Frequency Better Than Low Latency?
This is where we start to get into the geeky stuff. Let’s start with the short answer: While a higher data rate usually has a greater impact on measured performance, optimally “timed” memory kits such as DDR4-3200 CAS 14 can often outperform poorly timed kits such as DDR4-3600 CAS 20—despite the optimized kit’s lower data rate. (More about what “CAS” is in a moment.)
At the most basic level, frequency is the number of times anything happens over a certain period, while latency is the time it takes to catch up. Increasing the frequency of a data transfer will always increase the bandwidth of a continuous transfer, but because memory data is transferred in small packets, the delay between packets pushes bandwidth in the opposite direction. Latency is measured in nanoseconds but specified in clock cycles. Called “primary timings,” the four most significant of these are often indicated on a sticker on the memory module, or in its specifications list.
Memory timings: Sometimes, they’re right on the sticker.
Memory cells are organized in rows and columns in a similar fashion to spreadsheets:
CAS Latency (tCL) refers to the number of cycles required to access the cell in the correct column, when the correct row is already open.
RAS to CAS Delay (tRCD) refers to the amount of time it takes to open the correct row.
Row Precharge (tRP) refers to the amount of time it takes to close the incorrect row.
Row Active Time (tRAS) refers to the combined time required to close the incorrect row and open the correct row.
For most folks outside the overclocking crowd, this gets pretty deep in the weeds. This free-to-share video (by yours truly) gives a quick visual representation of these descriptions…
Just how fast is a clock cycle? Since frequency (operations per second) is the inverse of latency (seconds per operation), and since DDR4-3200 operates on a 1,600MHz bus clock, the answer at DDR4-3200 is 1 divided by 1600000000, or 0.625ns per cycle. The same calculations place DDR4-2400 at 0.833ns per cycle. And since 16 times 0.625 equals 10, and 12 times 0.833 also equals 10, DDR4-3200 CAS 16 has the same 10ns real-time latency as DDR4-2400 CAS 12.
Yes, that’s some in-the-weeds math. But this explains why in our lead example, DDR4-3600 CAS 20 (11ns) can underperform DDR4-3200 CAS 14 (8.75ns) in certain operations: It takes 2.25ns longer for DDR4-3600 CAS 20 to respond. Most memory buyers won’t get down to that level of granularity, but that explains why you can’t weigh just a single specification in assessing performance memory.
What Is XMP?
Intel’s Extreme Memory Profiles (XMP) are additional configuration sets, accessed via the system BIOS, that allow the motherboard to automatically apply overclocking values to match the needs of nonstandard memory. As an overclocking technology, XMP has some limitations: Some motherboards don’t support XMP at all, and some modules are programmed only with specific XMP values that exceed a given motherboard’s capabilities.
Turning on XMP in an Asus BIOS
It may be an Intel technology, but enthusiast-class AMD motherboards are also designed to support XMP. As motherboards are often programmed to slightly alter certain timings to further stabilize AMD’s different memory controllers, motherboard manufacturers have occasionally applied their own names to this setting, such as Asus and its D.O.C.P.
The usual drawback of XMP involves inadequate module programming. Many memory kits have only two automatic configurations—say, DDR4-3600 CAS 18 and DDR4-2133 CAS 15, where the motherboard will retain the CAS 15 setting when you manually select a middle value such as DDR4-3200. The manual configuration fails if the memory required CAS 16 to operate at DDR4-3200.
Different users can argue differently about the best memory product, but from an ease-of-use standpoint, it’s easier to argue, say, for a DDR4-3200 kit that contains a DDR4-2933 secondary XMP along with basic configurations of DDR4-2666, DDR4-2400, and DDR4-2133 than it is to argue against having those fallbacks. Overclocking is never a certainty, and it’s nice to know that the party won’t stop just because some other part of the system (such as the CPU’s memory controller) isn’t cooperating with an XMP setting that’s supposedly supported by the motherboard.
How Do Multiple Memory Channels Increase Performance?
A single channel of memory is 64 bits wide. Most modern systems support dual-channel memory architecture, which widens the memory pathway to 128 bits. With more cores being fed more data under heavier workloads, some High-End Desktop (HEDT) platforms, notably Intel’s Core X-Series (on socket LGA2066) and AMD’s Ryzen Threadripper (on sTR4) take this further, to 256 bits, with quad-channel memory arrangements.
Eight DIMM slots (for quad-channel operation) on an Asrock X299 Taichi motherboard
One thing to remember is that most systems require a matched pair of modules to run dual-channel mode, or four matching modules to operate in quad-channel mode. While past platforms have occasionally allowed for mixed modes using different modules, those didn’t perform optimally. That doesn’t necessarily mean that you have to ditch an old pair of modules when a pair of empty slots are available, as we’ve had good experience adding a new matched pair to an old matched pair of the same data rate, but doing so may make XMP mode unworkable. We’ve even added 2x 8GB kits next to 2x 4GB kits without breaking dual-channel mode, creating a 24GB (12GB per-channel) configuration as 8GB-4GB-8GB-4GB, by simply leaving the board at default (non-XMP) settings. You’ll just want to make sure the matched pairs are inserted into the proper paired DIMM slots recommended by the motherboard maker.
What Are Memory ‘Ranks,’ and Why Should I Care?
Each dual inline memory module (DIMM) has two 64-bit interfaces (one on each side) connected in series. Each interface supports one rank of memory, so that a single-sided module usually has one filled rank, and a double-sided module usually has both ranks filled. (Caveat alert: Though less common, some memory has through-paths, or “vias,” that connect both sides to a single interface.) Since the two sides of a dual-rank module are connected in series, one might not expect the added rank of memory ICs (integrated circuits, i.e. “chips”) to improve performance. This is where interleaving comes into play. Interleaving allows two different operations to occur simultaneously, such as accessing data on one rank while transferring data on the other.
The memory controllers of most consumer processors support up to four ranks of memory per channel, which is why so many dual-channel boards have four slots and why so many quad-channel boards have eight. If every module used in these boards was dual-rank, the memory controller would be “full.”
How does one determine whether a module is dual- or single-rank? Specs may tell you, but you can’t count on that. If not, physical examination is another way. A look under the edge of a module’s heat spreaders would reveal how many ICs are used. Since the ICs on most performance-oriented memory modules have an 8-bit interface, eight of those make up a 64-bit rank. (Some low-end memory uses four 16-bit ICs per rank. These “chips” tend to be rectangular.)
However, actually looking at RAM modules and peering under stickers or heat spreaders is not a realistic method for anyone ordering RAM online or trying to examine memory that’s packaged up in a store. Researching memory via memory reviews can help, but finding a review of the exact kit and speed/capacity flavor you are looking it is hit-and-miss. And even professional RAM reviews are relevant only if they’re very recent. Why? We’ve seen companies apply an old part number to a new product with half as many ICs (each at twice the density). Tweaking the actual on-module components can make all the difference.
Depending on what you do, it’s a legitimate strategy to buy a kit containing four DIMMs for a four-slot dual-channel motherboard, since you’re guaranteed to have at least one rank per module. But some motherboards are wired to overclock better with only two DIMM slots filled. If that is what you aim to do, you need to factor that in. Alternatively, kits that contain 32GB modules always have dual-rank DIMMs, since 16Gb is the current density limit for high-end consumer DDR4 ICs, and eight of those make a 16GB rank.
What Is the Best Memory Kit for Most Performance Enthusiasts?
Owners of enthusiast-class PC desktop motherboards have the advantage of multiple firmware settings to get their machine configured perfectly, but there are limits to what the hardware can support on a board-by-board level. Recent AMD architectures, and the latest Intel ones, clock the CPU’s memory controller at the same frequency as the memory, and most samples appear to hit limits somewhere between DDR4-3700 and DDR4-3900.
Both also allow the user to choose a memory-controller ratio other than 1:1 to reach even higher data rates, but doing so reduces performance by underclocking the memory controller. Motherboards using AMD’s X570 chipset will automatically reduce the memory-controller frequency (a spec called “FCLK”) at settings beyond DDR4-3600, and those based on Intel’s Z590 chipset with 11th Generation Core CPUs switch from what’s known on that platform as “Gear 1” (synchronous memory controller frequency) to “Gear 2” (half-speed) at settings above DDR4-3200. Overclocking motherboards allow AMD’s FCLK to be forced to 1:1 and Intel’s Z590 to Gear 1, but stability at synchronous data rates beyond DDR4-3600 is hard-fought.
Thus, the fastest practical kits for most performance enthusiasts will contain (and we’ll emphasize this with boldface!) dual-rank modules rated at DDR4-3600 CAS 14. (That is, unless you can find these specs at something lower than CAS 14.) Compatible platforms include recent mainstream AMD AM4 boards, along with most Threadripper (sTR4), Intel Core-X (LGA2066, LGA-2011v3), and mainstream Intel (LGA1200, and LGA1151), assuming the board is equipped with overclocking features.
Note that Intel’s 10th Generation and earlier processors ran memory asynchronously to the controller clock and thereby avoided controller frequency reduction, though performance gains were minuscule at data rates beyond DDR4-3600.
Let’s Get Granular! Our Platform-Specific RAM Recommendations
We’ve come up with an, ahem, “short” list of what you can (and/or should) use with specific desktop platforms, attempting to place these in rough chronological order (by release date, newest to oldest). For custom desktop PC builds, we recommend treating the statements of motherboard manufacturers regarding their memory support as theoretical limits and reading reviews to determine practical limits. Additionally, firmware limits set by system manufacturers usually cannot be exceeded, regardless of whether the machine is a notebook or desktop.
▶ Intel Z590, H570, and B560 Chipset Motherboards (With an 11th Generation Core “Rocket Lake” CPU)
The short version: Enough overclockers have shown the Intel 500 Series of chipsets stable at DDR4-3600 that we have no reservation recommending that class of DIMM to anyone with a 125-watt-TDP 11th Generation (“Rocket Lake”) processor like the Core i9-11900K, an adequate motherboard, and even the most modest tuning skills. Getting maximum performance from this data rate requires the memory controller to be overclocked by manually setting Gear 1 (synchronous memory controller frequency) mode.
Buyers who won’t or can’t overclock should stick to Intel’s guidelines to retain Gear 1 level performance, which are…
DDR4-3200 for the Core i9-11900K
DDR4-2933 for lesser 11th Generation Core i9, Core i7, or Core i5 chips
DDR4-2666 for Core i3, Pentium, or Celeron
The default switch from Gear 1 to Gear 2 when using DDR4-3200 with anything less than the Core i9-11900K is disabled on most retail motherboards, but we’ve yet to see the DDR4-2666 limit exceeded on budget processors, and Intel’s inclusion of memory overclocking in its H570 and B560 chipsets has not helped those with a DDR4-2666 limit.
Recommended by Our Editors
Lower-energy CPUs, such as Intel’s 65-watt TDP models, often power-throttle under heavy load, and the increased voltage of performance DRAM can amplify the problem. Manually configuring higher power thresholds is possible within the firmware of adequately provisioned (overclocking) motherboards. But once you’re in that zone, it’s a tricky push-and-pull of performance versus thermals, the likely reason you opted for a 65-watt CPU in the first place.
▶ Intel Z490, H470, B360, H410 Chipset Motherboards (With a 10th Generation Core “Comet Lake” CPU)
The short version: With no “Gear” modes to worry about on these previous-generation chipsets, there’s little to stop a mildly experienced builder from simply enabling a DDR4-3600 XMP profile on an overclocking-enabled Z490 motherboard. That’s a pretty safe bet for system builders working from a retail-sold DIY motherboard. Note, however, that some OEM systems might have the (ostensible) overclocking chipset, but not the firmware settings to actually accomplish this.
Without overclocking, Intel’s 10th Generation Core i9 and i7 processors support memory up to DDR4-2933, while its Core i5 and i3 versions top out at DDR4-2666. Intel never unlocked overclocking for its B or H 400-Series chipsets.
As with the Z590, it might be necessary to increase the power threshold of lower-energy (65-watt) processors to prevent a power-throttling increase. Check your motherboard firmware for these settings prior to memory selection.
▶ AMD TRX40 (Threadripper) and X570, B550, or A520 (Mainstream Ryzen) Chipset Motherboards
Even though these two are completely different platforms, both support DDR4-3600 at a synchronized FCLK. AMD recommended DDR4-3200 at the time of the launch of the Ryzen 3000 Series, and buyers who can’t afford DDR4-3600 at reasonable timings (CAS 18 or lower) might wish to consider this less-expensive option.
▶ Intel Z390, H370, B360, and Z370 Chipset Motherboards (With 8th and 9th Generation CPUs)
Intel’s Z-series chipsets are super-friendly to memory overclocking on adequately provisioned motherboards, so the same DDR4-3600 “best” and DDR4-3200 “alternative” recommendations apply for these chipsets catering to 8th Generation and 9th Generation CPUs. Unfortunately, H370 and B360 do not generally support anything beyond Intel’s official limits, which are DDR4-2666 for the Core i9, i7, and i5, and DDR4-2400 for the Core i3, Pentium, and Celeron.
▶ AMD X470 and B450 Chipset Motherboards (With Mainstream Ryzen CPUs)
Support for high RAM data rates across various motherboard models under these chipsets is mixed. Some easily exceed DDR4-3600; others barely go above DDR4-2933 when paired with a Ryzen 2000 Series CPU. The closest thing we’ve seen to consensus has been DDR4-3466, but again, we’ve had boards that topped out at far less.
The good news is a resurgence of AMD-compatible DDR4-2933 memory modules on the market following Intel’s addition of this speed to its 2020 desktop processor guidelines. Those who think that DDR4-2933 is unacceptably slow should dig a little deeper to find out what other people are running with the same motherboard and processor. Imitation can be far more than the greatest form of flattery—it can save you a heap of time and trouble!
▶ AMD X399 Chipset Motherboards (With First or Second Generation Ryzen Threadripper CPUs)
Remember that Threadripper X399 boards tend to have eight memory slots. Breaking from DDR4-3600 recommendations because many builders wish to fully populate these boards with eight dual rank modules (16 total ranks), stability at this setting is still common when using up to eight total ranks with the Ryzen Threadripper 2950X. Earlier processors can be fussier, though. DDR4-3200 is compatible across most Threadripper processor models and memory configurations, but AMD recommends only DDR4-2933 for second-gen Ryzen Threadrippers and DDR4-2666 for first-gen Ryzen Threadrippers.
▶ Intel X299 Chipset Motherboards (With LGA2066 Core X-Series CPUs)
Like Threadripper, Core X-Series boards gravitate to eight slots for quad-channel support. CPUs from the 9th and 10th Generation of Intel’s HEDT platform typically supported memory frequencies exceeding DDR4-3600 with up to four dual-rank modules, but DDR4-3200 became a far safer choice when deploying the platform’s 16-rank maximum configuration or when using a 7th Generation Core X-Series processor. For non-overclockers, Intel supported up to DDR4-2933 on 10th Gen and DDR4-2666 on 9th and 7th Generation Core X-Series CPUs.
▶ AMD X370, B350, A320 Chipset Motherboards (With Older Ryzen CPUs)
High data rates are a pipe dream for most users of the 300-Series AMD chipsets, and that’s mostly because of some major variation in the memory-controller stability of Ryzen 1000 Series CPUs. Some motherboard and CPU combos were good past DDR4-3466, while others couldn’t make it over DDR4-2400. Failures in trying to boost frequencies were more likely to occur as the number of ranks increased (for example, using dual-rank rather than single-rank DIMMs, or using four rather than two DIMMs).
Given this variation, we’d personally recommend DDR4-2933 that has a DDR4-2666 secondary XMP and DDR4-2400 SPD, such as Kingston’s(Opens in a new window) HX429C15PB3A (HyperX Predator RGB DDR4-2933) series. A set that this writer tested functioned properly on every older platform tried and blew past DDR4-4000 on newer platforms. So manual overclocking remains viable to those who find their CPUs exceeding our justified low expectations here.
▶ Intel Z270, H270, and B250 Chipset Motherboards (With 7th Generation Core CPUs)
Intel’s 7th Generation Core processors are getting on in years now. But they were (and remain) DRAM overclocking monsters, with many motherboards pushing data rates beyond DDR4-4000. Making DDR4-3600 run stably is usually no more difficult than just enabling XMP on overclocking-enabled Z270 motherboards.
DDR4-3200 could be a better choice for users who can’t afford DDR4-3600 at CAS 18 or lower latency, though. And given the age of these platforms, even slower (and thus, cheaper) memory could be appropriate. Investing in premium RAM for a venerable PC may not make sense within your budget, especially if you think you might upgrade the whole system before long.
Neither the H270 chipset nor the B250 supports memory overclocking, and DDR4-2400 is Intel’s default frequency limit across all 7th Generation Core CPUs.
▶ DDR3 Motherboards
DDR3 is the sign of a geriatric PC, and spending to the max on performance-minded RAM for a platform that is fast disappearing in the rearview mirror may be false economy. Most DDR3 motherboards supported at least DDR3-1600, with later examples such as the AMD 990FX and Intel Z97 often exceeding DDR3-2133 and DDR3-2800, respectively.
That said, check those specs carefully. Many early platforms limit you to installing only up to 4GB per module, while later ones might support 8GB with a proper firmware update. Challenging examples, such as the notebook shown in the screenshot at the start of this article, put additional pressure upon buyers to use the compatibility lists of various memory sellers to find better options than those available from the outdated support lists of system manufacturers. Indeed, if you are shopping for a RAM upgrade for a DDR3-only PC, spending the least money possible is your best value play.
Finally: When Buying RAM, What About Maximizing Value?
When it comes to a PC component as opaque as system memory, the idea of value-for-money often gets put aside when considering the “best” choice for a performance machine. But there’s a big statement that should be made here: Most programs see very little gain from high-performance memory with elite specs, and even the most memory-impacted programs we’ve used have showed less than 6% performance gain in going from ordinary DDR4 to an optimized configuration.
Moreover, most of that gain can simply be achieved by moving from one rank per channel to two, something that you might accomplish by simply adding another two matched-spec modules to a machine that has two empty slots. So bear that in mind as you shop the sales.
Popularity also drives memory-module availability to the point of affecting supply and demand. For example, DDR4-3200 CAS 16 memory represents some of the best current values we’ve found, at $80 for a pack of two 8GB modules. The sword cuts both ways, though. Take DDR4-3000. It became so popular that it virtually displaced DDR4-2933 from the market a few years ago, and that kind of memory is still commonly available, at less cost than DDR4-2933. It would be nice if people who really wanted DDR4-2933 could trust the slightly faster DDR4-3000 modules to self-configure at the slightly slower speed, but as outlined in the “What Is XMP?” section above, this isn’t usually the case. While some motherboards will allow users to pick a DDR4-3000 XMP profile and manually drop the data rate to 2933, others won’t. So your purchase, in terms of speed-versus-dollars, needs to be gauged against what you know your motherboard will play nice with.
For an extra bit of good news, consider this: DDR4-3600 CAS 18 is just as quick, has more bandwidth, and generally costs only 10% more than DDR4-3200 CAS 16. It might not be the CAS 14 pinnacle, but who among us, if we care about eking out performance at this level, wouldn’t find a way to afford so small a price difference?
That’s the kind of smart trade-off that you’re looking for in memory shopping. But ultimately, the simple luxury of having 16GB versus just 8GB at your PC’s disposal, or 32GB versus 16GB, will be what has the biggest real-world impact. So don’t let a sliver of specs get in the way of making that upgrade. Like that second slice of chocolate cake, extra RAM is one of those splurges that you’ll seldom regret making.
Like What You’re Reading?
Sign up for Tips & Tricks newsletter for expert advice to get the most out of your technology.