HiGame

Welcome to ExtremeTech’s comprehensive RAM guide, in which we’ll answer a broad range of questions related to how much system RAM you need these days, whether or not it’s worth it to upgrade older systems, and whether DDR3 or DDR4 (the new main types of system RAM) is a better investment option.

It’s interesting to look back and see how much things have changed over the past twenty years. People have been writing RAM guides for decades, but back when I was learning about computing, much more emphasis was paid to the specific technical implementation of any given RAM standard.* Fast Page Mode RAM, EDO RAM, SDRAM, DDR, and RDRAM are just a few of the standards that existed elbow-to-elbow, and which type of memory your system used often determined if it was worth upgrading.

Nowadays, things are simpler. While a few of you may still have DDR2-based equipment from 2005 to 2009, the majority of systems today are likely using DDR3. That’s the memory standard we’ll focus on; if you have DDR2-related questions you’re welcome to drop them in the comments.

How much RAM do you need?

How much RAM you need in a system depends on what you intend to do with it, how long you intend to keep it, and whether or not you can upgrade your memory post-purchase. This last point is important, as many high-end laptops have eliminated user-upgradeable RAM in order to reduce system thickness by roughly six nanometers.
Adding additional RAM to any laptop generally increases power consumption by a measurable (if small) amount, but this shouldn’t be an issue for most users. It’s also better to have a bit too much RAM than too little, as whatever you gain in power savings you’ll promptly lose to increased disk paging.

Apple’s MacBook Air offers 4GB of RAM, but most of the systems from Dell, HP, and other OEMs start at 8GB, and I think that’s the better sweet spot. That’s not to say you can’t get by on 4GB — you absolutely can — but 8GB gives you a bit more breathing room.

The MacBook Air ushered in the era of soldered DRAM. Everyone followed.

There’s at least some evidence that modern desktop applications have slowed the rate at which they demand more RAM. From 1990 to 2000, Photoshop’s minimum RAM requirement rose from 2MB to 64MB, a 32x increase in 10 years. It took another 16 years to match this early rate (from 64MB in 2000 to 2GB in 2016).
A lightweight system today can get by with 4GB of RAM. 8GB should be plenty for current and near-term future applications, 16GB gives you comfortable space for the future, and anything over 16GB is likely overkill unless you specifically know you need it (such as for video editing or audio post-production). This holds true for desktops as well as laptops.

DDR3 or DDR4?

Right now there’s plenty of DDR3 systems still being sold, but DDR4 has already begun to replace it on the mass market. If you’re building a new system and don’t have a specific reason to use DDR3, we’d recommend buying hardware that’s compatible with DDR4.
With that said, if your system does use DDR3, that’s not the problem that it used to be. In the old days, a computer stuck on, say, PC133 SDRAM was at an intrinsic performance disadvantage compared with systems that used DDR, particularly at higher clock speeds. That’s less true than it used to be, and it may make sense to upgrade a DDR3 system depending on what you have and when you bought it. The reason to use DDR4 at this point has more to do with long-term memory pricing trends and future compatibility than fundamental performance. We’ll explore current price and the performance question later in this guide.

Does faster RAM boost system performance?

Short answer: Sometimes, but not by much.
Medium answer: It depends on other system components, workload, and whether or not you’re using integrated graphics.
Longer answer: See below.
RAM performance is controlled by two metrics: Clock speed and access latency. Access latencies tend to fall much more slowly than clock rates — as this diagram shows, the memory cell cycle time of PC100 is roughly equivalent to DDR4-2133. DDR4 doesn’t match DDR3-2133 cycle times until you hit DDR4-4266.


RAM cycle times at various clock speeds

Conventional wisdom is that RAM latency has become relatively less important in recent years, thanks to a combination of factors. Back when L2 caches were small, memory controllers were off-die (and clocked at a fraction of CPU speed), and there were no L3 caches, memory latency had a larger impact on overall system performance. Modern CPUs are typically backed by 512 to 1MB of L2 (per core), and 1.5MB to 2MB of L3 cache (per core). Memory controllers are now integrated on-die and run at full processor speed. As a result, RAM latency simply doesn’t play as large a part as it once did in determining performance.
As for raw memory bandwidth, the same large caches that minimize the impact of RAM latency in most applications also limit the impact of memory bandwidth. Desktop applications are, for the most part, latency-sensitive, not bandwidth-sensitive.


These performance results are from Corsair, but they match extensive testing on the topic. AMD APUs love fast DRAM.

There’s one major exception to this rule: Integrated graphics performance. Both Intel and AMD integrated graphics see some benefit from higher-speed memory, but the gains are particularly large on the AMD side. This has proven true for every APU since at least Trinity, and will likely continue to be accurate for DDR4-based hardware. The advent of HBM2 in APUs will finally throw open the bandwidth floodgates — until then, integrated graphics will always be somewhat bandwidth-limited.

What about high-end gaming performance?

Until recently, I would’ve told you that high-speed RAM had very little impact on high-end gaming. A recent report from Digital Foundry, however, appeared to show otherwise.

more...