RAM is fairly integral to many of the functions on a given PC, smart device, gaming console, etc. It’s responsible for holding your applications in memory for easy access by the CPU and your other components.
Besides applications, it’s also important for holding large files in productivity software, though there is some crossover with VRAM when it comes to the specifics of how this is applied.
Largely, RAM and VRAM are separate technologies. But let’s talk about what VRAM actually is.
What is VRAM?
VRAM, or Video Random Access Memory, is the dynamic memory used specifically by your Graphics Processor (GPU) or integrated graphics (on an SoC [System on a Chip] or APU).
VRAM isn’t used for general computing purposes- as its name implies, it’s tied to specifically video-related tasks.
In gaming, VRAM manages the raw data needed to load the scene.
The amount of VRAM usage can scale directly to settings like resolution and texture quality, though it’s also somewhat common for games to take up as much VRAM as they can get, even if there isn’t a performance improvement from doing so.
For something like professional 3D rendering or video editing, VRAM works similarly, but not nearly as arbitrarily.
The amount of VRAM you have in professional software scales directly to the size and complexity of projects that you can manage.
Regular old RAM can be used in these scenarios too, but is much much slower than VRAM in the majority of scenarios, resulting in slowdowns and bottlenecks.
One important reason why GPU Render Engines slow down considerably when using RAM instead of VRAM (read: when a 3D Scene becomes too big to fit entirely into VRAM alone), is because RAM is much farther away physically (on the Motherboard), compared to VRAM which is soldered right onto the Graphic Card’s PCB next to the GPU Core. The closer the memory, the lower the latency, the higher your performance.
Types of RAM
DDR RAM, or Double Data Rate RAM, is what is used by PCs and smart devices as system memory.
It’s called “Double Data Rate” because its speed effectively doubles whenever you’re using two or more sticks, and this is actually required in order to achieve the rated frequency on the box.
For example, you aren’t going to be able to run your 3600 MHz RAM at 3600 MHz unless you have two RAM sticks, otherwise, your single RAM stick will run at half the advertised frequency.
There are several generations of DDR RAM, but at the moment the most prevalent options on the market are DDR4 and the newly-introduced DDR5.
DDR3 on the market today is mainly for upgrading older PCs, whereas DDR4 is compatible with the majority of PCs, even a number of current-generation platforms that can support DDR5.
Until DDR5 prices go down, I recommend sticking with DDR4 and DDR4-compatible boards so you don’t get gouged on memory.
To get an idea of how the DDR RAM standard has evolved over time, I’ll list each standard and its specifications below:
Generational Leaps in Memory DDR Generations
DDR RAM – Starts at 200 MHz, peaks at 400 MHz.
DDR2 RAM – Starts at 400 MHz, peaks at 1066 MHz.
DDR3 RAM – Starts at 800 MHz, peaks at 2133 MHz.
DDR4 RAM – Starts at 1600 MHz*, peaks at 3600 MHz.**
DDR5 RAM – Starts at 3200 MHz, peaks at 6400 MHz.***
* Most DDR4 RAM starts being sold at 2133 MHz or faster to be a clearer improvement over DDR3.
**This is not the fastest achievable speed with DDR4 RAM. Currently, up to DDR4-4600 RAM is being sold on the market, and G.Skill recently announced DDR4-5333 RAM.
However, achieving these much higher RAM speeds will require you to tweak XMP or your BIOS’ equivalent settings, and they may not work at the rated speeds on your system.
Be sure to do thorough research into your motherboard and CPU choice before getting high-end RAM!
***Since DDR5 is a new standard at the time of writing, this limit may increase over time.
Types of VRAM
GDDR RAM, or Graphics Double Data RAM, is the primary type of memory used by most graphics cards.
VRAM Modules on an Nvidia GPU PCB – Credit: Nvidia
Some graphics cards may also use HBM VRAM, or in particularly bad cases, DDR RAM.
Interestingly, there are situations when DDR VRAM actually isn’t that bad. More on that in its respective section below.
Like DDR (V)RAM, there are a variety of generations of GDDR VRAM.
Unlike DDR Kits, however, GDDR isn’t sold independently and doesn’t have differing speeds at a particular generation.
You can’t add or upgrade GPUs VRAM as they are soldered onto the Graphic Card’s Printed Circuit Board.
Any improvement to GDDR is either done on the user-end through overclocking or on a manufacturing level via a proper generational upgrade.
Bandwidth comparison on GDDR technologies – Credit: Micron
With that said, here are the generations of GDDR RAM:
GDDR RAM – Originally called DDR SGRAM.
GDDR2 RAM – 500 MHz memory clock with a transfer rate of 16 GB/s. Essentially the first “real” GDDR standard.
GDDR3 RAM – 625 MHz memory clock with a transfer rate of 19.9 GB/s.
GDDR4 RAM – 275 MHz memory clock with a transfer rate of 17.6 GB/s. While these numbers are lower than GDDR3, GDDR4 is actually much faster overall due to architectural improvements.
GDDR5 RAM – 625-1000 MHz memory clock with a transfer rate of up to 64 GB/s. This is where VRAM starts having the muscle for proper modern games.
GDDR5X RAM – 625-875 MHz memory clock with a transfer rate of up to 112 GB/s. This is a generational improvement on GDDR5, but still uses the same underlying architecture. Outside of bandwidth-limited scenarios, real-world performance improvements will be limited compared to GDDR5.
GDDR6 RAM – 875-1000 MHz memory clock with a transfer rate of up to 128 GB/s. Actually offers a significant performance improvement over GDDR5X due to architectural improvements outside of the light bump to raw transfer rate.
GDDR6X RAM – Not yet standardized at the time of writing. Expected to be a generational improvement over GDDR6 with significant improvements to temperatures and power consumption.
HBM, or High Bandwidth Memory VRAM, is primarily used by AMD’s Vega series of Graphics Cards and data center/server GPUs.
HBM is fairly unique and expensive to manufacture, so HBM isn’t seen very frequently in consumer graphics hardware at all.
Even AMD’s current-gen cards seem to be using GDDR6 instead, but this might change whenever HBM3 is standardized.
HBM VRAM – Up to 16 GB/s max bandwidth. While this sounds limited compared to GDDR, this version of HBM still outperformed the competing standard of GDDR4 when it launched in 2015.
HBM2 VRAM – Up to 32 GB/s max bandwidth. A significant improvement over HBM, this is where HBM becomes a fairly compelling option, offering raw speeds better than GDDR5.
HBM2E VRAM – Up to 38.3 GB/s max bandwidth. A generational improvement over HBM2 that makes it more competitive with GDDR5X and GDDR6.
HBM3 VRAM – Not yet standardized at the time of writing. Expected to offer significant improvements over 2E in line with previous HBM refreshes and be used for data center GPUs.
Last and actually least, we have DDR VRAM.
This is unfortunately a worst-case scenario where standard desktop DDR RAM is being used as VRAM, and since desktop RAM is much slower compared to VRAM in a given generation, it can be a pretty serious handicap.
This is the case for graphics cards that ship using DDR RAM, anyway- there is a slightly better case scenario for using DDR as VRAM.
This is when you are using an AMD APU or an Intel CPU with an integrated GPU.
Basically, you can get a lot more performance out of those little integrated graphics chips if you’re running on DDR4 or newer and XMP-rated speeds like 3600 MHz.
This is the only time in the desktop space where DDR VRAM is acceptable; if you’re buying a discrete graphics card, proper GDDR or HBM is a must.
How RAM is Used
Well, RAM is simultaneously used for just about everything on your PC but also doesn’t impact performance in the majority of applications, especially with modern standards of DDR3 and newer that are well-suited to your typical PC workloads.
The CPU needs Memory for process any type of task. It does have its own caches (L1, L2, L3 etc.) that are very fast and close to the CPU’s cores, but these caches are very small. Caches fill up fast and then spill over into your System’s RAM which is installed near the CPU on your Motherboard.
Of the other components in your PC, your RAM is most directly tied to your CPU and your motherboard.
The motherboard’s chipset will determine the maximum speed and timings you can achieve with your particular kit of DDR RAM.
Meanwhile, your CPU performance is improved whenever running your memory in a Dual-Channel configuration or better.
Especially if you’re looking to overclock or use a high-performance processor, opting for high-quality RAM is essential.
Only in rare cases will upgrading your RAM by itself impact your workload’s performance considerably:
A RAM capacity upgrade can increase performance if your workloads were previously spilling over onto your disc cache (Swap File)
Upgrade from a single RAM module to two Modules will double your frequency and potential bandwidth if your workloads can make use of this
Utilizing Quad or even Octa Channels on supported Motherboards will increase bandwidth that some workloads can make use of
Upgrading your low frequency, high latency RAM, to high frequency and low latency RAM can increase performance in some workloads
In most workloads, though, RAM upgrades will only net you a marginal performance uplift compared to upgrading the CPU or GPU.
How VRAM is Used
VRAM is used only for graphics hardware. That means you’ll only find it on graphics cards, laptops, gaming consoles, and other high-end PCs or smart devices.
In rendering-based productivity applications, VRAM is used to store things like 3D scenes, textures, or raytracing trees in fast, active, close-by memory directly accessible by the GPU.
VRAM is to the GPU as RAM is to the CPU; it serves as a vital cache for managing lots of information in real-time.
For productivity workloads like these, having lots of VRAM and processing power not only decreases render times, it also increases the scale of what can be done with a project without running into risk of crashes and bottlenecks.
When RAM is Used As VRAM
As discussed above, RAM used as VRAM is generally a big red warning sign of potential bad performance, especially on a discrete graphics card.
However, opting for the right RAM when you’re building an integrated graphics or APU-based PC setup can improve your experience significantly, and CPU-integrated graphics architectures actually work better with DDR RAM than a discrete graphics architecture does.
When VRAM is Used As RAM
VRAM is usually not used for general-purpose RAM.
When it is, it’s actually fairly fast, but you really only see manufacturers doing this in gaming hardware like recent PlayStation and Xbox consoles.
XBOX System on Chip, PCB and VRAM – Credit: Microsoft
It’s fairly inefficient to use VRAM as regular RAM due to the higher cost involved, but in the case of dedicated gaming machines with fixed architectures like those consoles, the hybrid approach actually makes sense.
Plus, since everything is on a single SoC (System on a Chip), the bandwidth concerns that would normally come from separating CPU and GPU on the motherboard aren’t in play here.
Unfortunately, you cannot use your graphics card’s VRAM as standard RAM in your PC.
How Much RAM Do I Need?
The amount of RAM you need truly depends on your workload.
If you’re just using your PC for web browsing and other lightweight tasks, something like 8 GB of RAM should serve you fairly well.
This also works for entry-level gaming builds, but some games are beginning to require more than 8GB of system RAM.
For higher-end gaming and rendering/editing builds, 16 GB of RAM is basically the starting point for having genuine flexibility with the work you’re doing.
You can pretty much keep scaling up from there in accordance with the scale of your workload, but 32 and especially 64 GB of RAM is definitely stretching practicality for non-server machines.
Want a more detailed look at RAM requirements? Check out the proper How Much RAM Do I Need? article written by Alex.
How Much VRAM Do I Need?
The amount of VRAM you need…well, isn’t necessarily clear cut, since VRAM by itself doesn’t determine how a GPU performs.
However, knowing what kinds of VRAM are good (GDDR and HBM), what the current generations are (GDDR6 and HBM2E) are good places to start. I’ve already covered that, but I haven’t really talked about the different capacities of VRAM.
The first thing to know about VRAM capacity is that its usage very much depends on what you’re doing.
VRAM capacity for professional purposes like 3D rendering pretty much scales indefinitely with the scale of whatever project you’re working on.
If you have a large 3D scene, having enough VRAM for all of it to fit inside at once is ideal to reduce bottleneck from system memory.
For gaming, VRAM capacity demands can change on a per-game basis. Some games can even be pushed to run at 4K with only 2GB of VRAM- I would know, as I’ve done it on a GTX 760 before.
However, for modern and cutting-edge AAA titles, the baseline for a good experience at 1080p resolution starts at around 4-6GB of GDDR5X/HBM2 or better VRAM.
Have more specific questions about VRAM requirements? Check out Alex’sHow Much VRAM Do You Need? article for even more detailed information!
What Kind of RAM Do I Need For Integrated Graphics?
Finally, let’s talk about what kind of RAM you need for integrated graphics.
This is, after all, the only scenario for a desktop PC where your RAM will also be behaving as VRAM.
Samsung B-Die Memory is particularly well-suited to memory overclocking, making it a favorite for enthusiasts.
More importantly, achieving high memory overclocks with Samsung B-Die Memory will not only increase your (AMD) CPU performance, but also serve to increase any CPU’s integrated graphics performance by giving it faster VRAM to work with.
Over to You
If you have any other questions about RAM and VRAM, let us know in the comments below or our forum!