Random-access memory (RAM) is an indispensable part of every computer system, as it allows the processor to store a ton of important data that it can then access virtually instantly, thus ensuring that the system works smoothly and without any jarring hiccups.
The system RAM is installed in the motherboard’s DIMM slots, independently of the CPU, and these days, it’s usually DDR4. However, when it comes to the GPU, the memory chips are located on the PCB itself, and there are a few more types of RAM that you will encounter when it comes to the various graphics cards that are available in 2020.
In short, those are GDDR5, GDDR5X, GDDR6, HBM, and HBM 2. So, how do they differ, and is it important to keep them in mind when choosing your ideal GPU?
Read on and find out!
Table of ContentsShow
First, we have GDDR SDRAM, which is currently the most popular type of graphics RAM and it’s what you’ll find in the vast majority of GPUs today. The abbreviation stands for Graphics Double Data Rate Synchronous Dynamic Random-Access Memory, so it is a type of DDR SDRAM (just like DDR4) but specialized for carrying out graphics-related tasks in tandem with a graphics processing unit (GPU).
As you might expect, there have been a few iterations of GDDR over the years. The original version was dubbed “DDR SGRAM”, and the SGRAM stood for “synchronous graphics RAM”, though it was renamed to GDDR for all the subsequent versions that followed: GDDR2, GDDR3, GDDR 4, GDDR5, GDDR5X, and GDDR6.
Now, as we have mentioned in the introduction, it is the last three that you’ll still encounter in graphics cards today, so how do they differ from one another?
First, we have GDDR5, which has been the dominant type of graphics RAM throughout the previous decade, as the majority of the GPUs that were released from 2010 onward used it.
It was followed by GDDR5X, which was a modified version of GDDR5 that had greater bandwidth. It was released in 2016 but only a handful of Nvidia’s Pascal-based GPUs ever actually ended up using it.
As for the differences between these three types of RAM, they mostly come down to speed and power efficiency. As you might have guessed, the newer technologies are both faster and they consume less power. GDDR5 has transfer speeds of 40-64 GB/s, GDDR5X is about twice as fast at 80-112 GB/s, and GDDR6 pushes this a bit further to 112-128 GB/s.
Now, the important question is, how much does this matter for gaming?
As you can see from the video above, when comparing the GDDR5-equipped and the GDDR6-equipped versions of the GTX 1650, the differences in actual in-game performance are generally marginal at best, coming down to only a handful of FPS in most games.
That said, the GPU’s processing power will make a much bigger difference than the type of RAM that it comes with.
Moreover, seeing as pretty much all of the newer GPUs are now equipped with GDDR6 memory, chances are that you won’t even find yourself in a situation where you’ll need to choose between GDDR5, GDDR5X, and GDDR6.
The only time this might happen is if you’re thinking about getting an older GPU model, and in that case, it’s always a good idea to check some benchmarks to see what kind of performance you can expect from it compared to some newer, similarly-priced graphics cards.
Next, we have HBM i.e. High Bandwidth Memory, and the name alone is a dead giveaway as to what this type of RAM has to offer compared to GDDR RAM – bandwidth. HBM memory takes up less space on the PCB, uses less power, and provides substantially more bandwidth than GDDR. All of this is achieved by “stacking” multiple (up to eight) DRAM dies on top of each other on the PCB.
The first thing that you might notice about HBM memory is its impressive memory bus that starts at 1024 bits per stack. That said, the more stacks there are, the wider the memory bus will be, and most HBM-equipped GPUs that were released so far come with either a 2048-bit or a 4096-bit memory bus.
Needless to say, this makes the memory bus on most GDDR configurations seem downright diminutive in comparison, as the majority of GDDR-equipped GPUs available at the moment have a memory bus that ranges from 128 to 392 bits.
Then, of course, there’s the bandwidth. As you might have guessed, HBM offers much greater bandwidth than GDDR, starting at 128 GB/s for HBM, 256 GB/s for HBM2, and a whopping 460 GB/s for HBM2E, although this is a new technology that hasn’t been implemented in any GPUs yet.
In practice, the overall bandwidth varies from GPU to GPU, and it can range anywhere from 307 GB/s in the more modest implementations of HBM to a whopping 1024 GB/s in cards such as the Radeon VII and some other workstation-oriented graphics solutions.
But now, the real question is – how much of a difference does HBM make when it comes to gaming?
In the video above, you can see an in-game performance comparison between three of Nvidia’s cutting-edge Titan GPUs that were released over the past few years– the GDDR5X-equipped Titan Xp, the HBM2-equipped Titan V, and the GDDR6-equipped Titan RTX.
As you can tell from the tests, the Titan RTX pulls ahead of the Titan V by about 5-10 FPS in most games, and there is a similar performance gap between the Titan V and its predecessor, the Titan Xp. That said, the differences in performance can mostly be chalked up to the advancements in GPU architecture and have very little to do with the graphics memory itself.
So, why doesn’t the massive bandwidth of HBM2 make a more noticeable impact?
Well, it’s simply because the majority of games in 2020 neither need as much bandwidth nor are they optimized to take full advantage of this technology’s capabilities. For the most part, GPUs that come with HBM memory are intended to be used in workstations and with memory-intensive software, not with games.
Not only that, but HBM memory is also extremely expensive to manufacture compared to GDDR, so combined with the limited benefits that it currently offers in regards to gaming, it’s not difficult to see as to why most modern gaming GPUs use GDDR memory instead of HBM.
So, in summary:
- GDDR6 is the fastest type of specialized GDDR graphics memory and it is used by most new GPUs in 2020. It offers more than twice the bandwidth of GDDR5 though it is only marginally faster than GDDR5X.
- HBM is a type of graphics memory that has much higher bandwidth than GDDR6 but its high manufacturing costs and limited benefits in terms of real-time in-game performance don’t make it particularly appealing for gaming in 2020. Currently, HBM2 and HBM2E are the two latest versions of HBM memory, although HBM2E has not been implemented in any GPUs yet.
That said, there are more important things to keep in mind when picking the right graphics card than the type of memory that it comes with.
As you’ve seen from the article, the GTX 1650 makes for a good test card, as there are two variants available – a GDDR5 and a GDDR6 one, and comparing the two shows that the differences in performance between these two types of memory are fairly minor in most modern games.
Besides, as mentioned before, the only scenario where you’d find yourself having to choose between GDDR5, GDDR5X, and GDDR6 is if you were thinking about going with an older GPU for whatever reason e.g. if you were trying to find a more budget-friendly solution that could potentially offer you better value for your money. In any case, it’s always a good idea to check some benchmarks and comparisons since on-paper specs are never a reliable way to estimate actual performance.
As for HBM2, getting a GPU with HBM2 memory would be a waste of money if you intend on using it solely for gaming, as the majority of HBM2-equipped GPUs are aimed at workstations and come with very hefty price tags, as most such GPUs do.
In 2020, the only GPU that comes with HBM2 memory and is viable for gaming is the Radeon VII, but only if you can get it at a price that’s lower than that of the AMD RX 5700 XT or the Nvidia RTX 2070 Super, as these two GPUs offer comparable in-game performance but usually at a much lower price.
In any case, check out our selection of the best graphics cards of 2020 if you’re currently in the market for a new GPU.