The eternal mystery: How much video memory is enough?

How much video memory should a graphics card have? How much video memory is needed to play games? After so many years in the industry, these questions are still unresolved and controversial. In particular, the video memory capacity of graphics cards of the same grade often varies greatly. The flagship card has as little as 4GB and as much as 12GB.

TechReport recently conducted another test, hoping to try to answer this question through a relatively simple investigation.

The only game they chose was Middle-Earth: Shadow of Mordor, because it is one of the most memory-intensive games currently available, and it can set a higher virtual resolution, downsample it after rendering and output it to the display resolution, similar to NVIDIA DSR and AMD VSR technology.

All high-definition materials for the game were installed during the test, and the image quality was set to the highest.

The game has a built-in benchmark program that can be repeated at different resolutions, including 2560×1440, 3840×2160, 5760×3240, and 7680×4320.

In this test, we will not look at how much video memory is consumed, but rather at the change in frame rate. If it drops abnormally, it means that the video memory has been maxed out.

The first ones tested were the R9 390X 8GB and R9 290X 4GB. The former has double the memory capacity, the frequency is increased from 5GHz to 6GHz, and the core frequency is 50MHz higher.

The R9 290X 4GB is sufficient even at 4K resolution, and the difference with the R9 390X 8GB is minimal. However, it is 4FPS lower at 5760×3240 resolution, and 8FPS lower at 7680×4320 resolution.

Obviously, 4GB of video memory is not enough at ultra-4K resolution, but on the other hand, the frame rate of R9 390X 8GB is only 20FPS, which is not enough to ensure the smoothness of the game. It's like the pot calling the kettle black.

Next, we have the new flagship Fury X 4GB. The video memory here is high-bandwidth HBM, but many players are also worried that 4GB is not enough, and it turns out to be true!

At a resolution of 5760×3240, its speed is already slower than that of R9 390X, and at a resolution of 7680×4320, it is only equivalent to that of R9 290X!

It seems that the first-generation HBM with limited capacity is indeed facing some embarrassment. Let's wait for the second generation next year, when the capacity will be easily doubled, and a single card can reach up to 32GB.

Let's look at the N card. When all four cards are used together, there is really no difference between the GTX 980 Ti 6GB and Titan X 12GB. Even at the highest resolution, they are exactly the same, which is enough to prove that 6+GB video memory is unnecessary.

In fact, 4GB of GTX 980 is enough, and the performance is normal at the highest resolution without any sudden drop in frame rate. It seems that its video memory utilization is more efficient.

The GTX 970 is quite special. Although it also has 4GB of video memory, due to the architectural design, the speeds of the front 3.5GB and the back 0.5GB are different. This has caused a lot of controversy before , and it is indeed prone to trouble. It cannot sustain high resolutions.

Next, comparing the performance of both, we can see that the Fury X 4GB is indeed a bit embarrassing. The 5760×3240 resolution is only equivalent to the GTX 980 4GB, and the 7680×4320 resolution is even worse.

A similar situation occurred in the Far Cry 4 test. The three A cards with 4GB video memory, R9 290X, R9 Fury, and Fury X, all had a large number of high-latency frames, while the R9 390X 8GB had no problem. On the other hand, all N cards had no problem, and the GTX 780 Ti 3GB was safe and sound.

Why is there such a big difference between A and N cards with the same amount of video memory? Even 4GB HBM doesn't work? There is no definite answer, but I highly suspect it has something to do with the way the driver manages the video memory. The efficiency of the catalyst is worrying.

But that's only one aspect; there's also the issue of bandwidth.

In the last test project, we tested compressible black textures and incompressible random textures on each card respectively. The larger the difference between the two, the higher the texture compression efficiency of the graphics card, thus saving more bandwidth and capacity.

The results are quite obvious. The Maxwell architecture GTX 980, GTX 980 Ti, and Titan X have very high compression efficiency, reaching 35-40%, while the Fury X is only 14%, and the R9 290X has no compression at all.

It seems that the Maxwell architecture bandwidth compression technology promoted by NVIDIA is really useful, and it is this that improves the utilization efficiency of video memory.

This article does not intend to compare the advantages and disadvantages of A-cards and N-cards, but only uses some simple examples to see how much video memory capacity is sufficient. At present, N-card players are more comfortable, 4GB is enough, 6GB and 12GB are wasted, just for appearance. For A-card players, 4GB is a bit tight when the resolution exceeds 4K, of course, you can also use two R9 390X 8GB to play.

As a winner of Toutiao's Qingyun Plan and Baijiahao's Bai+ Plan, the 2019 Baidu Digital Author of the Year, the Baijiahao's Most Popular Author in the Technology Field, the 2019 Sogou Technology and Culture Author, and the 2021 Baijiahao Quarterly Influential Creator, he has won many awards, including the 2013 Sohu Best Industry Media Person, the 2015 China New Media Entrepreneurship Competition Beijing Third Place, the 2015 Guangmang Experience Award, the 2015 China New Media Entrepreneurship Competition Finals Third Place, and the 2018 Baidu Dynamic Annual Powerful Celebrity.