Every time you change generations of the PCI-express interface, there is a lot of controversy that video cards that use the new version will not start on old motherboards, or will work much slower. At the disposal of the author was again the GeForce RTX 3080 version of Gigabyte Gaming. It, like all RTX 3000 representatives, supports the current version of the PCI-e 4.0 interface, and due to its performance and maintaining stable GPU frequencies under load, it can serve as a good marker for testing PCI-e 3.0 and PCI-e 4.0 interfaces.
Let’s see how many frames per second we lose if we use a PCI-e 3.0 card.
Option of the article for those who are too lazy to read
Of course, in the current situation with a shortage of video cards, the number of such stories has also become much smaller: but the Internet remembers how in more well-fed years, authoritative advice was often heard to change the platform to a new one for installing the GeForce GTX 750 Ti. “Put the card on a motherboard with PCI-e 2.0 – it will work 20-25% slower!” – then all sorts of experts lamented.
Earlier, the difference between the two current versions of the PCI-e interface was already considered using the example of the Radeon RX 5500 XT in versions with 8 and 4 GB of memory on board. But since then, the author has been pestered by curiosity: what will happen if you take a much more productive video card? Will the difference manifest itself in more tangible values, will it actually be two-fold, or at least close to it?
Test stand and testing methodology
The configuration of the test bench is already familiar to regular readers.
CPU: AMD Ryzen 9 3900X
CPU cooling system: ID-Cooling SE-207;
Thermal interface: Arctic MX-2;
Motherboard: Gigabyte X570 Aorus Master (BIOS F33F);
RAM: Kingston HX437C19FB3K2 / 32, 2x16gb;
Disk subsystem: SSD Gigabyte GP-ASM2NE6200TTTD + HDD Western Digital WD10EZRX-00A8LB0;
Case: Corsair Carbide 270R
Power supply: Cougar GX-F 750.
The frequency of the Ryzen 9 3900X processor during the tests was fixed at around 4200 MHz. The RAM was overclocked to a frequency of 3800 MHz with timings that can still be found in her review . The video card worked normally, no modifications were made on it.
During the tests, all background programs were turned off, except, of course, the launchers needed to run games.
Switching between PCI-e 4.0 and PCI-e 3.0
Last time, the choice of the AM4 platform caused wild delight among the commentators: “UUUU, tests on the X570, everything is clear!” …
But in the real world, the test platform allows you to manually set the operating mode of the PCI-e slots. And – separately for the x16 slots and other slots for expansion cards.
And this is by no means a fiction. Having installed the third generation of the interface for slots with 16 lines, using the same GPU-Z, you can see that the video card uses the interface version 3.0 under load, although both it and the test board still support PCI-e 4.0.
In idle mode, like all modern video cards, the interface version will decrease to 1.1, but under load it will not be higher than the one that was manually assigned.
Of course, the differences are not only in the readings of one graph in GPU-Z.
So, if you run the corresponding test from the 3Dmark package, you will notice that the interface bandwidth of version 3.0 for the GeForce RTX 3080 is only 13.07 gigabytes per second, and the number of fps in the test scene is only slightly different.
And if you switch to version 4.0, the peak bandwidth will increase to 26.29 gigabytes, and the frames per second will be noticeably higher.
That is, the theoretical difference in the interface bandwidth for the RTX 3080 is literally twofold – as, indeed, it was with the RX 5500 XT a year earlier.
But as with the RX 5500 XT, keep in mind that this is the difference in peak bandwidth that is speculative.
And the real meaning when we talk about video cards is the performance of the devices themselves in synthetics and real games.
So what’s the difference?
On the third version of the interface in 3Dmark Fire Strike, the test system knocks out 30,526 total points, and 42,268 points for graphics. Switch to the fourth version of the interface – and we get 30,696 overall points in 3Dmark and 42,719 points for graphics in the same test.
The result in Unigine Superposotion on PCI-e 3.0 is 21,894 points. If we return to the current version of the interface, the result in this test will increase to 21,969 points.
There is a difference and is observed in each of the three indicators – that is, it cannot be said that the matter is in the measurement error, the interface bandwidth also contributes.
However, for some reason, this difference is not only not twofold, but even a memetic 20-25% does not pull in any way.
But maybe these very “20 percent” will show games? Especially if you set the maximum or close to them graphics settings, and take measurements not only in FullHD, but also in QuadHD?
Disputes about the alleged incompatibility or the loss of a significant part of the performance of video cards when they are installed on a platform with an earlier version of the PCI-e interface is already an integral part of the story.
So it was during the transition from the first version to the second, from the second to the third, from the third to the fourth. Rest assured that with the advent of motherboards and video cards using PCI-e 5.0, absolutely nothing will change in the universe of comments.
No, no one here is going to argue with the fact that the theoretical bandwidth of the PCI-e interface increases with each generation. It is simply stupid to deny this, given that this is clearly indicated in the passport characteristics of the interface and is confirmed in practice, at least by synthetic tests.
But it must be borne in mind that this is the difference in the maximum throughput, that is, in the amount of data that, in principle, can be transmitted via this interface per unit of time.
But whether this characteristic will really affect the performance of a video card depends on how much data it actually operates with.
In this article, you could see an example of the GeForce RTX 3080. This card is far from top-end, but it is still very fast and most other models that support PCI-e 4.0 will definitely yield to it in performance.
And yet – even for the RTX 3080, the real difference between the two interfaces is not only not twofold, but even not even 20%, and even rarely where it will pick up 5%. And in many cases it will even be 1-2 frames. Which, given the general level of performance, looks more like a measurement error than an effect of the interface bandwidth.
Therefore, the main question that should be asked in this context is: if the difference is so insignificant for such a powerful card, why should it be larger for less efficient models?
And the answer is simple: she won’t be there anymore.
Of course, all this does not apply to those cases when a slower processor and RAM is also installed on a platform with an old version of the interface. For example, if we are talking about the Intel LGA 1150 platform , where the PCI-e 3.0 interface is adjacent to quad-core processors and DDR3 memory.
Or, if you like, when comparing systems on the Socket AM4 platform, one of which supports both Ryzen 5000 and PCI-e 4.0 processors, and the second is forced to be limited to models of the Ryzen 3000 and PCI-e 3.0 family.
In these cases, of course, the difference will be much more noticeable.
Only it will be caused not by the version of the PCI-e interface, but by much more obvious reasons. And in this case, we will have to talk not about the difference due to the PCI-e standard, but about the difference in the overall system performance.