PCIe 4.Zero vs. PCIe 3.Zero GPU Benchmark

Recently we talked about PCI Express 4.0 and its role in the new GeForce 30 series. We suspected that high-end parts like RTX 3080 and 3090 might have a small performance advantage when using PCIe 4.0 over 3.0, which is an increase of ~ 5%. We took a very quick look at this in our evaluation on the first day and found that it was at most a 3% increase, which honestly gives very little.

Even so, we had planned to take a closer look at PCIe performance with the RTX 3080. That's why we're going to do this today. We'll be looking at many different PCI Express configurations with the Ryzen 9 3950X on the X570 platform with the X570 Unify motherboard from MSI, and the Intel Core i9-10900K on the Z490 platform with the Z490 Unify from MSI. In both cases we are using 32 GB DDR4-3200 CL14 memory. The graphics card of choice is the RTX 3080 Founders Edition.

For all practical purposes, let's skip the conversation about the PCI Express bus, the bandwidth available in each mode, or something like that as we discussed this before. We'll assume you already know the details and we're mostly concerned about the numbers, so let's get started.

Ryzen 9 3950X PCI 3.0 vs PCIe 4.0 benchmarks

Starting with Death Stranding, we don't see any real performance difference between PCIe 4.0 and 3.0 when working at full bandwidth in x16 mode. In the unlikely scenario that you are running PCIe 3.0 x8 on an X570 or B550 motherboard, there is a performance dip, especially the 1% low performance where the frame rate drops by up to 11% at 1440p normally we see rather a drop in performance of 7%.

While Shadow of the Tomb Raider is a very visually impressive game, it doesn't consume a lot of PCIe bandwidth. Hence, we see virtually no change in performance even if we switch from PCIe 3.0 x8 bandwidth to PCIe 4.0 x16 bandwidth.

It is similar when testing with Rainbow Six Siege. When changing the PCIe version from 4.0 to 3.0, the performance drops by 2.5% at 1080p and by a further 2% when switching to x8 bandwidth. Really nothing to see here.

F1 2020 is another title that is not noticeably affected by PCI Express bandwidth. Only when we use PCIe 3.0 x8 do we notice a loss of performance at 1440p and 4K. Interestingly, again, it's 1440p that gets the most success, but we're only talking about a 6% drop for this unlikely configuration.

Horizon Zero Dawn has proven to be more PCIe sensitive than most of the other titles we tested, but even here we hardly see any difference between PCIe 3.0 and 4.0. In fact, we only noticed a reduction in the frame rate with PCIe 3.0 x8.

We know that Resident Evil 3 isn't CPU sensitive at all, at least within the bounds of reason, and doesn't appear very PCIe bandwidth sensitive either. We see practically no difference between PCIe 3.0 and 4.0.

The results of Doom Eternal are interesting as we get very high frame rates in this title. Often the greatest differences in performance occur at higher frame rates. This is because the bus transfers per picture are fairly consistent regardless of resolution. This means that the frame rate is the main driver of PCIe bandwidth.

The threshold for PCIe 3.0 x8 in this title seems to be a little over 300 fps, while x16 allows almost 360 fps and PCIe 4.0 x16 doesn't limit the 3950X and allows an average of 368 fps.

Wolfenstein: Youngblood is similar to Doom. While we are seeing a performance degradation with PCIe 3.0 x16, the margins are meaningless.

For those wondering if enabling a technology like DLSS doesn't change a lot, we're seeing margins similar to native performance.

We've also seen people claim that ray tracing uses a lot more PCI Express bandwidth. We're not sure what this is based on, but here are the DLSS + ray tracing numbers. If anything, this reduces the need for PCIe bandwidth as it lowers the frame rate.

Intel Core i9-10900K PCI 3.0 vs PCIe 4.0 benchmarks

Here is a brief overview of the RTX 3080's performance with the Z490 / 10900K combination with PCI Express 3.0 x16 and x8 bandwidth. Here we see a 5% reduction in average frame rate at 1080p and a 1% reduction in low power by 10%.

Even at 4K, we see a 6% decrease in average frame rate with a 10% decrease for the low 1% performance. So this is pretty significant, and a relevant statistic for some users that I'll get to in a moment.

As we saw with the Shadow of the Tomb Raider, this title doesn't require a lot of PCIe bandwidth, and we see again that if the PCIe 3.0 bandwidth is halved, performance drops by no more than 5%.

Rainbow Six Siege's PCIe bandwidth doesn't matter here, especially when playing 1440p and 4K games.

Halving the available PCIe bandwidth from PCIe 3.0 x16 to x8 results in a small performance penalty, down to a drop of 6% at 1080p, 5% at 1440p, and nothing at 4K.

We see similar performance margins at Horizon Zero Dawn. Here, reducing the PCIe 3.0 bandwidth has reduced performance by up to 5% at 1080p and 4% at 1440p without changing the 4K data.

In Resident Evil 3, we see a 5% drop in performance at 1080p with no significant change at 1440p and 4K.

Once again interesting results in Doom Eternal. With the 3950X, we previously found that the PCIe 3.0 x8 bandwidth limits the RTX 3080 to around 320 fps at 1080p, and we see the same thing with the 10900K. As a result, we see a massive 18% reduction in performance at 1080p, but little change in the 1440p and 4K data.

PCIe device discussion

Granted, much of this data wasn't all that surprising, but it's still good to double-check. Now there are a number of results that we would like to look at. These are often overlooked in these PCI Express tests for graphics cards. The PCIe 3.0 x8 data for the AMD system is a bit pointless to look at. They are only made in the name of science, although they may be useful information for users using a 300 or 400 series motherboard with multiple PCIe devices.

This information is much more relevant to Intel users as the PCIe 3.0 x8 bandwidth for your primary PCIe slot cannot be avoided if you ever want to use the secondary slot. We know a lot of people leave the second PCIe x16 slot empty. So this does not affect everyone or even the majority of Intel users. To find out how many of you are using a second or even a third PCIe device, we conducted a survey.

The vast majority of users install a graphics card in their system and that's it – no other expansion slots are used. About a quarter of users said they would install a second device, be it a sound card, additional storage, capture card, high-speed network, or other expansion card. It could be a PCIe card with a single USB connector, but when installed in the secondary PCIe x16 slot it cuts the bandwidth of the primary slot in half and forces you to use PCI 3.0 x8 on a flagship Z490 Motherboard to work. for example.

If you do this on a B550 or X570 card, you are limited to PCIe 4.0 x8 bandwidth when using the RTX 3080. This limits you to the PCIe 3.0 x16 bandwidth, so there is no performance penalty there. As we just said, this is not what most should be concerned about and we don't want to create an unrealistic scenario to limit performance on Intel systems. However, we are only going to show you what happens in this scenario for the smaller part of users who occupy this second PCIe slot.

Starting with Death Stranding, when both CPUs run in an x8 x x8 configuration for the primary and secondary PCIe slots, we see comparable performance on both platforms with slightly better 1% lows for the 3950X. At 4K, the 3950X won by a 3% lead, a meaningless difference, although it was consistently a few frames faster in these conditions.

AMD already had a small performance advantage in Shadow of the Tomb Raider. By limiting the 10900K by using PCIe 3.0 x8, the 3950X can deliver 6% more performance at 1440p and 5% more at 4K.

Intel enjoys a performance advantage in Tom Clancy's Rainbow Six Siege when using x16 mode, and that advantage is not completely negated by the limited x8 bandwidth. The 10900K still allows 8% more frames at 1080p and 3% more at 1440p, so the low performance of the 10900K saw when transitioning from x16 to x8 bandwidth isn't a big deal here.

Before that, the 10900K was 13% faster than the 3950X at 1080p in F1 2020, and here we see only a minor change in that margin, which has been reduced to 11%. The margin is completely eliminated at 1440p and 4K.

Due to a severe CPU bottleneck at 1080p in Horizon Zero Dawn with the Ryzen 9 3950X, reducing the PCIe bandwidth of the 10900K still offers 13% more performance, while at full x16 bandwidth it was 19% faster. That said, we're seeing identical performance at 1440p and 4K.

The Resident Evil 3 results do not change here, as playing with the PCIe 3.0 x18 bandwidth has no performance degradation and also has the same level of performance with the 3950X or 10900K.

The Doom Eternal results are interesting, at least the 1080p data. Unlimited by the PCIe bandwidth, we've seen before that the 10900K with the RTX 3080 is 6% faster than the 3950X at 1080p. With a second PCIe device installed, it's now 10% slower, giving AMD the win. Performance is the same at 1440p and 4K as before.

What we learned

In short, PCIe 4.0 does little to improve performance with the RTX 3080 right now. It is possible that this will change with future games, but for now this is not an issue.

The data we used when using the secondary PCIe x16 slot is interesting, even if only a portion of the users will. Ultimately, AMD fans cannot use this weapon as a weapon to gain all-important arguments on forums or Reddit. This is not a configuration that most users will be using and the performance degradation is not very noticeable.

From a different perspective, AMD fans can't have it both ways. Single digit margins can't be significant if they favor your preferred brand of hardware. It's even questionable how relevant the 3950X's 11% gain in Doom Eternal at 1080p is, given that it's 321 fps versus 357 fps.

If you want to get the most out of your PC's expansion options, use a platform that supports PCI Express 4.0 to avoid bandwidth bottlenecks. As we verified a few months ago, PCIe 4.0 allows the latest and greatest SSDs to reach amazing raw speeds, and the second generation of these drives is expected to arrive by the end of the year.

Purchasing links:
  • Nvidia GeForce RTX 3080 on Amazon
  • AMD Ryzen 9 3950X on Amazon
  • AMD Ryzen 9 3900X on Amazon
  • MSI X570 Unify motherboard on Amazon
  • Intel Core i9-10900K on Amazon
  • MSI Z490 Unify motherboard on Amazon

Leave a Reply

Your email address will not be published. Required fields are marked *