Radeon RX 9060 XT PCI Express 3.0, 4.0 & 5.0 Comparison (8GB vs. 16GB)

[Music] Recently, I checked out the 8 GB version of the 960 XT. You know, AMD's famous 1080p esports GPU that is officially

advertised to be neither of those things and instead is deceptively marketed on the back of the 16 GB model. So, you get that one. And as expected in many of the

that one. And as expected in many of the more modern games using settings that would otherwise be fine with enough VRAMm, the 8 GB model was effectively broken or at best much slower than it

should have been. This is bad for a few reasons. Firstly, in the examples we

reasons. Firstly, in the examples we showed, we were unable to unlock the true performance of the GPU because it's bottlenecked by its VRAMm, a thing that realistically should never happen with a

brand new GPU. This becomes of significant concern when you consider the fact that it is very easy to show these issues today using a product that presumably anyone planning on buying

will be keeping it for at least the next 3 to 4 years. But you know what isn't of significant concern? Today's sponsor

significant concern? Today's sponsor spot is brought to you by UG Green and their MagSafe 100 W 2in-1 power bank.

This desktop power bank allows 100 watts of fast charging via one of the three supplied USB outputs, enabling a 100% MacBook charge in just an hour and a half. It also offers a flip up 15watt

half. It also offers a flip up 15watt MagSafe dock for your MagSafe compatible devices and intelligent current matching to ensure more reliable battery safe charging. With the ability to charge

charging. With the ability to charge four devices simultaneously, excellent build quality, and compact design, this power bank is suitable for any and all of your mobile devices. We've been using

UG Green products in the office for a while now. And not only are they

while now. And not only are they reliable, but they also offer great value. So for more information, please

value. So for more information, please check the link in the video description.

So in 3 to 4 years from now, you can expect that games are going to become increasingly demanding using more VRAM than they do today. We've already heard from game developers that 8 GB GPUs have

become a nightmare for them over the past few years, making it extremely difficult to optimize nextG titles to work with such a severe hardware limitation. In almost all instances,

limitation. In almost all instances, they've had to waste a considerable amount of their development time optimizing games to work with 8 GB GPUs.

And yet, reading comments online, it appears that some gamers expect games to continue to evolve with these hardware restrictions in place. And despite all

of this, both AMD and Nvidia are still pushing 8 GB GPUs on the masses. So, the

situation continues to worsen. But the

good news is many gamers are now aware of the problem and are starting to push back. In fact, this VRAM issue is

back. In fact, this VRAM issue is something a lot of gamers have been painfully aware of for quite a long time now. I'd say before most reviewers

now. I'd say before most reviewers cottoned on. And this is because testing

cottoned on. And this is because testing VRAM can be extremely difficult. And in

many examples, a quick 30 to 60 second benchmark pass won't reveal the performance issues many of you face when playing games for extended periods of time. For example, in order to see

time. For example, in order to see performance related issues in Halo Infinite on an 8 GB GPU using the ultra settings at 1080p or 1440p, you generally have to play the game for an

extended period of time as it takes a bit of time to overwhelm the VRAM. That

said, it doesn't take long for textures to go missing, but you do have to be on the lookout for those problems. And again, they don't appear in benchmark graphs. And something else us reviewers

graphs. And something else us reviewers often overlook is the hardware configuration. For testing GPUs, you

configuration. For testing GPUs, you want to use the latest and greatest CPU in order to show the true uncapped graphics performance. And this is

graphics performance. And this is something we always do as it is best practice. However, when it comes to

practice. However, when it comes to VRAM, testing with the fastest CPU on the latest platform with high-speed system memory is without question a best

case scenario for exceeding the VRAM buffer. And I briefly demonstrated this

buffer. And I briefly demonstrated this in my 960 XT 8 GB video by running some tests on the Core i78700 K, a processor that is limited to

PCI Express 3.0 0 and use a slower DDR4 memory. And the results were pretty

memory. And the results were pretty shocking. Therefore, today we're mixing

shocking. Therefore, today we're mixing it up a bit. We're still taking my high-end 9800 X3D test system, but we're going to use it to test PCIe performance. This means this video is

performance. This means this video is isolating the PCIe bus and testing that specifically, giving us a clear idea of how much PCIe bandwidth can influence

performance when dipping into system memory. I had considered using older

memory. I had considered using older CPUs to represent the PCIe 3.0 and 4.0 configurations, but doing so ends up with a lot of moving parts, and I really just wanted to measure the effects of

the PCIe bandwidth. So, I'm sticking with the 9800X3D, but I will be manually adjusting the PCIe mode in the BIOS.

This means I'm able to measure the performance of the 8 GB 960 XT using PCIe 3.0, 4.0, and 5.0, while everything else remains the same. And I'll be

comparing that data to the 16 GB 960 XT using PCIe 3.0. And I'm really doing that to make a point. We could test it with 5.0, but I think testing the 16 GB

960 XT using PCIe 3.0 is more impactful. This data is extremely useful

impactful. This data is extremely useful for those of you without a PCIe 5.0 enabled system, as many CPUs and platforms are still limited to PCI 4.0

and even 3.0. So, let's get into the testing.

Starting with F125, we have the very high preset, the second highest quality preset in the game. Though, we'll also check out FPS performance using the high preset in a moment. Now, what's

interesting to note here is the 16 GB 960 XT averaged 73 FPS in this test with 1% lows of 51 FPS. So, that's decent

performance and certainly very playable.

Now, using PCI 5.0, 0. The 8 GB version was 18% slower, dropping down to 60fps, while we saw a 22% decrease for the 1%

lows. And again, this is comparing the 8

lows. And again, this is comparing the 8 GB model using PCIe 5.0 to the 16 GB model using PCIe 3.0. Now, if we downgrade the 8 GB card

3.0. Now, if we downgrade the 8 GB card to PCIe 4.0, performance tumbles, dropping by 30% to just

42FPS. Then, if we use PCIe 3.0,

42FPS. Then, if we use PCIe 3.0, the same mode used by the 16 GB model in our test. We're looking at just 35 FPS

our test. We're looking at just 35 FPS on average, which is a 17% performance loss when compared to the PCA 4.0

configuration and 42% slower than PCA 5.0. And worse still, when using PCA

5.0. And worse still, when using PCA 3.0, the 8 GB 96xt is 53% slower than the 16 GB model using the same PC bus

interface. Or put another way, the 16 GB

interface. Or put another way, the 16 GB model is 111% faster. Okay, so here's a look at F125

faster. Okay, so here's a look at F125 using the high preset again at 1440p with quality upscaling. The 16 GB 960 XT

is good for 190 fps on average in our test with 1% lows of 156 FPS. And this

performance was achieved while in the PCI Express 3.0 mode. In comparison, the 8 GB model was 11% slower using PCIe

5.0. However, when switching to PCIe

5.0. However, when switching to PCIe 4.0, we see a further 12% performance downgrade to 149 FPS on average and then another 13% performance loss when using

PCA 3.0. This means when both the 8 and

PCA 3.0. This means when both the 8 and 16 GB models are using PCA 3.0, The 16 GB version provided almost 50% greater average frame rate performance

while the 1% lows were boosted by a massive 63%. And again, this is with the

massive 63%. And again, this is with the high preset at 1440p with quality upscaling. So, very reasonable quality

upscaling. So, very reasonable quality settings. This is three presets off the

settings. This is three presets off the highest preset. And this means if you

highest preset. And this means if you have an older PC, you absolutely need to make sure you get the 16 GB card.

Moving on to Monster Hunter Wilds. This

game doesn't work on 8 GB graphics cards using the ultra preset at 1440p with quality upscaling. The 16 GB card was

quality upscaling. The 16 GB card was good for over 60 fps on average using PCI 3.0. Twice the performance we see

PCI 3.0. Twice the performance we see from the 8 GB card when using the same PCIe mode. But even with PCI 5.0, the 8

PCIe mode. But even with PCI 5.0, the 8 GB model struggles with terrible frame time performance resulting in 1% lows of 2 FPS. And what's interesting to note

2 FPS. And what's interesting to note here is that the increased PCIe bandwidth did boost the average frame rate of the 8 GB model, but it also reduced the 1% lows in the process,

creating more stuttering. It appears as though this is a consequence of rendering more frames overall while still running out of VRAM. The takeaway

here being that PCI bandwidth can't solve the issues of the 8 GB 960 XT in this title, even when it has all 16 lanes available.

Spider-Man 2 plays exceptionally well at 1440p with upscaling on the 16 GB 960 XT. We saw 112 FPS on average in our

XT. We saw 112 FPS on average in our test with 66 FPS for the 1% lows. That

made it 26% faster than the 8 GB model when using PCI 5.0 or 38% faster when looking at the 1% lows. However, if we

switch the 8 GB model to PCIe 4.0, 0 we see a massive 20% decline in performance and then an even bigger 27% performance loss when dropping down to PCI 3.0 as

the average frame rate dropped to just 52 fps. This means when playing

52 fps. This means when playing Spider-Man 2 in a PCI Express 3.0 O system using the settings shown here.

The 16 GB version of the 960 XT will deliver at least 115% greater average frame rate performance and 175% greater 1% lows.

But even if you have a PCI 4.0 system, the 16 GB model will still be at least 58% faster for the average frame rate and 74% faster for the 1%

lows. Moving on to Ratchet and Clank

lows. Moving on to Ratchet and Clank Rift Apart. In this test, we're again at

Rift Apart. In this test, we're again at 1440p with quality upscaling using the very high preset. The 16 GB 960 XT using

PCI 3.0 was able to average 128 FPS with a 1% low of 97 FPS. So, excellent

performance overall. When compared to the 8 GB version under the same conditions using PCI 3.0, know the 16 GB model was 103% faster when comparing the

average frame rate and 155% faster for the 1% lows. Upgrading to PCI 4.0 improved the 8 GB model's performance by

30%. But even so, the 16 GB card was

30%. But even so, the 16 GB card was still 56% faster. Then when going from PCI 4.0 to 5.0, The 8 GB model saw a

further 32% performance boost and now it is very usable despite the 16 GB model still offering around 20% better performance. Next up, we have Indiana

performance. Next up, we have Indiana Jones and the Great Circle. And here the 16 GB 960 XT using PCI Express 3.0 was able to

render 135 FPS on average with a 1% low of 113 FPS. So again, very good performance overall.

Now, under the same conditions, the 8 GB model, again using PCI 3.0, was only able to average

45fps, making the 16 GB card 200% faster and 223% faster when comparing the 1% lows. Increasing the PCI bandwidth with

lows. Increasing the PCI bandwidth with PCI 4.0 improved the average frame rate of the 8 GB model by a massive 44% allowing for 65 fps on average, which is

still less than half the performance seen from the 16 GB model. Then with PCI 5.0, we saw a further 42% performance increase and now the overall performance

was good. Even so, the 16 GB card was

was good. Even so, the 16 GB card was still 47% faster. And again, it's able to provide this performance boost while using PCI Express 3.0. Dragon Age: The Valgard doesn't

3.0. Dragon Age: The Valgard doesn't play well on 8 GB graphics cards using the ultra preset at 1440p, even with the help of upscaling. And regardless of the PCI Express mode used, performance was

bad. When using PCI 3.0, for example,

bad. When using PCI 3.0, for example, the 16 GB model was 174% faster, then 70% faster when switching the 8 GB model

to PCI 4.0, and finally 31% faster when using PCI 5.0. But the 1% lows are still a disaster on the 8 GB model regardless

of the PCIe mode used. Now, wrapping up this testing, I installed these GPUs in my Ryzen 7 5800X 3D test system, which is limited to PCI Express 4.0 as that

was the highest spec interface the AM4 platform supported. What's interesting

platform supported. What's interesting to note here is that the 16 GB version of the 960 XT delivered virtually identical performance using either the

5800X 3D or 9800 X3D as the results are limited by the GPU's compute performance and nothing else. So using either of these processors, you can unlock the

full potential of the 960 XT, at least using these quality settings. The 8 GB model though, as usual, is more of a

mixed bag. Using PCI 4.0 with the

mixed bag. Using PCI 4.0 with the 9800X3D, the 16 GB model was already 27% faster than the 8 GB version. But when

we move to the 5800 X3D, it ends up delivering 44% more performance. This is

because we saw a further 11% performance regression on the AM4 platform, and this is likely due to the use of slower DDR4 3600 memory.

However, it is worth noting that this issue isn't always seen. Take a Ratchet and Clank Rift Apart for example. Here,

the 8 GB 960 XT using PCI 4.0 was much the same when paired with either the 5800 X3D or 9800 X3D. Though, the 1% lows, which were already bad on the 9800

X3D, were even worse when using the 5800 X3D. So, the takeaway here being that at

X3D. So, the takeaway here being that at best you will see the results provided by the 9800 X3D when using an older, slower CPU with PCI Express 3.0 or 4.0,

but more often than not, the margin between the 8 GB and 16 GB models will be even greater. So, there you have it.

As we saw previously with the Core i78700 K, you absolutely do not want to run out of VRAM on a PCI Express 3.0 system, even when all 16 lanes are

available. Now, those of you with PCIe

available. Now, those of you with PCIe 4.0 systems, you will fare a lot better.

But even as we saw with PCIe 5.0 enabled, you can't always avoid serious performance hits. In virtually all the

performance hits. In virtually all the games tested, the 8 GB 960 XT was unusable when limited to PCI 3.0. F125

using the high preset was really the only exception here, though performance overall was still significantly down on the 16 GB card. And this all points to the fact that performance in general,

even with more modest quality settings, can be much lower on these older platforms, as you only have to be exceeding the local video memory capacity ever so slightly to see serious

performance issues. Issues that would

performance issues. Issues that would largely go unnoticed on more modern platforms. Another problem to consider is the fact that many motherboards, especially the more budget focused

boards, will reduce the PCIe lane count to the primary slot when the secondary slot is in use. So if you install quite literally anything in the secondary PCIe

x6 slot, both the primary and secondary PCIe x16 slots will run in a time 8 x 8 configuration and this is extremely

common. Doing so will have the available

common. Doing so will have the available PCIe bandwidth to the graphics card. For

example, in a PCI 4.0 system, when all 16 lanes are available, there is about 64 GB of total bandwidth. But with just eight lanes, that configuration gets

halfved to 32 GB, which is the same amount of bandwidth seen when using PCA 3.0* 16. Essentially, what all of this means

16. Essentially, what all of this means is you never want PCI bandwidth to be a performance limiting factor. You don't

want it to have a chance to influence performance. And the only way to ensure

performance. And the only way to ensure this is to have enough VRAM. And at this point in time, 16 GB does look to be enough VRAM today and likely into the future, at least the realistic lifespan

of these products. But clearly, 8 GB just isn't enough as we're able to demonstrate that today. Still, having

the full 16 lanes available does help the 8 GB 960 XT quite a bit. So, I'll be redoing all of this testing next week, but with the 5060 Ti 8 GB to see how

that model performs when limited to PCI 3.0 and 4.0, as that model does only have eight lanes available. There is no 16 lane version. So, that should make for a pretty interesting comparison. In

the meantime, if you appreciate all the testing that went into this video, then please do give it a like, subscribe for more content because, like I said, there's more of this coming up. And of

course, we have other nonVRAM related content coming up on the channel. You

will see some of that next week as well.

But yeah, you can also join. Patreon's a

thing if you want to get more access to hardware box goodness. We have a Discord server, monthly live streams, Q&A stuff, and behind the scenes content. So check

that out if you're interested. But if

not, that's perfectly fine. And I would like to thank you for watching this video. I'm your host, Steve. See you

video. I'm your host, Steve. See you

again next time.

[Music] [Music]