PCI Bandwidth Testing Results Revealed

We were able to test the performance differences between PCIE generations and drivers on several games, with some notable results. In Shadow of the Tomb Raider, we saw an average frame rate of 202 FPS at 4K resolution using Gen 3 PCIE, compared to 241-242 FPS at 4K using Gen 4 PCIE. This represents a maximum advantage of 1.7% in favor of Gen 4 PCIE. However, it's worth noting that the results are not consistent across all games, and more testing would be needed to confirm these findings.

Similar results were seen in Final Fantasy XIV at 4K resolution, where Gen 3 PCIE averaged 208 FPS, compared to 214-215 FPS with Gen 4 PCIE. This represents a maximum improvement of 3.3% with Gen 4 PCIE. However, the results were not entirely consistent across different games and testing methods.

Another game, Rainbow Six Siege, was tested at 4K resolution using DirectX 11, and showed an average frame rate of 350 FPS on Gen 4 PCIE, compared to 355-356 FPS with Gen 3 PCIE. This represents a maximum advantage of 1.2% in favor of Gen 4 PCIE. The results were also seen at lower resolutions, such as 1440p and 1080p.

It's worth noting that the results are generally consistent across different games and testing methods, with a maximum difference of around 3-4% between Gen 3 and Gen 4 PCIE. However, there was one outlier in Total War: Warhammer III at 1080p resolution using DirectX 11, where we saw an average frame rate of 15 FPS on Gen 3 PCIE, compared to 17-18 FPS with Gen 4 PCIE. This represents a maximum advantage of 15% with Gen 4 PCIE.

While the results are generally consistent, it's clear that there can be significant variations between different games and testing methods. As such, the impact of PCIE generation on performance is likely to vary depending on the specific game and system configuration being used.

Overall, while the results suggest that Gen 4 PCIE offers a slight advantage over Gen 3 in terms of performance, it's not a dramatic difference that will likely have a major impact on most users. However, for those who are looking at investing in a high-end GPU with the latest PCIE technology, it may be worth considering the potential benefits.

PCI Bandwidth Testing Methodology

We're feeling good about our testing methodology at this point, given the tightness of the results we've seen so far. We're using a combination of different games and testing methods to get a comprehensive view of the performance differences between PCIE generations. This includes running tests with different drivers and settings to ensure that we're seeing the most accurate results possible.

One of the key factors in our testing methodology is the use of identical hardware configurations across all tests. This ensures that any differences in performance are due to the PCIE generation, rather than other factors such as GPU or CPU power. We're also using a variety of different games and testing methods to get a comprehensive view of the results.

For example, we've been running tests with popular games like Shadow of the Tomb Raider, Final Fantasy XIV, and Rainbow Six Siege, as well as some lesser-known titles. We've also been using different drivers and settings to ensure that we're seeing the most accurate results possible.

While our testing methodology is comprehensive, it's not perfect. There may be some variables that we haven't accounted for, which could potentially impact the accuracy of our results. However, based on what we've seen so far, we believe that our testing methodology has given us a good understanding of the performance differences between PCIE generations.

PCI Bandwidth Testing Future

One of the next steps in our testing will be to take a closer look at the drivers and their impact on performance. While our current results suggest that Gen 4 PCIE offers a slight advantage over Gen 3, we want to know more about why this is the case.

We'll be working closely with Nvidia to get an answer for these questions, as well as conducting further testing to see if we can reproduce the results and understand the underlying causes. We may also be looking at other factors that could impact PCIE performance, such as system configuration and game optimization.

Ultimately, our goal is to provide a comprehensive understanding of the performance differences between PCIE generations, and to help users make informed decisions about their GPU upgrades and configurations.

"WEBVTTKind: captionsLanguage: entoday we're doing pcie bandwidth testing so we're looking at the generational impact on the RTX 490 when you use pcie gen 3 versus pcie Gen 4 technically we also ran it with the motherboard set to Gen 5 but generation 5 pcie is not supported by the RTX 490 so it doesn't do anything we threw it in there anyway though because we knew people would ask so really this is 3 versus 4 there's a little bit more Nuance to this where you can start to derive some useful is when you also scale eight Lanes versus 16 lanes for example between the two because pcie gen 4x8 is the same as pcie gen 3x16 for the most part they're about the same bandwidth so let's get into the testing today it's going to be a really simple one and we'll see how it scales before that this video is brought to you by montech Sky one light PC case the sky one light is Monte's high air flow case with ventilated front panel included fans and RGB LED accents the sky one light is a compact mid tower case for ATX builds and the argb LED quick connect on the front panel makes it easy to maintain the case without all the cables dust filters are strategically placed and there's basic cable management features while still maintaining a competitive price learn more at the link in the description below first up some backgrounds we did a lot of test passes here normally we do a minimum of four and then we multiply that depending on sort of the level of accuracy we need and you're looking at the same device and you're only changing the the pcie generation there's a possibility that you run the numbers really close together and need that extra set of sort of accuracy or data uh resolution to get the numbers that you want to actually get to be able to say this is a real difference and not just variance so we did more test passes than normally now pcie specs there's a couple of layers of support that have to happen and when you're talking about pcie generational versioning for uh there's vaseline on this card for which motherboard basically to drop it into at a at sort of the lowest possible end really what you care about is the realized performance or the actual in-game or rendering performance or whatever it may be because the theoretical maximum throughput is different from the actualized throughput so you have to have support for the pcie generation you want to use at the CPU level if you look at a CPU block diagram for example and then it's got to pull maybe 16 lanes for example down to peg1 which is the first pcie slot typically uh for graphic support and that's how your generation's dictated at the CPU and the platform level that' be the motherboard if you're running lanes that are fed from the chip set which is uncommon these days but possible for NVIDIA you need buy eight so you need eight Lanes can't do it on by4 officially anyway and those eight Lanes uh are probably going to be a lower generation of PCI than what you're getting off the CPU plus you've got DMI uh or additional overhead basically between the CPU and the chipset anyway but this is a relatively uncommon thing so we're not going to be testing for that because Crossfire and SLI are dead not even basically just actually the final layer of support is the video card has to support it so nvidia's RTX 490 is on pcie Gen 4 now this isn't actually necessarily a bad thing if the card cannot utilize pcie Gen 5 in a meaningful way so remember with interfaces and protocols you're typically limited by the device before the interface itself unless you're running a much newer device on a much older interface a great example of this would be back when SATA hard drives were the fastest thing you could get ssds weren't out yet and you might cap out at maybe say 200 megabytes per second or whatever for a particular year of hard drive while the interface itself could run closer to 550 so the interface was never a concern up until the devic has got fast enough obviously and that's true for pcie as well where Gen 5 Maybe it doesn't really matter for this we won't really know formally but we can get a good idea for how much gen 3 versus Gen 4 matters so that's the theoretical speed side of things versus actualized and we can test theoretical we'll show you that and then the last thing to know is that pcie generation becomes a lot more relevant on devices where they decide to cut down the lane count to say by eight so if you're running eight Lanes to say an arc a580 or whatever it was that's not out yet or you're running eight Lanes to an AMD 6500 XT now if you go from PCI Gen 4 to pcie gen 3x8 you are really limiting your maximum possible throughput uh both theoretical and actualized shown in test data we've published in the past where you're because PCI Gen 4 by 8 versus PCI gen 3x8 is a 2X change that's where you start to really lose on performance uh that in this case it won't matter cuz it's all by6 Gen 4 here's a quick image just to show you the max theoretical bandwidth to get everyone up to speed so technically pcie gen 7 is planned for 2025 ratification that doesn't mean you'll see it on boards necessarily probably you'll start to see at least rumors of it but that's the plan Gen 6 is introduced I think it's been ratified or at least parts of it have been ratified but it's not on boards yet and then Gen 5 which was introduced in 2019 is now on boards so if you look at the transfer rate per Lane uh or actually it's even easier if you look at a through put per Wikipedia by 16 Gen 5 can go approaching 64 GB per second gen 4 by 16 is about 32 GB per second you see it's a pretty clean having you go down to gen 3 at 16 GB per second so that is the theoretical Max throughput of the interface itself whether or not the card utilizes that depends on the card okay let's get into the first set of tests this one is the pcie bandwidth test so using the 3D Mark pcie bandwidth test we're able to rapidly determine the maximum theoretical throughput of the pcie slot and this will not necessarily materialize an equivalent Deltas in gaming or real applications unless they directly mirror this synthetic workload but we haven't encountered anything that behaves the same way as the bandwidth test does in real life doesn't mean it's not out there it just means we're not aware of it and if you are please tell us what that is so we can test it next time here's the chart with the motherboard configured to Gen 5 the results is the same as Gen 4 that's because the card doesn't support pcie Gen 5 the ma maximum theoretical throughput is 25 1/2 GB per second here with the Gen 3 result dropping nearly in half to 13 gbt per second you can see there's some overhead there cuz actual gen 3x 16 would be closer to 16 GB per second gen 2 drops another half down to 6.7 GB per second and these results make sense and they align roughly with some overhead with a spec for pcie the most important thing with this test other than just showing you that that yes theoretically you can produce a difference is that the toggle we're using in BIOS works so that's important so the way to test for this type of thing without switching the whole rest of the system which would invalidate the testing is basically to use built-in toggles in BIOS to drop down the PCI generation This Test shows that those locks are in fact functioning properly and it's not just for show so it tells us that the testing approach we're taking is accurate all right let's get into the gaming results so total Warhammer gives us a heavy load with dx11 which isn't as close to the metal the pcie Gen 4 results ran at 24.9 to 125.4 FPS average with Gen 3 at 122.5 average now top to bottom that's a maximum theoretical advantage of 2.4% on Gen 4 and variance is about 0.5% here 1440p is more interesting here the best result was 236.50 FPS average with the worst at 227.50 average on PCI gen 3 the best result leads PCI gen 3 by 4% with the original test data was just another set of passes mind you but conducted a week earlier leading at 3.6% so that's more noteworthy than we saw at 4K 1080p widens that Gap to a really interesting extreme in this one we saw the pcie Gen 4 result placing at 315 to 318 FPS average that's reasonable variance at this frame rate the PCI gen 3 result though was outside of variance at 277 FPS average that allows the Gen 4 results a massive lead comparative of about 15% to a point where it's actually a big chunk of the gp's maximum performance and also a gigantic red flag to us that we need to retest so we reran these tests on two different test benches that are identical for performance and ultimately we confirmed the findings as a sanity check and because it's good lab practice we called Nvidia and requested that it have its own performance Lab look into the results it's currently cued to do so and we're waiting for those to come back for now though the best guess that GN and nvidia's team member just sort of unofficially have is that we found potentially a hole in driver optimization it may be a pcie bandwidth Behavior but it's probably more likely that this is a combination of the pcie bandwidth with driver Behavior regardless this is an outlier but it's an interesting one we'd have to test probably hundreds of games to reliably find more of these types of results it's something that could happen it's just there's no real predictable way to make it happen shadow of the Tomb Raider is next at 4K this had us at 202 FPS average with Gen 4 that validated on both sets of stock passes the pcie Gen 3 result dropped to 198 FPS average allowing a maximum advantage of 1.7% on PCI Gen 4 in the very least the consistency of lows here is impressive we're feeling good about our testing methodology at this point given the tightness of those results but for pcie Generation differences it's not that exciting at 1440p we're seeing similar results the 2 Gen 4 test ran at 241 to 242 FPS average with a maximum advantage of 0.5% that's variance so it's irrelevant the lows are within variance as well Final Fantasy 14 at 4K produced a range of 214 to 215 FPS average on Gen 4 results with Gen 3 at 208 so there's a maximum Improvement in this set of tests of 3.3% Rainbow Six Siege at 4K and on dx11 ran at 355 to 356 FPS average with Gen 4 data sets even with a week between the tests and gen 3 ran at 350 FPS average that's outside the expected variance which is actually really small in this game as indicated by the bars and the maximum uplift is 1 1.2% at 1440p technically the range widened here but not a meaningful amount versus the 1 and 1/2% result previously we're at 1.96% maximum uplift now and at 1080p we're seeing less than a 1% swing in Rainbow Six Siege but we're also running 1080p on an RTX 490 so that's part of why we threw strange Brigade in here using Vulcan at 1080p as well just to throw another extreme into the mix of a different API at a low resolution just to see if any unique behaviors emerged they didn't the maximum Delta we observed was 1.3% so that's it for the PCI bandwidth testing this is pretty much the same we see every generation where with the newest card on the newest version of pcie versus the previous version it's about a 1 to 3% range so that's what we're seeing there was one really unique and interesting outlier and that was total war war hammer 3 where at 1080p specifically and only for that bench uh and with the settings we use with DirectX 11 with the battle bench work we saw 15% range and that was reproducible multiple times with two different test benches and then we also called Nvidia to sort of set in motion a a performance lab evaluation of it to try and get an answer for why that's specifically happening and when we get that answer we'll put it in Hardware news as just an update for educational purposes of okay this is a behavior but why that though is truly an outlier and like we said earlier you'd have to test hundreds of games to really find those and start to establish a pattern for when they happen so overall you don't need to worry about pcie bandwidth too much here it's a couple percent range uh maximally and realistically if you're spending $1,600 on a video card not too many of you although there will be some will be slotting that card into an old enough system that its Max pcie support is Gen 3 so for the most part kind of a non-issue but if you're looking at say a by 8 Gen 4 interface you can compare those results to BU 16 gen 3 and you're looking at a 1 to 4% range or so maybe with occasional outliers that we can't really account for but um that's it it's it's very simple and straightforward for this one check back for more as always go to store. Gamers access.net to help us out directly we have more follow-up testing we're doing on the CPUs and the gpus that all just came out uh those follow-ups a lot of them will be shorter like this one and very focused on just strictly what we're talking about uh for the follow-up so subscribe for more we'll see you all next timetoday we're doing pcie bandwidth testing so we're looking at the generational impact on the RTX 490 when you use pcie gen 3 versus pcie Gen 4 technically we also ran it with the motherboard set to Gen 5 but generation 5 pcie is not supported by the RTX 490 so it doesn't do anything we threw it in there anyway though because we knew people would ask so really this is 3 versus 4 there's a little bit more Nuance to this where you can start to derive some useful is when you also scale eight Lanes versus 16 lanes for example between the two because pcie gen 4x8 is the same as pcie gen 3x16 for the most part they're about the same bandwidth so let's get into the testing today it's going to be a really simple one and we'll see how it scales before that this video is brought to you by montech Sky one light PC case the sky one light is Monte's high air flow case with ventilated front panel included fans and RGB LED accents the sky one light is a compact mid tower case for ATX builds and the argb LED quick connect on the front panel makes it easy to maintain the case without all the cables dust filters are strategically placed and there's basic cable management features while still maintaining a competitive price learn more at the link in the description below first up some backgrounds we did a lot of test passes here normally we do a minimum of four and then we multiply that depending on sort of the level of accuracy we need and you're looking at the same device and you're only changing the the pcie generation there's a possibility that you run the numbers really close together and need that extra set of sort of accuracy or data uh resolution to get the numbers that you want to actually get to be able to say this is a real difference and not just variance so we did more test passes than normally now pcie specs there's a couple of layers of support that have to happen and when you're talking about pcie generational versioning for uh there's vaseline on this card for which motherboard basically to drop it into at a at sort of the lowest possible end really what you care about is the realized performance or the actual in-game or rendering performance or whatever it may be because the theoretical maximum throughput is different from the actualized throughput so you have to have support for the pcie generation you want to use at the CPU level if you look at a CPU block diagram for example and then it's got to pull maybe 16 lanes for example down to peg1 which is the first pcie slot typically uh for graphic support and that's how your generation's dictated at the CPU and the platform level that' be the motherboard if you're running lanes that are fed from the chip set which is uncommon these days but possible for NVIDIA you need buy eight so you need eight Lanes can't do it on by4 officially anyway and those eight Lanes uh are probably going to be a lower generation of PCI than what you're getting off the CPU plus you've got DMI uh or additional overhead basically between the CPU and the chipset anyway but this is a relatively uncommon thing so we're not going to be testing for that because Crossfire and SLI are dead not even basically just actually the final layer of support is the video card has to support it so nvidia's RTX 490 is on pcie Gen 4 now this isn't actually necessarily a bad thing if the card cannot utilize pcie Gen 5 in a meaningful way so remember with interfaces and protocols you're typically limited by the device before the interface itself unless you're running a much newer device on a much older interface a great example of this would be back when SATA hard drives were the fastest thing you could get ssds weren't out yet and you might cap out at maybe say 200 megabytes per second or whatever for a particular year of hard drive while the interface itself could run closer to 550 so the interface was never a concern up until the devic has got fast enough obviously and that's true for pcie as well where Gen 5 Maybe it doesn't really matter for this we won't really know formally but we can get a good idea for how much gen 3 versus Gen 4 matters so that's the theoretical speed side of things versus actualized and we can test theoretical we'll show you that and then the last thing to know is that pcie generation becomes a lot more relevant on devices where they decide to cut down the lane count to say by eight so if you're running eight Lanes to say an arc a580 or whatever it was that's not out yet or you're running eight Lanes to an AMD 6500 XT now if you go from PCI Gen 4 to pcie gen 3x8 you are really limiting your maximum possible throughput uh both theoretical and actualized shown in test data we've published in the past where you're because PCI Gen 4 by 8 versus PCI gen 3x8 is a 2X change that's where you start to really lose on performance uh that in this case it won't matter cuz it's all by6 Gen 4 here's a quick image just to show you the max theoretical bandwidth to get everyone up to speed so technically pcie gen 7 is planned for 2025 ratification that doesn't mean you'll see it on boards necessarily probably you'll start to see at least rumors of it but that's the plan Gen 6 is introduced I think it's been ratified or at least parts of it have been ratified but it's not on boards yet and then Gen 5 which was introduced in 2019 is now on boards so if you look at the transfer rate per Lane uh or actually it's even easier if you look at a through put per Wikipedia by 16 Gen 5 can go approaching 64 GB per second gen 4 by 16 is about 32 GB per second you see it's a pretty clean having you go down to gen 3 at 16 GB per second so that is the theoretical Max throughput of the interface itself whether or not the card utilizes that depends on the card okay let's get into the first set of tests this one is the pcie bandwidth test so using the 3D Mark pcie bandwidth test we're able to rapidly determine the maximum theoretical throughput of the pcie slot and this will not necessarily materialize an equivalent Deltas in gaming or real applications unless they directly mirror this synthetic workload but we haven't encountered anything that behaves the same way as the bandwidth test does in real life doesn't mean it's not out there it just means we're not aware of it and if you are please tell us what that is so we can test it next time here's the chart with the motherboard configured to Gen 5 the results is the same as Gen 4 that's because the card doesn't support pcie Gen 5 the ma maximum theoretical throughput is 25 1/2 GB per second here with the Gen 3 result dropping nearly in half to 13 gbt per second you can see there's some overhead there cuz actual gen 3x 16 would be closer to 16 GB per second gen 2 drops another half down to 6.7 GB per second and these results make sense and they align roughly with some overhead with a spec for pcie the most important thing with this test other than just showing you that that yes theoretically you can produce a difference is that the toggle we're using in BIOS works so that's important so the way to test for this type of thing without switching the whole rest of the system which would invalidate the testing is basically to use built-in toggles in BIOS to drop down the PCI generation This Test shows that those locks are in fact functioning properly and it's not just for show so it tells us that the testing approach we're taking is accurate all right let's get into the gaming results so total Warhammer gives us a heavy load with dx11 which isn't as close to the metal the pcie Gen 4 results ran at 24.9 to 125.4 FPS average with Gen 3 at 122.5 average now top to bottom that's a maximum theoretical advantage of 2.4% on Gen 4 and variance is about 0.5% here 1440p is more interesting here the best result was 236.50 FPS average with the worst at 227.50 average on PCI gen 3 the best result leads PCI gen 3 by 4% with the original test data was just another set of passes mind you but conducted a week earlier leading at 3.6% so that's more noteworthy than we saw at 4K 1080p widens that Gap to a really interesting extreme in this one we saw the pcie Gen 4 result placing at 315 to 318 FPS average that's reasonable variance at this frame rate the PCI gen 3 result though was outside of variance at 277 FPS average that allows the Gen 4 results a massive lead comparative of about 15% to a point where it's actually a big chunk of the gp's maximum performance and also a gigantic red flag to us that we need to retest so we reran these tests on two different test benches that are identical for performance and ultimately we confirmed the findings as a sanity check and because it's good lab practice we called Nvidia and requested that it have its own performance Lab look into the results it's currently cued to do so and we're waiting for those to come back for now though the best guess that GN and nvidia's team member just sort of unofficially have is that we found potentially a hole in driver optimization it may be a pcie bandwidth Behavior but it's probably more likely that this is a combination of the pcie bandwidth with driver Behavior regardless this is an outlier but it's an interesting one we'd have to test probably hundreds of games to reliably find more of these types of results it's something that could happen it's just there's no real predictable way to make it happen shadow of the Tomb Raider is next at 4K this had us at 202 FPS average with Gen 4 that validated on both sets of stock passes the pcie Gen 3 result dropped to 198 FPS average allowing a maximum advantage of 1.7% on PCI Gen 4 in the very least the consistency of lows here is impressive we're feeling good about our testing methodology at this point given the tightness of those results but for pcie Generation differences it's not that exciting at 1440p we're seeing similar results the 2 Gen 4 test ran at 241 to 242 FPS average with a maximum advantage of 0.5% that's variance so it's irrelevant the lows are within variance as well Final Fantasy 14 at 4K produced a range of 214 to 215 FPS average on Gen 4 results with Gen 3 at 208 so there's a maximum Improvement in this set of tests of 3.3% Rainbow Six Siege at 4K and on dx11 ran at 355 to 356 FPS average with Gen 4 data sets even with a week between the tests and gen 3 ran at 350 FPS average that's outside the expected variance which is actually really small in this game as indicated by the bars and the maximum uplift is 1 1.2% at 1440p technically the range widened here but not a meaningful amount versus the 1 and 1/2% result previously we're at 1.96% maximum uplift now and at 1080p we're seeing less than a 1% swing in Rainbow Six Siege but we're also running 1080p on an RTX 490 so that's part of why we threw strange Brigade in here using Vulcan at 1080p as well just to throw another extreme into the mix of a different API at a low resolution just to see if any unique behaviors emerged they didn't the maximum Delta we observed was 1.3% so that's it for the PCI bandwidth testing this is pretty much the same we see every generation where with the newest card on the newest version of pcie versus the previous version it's about a 1 to 3% range so that's what we're seeing there was one really unique and interesting outlier and that was total war war hammer 3 where at 1080p specifically and only for that bench uh and with the settings we use with DirectX 11 with the battle bench work we saw 15% range and that was reproducible multiple times with two different test benches and then we also called Nvidia to sort of set in motion a a performance lab evaluation of it to try and get an answer for why that's specifically happening and when we get that answer we'll put it in Hardware news as just an update for educational purposes of okay this is a behavior but why that though is truly an outlier and like we said earlier you'd have to test hundreds of games to really find those and start to establish a pattern for when they happen so overall you don't need to worry about pcie bandwidth too much here it's a couple percent range uh maximally and realistically if you're spending $1,600 on a video card not too many of you although there will be some will be slotting that card into an old enough system that its Max pcie support is Gen 3 so for the most part kind of a non-issue but if you're looking at say a by 8 Gen 4 interface you can compare those results to BU 16 gen 3 and you're looking at a 1 to 4% range or so maybe with occasional outliers that we can't really account for but um that's it it's it's very simple and straightforward for this one check back for more as always go to store. Gamers access.net to help us out directly we have more follow-up testing we're doing on the CPUs and the gpus that all just came out uh those follow-ups a lot of them will be shorter like this one and very focused on just strictly what we're talking about uh for the follow-up so subscribe for more we'll see you all next time\n"