NVIDIA GeForce RTX 4090 PCI-Express Scaling

[mechtech]

Nice. Just gotta hacksaw out the ends of my 1x slots so I can put a 16x card in there

 

[defaultluser]

I said you don't need PCI 4.0 or more even for 4090... 3.0 seems enough

agreed

you have to castrate the ever-loving-shit out of my ivy bridge h77 system by using the south-bridge-only second video slot at2.0 x4 \

or install a 2500k in place of my 3570lk

 

[BigMack70]

Good to know I don't need to upgrade from my 9900ks yet... Was hoping that would be the case

 

[Hxx]

thank you for posting such great work!!
As a 10th gen intel owner i am relieved i dont need to upgrade my platform lol... for at least two more years.
this is excellent news for all those still on pcie 3.0.

looks like there's finally a point to 4.0x16 for high refresh.
interesting.

yes 2% lmao. massive improvement

 

[Sora]

this testing is missing horizon zero dawn, where massive differences were noted.

 

[W1zzard]

this testing is missing horizon zero dawn, where massive differences were noted.

Just bad programming

 

[Mussels]

Does halo infinite have issues with the 40 series? whats up with the 1080p results getting worse than 1440p?

Yes, but any MB that is Gen3 probably has a CPU that will be holding it back.

*Slaps 5800x3D on x370*

Yeaaaaah booooooi.
To be fair that's about the only possible example it could work, Zen3/3D on 300 series boards - otherwise i'm totally agreeing with you.





I think people are missing that these results are showing that even an 8x 3.0 or 4.0 link will be fine for 99% of users, which is great when a lot of motherboards have issues with bandwidth being yoinked from the GPU slot for NVME and other PCI-E slots.

 

[W1zzard]

Does halo infinite have issues with the 40 series? whats up with the 1080p results getting worse than 1440p?

CPU limited, and the higher CPU overhead of Ada causes it to fall behind older gen while CPU limited, because the driver eats a few extra CPU cycles that could otherwise go to the game to eke out of a few more frames

 

[Calmmo]

CPU limited, and the higher CPU overhead of Ada causes it to fall behind older gen while CPU limited, because the driver eats a few extra CPU cycles that could otherwise go to the game to eke out of a few more frames

Here's hoping AMD releases the 3D variants soon after the 13900k...
even if it only manages to roughly match the 13900k id rather buy into a platform that will be getting at least one more launch

 

[Bjorn_Of_Iceland]

I just wish they would add an Unreal Engine 5 based app to benchmark, as this would also indicate recent upcoming title performance.

 

[mahirzukic2]

Nice. Just gotta hacksaw out the ends of my 1x slots so I can put a 16x card in there

You don't need to hacksaw anything. Just get a PCIe riser.

 

[AnarchoPrimitiv]

It's a shame that USB4 was just released, because a Thunderbolt 5/USB5 based on a PCIe 5.0x4 link would have been awesome since in terms of bandwidth it equates to 3.0x16, which, with the proper chips, could mean an eGPU enclosure could offer 3.0x16 and basically make it so an external GPU would be just as performant as a GPU slotted into the motherboard.

And I guess mostly AMD fans have trashed NVIDIA for not supporting PCIe 5.0.

Considering there's on average a 2% difference between PCIe 3.0 and 4.0 there would be a 0% performance inrovement from using PCIe 5.0.

DP 2.0 for RTX 4090 - that's relevant for some. PCIe 5.0? The time has yet to come.

Although I'm not a fan of AMD, I do cheer for them as any gains for AMD at Intel's expense equates to a better situation for consumers with the idea being that if they reach a 50/50 marketshare, they will reach a balance of power and be forced into the most competetive situation possible and therefore result in the best possible market for consumers.....but anyway, I think anyone who knows anything about PCIe knew that 5.0 wouldn't really provide any tangible benefits over 4.0 in the consumer space with respect to GPUs. And thus far I uavent heard or read anyone, regardless of their allegiance, making a big deal about it. Regardless, there'll always be defenders on either side trying to pick on the perceived shortcomings of others....

Just like people trying to negatively point out that AM5 is a brand new platform and will therefore have associated costs with adopting a brand new platform.....and the sky is blue ...that's always going to be true with any new platform or any consumer building a computer from scratch and requiring all new hardware even if they're using an older platform....it's a reality that can never be avoided so I don't understand the reason behind pointing it out, especially in the summary/conclusion of a review on Ryzen 7000 when it's listed under "negatives", it implies that a company can do something to avoid such a situation, and besides artificially lowering prices and cutting into profit, something a publicly traded company can be sued for by their shareholders, there's nothing really that can be done....it's just the inherent nature of the situation, but by hypothetically listing with things such as, "DDR5 only, no choice of DDR4", it makes it seem like it was a design choice by AMD like the example I just gave of DDR5.

 

[W1zzard]

I just wish they would add an Unreal Engine 5 based app to benchmark, as this would also indicate recent upcoming title performance.

I've been thinking about make an UE5-based workload for various testing, but I feel it's too early for a rendering workloads and it might misrepresent what UE5 can do eventually. I added an UE5 developer workflow in my CPU benchmarks recently

 

[Quicks]

looks like there's finally a point to 4.0x16 for high refresh.
interesting.

are we reading the same charts?

 

[mechtech]

You don't need to hacksaw anything. Just get a PCIe riser.

but that would be a better idea though

 

[lightning70]

As a result, PCI Express 5.0 is not yet required for graphics cards as there is almost no difference between 3.0 and 4.0.

 

[Denver]

Would you consider breaking down the three resolutions into SRD and HDR gaming performance? I could not find information on the testing page whether games are tested for SDR/HDR output. I can see "highest quality settings". Does that include HDR being on in each game?

As shown by Tim from HUB, some gaming engines deal with HDR differently and need more processing when enabled, which lowers frame rates on HDR monitor. This would mean that games need to be tested tested in SDR and HDR, to show differences in GPU performance in those scenarios. A game achieving say 70 fps on 4K SDR monitor could achieve 63 fps on 4K HDR monitor. Many games perform similarly in both modes, but it would be useful to identify the outliers, so that published charts are even more accurate.

This is irrelevant. He was testing PCIe bus throughput, not which CPU is better performing.


Someone would need to test several workflows at the same time to tell us where saturation point is. This somebody would need to have a lot of time to do that.


Please make pictures smaller before posting, as they appear gigantic in the comment section! Try to drag picture angle diagnonally towards its centre.


It depends on what game people play on what kind of gear. Here, average performance is a meaningless measure. If I play Flight Simulator and Cyberpunk in ultra settings only, which is very demanding on GPU, I will not need more than 4K/90hz HDR display. Even mighty 4090 cannot give more than 73 fps in Flight Simulator in this case. Do we need 73 fps in Flight Simulator? Not necessarily. 60 is plenty. It's a slow pace game mostly, unless someone enjoys fighter jet action. It's all about use scenario.

4090 is a halo product that unlocks higher frame rates on 4K display and could be purchased specifically for that purpose by those who need it to do that. Anyone playing 1080p or 1440p should never consider this card, unless needing insane frame rates. A total overkill. Anyone who needs future-proof DisplayPort 2.0 connectivity should never buy 4000 series cards.

This is very relevant, how do you want to measure a possible PCIe bottleneck when there is already a CPU bottleneck?

 

[The Von Matrices]

As a result, PCI Express 5.0 is not yet required for graphics cards as there is almost no difference between 3.0 and 4.0.

It isn't needed for performance but PCIe 5.0 would offer extra flexibility. If the card supported PCIe 5 then you could run it in an x8 slot with a PCIe 5 SSD on Intel platforms without losing GPU performance.

 

[Tek-Check]

As a result, PCI Express 5.0 is not yet required for graphics cards as there is almost no difference between 3.0 and 4.0.

PCIe 5.0 is not required, but it is desirable for future-proofing PC systems. Nvidia was anti-consumer by not offering PCIe 5.0 and DisplayPort 2.0 on 4000 GPUs.

PCIe 5.0 is desirable for the entire PC ecosystem, including GPU. Although is makes motherboard and peripherals design a bit more expensive initially, it will greaty improve connectivity and expansion slot provision for peripherals, like never before. You can run GPUs in x8 or even x4 mode on 5.0 and free up another x8 or x12 lanes for other devices with direct traffic to CPU.

Having 24 Gen5 usable lanes on 7000 Ryzen CPUs, is like having 48 Gen4 lanes, practically entry workstation platform. First generation of AM5 motherboards is a safe, mainstream design, but I do expect some vendors to be more adventurous in new edition of boards next year. We can expect some developments:

- PCIe switch chip Gen5-to-Gen4 x4 (previosuly known as PLX), that will double downstream lanes. Current Promo21 chip uses PCIe 4.0 x4 link to CPU at 64 Gbps, so half of 128 Gbps that CPU can provide. A new double chipset wil be able to connect to CPU via PCIe switch chip independently, gaining x4 lanes for another NVMe drive or other peripheral. Current daisy-chaining consumes those x4 lanes.

- or new edition of AM5 chipset Promo21 from AsMedia will be PCIe 5.0 capable itself, taking all 128 Gbps from CPU chipset link

- several boards currently have two x16 Gen5 slots with ability to bifurcate lanes into x8 x8 configuration. This allows new GPUs to operate in x8 Gen5 slot without any performance loss and frees another x8 Gen5 lanes for AIC peripheral, such as NVMe RAID array or other solution.

- three x16 slots can trifurcate 16 Gen5 lanes too, x8 or x4 electrically for GPU, and another x8 and x12 respectively for other peripherals.

There are many options, depending on how many PCIe slots vendors want to provide, how to wire electrically those slots and where to install PCIe switch chip. PCIe switch can also link to those bifurcarted lanes from GPU, say x8 Gen5, and provide another x16 Gen4 lanes for other non-AIC peripherals.

Therefore, PCIe 5.0 brings a lot of flexibility to motherboard design and connectivity without compromising on performance or need to bi-wire two devices on one set of lanes, which has often been the case. For example, many boards have mutually exclusive option to either connect M.2 drive or PCIe x4 device. Both cannot work in the same time. With PCIe 5.0, this is not necessary anymore.

This is very relevant, how do you want to measure a possible PCIe bottleneck when there is already a CPU bottleneck?

I corrected myself on this one already in another post.

 

[AlwaysHope]

Hah! PCIe 2.0 still being fine almost 16 years later

Sure, if you want that last 5% then you'll need PCIe 4.0 but these articles always prove that the PCIe race really isn't that necessary unless you're already chasing diminishing returns.

Yes, that is interesting. 4.0 will set you up for rest of this decade, at least on a single monitor.

 

[wheresmycar]

PCIe 5.0 is not required, but it is desirable for future-proofing PC systems. Nvidia was anti-consumer by not offering PCIe 5.0 and DisplayPort 2.0 on 4000 GPUs.

PCIe 5.0 is desirable for the entire PC ecosystem, including GPU. Although is makes motherboard and peripherals design a bit more expensive initially, it will greaty improve connectivity and expansion slot provision for peripherals, like never before. You can run GPUs in x8 or even x4 mode on 5.0 and free up another x8 or x12 lanes for other devices with direct traffic to CPU.

Having 24 Gen5 usable lanes on 7000 Ryzen CPUs, is like having 48 Gen4 lanes, practically entry workstation platform. First generation of AM5 motherboards is a safe, mainstream design, but I do expect some vendors to be more adventurous in new edition of boards next year. We can expect some developments:

- PCIe switch chip Gen5-to-Gen4 x4 (previosuly known as PLX), that will double downstream lanes. Current Promo21 chip uses PCIe 4.0 x4 link to CPU at 64 Gbps, so half of 128 Gbps that CPU can provide. A new double chipset wil be able to connect to CPU via PCIe switch chip independently, gaining x4 lanes for another NVMe drive or other peripheral. Current daisy-chaining consumes those x4 lanes.

- or new edition of AM5 chipset Promo21 from AsMedia will be PCIe 5.0 capable itself, taking all 128 Gbps from CPU chipset link

- several boards currently have two x16 Gen5 slots with ability to bifurcate lanes into x8 x8 configuration. This allows new GPUs to operate in x8 Gen5 slot without any performance loss and frees another x8 Gen5 lanes for AIC peripheral, such as NVMe RAID array or other solution.

- three x16 slots can trifurcate 16 Gen5 lanes too, x8 or x4 electrically for GPU, and another x8 and x12 respectively for other peripherals.

There are many options, depending on how many PCIe slots vendors want to provide, how to wire electrically those slots and where to install PCIe switch chip. PCIe switch can also link to those bifurcarted lanes from GPU, say x8 Gen5, and provide another x16 Gen4 lanes for other non-AIC peripherals.

Therefore, PCIe 5.0 brings a lot of flexibility to motherboard design and connectivity without compromising on performance or need to bi-wire two devices on one set of lanes, which has often been the case. For example, many boards have mutually exclusive option to either connect M.2 drive or PCIe x4 device. Both cannot work in the same time. With PCIe 5.0, this is not necessary anymore.


I corrected myself on this one already in another post.

no doubt there will be a string of buyers who will benefit from the uncompromising 5.0 flex, more-so specific workstation-class builds but for the wider majority (gamers/general productivety/office work/etc) i can't see most of us needing anything above 3.0 and yet 4.0 is a probable long term blessing in itself.

For most by the time 5.0 becomes an essential piece of the puzzle current platforms will be, from an overall performance enthusiast perspective, ANCIENT! Saying that, as a gamer i'm still keeping my options open and might end up pulling the trigger on Zen 4... not necessarily for PCIe 5.0 but socket longevity.

 

[Wasteland]

no doubt there will be a string of buyers who will benefit from the uncompromising 5.0 flex, more-so specific workstation-class builds but for the wider majority (gamers/general productivety/office work/etc) i can't see most of us needing anything above 3.0 and yet 4.0 is a probable long term blessing in itself.

For most by the time 5.0 becomes an essential piece of the puzzle current platforms will be, from an overall performance enthusiast perspective, ANCIENT! Saying that, as a gamer i'm still keeping my options open and might end up pulling the trigger on Zen 4... not necessarily for PCIe 5.0 but socket longevity.

Agreed. PCIe 5.0 isn't worth the apparent extra cost (on motherboards) to the average user. Heck, to most users, the only practical difference between PCIe 3.0 and PCIE 4.0 SSDs is that the latter run hotter. No lie, I actually bought a secondary PCIe 4.0 SSD by accident recently, because the price was low enough that I assumed it was a 3.0 model. Learned otherwise when I saw the temperature after installing it, though, lol.

Not to say that there isn't merit in pushing these newer technologies to the consumer space, but I just don't have any compelling reason to care about what they offer. W1zzard's excellent testing here reinforces the point.

 

[dicobalt]

This makes me more annoyed at PCIe 5 and how expensive it is. Just put the damn m.2 in a 8x slot instead. You don't need to explode the platform cost for the sake of a form factor.

 

[samih22]

so I pre-ordered Raptor-lake , and already gotten myself a 4090.

my question is, should I leave the GEN 5 m.2 slot empty ? or is it fine if I use a GEN4 m.2 on it ? if I do , I would also lose half the lanes?

I do not want to lose 1% or 2%. I will just use the other m.2 slots , I just wish if I could use all the slots on my board. Thanks

 

[Godrilla]

Good to know I don't need to upgrade from my 9900ks yet... Was hoping that would be the case

I have the same setup against all odds, 9900ks, itx, 750 watt sfx platinum psu by corsair, Pcie 3.0 and 4090 Suprim liquid to replace the 3090 xc3 ultra hybrid. I will keep you posted on the outcome. ( probably upgrade to zen 4 3d to improve frame variance and better 0.1 % lows)
I was hoping they test which cpu generation shows the significant cpu bottlenecks at a resolution that scales best for the 4090 like 4k also will using DSR to 8k on 4k display while mitigating the cpu bottlenecks and still improving image quality. This is where I see the next trend in upscaling technology going upscaling to a higher resolution than your monitor to mitigate bottlenecks when there is performance on the table and improve image quality. Aka Dldsr.