NVIDIA RTX 4090 PCIe 3.0 vs. 4.0 x16 & '5.0' Scaling Benchmarks
video description
Date: 2022-11-01
Related videos
Comments and reviews: 14
Max
It occurs to me, that, (along with having SUPER toxic corporate working atmospheres) the Industry design of motherboards is driven by meming edge-lord breakneck mentality. (to be fair, that is a use case), BUT, just like there are servers, and workstations, gaming computers don't ALWAYS need to be pushing the envelope. Let's talk about E-WASTE. If a PCIE Gen 5 MOTHERBOARD were retro-compatible with Gen 4 components, it would alleviate alot of the bottlenecks existing in the previous Gen, create more resale value, lower E-waste amounts, and give second life to old systems. The profits would be made because the motherboards will get sold, and GAMERS who want to push MAXIMUM PERFORMANCE will still be there and in relatively the same amount of numbers. But Mid Tier and hobbiest configurations might BOOM. First Motherboard company that gives us retro-compatibility or even universal (?? socket types etc) adaptability gets a nice fat sack of MONEY.
reply
It occurs to me, that, (along with having SUPER toxic corporate working atmospheres) the Industry design of motherboards is driven by meming edge-lord breakneck mentality. (to be fair, that is a use case), BUT, just like there are servers, and workstations, gaming computers don't ALWAYS need to be pushing the envelope. Let's talk about E-WASTE. If a PCIE Gen 5 MOTHERBOARD were retro-compatible with Gen 4 components, it would alleviate alot of the bottlenecks existing in the previous Gen, create more resale value, lower E-waste amounts, and give second life to old systems. The profits would be made because the motherboards will get sold, and GAMERS who want to push MAXIMUM PERFORMANCE will still be there and in relatively the same amount of numbers. But Mid Tier and hobbiest configurations might BOOM. First Motherboard company that gives us retro-compatibility or even universal (?? socket types etc) adaptability gets a nice fat sack of MONEY.
reply
Mr.
I would like to remind people of a DirectStorage feature: memory-to-memory. That means DRAM compressed game assets get decompressed directly to VRAM using GPU hardware acceleration which bypasses the CPU. DirectStorage 1.1 releases by the end of this year. I would also like to remind people that DDR4 3200 blows away even the high powered PCIe5.0 SSDs transfer speeds (25,600MB/s vs 16,130MB/s). So no, PC gamers will not need expensive SSDs running super hot crammed up next to a steaming GPU to use DirectStorage. PC users will be far better off with 32-64GB of DRAM and a moderately fast SSD. Games will load the biggest most common stuff near the user into RAM. There is a whole asset management code base associated with implementing DirectStorage. Remember the game doesn't need to load every single asset in the game at any time, people often make that assumption. Open world games today already demonstrate how that can be managed.
reply
I would like to remind people of a DirectStorage feature: memory-to-memory. That means DRAM compressed game assets get decompressed directly to VRAM using GPU hardware acceleration which bypasses the CPU. DirectStorage 1.1 releases by the end of this year. I would also like to remind people that DDR4 3200 blows away even the high powered PCIe5.0 SSDs transfer speeds (25,600MB/s vs 16,130MB/s). So no, PC gamers will not need expensive SSDs running super hot crammed up next to a steaming GPU to use DirectStorage. PC users will be far better off with 32-64GB of DRAM and a moderately fast SSD. Games will load the biggest most common stuff near the user into RAM. There is a whole asset management code base associated with implementing DirectStorage. Remember the game doesn't need to load every single asset in the game at any time, people often make that assumption. Open world games today already demonstrate how that can be managed.
reply
ManyMirrors
Well, it is better to have a platform that is ahead of it's time rather than having hardware that is too fast for that platform. If nvidia wanted to, they could design an RTX 4090 which can run on only 4x gen 5.0 PCIe lanes at almost no cost to performance.
Imagine a hypothetical situation where one PCIe x16 slot was too slow for the newest Nvidia GPU. We'd be overclocking our PCIe slots and motherboard manufacturers would be releasing motherboards with overclocked PCIe slots for even crazier prices.
Oh yeah and as someone already stated, you can fit more bandwidth in fewer lanes which is a big plus considering how few PCIe lanes we have directly connecting to the consumer level CPUs nowadays.
reply
Well, it is better to have a platform that is ahead of it's time rather than having hardware that is too fast for that platform. If nvidia wanted to, they could design an RTX 4090 which can run on only 4x gen 5.0 PCIe lanes at almost no cost to performance.
Imagine a hypothetical situation where one PCIe x16 slot was too slow for the newest Nvidia GPU. We'd be overclocking our PCIe slots and motherboard manufacturers would be releasing motherboards with overclocked PCIe slots for even crazier prices.
Oh yeah and as someone already stated, you can fit more bandwidth in fewer lanes which is a big plus considering how few PCIe lanes we have directly connecting to the consumer level CPUs nowadays.
reply
Olav
I know that in some cases, like if you want a lower input lag - you usually can with a high end GPU set a framerate limit.
This makes the GPU do less , but do enough . Can this also be the case with the PCI-E? Maybe that while a higher gen means it can load data more quickly, maybe it doesent really need the data that quickly and then the loading of so much data might affect the general performance/headroom?
Maybe a silly theory, but it just came to the top of my head as its what you see if you let the GPU produce all the frames it can possible do (higher input lag).
reply
I know that in some cases, like if you want a lower input lag - you usually can with a high end GPU set a framerate limit.
This makes the GPU do less , but do enough . Can this also be the case with the PCI-E? Maybe that while a higher gen means it can load data more quickly, maybe it doesent really need the data that quickly and then the loading of so much data might affect the general performance/headroom?
Maybe a silly theory, but it just came to the top of my head as its what you see if you let the GPU produce all the frames it can possible do (higher input lag).
reply
Andrew
Just a hypothesis, but Total Warhammer looks like it has tons of vertex updates going on in those models. Those updates are happening on the CPU and being transferred to the GPU every frame, which is part of what that PCI test you did earlier does. Dropping down the resolution for total warhammer may have hit the threshold between rendering as a bottleneck and vertex transfer as the bottleneck, especially if the model quality was kept the same across resolutions. You might find a similar effect in other games where there are tons of visible dynamic models
reply
Just a hypothesis, but Total Warhammer looks like it has tons of vertex updates going on in those models. Those updates are happening on the CPU and being transferred to the GPU every frame, which is part of what that PCI test you did earlier does. Dropping down the resolution for total warhammer may have hit the threshold between rendering as a bottleneck and vertex transfer as the bottleneck, especially if the model quality was kept the same across resolutions. You might find a similar effect in other games where there are tons of visible dynamic models
reply
ZeroLife
test it on gen 2.
as clearly gen 3 isn't even a problem there. Or ran it with gen 3 on 4x or something. Imagine your main 16x PCIe lane broke (it happens, really), and you had to use the secondary which isn't running at full 16x and you might be thinking hmm, my gpu must be running really slow with low multipler . When in fact... it might run just fine with barely noticeable differences.
I accidentally ran some older gen GPu even on PCIe2.0 2X... and I didn't even noticed it for months...
reply
test it on gen 2.
as clearly gen 3 isn't even a problem there. Or ran it with gen 3 on 4x or something. Imagine your main 16x PCIe lane broke (it happens, really), and you had to use the secondary which isn't running at full 16x and you might be thinking hmm, my gpu must be running really slow with low multipler . When in fact... it might run just fine with barely noticeable differences.
I accidentally ran some older gen GPu even on PCIe2.0 2X... and I didn't even noticed it for months...
reply
EVPointMaster
I noticed that the game Flower is actually extremely bandwidth sensitiv and also that the grass density setting impacts this a lot.
I was curious so I tested with different pcie bandwidths. I used a 3080, though my board only supports 3.0
The frame rate scales almost linearly with the bandwidth!
1.0 = 35.6fps
2.0 = 68.6fps
3.0 = 118.1fps
It's possible that the bottleneck shifts at 4.0 or even 5.0, but even then it could be interesting to test for cards that are only x8.
reply
I noticed that the game Flower is actually extremely bandwidth sensitiv and also that the grass density setting impacts this a lot.
I was curious so I tested with different pcie bandwidths. I used a 3080, though my board only supports 3.0
The frame rate scales almost linearly with the bandwidth!
1.0 = 35.6fps
2.0 = 68.6fps
3.0 = 118.1fps
It's possible that the bottleneck shifts at 4.0 or even 5.0, but even then it could be interesting to test for cards that are only x8.
reply
Gustavo
I understand that this affects a lot when you have high fps. But does it affect when you are with fps below 100 because the game is very demanding? and I also understand that this affects the high end graphics 3080, 3090, 4080 and 4090. But does it affect the mid-range or low-end graphics? The answer to the first question I have no idea, but I think the answer to the second question is pretty sure no. although I would like a verification of that since they are the most used graphs by all.
reply
I understand that this affects a lot when you have high fps. But does it affect when you are with fps below 100 because the game is very demanding? and I also understand that this affects the high end graphics 3080, 3090, 4080 and 4090. But does it affect the mid-range or low-end graphics? The answer to the first question I have no idea, but I think the answer to the second question is pretty sure no. although I would like a verification of that since they are the most used graphs by all.
reply
Michael
This will be pertinent to everyone building on intel 13th gen as you lose 8 lanes to the pci5 m.2 slot if occupied. Intel 13th gen only supports pci5x8 slot when using the gen5 m.2 slot. So lots of people building intel 13th gen platforms will be limited to pci4x8 on the pci slot so good to know it seems to be only a few percent lost. I changed my latest 4090 build to amd from intel because of this point in addition to the thermal issues with the 13900k!
reply
This will be pertinent to everyone building on intel 13th gen as you lose 8 lanes to the pci5 m.2 slot if occupied. Intel 13th gen only supports pci5x8 slot when using the gen5 m.2 slot. So lots of people building intel 13th gen platforms will be limited to pci4x8 on the pci slot so good to know it seems to be only a few percent lost. I changed my latest 4090 build to amd from intel because of this point in addition to the thermal issues with the 13900k!
reply
Max
Know what I wanna see from motherboards? A HEFTY, user-friendly UI. I want to be able to TASK, individual lanes, individual cores, prioritize APPs and Software (banish telemetry to the abyss)(end bloatware jacking up my thermals even when running in the background).If a game is CPU heavy, I want to see it, and if I feel like it, I WILL give it 7 of 8 cores, or 3 of 4, that's my right and should be my choice etc. Consider the UI, my need-to-know
reply
Know what I wanna see from motherboards? A HEFTY, user-friendly UI. I want to be able to TASK, individual lanes, individual cores, prioritize APPs and Software (banish telemetry to the abyss)(end bloatware jacking up my thermals even when running in the background).If a game is CPU heavy, I want to see it, and if I feel like it, I WILL give it 7 of 8 cores, or 3 of 4, that's my right and should be my choice etc. Consider the UI, my need-to-know
reply
SundowN
Comment at 3:00 about chipset overhead.
I'm looking for the documentation but as far as I know Nvidia SLI uses (or rather, SLI certification requires the use of) CPU PCIe lanes. Meaning in the case of basically all consumer platforms in recent history, bifurcation of the 16 CPU PCIe lanes into two 8x. I think Maxwell SLI allowed bifurcation down to four 4x slots, nevertheless, SLI requires CPU PCIe lanes, and does not use chipset lanes.
reply
Comment at 3:00 about chipset overhead.
I'm looking for the documentation but as far as I know Nvidia SLI uses (or rather, SLI certification requires the use of) CPU PCIe lanes. Meaning in the case of basically all consumer platforms in recent history, bifurcation of the 16 CPU PCIe lanes into two 8x. I think Maxwell SLI allowed bifurcation down to four 4x slots, nevertheless, SLI requires CPU PCIe lanes, and does not use chipset lanes.
reply
irorules
Hey just a comment on the Case ADs. I totally get ads and have no problem, but i would prefer if you only advertised cases that have been tested. They dont have to be high performing results, but being able to see a GN level video for the case would be more helpful than just an ad spot. i also understand conflicts of interest, but in some case Ads (bequiet), you mention it performing well in testing. Love the work you guys put in!you
reply
Hey just a comment on the Case ADs. I totally get ads and have no problem, but i would prefer if you only advertised cases that have been tested. They dont have to be high performing results, but being able to see a GN level video for the case would be more helpful than just an ad spot. i also understand conflicts of interest, but in some case Ads (bequiet), you mention it performing well in testing. Love the work you guys put in!you
reply
James
PCIe advancement is mainly Lower Latency Connections ONLY = However = If The OS and Service Provider does Not Provide Support for Such = it has Absolutely No Affect , well maybe a Tiny bit , and In Most Cases makes it Worse . So the Manufactures got together and Lobbied Each Other To Force The New win11 Mandatory Upgrade on New DDR5 Platform . and Will Not Reverse Compatibility .
reply
PCIe advancement is mainly Lower Latency Connections ONLY = However = If The OS and Service Provider does Not Provide Support for Such = it has Absolutely No Affect , well maybe a Tiny bit , and In Most Cases makes it Worse . So the Manufactures got together and Lobbied Each Other To Force The New win11 Mandatory Upgrade on New DDR5 Platform . and Will Not Reverse Compatibility .
reply
Tony
I'm not upgrading my PCIe 3.0 PC yet, got a 5950x/3070 and soon going over to AMD. I still game at 1080p with my 360hz panel, I love the high FPS for now later on I'll go 1440p But if the FPS are already in the 100s or 200s like in the benchmarks here at 1080p then we are good for quite a while What I want to see is how future games will perform like games using UE5 for example.
reply
I'm not upgrading my PCIe 3.0 PC yet, got a 5950x/3070 and soon going over to AMD. I still game at 1080p with my 360hz panel, I love the high FPS for now later on I'll go 1440p But if the FPS are already in the 100s or 200s like in the benchmarks here at 1080p then we are good for quite a while What I want to see is how future games will perform like games using UE5 for example.
reply
Add a review, comment
Other channel videos
