PCIe 4.0 vs. PCIe 3.0 GPU Benchmark, Feat. GeForce RTX 3080 FE

hahahanoobs

Posts: 3,036   +1,209
If you're using second x16 slot for a x1 to x8 device, you're doing it wrong, and easily fixable. Additional PCIe lanes coming from the chipset exist for this exact reason. Motherboard manuals exist for this reason.

Not sure why over a paragraph was spent on a very odd hypothetical scenario.
 
  • Like
Reactions: Shadowboxer

pcnthuziast

Posts: 895   +538
Looks to me that while it may not matter much at this moment, it's going to very soon and be quite significant in 2021.
 
The only two games I’m interested in seeing benchmarks for right now are Crysis Remastered and Microsoft Flight Simulator.

They are the most demanding.
I agree with Flight Sim 2020.

Crysis Remastered appears to be CPU limited. Digital Foundry had a great video covering this. In short it looks like even with an Intel 10900K CPU or Ryzen 3900X CPU, it is essentially using just one CPU core for most of the heavy hitting.

 

Maxiking

Posts: 132   +150
Zen2 gaming performance is pathetic. Can not beat a 6 year old architecture heavily slowed down by security patches. Jesus Christ, if there had been no patches, Intel would have been almost on par in productivity with Amd.
 
  • Like
Reactions: Shadowboxer

DonquixoteIII

Posts: 7   +6
Here's a non-hypothetical... I will be running a 3090 AND and raid 0 nvme x 4 board, with either Samsung 980 Pros or Sabrent Rocket 4 plus drives. (I would like to go with two highpoint controller cards and 8 NVME drives... But I think that is pushing my luck, even for server CPUs and Chipsets.) I might go with TWO 3090's if it will improve rendering to a significant degree. Don't think Intel gives me that choice, even in PCI/e 3 unless I go with TWO Xeon's and a server board... And, on the AMD side I am limited to whatever socket the gen 4 Threadrippers (Zen 3, supposedly, will use the sTRX4...) use, or go with an EPYC server chip and board.

I wish that the CPU makers would start to take PCI/e seriously.
 
  • Like
Reactions: Mister_K

gamerk2

Posts: 432   +303
Whats impressive is how well PCI-E 3.0 x8 holds up. Remember that PCI-E 3.0 x8 offers the same bandwidth as PCI-E 2.0 x16, so even some really aged motherboards can still keep up relatively well.

Yes, more bandwidth helps, but you see it isn't that a huge difference.
 
  • Like
Reactions: nismo91

Digitalzone

Posts: 135   +75
Crysis Remastered is crap. They intentionally poorly optimized it so they got the public attention and people would ask again: Will it run Crysis. The game doesnt even look good to have even half of HW demands.
 

ubronan

Posts: 57   +9
I constant used many cards but found when using them you got little to no speed improvements.
When I started investigating why I constant had issues witt mutiple cards installed if I used 2, 3 or 4 cards the each step lower performance gain made me decide to stop using more cards.
Especially because the software developers did not really had an interest in making this work better, for them it was just some silly people using such amounts of cards.
In most cases the slots are reduced to X8 or even less often some of the older boards claimed to have full speed but in reality most of the time you never got them working at that speeds for real.
Anyway most motherboards switch to lower speeds on the bus when using more cards.
In reality when pci-e 3.0 came out I already noticed hardly any difference if the slot ran in pci-e x8 speeds or with full x16 the difference was so minor that it made for me the conclusion that the cards are not really needings those lanes for real. Maybe there are some limits into play which I actually never have investigated. However I found on almost everything in consumers boards that they often have alot of limits or use alot of pci-e lanes sharing.
For instance when you want to use alot of ssd's to make a raid setup you often kinda end up in a recipe of disaster. So the thing is with all my experiments I ended up that I actually need to buy a very expensive PRO board for what I like todo. But even those often have huge issues when you want do what I do.
The same goes for NVME I bought a board with 3 nvme slots guess what all 3 share 1 pci-e lane and worse is that the sata devices use the same freaking pci-e lane.
And actually after trying to find a game board which really can do what I want I ended up with nothing NO board is able to run a sata ssd raid and 2 or 3 nvme at the same time, all consumer boards switch of the sata ports or you have to remove at least 2 nvme drives or all sata drives.
But worse is that the pci-e slots often are very limited as well, they sell these for high prices but they actually all switch back to x8 or less if you start using more slots.
And no I do not consider myself as a heavy user, but yes I like to do weird things.
People keep blabbing that nvme is so fast, ITS not only in certain multi user environments they are fast not in a single user environment. I have been telling thousands of people that sata is still the faster gaming drive, but nobody believed me even if they saw it with their own eyes. People mocked me and called me many bad names but fact is I am right a normal SATA SSD is in most cases often faster with games and such. Hell even with many heavy workloads sata does beat the crap out the most overhyped so called super fast drives. One thing is for sure I love the tiny cards and the way can be be build into your system and take so little space thats the major point I love them. Performance wise they still are slower than several good sata based ssd.
I can not tell you why but I really have tested it alot and my games run better on the fast sata drives I have. Yes they are so called slower in synthetic benchmarks but those are idoit products which do not really show anything else than a hyped potential sell product.
In reality the numbers are often bleak for new products. Sata ssd was not and still are not such a fake upgrade they are king in performance. Funny enough I hate several reviewers and one of them kinda proved what I am telling people for years, he made a video about and there is now worldwide proof about what I been telling people for many years when nvme came out.
Nobody in a single user environment can make use of these so called superspeeds if your gaming nor could anyone show me a real life scenario where the so called 3500 Mb/s was actually used in normal work.
But my main point still is games never ever ran better or faster and thats a fact I been telling after nvme was released. Real world performance is often lower than the best sata ssd's can perform at. Now we need to know why the so called speed kings fail to be as fast or why they can not come near that speed when we are gaming.
I read that some game developers are busy trying to improve game loading performance but still its a very difficult subject, there are many limits in play here so making games load faster needs some real serious thinking and probably new tech to actually make improvements.
For now SATA ssd is really still the best for loading games [ PERIOD ]
Now we are at PCI-e 4 and I see actually again the same there is little to no gain in the new standard slots, nobody has showed me a real life setup where the PCI-e 4 slot shows a superb speed increase at all. Again I do not want to buy a os called PRO board because A) they are overpriced B) need a very overhyped plenty of cores cpu are super prices which I do not need or ever use. C) actually turns out to be often the same big pain in the behind with conflicts and problems. And do not really add something for gaming in fact its a fact they are almost allways slower for gaming.
Lets make one thing clear there are hardly any games out there who people actually play and I mean not the ones being used as a benchmark ... (hint 1) which is actually almost the only DX12 game you always see as a benchmark (hint 2)
Back on the point no games actually really make use of more than 1 to 4 cores some people showed me some results but they actually are fake the os and other tasks being used are showing usage and NOT the game is actually using more than 4 Threads. Besides the above mentioned benchmark toy, which hardly is played by people anymore. There are hardly any games which need more than 4 cores for real so again the same story what I need a 12c/24t cpu for I do not know I can not think of any real world scenario besides Pro work which I do not need nor do anymore.
Thats why I did not upgarde really from my 6700k there is no real gain, I even proved to many friends my 6700k ran better than their expenssive new toys in most cases. And the 77ook ran slower in my games so that was a no go either. However I was hoping that a newer board would solve my problems with my high demands and allowed the setup I wanted.
But in reality I ended up getting yet another cold shower, the on paper should work motherboard did again did not run what I want as all other before.
Alot of promises and sell arguments again being nonsense and lies again from the hardware sellers....
Do not believe the companies on delivering what they promise or claim if your a high end user, thats my only advise I can give, and above all do not believe anything so called benchmarks tell you.
They are often BS [ period ]
 
Last edited:

mAdmAnDingo

Posts: 51   +44
I constant used many cards but found when using them you got little to no speed improvements.
When I started investigating why I constant had issues witt mutiple cards installed if I used 2, 3 or 4 cards the each step lower performance gain made me decide to stop using more cards.
Especially because the software developers did not really had an interest in making this work better, for them it was just some silly people using such amounts of cards.
In most cases the slots are reduced to X8 or even less often some of the older boards claimed to have full speed but in reality most of the time you never got them working at that speeds for real.
Anyway most motherboards switch to lower speeds on the bus when using more cards.
In reality when pci-e 3.0 came out I already noticed hardly any difference if the slot ran in pci-e x8 speeds or with full x16 the difference was so minor that it made for me the conclusion that the cards are not really needings those lanes for real. Maybe there are some limits into play which I actually never have investigated. However I found on almost everything in consumers boards that they often have alot of limits or use alot of pci-e lanes sharing.
For instance when you want to use alot of ssd's to make a raid setup you often kinda end up in a recipe of disaster. So the thing is with all my experiments I ended up that I actually need to buy a very expensive PRO board for what I like todo. But even those often have huge issues when you want do what I do.
The same goes for NVME I bought a board with 3 nvme slots guess what all 3 share 1 pci-e lane and worse is that the sata devices use the same freaking pci-e lane.
And actually after trying to find a game board which really can do what I want I ended up with nothing NO board is able to run a sata ssd raid and 2 or 3 nvme at the same time, all consumer boards switch of the sata ports or you have to remove at least 2 nvme drives or all sata drives.
But worse is that the pci-e slots often are very limited as well, they sell these for high prices but they actually all switch back to x8 or less if you start using more slots.
And no I do not consider myself as a heavy user, but yes I like to do weird things.
People keep blabbing that nvme is so fast, ITS not only in certain multi user environments they are fast not in a single user environment. I have been telling thousands of people that sata is still the faster gaming drive, but nobody believed me even if they saw it with their own eyes. People mocked me and called me many bad names but fact is I am right a normal SATA SSD is in most cases often faster with games and such. Hell even with many heavy workloads sata does beat the crap out the most overhyped so called super fast drives. One thing is for sure I love the tiny cards and the way can be be build into your system and take so little space thats the major point I love them. Performance wise they still are slower than several good sata based ssd.
I can not tell you why but I really have tested it alot and my games run better on the fast sata drives I have. Yes they are so called slower in synthetic benchmarks but those are idoit products which do not really show anything else than a hyped potential sell product.
In reality the numbers are often bleak for new products. Sata ssd was not and still are not such a fake upgrade they are king in performance. Funny enough I hate several reviewers and one of them kinda proved what I am telling people for years, he made a video about and there is now worldwide proof about what I been telling people for many years when nvme came out.
Nobody in a single user environment can make use of these so called superspeeds if your gaming nor could anyone show me a real life scenario where the so called 3500 Mb/s was actually used in normal work.
But my main point still is games never ever ran better or faster and thats a fact I been telling after nvme was released. Real world performance is often lower than the best sata ssd's can perform at. Now we need to know why the so called speed kings fail to be as fast or why they can not come near that speed when we are gaming.
I read that some game developers are busy trying to improve game loading performance but still its a very difficult subject, there are many limits in play here so making games load faster needs some real serious thinking and probably new tech to actually make improvements.
For now SATA ssd is really still the best for loading games [ PERIOD ]
Now we are at PCI-e 4 and I see actually again the same there is little to no gain in the new standard slots, nobody has showed me a real life setup where the PCI-e 4 slot shows a superb speed increase at all. Again I do not want to buy a os called PRO board because A) they are overpriced B) need a very overhyped plenty of cores cpu are super prices which I do not need or ever use. C) actually turns out to be often the same big pain in the behind with conflicts and problems. And do not really add something for gaming in fact its a fact they are almost allways slower for gaming.
Lets make one thing clear there are hardly any games out there who people actually play and I mean not the ones being used as a benchmark ... (hint 1) which is actually almost the only DX12 game you always see as a benchmark (hint 2)
Back on the point no games actually really make use of more than 1 to 4 cores some people showed me some results but they actually are fake the os and other tasks being used are showing usage and NOT the game is actually using more than 4 Threads. Besides the above mentioned benchmark toy, which hardly is played by people anymore. There are hardly any games which need more than 4 cores for real so again the same story what I need a 12c/24t cpu for I do not know I can not think of any real world scenario besides Pro work which I do not need nor do anymore.
Thats why I did not upgarde really from my 6700k there is no real gain, I even proved to many friends my 6700k ran better than their expenssive new toys in most cases. And the 77ook ran slower in my games so that was a no go either. However I was hoping that a newer board would solve my problems with my high demands and allowed the setup I wanted.
But in reality I ended up getting yet another cold shower, the on paper should work motherboard did again did not run what I want as all other before.
Alot of promises and sell arguments again being nonsense and lies again from the hardware sellers....
Do not believe the companies on delivering what they promise or claim if your a high end user, thats my only advise I can give, and above all do not believe anything so called benchmarks tell you.
They are often BS [ period ]
That is why Microsoft has developed "Direct Storage" API (Series X/S also uses DS API) which the new RTX 3000 (and likely RDNA2) support.
https://www.nvidia.com/en-us/geforce/news/rtx-io-gpu-accelerated-storage-technology/

It does require game devs to actually program their games to make use it, but that shouldn't be an issue with Micorsoft's next gen consoles using the new API as well. And it should massively reduce CPU usage (in relevant workloads) and make much better use of NVME and even PCIE-4.0 (they say it could actually surpass even PCIE-4.0 bandwidth).

So just give it some time, as next gen games start releasing in the next year or two, we should finally have games making use of our NVME and PCIE-4.0 PCs. In fact, it might become mandatory for a "great" experience in a few years time, if every new demanding next gen console game actually supports it and is designed for it from the get go. Your games will probably still work if your PC does not support it (no DS API enabled GPU), but it will probably not be an ideal or most desirable experience.

NVME PCIE-3.0 might be enough for a while as well, and it might even be good enough for the entire next gen console generation. I am not sure, but considering both PS5 and Series X/S are using PCIE-4.0 and extremely fast storage, PCIE-4.0 may become required for an optimal experience as the gen progresses, I am not sure though.

But, we'll see. We will take another look in 2 years and see how things are coming along. It would be great for it to take off and be adopted quickly though, I for one would be thrilled with all the benefits the new API could bring to gaming. As you said yourself, games are barley making use of our current storage subsystems and bandwidth, we do need a drastic improvement on this front. And DS API does look to rectify much of it, and then some.

Our current games have been largely held back by the current gen consoles (PS4,Xbone) and that is why games have not really taken advantage of faster storage such as NVME (because the current consoles have HDD, so games are designed for them specifically). But because PS5/Series X consoles have fast PCIE-4.0 NVME storage, and also 8c/16t CPUs and 10-12 teraflop RDNA2 GPUS (with ray tracing support), we can expect game requirements to climb quite a bit as the gen progresses.

Which is great, games will finally be taking advantage of our hardware (because next gen console are much more powerful), and we will no longer be held back, as the current consoles have been holding game engine potential (and us) back. Once again, just give it a year or two to really take off. But I can already see that the "recommended" specs for "optimal" performance in upcoming games are quite a bit higher than previous gen games (for console matching or console beating performance). Next gen will take a year or two to really get going, but it will get going.

I also just want to be clear, I am not recommending that everyone run out and purchase new many-core CPUs, PCIE-4.0 platforms and new NVME drives with DS API GPUs. I am not saying that in the slightest. It only makes sense to upgrade if and/or when your PC no longer runs to your satisfaction, so I am not advising anyone to run out and purchase new PCs/Components. I am just talking about what I see as the likely direction that PC gaming is going over the next console gen, according to all the information at hand. I am just looking into my crystal ball, so best to think of it that way.
 
Last edited:
  • Like
Reactions: Mister_K

richcz3

Posts: 43   +29
I agree with Flight Sim 2020.

Crysis Remastered appears to be CPU limited. Digital Foundry had a great video covering this. In short it looks like even with an Intel 10900K CPU or Ryzen 3900X CPU, it is essentially using just one CPU core for most of the heavy hitting.
Yep, DF did a great break down. Was surprised to see how many visuals did not translate well.

One of the more interesting points they made was about the build they used in the Remaster. It was the latest build used for the console ports. Although that makes sense from a development stand point, it's essentially a Remastered Console port.
 
  • Like
Reactions: xenonxenonxenon

Kshipper

Posts: 248   +40
TechSpot Elite
Would choosing a PCIe NVME GEn 4.0 SSD like the Samsung 980 Pro help frame rates or hurt frame rates in anyway? What if a user chose the B550 chipset would that also be a factor (I can see the B550 chipsets starting to sneak into builds)
 

pencea

Posts: 227   +176
Zen2 gaming performance is pathetic. Can not beat a 6 year old architecture heavily slowed down by security patches. Jesus Christ, if there had been no patches, Intel would have been almost on par in productivity with Amd.
Well over 100fps in the majority of games is far from pathetic. Also believe or not there are people who actually use their computers to do more than just gaming mate.
 

EdmondRC

Posts: 20   +14
This is good to know because I plan on upgrading to a RTX 3080 or 3070 (depending on available funds) sometime over the next year, when you can actually find them in stock. I have a i7-9700k which of course has a PCIe 3.0 x 16 motherboard. I was worried that the lack of PCIe 4.0 might hamper my experience with the 3080, but it looks like it will be just fine for a while.