nub 2023 年 5 月 8 日 上午 10:32
PCI E slot question
So I have PCI E 3, but the graphics card I have is for PCI E 4. Does it make a huge difference?
< >
目前顯示第 1-14 則留言,共 14
Cathulhu 2023 年 5 月 8 日 上午 10:45 
Depends. If it uses PCIe 3.0 x16, then most likely you will not notice a difference.
If it is a x8 card, that only uses 8 lanes, then yes, you will notice a difference.

What videocard do you have?
最後修改者:Cathulhu; 2023 年 5 月 8 日 上午 10:45
nub 2023 年 5 月 8 日 上午 10:54 
What is x16 and x8?
plat 2023 年 5 月 8 日 上午 10:55 
I have a PCI-e v. 3 board with a v.4 card. It's fast, much faster than the prev. Pascal card in that slot and way more energy-efficient.

Would it be faster still in the v. 4 board? Yes but I have no idea how much. Maybe some have run benchmarks on setups with both versions and will share this info. Nevertheless, I am very satisfied with the outcome here, on my lower-end z390 board. That's what counts.

Bottom line: It's perfectly fine to run a v. 4 gpu on a v. 3 board, provided the power supply and board specs can handle it. You just may not see the card's fullest potential.
nub 2023 年 5 月 8 日 上午 10:55 
Oh mine is x8.
GOD RAYS ON ULTRA™ 2023 年 5 月 8 日 上午 11:11 
List your specs!
Overseer 2023 年 5 月 8 日 上午 11:19 
The x8 and x16 are actual physical lanes that are used for data transfer in parallell. The standard from 3 to 4 basically just doubles the frequency. So 8 lanes at PCIe 4 are basically the same as 16 lanes at PCIe 3.
There should be some benchmarks around if you can't test it yourself.
Illusion of Progress 2023 年 5 月 8 日 上午 11:36 
No, it doesn't make a huge difference. It barely makes any difference at all most of the time.

There's like two GPUs in particular where it matters to any meaningful extent (the RX 6400 and 6500 series when put in a PCI Express 3.0 or especially 2.0 board). For any other GPU, it's a non-issue. You don't need to worry about it.

The PCI Express version largely just determines the maximum amount of bandwidth between the CPU and GPU. And that amount of bandwidth tends to far, far, far exceed what is often necessary for games.

So here's a hypothetical question, if a particular thing is nowhere near your limitation, and you double that particular thing, what happens? It's obvious; a big lot of nothing happens. You don't gain performance because it's not your limitation. So the inverse is true; you're not missing performance not to have it.

Products (in this case, graphics cards) simply tend to gravitate towards supporting the newest version when they are designed, but that doesn't mean it's "under-performing" if you put it in a system that doesn't support that same, nor does it mean much when games will not need it all (this is actually the key part, because people focus on the GPU like it's some constant thing but it's going to depend on what the game needs). Most games/uses will not push anywhere near the amount of bandwidth through that link to the point that it saturates the latest versions. And for the few edge cases that can, they often aren't very "real world applicable, because you'll almost always have something else be a bottleneck first.

Look at the PCI Express link utilization in Afterburner if you want to check yourself for your own reassurance. You might be surprised how low it often is.

There might be some split moments it registers a difference that a benchmark will reflect, so you may find benchmarks show a difference (usually only with top end cards, and even then usually not far off from margin of error, so you definitely won't notice it blindly), but in real world use and in the vast majority of cases, no it's not a major difference. The "lost potential" is a non-issue.

A PCI Express 4.0 graphics card will perform fine in a 3.0 system. Any limitation it faces will quicker come from other things, such as the CPU itself, sooner than it will from the strict lack of bandwidth between the CPU and GPU because of the PCI Express version. The level of bandwidth we have there is just so far in excess of what is typically necessary.
_I_ 2023 年 5 月 8 日 下午 12:28 
引用自 Tiberius
https://www.techpowerup.com/review/nvidia-geforce-rtx-4090-pci-express-scaling/28.html

It barely matters
the 4090 has all 16 lanes
even on a much weaker gpu with 4-8 lanes, it will make a huge difference

pci-e 4.0 is only really useful when the cpu has fewer lanes or the device cannot use as many lanes
Bad 💀 Motha 2023 年 5 月 8 日 下午 1:30 
How is your PCIE only an X8?
The top most PCIE X16 length slot should function as a full speed X16 slot. UsingPCIE 4.0 GPUs on PCIE 3.0 Motherboard is not a problem at all.
Tiberius 2023 年 5 月 8 日 下午 2:13 
引用自 _I_
引用自 Tiberius
https://www.techpowerup.com/review/nvidia-geforce-rtx-4090-pci-express-scaling/28.html

It barely matters
the 4090 has all 16 lanes
even on a much weaker gpu with 4-8 lanes, it will make a huge difference

pci-e 4.0 is only really useful when the cpu has fewer lanes or the device cannot use as many lanes

that site alr did the test with so many games
_I_ 2023 年 5 月 8 日 下午 4:38 
the 6500xt is pci-e 4.0 x4

if the cpu has few lanes, it would not be able to put all 16 lanes to the gpu slot

or fi the gpu is stuck a x16 slot with fewer lanes
Bad 💀 Motha 2023 年 5 月 8 日 下午 4:47 
That is a specific issue for some of the lower end AMD GPUs where if it's designed specifically for PCIE 4.0 x4 or x8 it is designed to run even slower if you out it into a PCIE 3.0 slot. Why AMD does that makes no sense.

A majority of NVIDIA GPUs they will still operate at X16 even when put into a PCIE 3.0 X16 length slot.

However to avoid a possible issue you want to then limit any extra PCIE cards when doing this.
Illusion of Progress 2023 年 5 月 8 日 下午 5:29 
引用自 Bad 💀 Motha
That is a specific issue for some of the lower end AMD GPUs where if it's designed specifically for PCIE 4.0 x4 or x8 it is designed to run even slower if you out it into a PCIE 3.0 slot. Why AMD does that makes no sense.

A majority of NVIDIA GPUs they will still operate at X16 even when put into a PCIE 3.0 X16 length slot.

However to avoid a possible issue you want to then limit any extra PCIE cards when doing this.
It's not just AMD that uses less than 16 lanes on graphics cards. The reason AMD is prominent with the practice right now is because they did it to a great extent by giving the 6400 but especially the 6500 only 4 lanes, when they definitely warranted 8 (at least the 6500 did). nVidia does it too with their lower end stuff though. The x30s are commonly 4 lanes cards, and the RTX 3050 is 8 lanes, for example. The upcoming RTX 4060 Ti is even going to be an 8 lane card (and 8 GB VRAM, and 128-bit bus), but that's no surprise as it represents a quarter of the flagship, something that was traditionally called an x50.
最後修改者:Illusion of Progress; 2023 年 5 月 8 日 下午 5:30
< >
目前顯示第 1-14 則留言,共 14
每頁顯示: 1530 50

張貼日期: 2023 年 5 月 8 日 上午 10:32
回覆: 14