PCIe [4.0] confusion?

[Mark Kaine]

So with all the talk about older motherboards not supporting pcie 4 (?) i just saw in GPUz my GPU runs at pcie x16 4.0.

 

 

So im really just wondering reading stuff like this 

how come my GPU does run at that speed when the motherboard [MSI B350m] supposedly doesnt  support it? Or is that just a myth after all, or im just not understanding something right?

 

Tell you what ,i was about to buy a new mobo just now after seeing my 3070 running at mere pcie x16 1.1, but turns out its actually running at 4.0 under load ...  dodged a bullet there didnt I?

 

 

 

[Nathanpete]

My PCIe 4.0 compatible system was running at 3.0 speeds on my GPU because I choose so. Both are PCIe 4.0 capable, don't ask me why I did it like this. But GPU-Z still claimed it was running at PCIe 4.0 x16 when it was really PCIe 3.0 x16.

 

@Mark Kaine, I don't think it is running at that speed it is probably actually running at Gen 3 speeds and GPU-Z is mistaken. 

[Mark Kaine]

4 minutes ago, Nathanpete said:

My PCIe 4.0 compatible system was running at 3.0 speeds on my GPU because I choose so. Both are PCIe 4.0 capable, don't ask me why I did it like this. But GPU-Z still claimed it was running at PCIe 4.0 x16 when it was really PCIe 3.0 x16.

 

@Mark Kaine, I don't think it is running at that speed it is probably actually running at Gen 3 speeds and GPU-Z is mistaken. 

Ah, i see , that makes sense, i knew that didnt seem right but i didnt think gpuz is just reporting nonsense either...  and i really wasnt sure if not some BIOS update added support , i know thats kinda like downloading more RAM, but only kinda  

 

Hmm, so am I missing out much by running at (supposedly) 3.0 speed?

 

It couldn't be too much i imagine since im getting roughly above average results in benchmarks ?(actual benchmarks like 3dmark)  

[Nathanpete]

1 minute ago, Mark Kaine said:

Ah, i see , that makes sense, i knew that didnt seem right but i didnt think gpuz is just reporting nonsense either...  and i really wasnt sure if not some BIOS update added support , i know thats kinda like downloading more RAM, but only kinda  

 

Hmm, so am I missing out much by running at (supposedly) 3.0 speed?

 

It couldn't be too much i imagine since im getting roughly above average results in benchmarks ?(actual benchmarks like 3dmark)  

Depends on what your GPU is. If it is Navi the difference is miniscule and you can look up videos on youtube comparing the two, but if it is Big Navi or Ampere the difference is bigger, but still not substantial. I know LTT once advertised an upcoming video comparing the two, but that was a while ago and I don't remember if he uploaded it or not. But other youtubers have also done it. Look up "(Your RTX or RX card) PCIe 3.0 vs 4.0 comparison" and you will get some stats. 

[Mark Kaine]

1 minute ago, Nathanpete said:

Depends on what your GPU is.

Still the old trusty RTX 3070 ... (cant believe its almost half a year old by now o.o)

 

1 hour ago, Nathanpete said:

Look up "(Your RTX or RX card) PCIe 3.0 vs 4.0 comparison" and you will get some stats. 

Alright, thanks for the tip.

[Mark Kaine]

Ok, i checked some benchmarks, written, not on youtube as videos hold about zero credibility to me, even GN, who as much i like him always muddies the waters with "memory tuning" and similar shenenigans (sorry) and the difference seems to be indeed marginal to non existent depending on title and resolution, difference between intel vs amd seems to have a larger impact, but then how do you even compare cpus meaningfully, a better  cpu will have better performance is what it boils down to, duh.

 

[Mark Kaine]

@Nathanpete wait, now Im even more confused hwinfo64 also says pcie4!?

 

 

16 GT/s, and when its idling it says 2.5 GT/s?

 

Well maybe it doesnt matter but pcie4 has quite more features than pcie3 I think, thats why Im wondering. Weird bug if it is one .

 

Edit: Ok my BIOS says pcie3 is max, but that is still weird, I would think the GPU should run at that speed then and not at 16GT/s?

[Nathanpete]

23 hours ago, Mark Kaine said:

@Nathanpete wait, now Im even more confused hwinfo64 also says pcie4!?

 

 

16 GT/s, and when its idling it says 2.5 GT/s?

 

Well maybe it doesnt matter but pcie4 has quite more features than pcie3 I think, thats why Im wondering. Weird bug if it is one .

 

Edit: Ok my BIOS says pcie3 is max, but that is still weird, I would think the GPU should run at that speed then and not at 16GT/s?

Remind me what are the gigatransfer caps of PCIe generations. Like I know DDR3-2133 is both 1067 Megahertz and 2133 MT/s. But that is over a memory bus, not a PCIe bus. Is that the same for GDDR6? Like I don't think it is on GDDR6X because of PAM4 signaling. Like I know your GB/s memory bandwidth is determined by your memory type and frequency. Like my 5600XT, also with GDDR6, is clocked to 1810 MHz, so 40 MHz lower than you, so a itty bitty less bandwidth, but I also believe I have a smaller bus, so does that also affect memory throughput?

 

I tried to calculate all this stuff when researching the upcoming consoles like 12 months ago but got so fucking lost, and this is reminding me that I need to look it up again. All I can confirm is that GDDR5 and up is so much faster than DDR4 no matter the frequency. 

 

 

[Mark Kaine]

9 minutes ago, Nathanpete said:

Remind me what are the gigatransfer caps of PCIe generations. Like I know DDR3-2133 is both 1067 Megahertz and 2133 MT/s. But that is over a memory bus, not a PCIe bus. Is that the same for GDDR6? Like I don't think it is on GDDR6X because of PAM4 signaling. Like I know your GB/s memory bandwidth is determined by your memory type and frequency. Like my 5600XT, also with GDDR6, is clocked to 1810 MHz, so 40 MHz lower than you, so a itty bitty less bandwidth, but I also believe I have a smaller bus, so does that also affect memory throughput?

 

I tried to calculate all this stuff when researching the upcoming consoles like 12 months ago but got so fucking lost, and this is reminding me that I need to look it up again. All I can confirm is that GDDR5 and up is so much faster than DDR4 no matter the frequency. 

 

 

Gen 3 is 8GT/S and gen 4 is 16GT/S.

But, I think something is wrong with these readings after all - which makes me trust these monitoring programs a lot less, when they even dont get the basic stuff right...

 

 

My mobo (b350m mortar) doesnt support gen4 pcie, yet all these programs say it does? Certainly a bit weird and confusing.

Seems they just read what the GPU reports, and even then its odd that it goes to 16GT/S when under load, i would expect the GPU to throttle at 8 (the theoretical max the motherboard supports) ¯\_(ツ)_/¯

[Nathanpete]

46 minutes ago, Mark Kaine said:

Gen 3 is 8GT/S and gen 4 is 16GT/S.

But, I think something is wrong with these readings after all - which makes me trust these monitoring programs a lot less, when they even dont get the basic stuff right...

 

 

My mobo (b350m mortar) doesnt support gen4 pcie, yet all these programs say it does? Certainly a bit weird and confusing.

Seems they just read what the GPU reports, and even then its odd that it goes to 16GT/S when under load, i would expect the GPU to throttle at 8 (the theoretical max the motherboard supports) ¯\_(ツ)_/¯

Sometimes it is possible to find out your exact motherboard model with the command prompt with this command 

wmic baseboard get product, manufacturer, serialnumber, version

but for example, when I do this, I get back that it is an Intel board with a serial number and board number. It doesn't say that it is a mini-ITX H310 board from Asus. In fact, W10 even reports that it has 4 RAM slots when it is mini-ITX, meaning two slots. Building a database with the manufacturer, product, and version for every single motherboard since the core 2 duo would be so big that I would consider it too hard to try, especially for a piece of freeware. 

 

I'm interested what you result is when you run the above command yourself. Like the I know the Maximus boards self identify as a Maximus board with this command, but if you are running a low end mobo like a B350 it may not. 

 

The best chance for this to work would be for a program that is able to extract the BIOS ROM to the hard drive, and decrypt it, taking info to piece together the features, compare to a database, just to report the correct PCIe generation. Or maybe a test that attempts to fill the PCIe bus connected to the GPU. But if it is connected to a bus that goes through the chipset, then something like a large file transfer in the background can make it harder to pinpoint your PCIe generation. Either way, all three options are going to be way more intrusive that HWiNFO already is, or just too big for the devs to make. The only way is to get every major mobo manufacturer to make the command above return exactly what is needed to compare to a common mobo database like that on PCPP. But it won't be happening retroactively because there is no way that you are going to be any firmware team to push out BIOS updates just to make HWiNFO work correctly. 

 

But yeah it is 100% just pulling from the information from the GPU, which is a lot more readily available because so many graphics cards use the reference board, which probably has all the data GPU-Z needs saved to a ROM chip. (INB4 scammers replace the ROM chip to pass off a RTX 3070 as an RTXXX 6969 in 6 years.) 

[Mark Kaine]

On 5/17/2021 at 7:16 PM, Nathanpete said:

But yeah it is 100% just pulling from the information from the GPU, which is a lot more readily available because so many graphics cards use the reference board, which probably has all the data GPU-Z needs saved to a ROM chip.

So, funny thing, i updated my mobo bios - the old one clearly wasnt working right (lots of small glitches, missing features etc) and now everything reports the GPU as "8GT/s"...  

So it was probably just an odd bug.

 

Theres some other things I havent seen , for example "AMD overclocking" which i think might enable Ryzen master to work, not sure, i just know RM never worked on this board.  Could also be "MSI overclocking"  though (that *used* to work) i cant tell... MSI documentation is terrible to non existent.

 

[Nathanpete]

5 hours ago, Mark Kaine said:

Theres some other things I havent seen , for example "AMD overclocking" which i think might enable Ryzen master to work, not sure, i just know RM never worked on this board.  Could also be "MSI overclocking"  though (that *used* to work) i cant tell... MSI documentation is terrible to non existent.

AMD Overclocking is where you enable PBO, which is what Ryzen Master modifies, so yes.