Secondary NVMe drives in M.2 slots running off PCH - 50% write speed @ PCI 4.0 4x

[Sagath]

I got around to looking at mine. Not sure if it really adds anything to the conversation.

This is a 970 evo, in M2.2;



I did verify that the AX200 is running in 128byte mode. Heres the 970evo performance. Seems mostly in line with your 10% on writes (should be closer to 2800).



The WD 850 Black, in M2.1 slot. Of course, no issues here since its right to the CPU, and, well, its a WD with the updated firmware..



And, yep, of course its at 512b

 

[Solace]

The only value would be if the SN850 was slot 2 and if the latest firmware did fix the write speeds. I tested the 960 PRO 2tb before I just sold it today and concluded NVMe 3.0 x4 drives are not fast enough to be bottled necked by the 128byte link. Slot 1 for a 4.0 drive is 100% for all.

 

[Shadowarez]

mis the days when we got 64 lanes my old 4960x was a beast in its day.

 

[Sagath]

Disabling the WIFI did not improve my speeds on the 970 evo.



So I think this is a negotiation limitation actually, based on that, and what HWinfo is saying. Why?

Now that I disabled the AX200, all my devices are running at 256. I was expecting that based on conversation here. However, the SSD is the only 'max payload' of 256 that is on this PCI Express x8 bus. The other devices are all 'max payload' listing at 512. Which would explain why Shadow is getting 512, because he probably has one thats negotiating it properly. Whatever is lowest wins so to speak.

So, the BLUF is it looks like whatever is on the same BUS (#2 for me) will be limited to the lowest Payload size, either 128, 256 or 512.

Edit; Correction, the USB controller is 256 too, so this doesnt really prove that this is the case, although I still think this is the case. I'd be very curious to see @Shadowarez HWinfo screen cap to see how his is laid out, if he is actually getting 512. If my theory is correct, I dont believe he is getting 512.

 

[clshades]

I'm.gonna invade your houses and put y'all back on spinpoint drives

 

[Solace]

Disabling the WIFI did not improve my speeds on the 970 evo.

View attachment 33742

So I think this is a negotiation limitation actually, based on that, and what HWinfo is saying. Why?

Now that I disabled the AX200, all my devices are running at 256. I was expecting that based on conversation here. However, the SSD is the only 'max payload' of 256 that is on this PCI Express x8 bus. The other devices are all 'max payload' listing at 512. Which would explain why Shadow is getting 512, because he probably has one thats negotiating it properly. Whatever is lowest wins so to speak.

So, the BLUF is it looks like whatever is on the same BUS (#2 for me) will be limited to the lowest Payload size, either 128, 256 or 512.

Edit; Correction, the USB controller is 256 too, so this doesnt really prove that this is the case, although I still think this is the case. I'd be very curious to see @Shadowarez HWinfo screen cap to see how his is laid out, if he is actually getting 512. If my theory is correct, I dont believe he is getting 512.

The evo will not see any gains, 128byte bus will be sufficient for PCIe 3.0 4x speeds, PCIe 4.0 drives seem to bottle neck around 3000+ from most other user posts.

Yes the lowest device will set the limit like legacy devices on wifi networks in the older days

The same reason I asked for photos after hearing 512byes on both bus, seems unlikely unless he clicked the wrong one in the wrong tree and it was the GPU instead of PCH bus. If it really is something else then a specific windows version might have patched it.

I bought 2 new wireless NICs to try, both are AX210, one PCIe, the other to replace the M.2 onboard chip. I'm thinking considering its supporting 6e tri band that the throughput will be higher and 256bytes minimum. Apparently realtek wirless devices set this at a minimum. Likely vendors failed to realize the bottle neck and no one has any aptitude around testing methodologys and lack resources to test stuff.

 

[Shadowarez]

read it wrong from the looks of this more in depth the first pcie drive is 512kb everything else inc the gpu is at 256. read the top lines as 512kb supported but right under it 256.

 

[Solace]

read it wrong from the looks of this more in depth the first pcie drive is 512kb everything else inc the gpu is at 256. read the top lines as 512kb supported but right under it 256.

Ah yes I missed the earlier post where you said you had all other stuff disabled already, if you were to enable the AX200 wifi module and re-test then likely be 128 instead of 256. It would be a good piece of intel as you have Sabrent drives to see if they have the same issue. You can still use LAN internet, you do not need to actually connect to WIFI or anything of such. Its just on/off in the bios determines the negotiation for the bus.

 

[Shadowarez]

I know wonder if all thes drives that use the same phison controller suffer the same issues. All the 7000+ 6500+ read / write all use the same controllers.

 

[Solace]

Pending your results ,

I know wonder if all thes drives that use the same phison controller suffer the same issues. All the 7000+ 6500+ read / write all use the same controllers.

WD/SanDisk NVMe G2 is used for the SN850 which was the first drive expressing the issue. I wish I had an older SN850 to test this with, or if someone else did. There may have been issues with the drive negotiation being set to 128bytes rather than another device tanking the setting such as the wifi adapter.

Ill at least be testing the AX210 to see if its also still 128bytes this weekend as a workaround. This is likely a niche issue and would just be gone in the next revision of wireless chips or whatever else is intergrated onto a motherboard. Supposedly thunderbolt ports cause the same problem as well if you have one enabled.