Zenith Extreme & 1950X ... 3rd slot stuck at 8 lanes

Hardware Motherboards
1

This has been quite frustrating. For a while now (and possibly since the very beginning) I have known this system to be very stable, with one quirk. The 3rd slot (which is a x16 slot) only connects to the GPUs I put in it at 8x.

The problem here is the standard GPU slot (slot 1) is completely obstructed by the NH-U14S TR4-SP3 cooler. Yes I have adjusted its mount with the 6mm offset it provides, and this still is short by probably 4mm for a typical backplate-having GPU to fit in that slot. I’d need to rig some janky kind of riser dealio to make it fit, which it really won’t given the case I have this in now (or would intend to put it in).

I’m not sure what I’m going to do once I reseat the CPU and this issue persists. Yes I have updated the mobo’s bios. Yes I have reset CMOS and lost all my clock speed tweaks. I will probably have to just accept it for what it is. I just really don’t like this because i’ve invested a good amount into the machine. It’s got 96GB of RAM. I thought I was being clever getting ECC Samsung B-die memory for it. now I realize i got slapped with 1/2 the expected PCIe bandwidth on the system optimized for I/O.

What could cause it? I have carefully inspected the slot on the motherboard. Multiple GPUs all run at 16x in other machines and at 8x in this slot on this machine. What would be nice is if I can use a pcie riser to confirm that the first slot runs at 16x…

image
image1280×958 245 KB

2

what GPU is it

3

How thick is the backplate on your GPU?

I also have a Zenith Extreme and the U12S TR4 cooler, and I have my GTX 1080 in the first slot.

4

Clear out the motherboard PCB face up, then follow this ~3:30 minute mark https://www.youtube.com/watch?v=M1b_b89ZEvs Grab a Multi-Meter dial into continuity through Resistence (KOhm) mode and check the trace above the third “16X” PCIe slot. (Additionally, video is in Russian. I used CC + auto translate into Anglais).

If such is the case that the PCB spots are not reporting back the 6.48 Kohms, you may have a short. Another advisement, this was performed on a PCIe 2.0 spec board. But I am taking an educated guess that if you are getting varying readings at any point motherboard may have a short. If you cannot RMA the board, change your CPU cooler to fit into the first slot or sell at the board (describing the issue to the new prospective buyer) and take a hit and be more mindful in the future. Good Luck.

5

Maybe you can manually set the pci-e link speed on the said slot to 16x in the bios.

Another thing you could check out when booted into the Os,
run GPU-Z, and start a Unigin Valley / timespy or whatever benchmark,
and check if the pci-e link speed then goes to 16x.
Because when a gpu isn’t really used much, it can also default back to 8x speeds,
this is a power saving feature.
Because in most normal cases the pci-e link speed to a x16 slot is set to automatic in bios.

6

reseating the CPU could help, as having it off slightly will disable PCI-E lanes and memory channels but still partially work

2 Likes
7

Thanks for the suggestions. Tested with both a 980ti and a 1080ti (both have solid backplates) and I have another 1080ti, as well as a 670 that I can test with. But it is already clear that the problem is mobo side, as the two tested GPUs all work at 16 in my x99 system.

I’m going to be doing a 4950x/3080 build soon which will obsolete this particular threadripper in a sense but that is entirely beside the point. I want to be using this zenith extreme for NAS/VM/CI/etc I/O heavy purposes far into the future, it’s not like it was cheap or anything.

8

U14S cooler. The U12S had better provide the clearance…

9

It’s my bad, I mistyped the model in my OP. Corrected.

1 Like
10

Can you get the board RMA’d/ contact the support?

11

Yeah I’m thinking that it is still covered under teh 3 year warranty. We are just over 2 years into it.

12

I have now completed the CPU reseat. There is no difference, still connecting at 8 lanes on the third PCIe slot.

I also inspected the CPU pads and although there are 4 thousand pins i am pretty sure i scanned over all of them and all appeared to have a little mark, indicating contact with a pin. I carefully inspected the LGA as well and it has no problematic looking pins. Screwing the retention clip on (the step where you screw 3 screws using the special screwdriver) was just as difficult as the first time. it’s hard to get the screws to catch from both sides. But it was manageable.

The ASUS site is pretty weird. One time when entering in the RMA page, I got it to tell me that the warranty expired on august 30 2020, which does not make sense since I bought this thing on June 2018. I sent an email through the site, we’ll see what they say when they get back to me. i should be in no hurry as I have other machines to use, and in fact may continue to run it as it is certainly quite stable. However this is unlikely. since I am not satisfied with this situation on a $450 board I’m preparing to be able to send it to get fixed so I have transferred the m.2 drive to my 8700K machine for now.

13

I got this response:

I appreciate your patience and for giving us time to come up with a resolution for your case.
I am writing this email to provide you an update about your ongoing case. Here’s the recommendation of our engineer.

We strongly recommend not to use PCIEX16_3 as primary display slot. Unfortunately, the PCIEX16_3 slot runs at 8x if GPU is installed as primary.

I don’t find this response particularly satisfying. I replied with this:

Thank you for the clarification. Yes I think I was having trouble finding anyone else on the internet to report on single GPU usage with the 3rd slot.
Could you provide further confirmation for me? Based on this information, does it mean that I can expect the 3rd slot to engage at 16x ONLY IF the first slot is also populated? When I populated the first x16 slot with a low profile adapter for a (4 lane) m.2 NVMe adapter, the 3rd slot was also still limited at 8 lanes. I will obtain a PCIe riser extension in order to use the top x16 slot in any case.

14

Well that is just Odd… I could understand some boards switching a lower slot to fewer lanes if a higher slot is active, but that shouldn’t even be a thing with the number of lanes coming from the CPU.

15

Well… I’ve done some testing.

I got a x16 pcie3 riser cable from Amazon (Asiahorse, was $20). I used that with the GTX 670 card on the top slot, and this connected at x16.

Once I put the 2nd GPU (1080ti) in the 3rd slot, it still connected at x8 there. However, the BIOS page does not show it in red anymore. This is weird.

image
Yes I put a folded microfiber cloth under that GPU so it would not short against the HSF.

image
image1280×958 231 KB

Very sorry. no idea why iOS shows me the photo the right way up but it ends up upside down when uploaded.

I simply cannot work out what’s going on with this. All I want to know is if i can run a bifurcated x4/x4/x4/x4 nvme card in one or both of these slots… It is looking like i would need to test it.

Also the 670 appears to no longer be stable. it is acting really weird. Freezing in windows every few seconds and preventing booting into Ubuntu. Could be a coincidence that this is the beginning of the end of that card. But I’m now actually a bit afraid to test the 1080ti with this riser cable on the top slot now. Not to mention without designing some kind of mount, having the graphics card just hanging out is not ideal at all.

I am pretty sure that i’m getting nowhere with the ASUS email support. I am just really tired of dealing with it by now. I sent a detailed email, they told me to use either the first or third slot (proving they didnt read my email), I reiterated, they sent me a word doc requesting the same details I already provided, and I complied with that.

What I posted already was the first response, to repeat it here again:

We strongly recommend not to use PCIEX16_3 as primary display slot. Unfortunately, the PCIEX16_3 slot runs at 8x if GPU is installed as primary.

I replied:

Thank you for the clarification. Yes I think I was having trouble finding anyone else on the internet to report on single GPU usage with the 3rd slot.
Could you provide further confirmation for me? Based on this information, does it mean that I can expect the 3rd slot to engage at 16x ONLY IF the first slot is also populated? When I populated the first x16 slot with a low profile adapter for a (4 lane) m.2 NVMe adapter, the 3rd slot was also still limited at 8 lanes. I will obtain a PCIe riser extension in order to use the top x16 slot in any case.

The next day I received:

If 1st slot is occupied, the 3rd slot will switch to 8x

My response:

Thank you for your response. This information cannot be correct as it contradicts information on the internet as well as the ZENITH EXTREME manual itself.
The threadripper (gen 1) platform provides 64 PCIe lanes, and the slot distribution is x16, x8, x16, x8, in that order, for the 4 full size slots on the board. The third slot (which I am trying to use) is a full x16 slot, and should run at full 16 lanes, ESPECIALLY if the slot 1 is also populated with a x16 card.
I am currently testing and will report back soon with whether or not the first x16 slot will connect a GPU at x16. And we shall then see (once I come up with a way to fit things in the case safely) if the third slot can connect at x16 if the first slot is filled.
I am disappointed in the product support service that I have received thus far. I hope that we can escalate this discussion further.
Thank you.

This whole thing is just going nowhere. I doubt over 2 years after release anyone over there really cares about this board anymore. And I’m starting to feel the same way myself. They’re also making a mess of the email ticket numbers. I think I have 4 different ticket numbers. I reply to the email like they ask, and then some different rep ends up replying to me on a new ticket.

In a few weeks we will get AMD’s announcements, and I am sure that I will feel better once I do a 5950X build (I may also entertain the idea of a threadripper 5970X build) so that I can get rid of this 1950X setup…

Before I was annoyed that it seems to not work as promised, but at this point I’d rather just use all these cores on this CPU, machine is stable anyway and all ram slots work… I had swapped over to my delidded 8700K running 5Ghz but going from 16 okayishly fast cores to 6 quite fast cores is a downgrade overall for my work (which involves a lot of code compiling sometimes). The other thing was some driver annoyances with running the OS drive in the wrong system (does not have a performance impact though, been gaming a little here and there).

I dunno. I need to unload some of my hardware so I can spend more time focusing on the rest of my life because it is too easy to tinker with hardware especially with 3 PCs like I have right now. The other one is a 5820K that is humming along as a NAS. I was planning to upgrade that with the threadripper but I honestly just don’t “trust” it as a platform at this point.

I think once I build a Zen 3 setup I’m going to want to liquidate both of these machines.

Thinking about it a bit more, it is potentially ideal to have a 3 PC setup. One is a NAS and mostly runs headless, one with many cores is for work, and one with faster cores is for games. And preferably that last one is SFF.

If I can make SFF work with zen 3 threadripper I will probably go down that path and end up with a single PC for work and play. The main problem with this is it’s annoying to wait for reboot and starting things back up… so I’ll be looking into that VFIO goodness. I really hope I’m not hyping Zen 3 up too much to turn it into a disappointment.

16

Are they going to amend their manual to version 1.01.pdf? If their own conclusion is that the third slot becomes an x8 electrical device if first slot is populated needs to be known (ask for a refund due to product not fitting the needs of your situation and misrepresention of features).

17

Yeah the third slot never achieving 16 lanes proves that it’s defective. But I just feel like there is a slight possibility that the CPU could be at fault.

18

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.