Lenovo Thinkcentre/ThinkStation Tiny (Project TinyMiniMicro) Reference Thread

[besseddrest]

Has anyone successfully got a GPU to work in the M90q Gen 3?

The RX 6400 I ordered arrived today, however when in the M90q Gen 3 the fan on the RX 6400 spins up when powered on, it will never output anything on the Bios and is not detected in Windows. (The Bios / Windows will still output on the iGPU with the RX 6400 present).

I’ve attempted changing the active video from Auto to PEG for PCIe, along with changing the PCIe version in the Bios between 3 and 4.

• 5C50W00933 / BA7K78 Rev1.2 PCIe x16 riser purchased directly from Lenovo
• XFX SWFT105 Radeon RX 6400 Low profile Gpu
• 230w Lenovo power supply

The RX 6400 does work in another machine to rule out that been faulty.

Hey bud I know this is an old post but, check out some of the convo on page 153-154 - i basically had the same issue and another user wanted me to test a theory on how to get this to work. I got my RX 6400 to work on my P3 Tiny Gen 1

 

[Envy907]

I'm kinda lost on which SFP+ NIC to choose for my p330. There's way too many options. What are the popular picks these days?

 

[WifiCable]

i mean, i guess you can say it was difficult; but it's more like i could see this being extremely frustrating for those who are a bit more clumsy. After you cut something so small even the force of your cutting tool can flick away the piece you were targeting, lol. Eventually i managed a technique with the pointy ends of some spudgers

Thanks for the feedback and pictures on that, I guess LGA1700 pins are a lot smaller than previous generations. Your picture does make it look quite tiny.

Thanks again - i actually forked your repo and i can commit an image for you there, i was going to add a line here and there for m75q gen 2, maybe any notes i have re: P3, dGPUs

Sure, any useful notes you have are welcome. If you submit a pull request to the repo it might take me a while to get it through, I'm still no expert on github's more complex features and I'll probably want to do some rewording and rearraning to keep things consistent.

But I'd be curious to know what extra features you have in mind, as I'd been looking out for custom risers for the P360.

For now, just the usual, likely same as Tinyriser6. PCIe slot, NVMe slot, USB port, a breakout for whatever random io happens to be in the riser slot. I have some fitment issues to figure out regarding the NVMe slot, the same place as it has been on previous versions isn't going to work. There isn't enough space available between the end of the riser slot and the chassis intrusion switch. I'm also gonna have to learn how to build a 20 to 12V buck regulator, that's not something I want to do without really knowing what I'm doing as a bad design risks frying expensive PCIe cards.
I kind of would like to be able to get some more community feedback while building stuff like this, I've been considering setting up a discord server or something for discussing Tiny mods/development and other stuff I'm working on, but I'm still thinking about how to handle that. Maybe you'll see it in the future.

If I try this out, will higher bandwidth still be available for the B50 to use? I haven't tried a different card in this yet, but your explanation gives me some hopium to huff.

I don't know if this will solve your particular issue, but it shouldn't affect 'higher bandwidth availability' at least.

 

[jja2000]

I don't know if this will solve your particular issue, but it shouldn't affect 'higher bandwidth availability' at least.

Thank you for the reply, I guess it's worth a try when I get to it. If it ends up working out, I'll shoot Intel an email. Maybe they can improve handling of lane reversal with a firmware update (they ship those out of LVFS which is pretty neat).

The B50 is a good card for not that bad of a price. Better than shelling out 1000€ for a similar Nvidia card, but I guess it comes with these growing pains...

 

[Bacterium]

Hello everyone, I'm looking for some volunteers who might want to test a theory on a Tiny8+ machine.
The boardview for the M90q Gen 3/P360 Tiny (NM-D581) recently became available, and after me and some others looked through it for a while we found out the possible cause of the issue where PCIe x4 and x1 cards don't work in the x16/x8 riser on these machines: Tiny8 riser pinout · Issue #1 · qq8322302/P3Tiny-PCIE-Card
Also discussing it here: https://forums.servethehome.com/ind...-2-connector-and-much-more.52689/#post-503708

It turns out that Lenovo has activated lane reversal on the CPU's PCIe lanes for no good reason, which is causing the x8 link on the riser slot to be set up in reverse lane order. For most x8+ cards this seems to work fine as they have their own lane reversal support built in, but for example, if you install a x4 card this causes it to be connected to lanes 7-4 of the x8 link instead of 0-3, which doesn't work.

There are a few ways to address this:
- Remove the pulldown resistor on the CPU's CFG2 line from the motherboard.
- Mask the CFG2 pad on the CPU
- Build a riser that routes the PCIe lanes in reverse

I feel a bit iffy about designing a riser that connects the lanes in reverse, because:
1: There's at least one example of a x4 card working in the x16 riser: youtube.com/watch?v=W4wt1ioIkY4
So it's possible not all machines are affected by this and would instead manifest this issue with a riser wired in reverse.
2: Future machines using the same riser slot might not have the link configured in reverse

So far I've looked at available board images from Lenovo's parts catalog and all pictures I could find of M90q Gen 3, Gen 4 and Gen 5, as well as the P360 Tiny, P3 Tiny and Neo Ultra's boards have the strap resistors set to activate lane reversal, but not all FRUs have images available, so I can't be sure that definitely every board with the Tiny8 riser slot uses lane reversal. This is especially the case for the latest M90q Gen 6, P3 Tiny Gen 2 and Neo Ultra Gen 2 models because on that generation Intel switched to using firmware straps for setting lane reversal, so there's no way to tell visually.

In conclusion, since I don't have a Tiny8 riser slot machine to test with, I'm asking the community for 2 things.
If you own an M90q Gen 3, Gen 4, Gen5, a P360 Tiny a P3 Tiny (Gen 1) or a Neo Ultra (Gen 1),

1: Would you be willing to test if either removing the marked resistor here (invasive): View attachment 48756
or masking this pad on the CPU (socket perspective in the picture, less invasive): View attachment 48757
fixes compatibility with x4 and smaller cards when using the x16 riser?

2: If you don't wanna test an unverified mod on your machine, it would still be very helpful if you could take a picture of the CFG resistor area next to the CPU socket (the area in the resistor removal picture above), I'm hoping to collect pictures from as many different machines as possible to find if there are any outliers which might have them set differently than I've seen so far.


This information would both help people who want to use x4 and x1 cards in the original x16/x8 riser, and would also be very useful for me and other custom riser developers to choose how to proceed.
Thanks a lot in advance!

Photo of my P360 Tiny:


@WifiCable

Testing so far...

System: Lenovo ThinkStation P360 Tiny
Riser: Lenovo PCIe riser 5C50W00910

Device	Without Masked Pad	With Masked Pad
Nvidia T400
Intel X710-T4L
NVMe drives (Sabrent M.2 to PCIe adapter)

 

[danman218]

After some further tests and catching back up with the thread here:

It seems my x16 riser in the first test has bit the dust, and was why the dGPU wasn't showing up. Oof (probably from messing around with the enable pins a while ago)

Testing with a known good riser, the T1000 worked again,
but the x8 network card would have the system get stuck on the Lenovo splash screen on boot.

I removed the card, went into BIOS and disabled ASPM (credit @jja2000 for this reminder), saved and shut down, reinserted the card, it then booted up just fine and showed up in Proxmox.

The associated network adapters showed up as enp2XXXX instead of enp3XXXX when the card was working on the x4 riser, which should confirm that it is on the correct x16 PCIe bus now.

This is with the masked pin being 11 along the longer side of the CPU and 9 in from the shorter side.

Sorry for the blur, camera effects. the other should be decent.

 

[Wasmachineman_NL]

I'm kinda lost on which SFP+ NIC to choose for my p330. There's way too many options. What are the popular picks these days?

Forget about Mellanox anything if you want proper ASPM support. I myself am running a X550-T1 in the NASeedbox and it works just fine on 10G, and still goes to C10.

(still want a X(XV)710-DA2 so I can use the SFP+ port on my BE88U)

 

[Squoblat]

Some more mad science experiments aren't going so well - I've got an M75q Gen5 and have just fitted an Oculink port to the spare NVMe slot. GPU in a cradle powers up with the M75q, but it isn't detected in hwinfo or lspci. BIOS has no options for allowing external video, PCIe BIOS settings are at "auto" but it isn't detecting. PCIe messages are obviously getting to the Oculink dock, as it's powering up on command.

I've cycled through every option in the graphical BIOS, not entirely sure where to go from here. Anyone tried fitting an Oculink to an M75q gen5 before?

Edit: Possible cause - looks like both of the M2 slots are provided by the chipset, rather than the CPU. Can anyone confirm this?

Edit 2: M2 ssd slot 1 doesn't work with the oculink adapter at all, M2 ssd slot 2 leaves me back at square one. It will communicate with the oculink dock, power up the GPU but still nothing in terms of detection on the PC itself.

Edit 3: after watching some Jeff Geerling, I've ordered an adapter with a ReDriver chip. Will report back on Thursday.

 

[muetto]

Hi everyone,

Inspired by the awesome Project TinyMiniMicro threads here, I'm currently working on upgrading a Lenovo Tiny into a high-performance networking node. I want to make sure I have the exact right parts before assembling everything, and I've hit a roadblock regarding the rear PCIe bracket (baffle).

Here are the precise details and links for my hardware setup:

• System: Lenovo ThinkCentre M90q Gen 5 Desktop (Type: 12TH)
• Machine Model: 12THCTO1WW
• Serial Number: GM0S1WNK
• Product Reference: Lenovo PSREF - Product Specifications Reference
• PCIe Riser: Official Lenovo Taisol Riser card AIC_PCIEx8 (Part Number: 5C50W32125)
• NIC: Intel Ethernet Network Adapter X710-T2L (Dual-Port 10GBASE-T RJ45)
• NIC Specifications: Intel® Ethernet Network Adapter X710-T2L - Product Specifications | Intel

I have the PC, the riser, and the network card ready, but I am missing the correct rear PCIe baffle to secure the X710-T2L and cleanly close the back of the chassis. Since the Intel X710-T2L features dual RJ45 ports, I need a bracket that matches this specific layout for the Gen 5 chassis.

I would really appreciate some help with the following questions:

1. Does anyone know the exact official Lenovo FRU part number for a dual-port RJ45 baffle that is fully compatible with this specific machine type and hardware combination?
2. If an official part is too difficult to source, does anyone have a link to a verified 3D-printable .STL file or a reliable aftermarket seller that makes a dual-port bracket known to fit this exact setup?

Any precise guidance, verified part numbers, or direct solutions from anyone who has experience with this hardware configuration would be massively appreciated!

Thanks in advance!

 

[MBastian]

NIC: Intel Ethernet Network Adapter X710-T2L (Dual-Port 10GBASE-T RJ45)

This card is 167mm long. Are you sure it fits? I have a m90q Gen 5 too and there is an internal bracket that would need to be removed for anything over 154mm

 

[loheiman]

Curious, has anyone found a 10in rack mount that has this tiny pc facing backwards, rear out? I use mine with a 4 NIC card and would love easy access so its ethernet ports.

 

[muetto]

This card is 167mm long. Are you sure it fits? I have a m90q Gen 5 too and there is an internal bracket that would need to be removed for anything over 154mm

I didn't think there would be any length issues because I opened up the PC and it looked clear of any obstructions.
I had also considered using the Intel X550T2 network card, which is even shorter, but I saw that the chip is at end of life and the card itself will soon be at end of life as well, so I wanted to get something more recent and modern.
Are you sure the 167-millimeter network card is too long?

 

[Bacterium]

I didn't think there would be any length issues because I opened up the PC and it looked clear of any obstructions.
I had also considered using the Intel X550T2 network card, which is even shorter, but I saw that the chip is at end of life and the card itself will soon be at end of life as well, so I wanted to get something more recent and modern.
Are you sure the 167-millimeter network card is too long?

I can fit an Intel X710-T4L into my Tiny P360. I did have to remove a bracket near the front of the system.

It refuses to boot with it installed however!

 

[MBastian]

Are you sure the 167-millimeter network card is too long?

It is there and will block your card. Just open your m90q and try it out. It is removable but personally I am loathe to do it. For now I use an i350-T4 but would like to switch to an X710-da2 in the future. Problem for me is that no EU available card that would fit without removing the bracket supports ASPM.

 

[MBastian]

The i350 is the opposite end, it holds the link active and keeps the package in a shallow C-state no matter what the rest of the box wants to do, which is why an i350-T4 in a ~12W M720q can push you into the high teens at the wall.

I can't confirm that. With an i350-T4 my m920x and m90q Gen5 idle at around 6W, add ~0.3-5 watt per active Ethernet Interface. Just the plain box can reach slightly below 4W at idle with a running Linux OS

Full disclosure, I put it together, so it's mine. It's free and there's no signup. Since most of the numbers came out of this community to begin with, corrections are kind of the whole point, so if an idle figure is wrong for a box you actually own, reply here or open an issue and I'll fix the row.

Oh my...

Update: Aaaand the post was deleted and the user got booted. Good riddance to that low effort LLM fueled crap.

 

[Wasmachineman_NL]

On the ASPM angle, since it keeps coming up with these 10G cards: the thing that actually decides idle in one of these Tinys is whether the card lets the platform keep hitting its deep C-states. Wasmachineman_NL's X550-T1 still reaching C10 is the proof of that, a card with working ASPM L1 barely moves the package power floor. The i350 is the opposite end, it holds the link active and keeps the package in a shallow C-state no matter what the rest of the box wants to do, which is why an i350-T4 in a ~12W M720q can push you into the high teens at the wall. Most ConnectX-3/4 behave the same way. So for Envy907's "which SFP+ for my p330" question, the boring answer that keeps idle sane is the Intel X710 / X550 family, then double-check ASPM is actually enabled in BIOS after fitting it (per danman218's reminder above).
Semi-related and hopefully useful here: I've been collecting the wall-meter idle numbers people post for these boxes into one open dataset, partly because a lot of the good measurements are buried 150 pages deep in this exact thread and I got tired of re-finding them. It's about 50 Tiny/Micro/Mini models with measured idle watts, the onboard NIC chipset (the i219 boxes vs the Realtek 8111 ones, which matters for the Linux/ASPM side), plus ECC and Proxmox passthrough notes, and every row links back to wherever the number was actually measured. Searchable here: IdleWatt — The homelab mini-PC finder for measured idle power and the raw CC-BY data is on GitHub: GitHub - SolvoHQ/homelab-mini-pc-dataset: Open, cited dataset of homelab mini-PC / SFF deciding facts: MEASURED idle watts (not TDP), ECC, NIC chipset, IOMMU/passthrough, AMD reset-bug. Filterable finder: idlewatt.vercel.app
Full disclosure, I put it together, so it's mine. It's free and there's no signup. Since most of the numbers came out of this community to begin with, corrections are kind of the whole point, so if an idle figure is wrong for a box you actually own, reply here or open an issue and I'll fix the row.

I can confirm that even with a X710-DA2 my Precision 3240 Compact still hits C10

 

[Squoblat]

Some more mad science experiments aren't going so well - I've got an M75q Gen5 and have just fitted an Oculink port to the spare NVMe slot. GPU in a cradle powers up with the M75q, but it isn't detected in hwinfo or lspci. BIOS has no options for allowing external video, PCIe BIOS settings are at "auto" but it isn't detecting. PCIe messages are obviously getting to the Oculink dock, as it's powering up on command.

I've cycled through every option in the graphical BIOS, not entirely sure where to go from here. Anyone tried fitting an Oculink to an M75q gen5 before?

Edit: Possible cause - looks like both of the M2 slots are provided by the chipset, rather than the CPU. Can anyone confirm this?

Edit 2: M2 ssd slot 1 doesn't work with the oculink adapter at all, M2 ssd slot 2 leaves me back at square one. It will communicate with the oculink dock, power up the GPU but still nothing in terms of detection on the PC itself.

Edit 3: after watching some Jeff Geerling, I've ordered an adapter with a ReDriver chip. Will report back on Thursday.

No luck with the new ReDriver adaptor, same symptoms. Nothing in lspci or hwinfo. I suspect the problem is in the spi5100_piix4 driver, it's showing as disabled at the top of hwinfo for some reason.

Some more head scratching. I might try replacing the Oculink cable on the off chance it's that. I think I'm at the point of installing the hardware into a different machine and seeing if it works.

Edit: I also have a sneaking suspicion that Lenovo have a whitelist on the M2 slots, but I haven't got anything else I can test that theory with.

 

[joeribl]

No luck with the new ReDriver adaptor, same symptoms. Nothing in lspci or hwinfo. I suspect the problem is in the spi5100_piix4 driver, it's showing as disabled at the top of hwinfo for some reason.

Some more head scratching. I might try replacing the Oculink cable on the off chance it's that. I think I'm at the point of installing the hardware into a different machine and seeing if it works.

Edit: I also have a sneaking suspicion that Lenovo have a whitelist on the M2 slots, but I haven't got anything else I can test that theory with.

I know that there is mention of a chipset in the specs of the Lenovo M75q. However if i look at the pictures of the M75q mainboards, there is no chipset in sight. And if you look at the block diagrams you can find on the internet of the M75q gen 1 and 2, everything is directly connected to the CPU. There is a SuperIO chip for some basic functions not covered by the CPU. For the M75 gen 2, the second M.2 slot is connected to a PCI Switch, shared with the (unpopulated) PCIe x4 slot.

 

[Squoblat]

I know that there is mention of a chipset in the specs of the Lenovo M75q. However if i look at the pictures of the M75q mainboards, there is no chipset insight. And if you look at the block diagrams you can find on the internet of the M75q gen 1 and 2, everything is directly connected to the CPU. There is a SuperIO chip for some basic functions not covered by the CPU. For the M75 gen 2, the second M.2 slot is connected to a PCI Switch, shared with the (unpopulated) PCIe x4 slot.

I have also had a deeper look, the NVMe drives don't appear to be on a chipset. I've tried both slot 1 and 2, slot 1 results in absolutely nothing happening and slot 2 results in the GPU powering on but not being recognised. Not sure why, as far as I can tell the two NVMe slots on the gen 5 should be identical.

 

[joeribl]

I have also had a deeper look, the NVMe drives don't appear to be on a chipset. I've tried both slot 1 and 2, slot 1 results in absolutely nothing happening and slot 2 results in the GPU powering on but not being recognised. Not sure why, as far as I can tell the two NVMe slots on the gen 5 should be identical.

What Oculink dock are you using?