A good Ryzen server motherboard at last?

[Ericloewe]

Now that you mention it, the CPU power delivery stuff does look somewhat puny. Maybe a step below what is typical for Intel C2xx boards, even though higher power levels are expected.

 

[Matthew Dillon]

Now that you mention it, the CPU power delivery stuff does look somewhat puny. Maybe a step below what is typical for Intel C2xx boards, even though higher power levels are expected.

Yah, it doesn't look all that good. A quick google didn't give me the skinny on the VRM configuration but it really looks quite wimpy.

Ok, I decided not to wait for the second 3900X to come in. I ripped the one I had out and stuffed it in there. I realized that the BIOS settings for the power and clock management with the 2700X stuffed were completely different than what the would be for a Zen 2 chip. So maybe the BIOS code for the Zen 2 chips would work better. And, indeed, I am getting much better results. I am able to run the 3900X with 'Platform Power Setting' in CBS/NBIO/SMU set to 120W (and not messing with cTDP which might be more of a failsafe field). The watt meter is showing 150-160W at the wall under full load and it seems to be sticking there without any need for excess cooling.

Unfortunately I can't get exact temperatures because the BIOS/ACPI isn't providing them on the Zen 2 chips yet... its still an early beta-like BIOS for the Zen 2 support. But the IR thermometer seems to indicate reasonable temps.

--

I did one other thing. I actually removed the Noctua low-profile cooler and put in a Stock 3600X cooler (the lower-end one... the high-end one that comes with the 3900X is too tall for a 2U case). I had to saw-off a bit of the plastic cowling to get it to fit without hitting the DRAM, and the heat sink still impinges on the first DDR slot a little from the side, but actually a bit less so than the Noctua did and the fan has more headroom because the cooler is shorter, overall. This appears to work ok with the platform power setting set to 120W. It's not the best cooler in the world but it can handle 120W in the socket as long as case airflow is reasonable.

The advantage of the stock cooler is that it is a box cooler that pushes air directly onto the VRM heatsink from the side, so it looks like the VRMs are not overheating anymore. I'm running with the box cooler and four 80mm fans in a 2x2 configuration (CORRECTION ON PREVIOUS POST: I meant to say 80mm, not 40mm in the previous post).

I am doing a 24-hour burn-in now. I'm still not enamored with this board, particularly the primitive state of the IPMI and the very early BIOS, but... it might actually work for us. If it survives the burn-in without catching fire .

-Matt

 

[Ericloewe]

I just noticed there's a new variant with 10GbE onboard: https://www.asrockrack.com/general/productdetail.asp?Model=X470D4U2-2T

The only compromise I can identify in a quick read-through is that they got rid of the extra SATA controller, which is of dubious value anyway.

 

[Constantin]

I just noticed there's a new variant with 10GbE onboard

That’s nifty though I prefer SFP+ over a copper 10Gbe interface for a number of reasons. But AsRock may have run out of space to accommodate a SFP+ module!

 

[Matthew Dillon]

Oooh, very nice, though they still haven't fixed the weak VRMs or the DDR4 socket spacing being too close to the CPU. It does look like all they did was pull a few PCIe lanes away from the SATA to plug the 10Gbe in. It's unclear whether that BIOS for the new mobo supports Zen 2. The original went through a 1.50 and then a 3.04 BIOS to get Zen 2 support. The new board is on 1.40 BIOS but the BIOS is dated more recently than the 3.04 Zen 2 BIOS for the old board.

My burn-in tests with the X470D4U succeeded. With the 3900X in there the CPU cooler more or less needs to run at full speed for anything that isn't full idle (keeping temps down is very important to maximize Zen 2 performance). The VRMs did ok with 120W in the socket (160W at the wall) though they definitely require the direct airflow they get from the downdraft box cooler I'm running atm. Basically the burn-in test consists of running a bulk package build with synth on DragonFlyBSD... basically building the FreeBSD ports tree with maximal concurrency. Around 30000 packages and millions of files. Takes around 24 hours to complete. It's a great test because if anything at all goes wrong it's really obvious in the error logs whether it is related to CPU/stability or not.

Obviously OC'ing the CPU is IXNAY, the VRMs aren't beefy enough, but the 3900X does run close to its maximal power-performance point at 120W in the socket so I'm happy. In contrast, the mobo actually had a much harder time with the 2700X than with the 3900X due to not having the more advanced thermal/power management algorithms for Zen+ that it has for Zen 2.

Since OCing memory is really easy to do and costs basically nothing, I ran a second burn-in with 64G worth of 2133 ECC UDIMMs OC'd to 1.34V @ 2666C15 and it ran without issue, boosting general performance by around 10% (which is expected for Zen/Zen+/Zen2 CPUs... they really like fast memory), and memory performance by 28% from 2.8 MFault/sec to 3.6 MFault/sec on an all-cores zero-fill fault test. The BIOS is still really primitive, it took a while to figure out the right OC settings for the memory. In this BIOS, under 'AMD Overclocking' set the actual memory frequency (which is half the marketed frequency)... so 1333 in order to get 2666 equivalent, and the voltage setting in manual mode to '10' assuming roughly 14mV per division. 1.2V + 10 * 0.014V = 1.34V. The BIOS is unclear but those settings seemed to work. Be careful not to OC memory above 1.35V.

-Matt

 

[jmpage2]

Hi all,

I registered specifically to discuss this motherboard as it's pretty new and there aren't too many sources of discussion.

I am strongly considering this as an alternative to the refurbished Intel based server I have been considering for some time. This appears that it would be more powerful for my needs (hypervisor with strong single thread performance), less expensive and less power consumption (I would couple with Ryzen 3600 CPU).

Is there any reason that a Dell Perc H330 RAID controller would not work in this motherboard? Can anyone make specific recommendations around ECC compatible RAM that is performing well? I would need to passthrough a Quadro P400 or similar GPU for Plex transcoding and am interested if I could get it to work under a hypervisor.

The box would be replacing an aging HP i7 based workstation that currently runs Windows.

Thanks

 

[MikeyG]

@jmpage2 I will be receiving one of these in about a week or so, and pairing it with a Ryzen 1700, Crucial ECC memory, and an LSI 9305. I will check back in when (hopefully not if) I get it up and running in a few weeks.

 

[jmpage2]

@jmpage2 I will be receiving one of these in about a week or so, and pairing it with a Ryzen 1700, Crucial ECC memory, and an LSI 9305. I will check back in when (hopefully not if) I get it up and running in a few weeks.

Awesome! Just don’t know how long I can wait. Trying to find the “perfect” server chassis might keep me occupied for awhile.

 

[Matthew Dillon]

Hi all,

I registered specifically to discuss this motherboard as it's pretty new and there aren't too many sources of discussion.

I am strongly considering this as an alternative to the refurbished Intel based server I have been considering for some time. This appears that it would be more powerful for my needs (hypervisor with strong single thread performance), less expensive and less power consumption (I would couple with Ryzen 3600 CPU).

Is there any reason that a Dell Perc H330 RAID controller would not work in this motherboard? Can anyone make specific recommendations around ECC compatible RAM that is performing well? I would need to passthrough a Quadro P400 or similar GPU for Plex transcoding and am interested if I could get it to work under a hypervisor.

The box would be replacing an aging HP i7 based workstation that currently runs Windows.

Thanks

Any PCIe card should work but in terms of pass-through that really depends on IOMMU partitioning provided by the hardware and BIOS. I don't know the answer to that on this mobo but it is often the case that early BIOSes do not set up the IOMMU partitioning in a reasonable way, so YMMV. Zen 2 itself should have pretty good IOMMU and SRIOV support. This is one of those situations where the only way to know for sure is to throw the stuff together and see what actually works.

-Matt

 

[MikeyG]

Update: Got the X470D4U installed. Paired with Ryzen 1700, 4X Crucial CR16G4WFD824, LSI 9305, and Chelsio T520-SO-CR.

Had issues at first with stability - it would not boot further than the BIOS with all 4 sticks installed. The memory is rated for 2400, but I set it manually down to 2100, and then it was fine. I believe the user guide does mention max speeds around here for fully populated memory, so not too surprising.

Only been up for about 48 hours now, so a little early to tell, but so far no problems.

As others report, using stock AMD cooler the memory is pretty close to the heat sink, but my memory modules don't have heat sinks themselves and so they don't touch. I would not recommend using memory with heat sinks on this board.

I have 2x nvme drives connected. An adata and a Samsung 970 pro. Both are limited to a little less than 2GBps throughput which is per spec from the user guide. For my use, not an issue but something to keep in mind.

IPMI looks fine to me. A little bit prettier than Supermicro. No issues there.

If I run into anything interesting I will update this thread.

 

[JedixJarf]

Update: Got the X470D4U installed. Paired with Ryzen 1700, 4X Crucial CR16G4WFD824, LSI 9305, and Chelsio T520-SO-CR.

Had issues at first with stability - it would not boot further than the BIOS with all 4 sticks installed. The memory is rated for 2400, but I set it manually down to 2100, and then it was fine. I believe the user guide does mention max speeds around here for fully populated memory, so not too surprising.

Only been up for about 48 hours now, so a little early to tell, but so far no problems.

As others report, using stock AMD cooler the memory is pretty close to the heat sink, but my memory modules don't have heat sinks themselves and so they don't touch. I would not recommend using memory with heat sinks on this board.

I have 2x nvme drives connected. An adata and a Samsung 970 pro. Both are limited to a little less than 2GBps throughput which is per spec from the user guide. For my use, not an issue but something to keep in mind.

IPMI looks fine to me. A little bit prettier than Supermicro. No issues there.

If I run into anything interesting I will update this thread.

Are the IOMMU groups split up properly? Are you able to pass GPU's through to VM's?

 

[MikeyG]

Are the IOMMU groups split up properly? Are you able to pass GPU's through to VM's?

I'm afraid I'm not super familiar with IOMMU other than that it's a thing for memory addresses in virtualization. I think SVM is actually turned off at the moment, as I'm not using any virtualization functionality.

If there is something you'd like me to check in IPMI or FreeNAS (don't really want to reboot to get into the BIOS) though let me know!

 

[firsway]

Just seen this thread. For info, I've got my FreeNAS 11.2-U6 working quite nicely, using the following configuration;

Asrock X470 Master SLI
Ryzen 5 3600 CPU with the default cooler
32GB Samsung ECC unbuffered RAM (across 2 slots)
5 x 6TB 7200rpm SATA3 Toshiba
1 x M.2. 128GB NVMe for SLOG

The board had a bit of a wobbly at first, took ages to POST, but then discovered that it could have been because a 2nd M.2. NVMe that I had configured as a cache (mounted on a PCIe x4 converter and placed in one of the x16 slots) for some reason showed some issues. Not got to the bottom of that just yet. It could be damaged, or it might have started having trouble working alongside another PCIe x 16 mounted Graphics card.
Working just nicely now though, although a known bug (resolved in 11.3) prevents you from seeing the CPU temperatures at present.

 

[CompuGlobalHyperMegaNet]

The (fantastic) youtube channel Gamers Nexus has just released the first part of a two part video where they build a NAS with a little help from Wendell of (the also fantastic) Level1Techs. The build is based around the X470D4U2-2T. The video does cover some of the features of the board and is well worth a watch in it's entirety but the linnk below takes you to the point in the video where Steve and Wendell start talking about the mobo.

Building the Ultimate DIY NAS on AMD, Ft. Level1Techs, Unraid, & ZFS (Part 1/2)

We're finally building our own NAS / media server, and we're using Unraid with ZFS to do it. This first part is all about the hardware. Wendell from Level1Te...

youtu.be

 

[CompuGlobalHyperMegaNet]

[EDIT] The second part of the video, linked below, does not contain any further info about the X470D4U2-2T. I'm going to leave this comment up though, to save you time just in case you're only interested in watching the videos purely for info on the motherboard.

Part 2 of the video is up. I haven't watched it yet though so I can't comment on it's content.

Part 2: DIY AMD NAS with Unraid & ZFS Software Setup, ft. Level1Techs

This is part 2 of our journey embarking on a custom DIY NAS, this one focusing on software setup for Unraid & ZFS (ft. Level1Techs). Sponsor: Get Linode Clou...

www.youtube.com

 

[NickF]

Just as an FYI. I have this board, the X470D4U variant, and I have been having nothing but issues. I've done various configurations with 3 separate kits of RAM with different speed settings and 3 separate processors.

-The system was unstable and would crash periodically every 24 hours or so with 4*8= 32GB of DDR4 non-ecc 2400 single rank Samsung e die and a Ryzen 7 1700. This memory and cpu are now stable and have been running for about a week on an MSI X370 gaming Pro Carbon running my Hypervisor.

-The system was unstable and would crash with that same memory and a Ryzen 5 3600. This further confirms that the Ryzen 1700s weaker IMC is not to blame.

-The system was more unstable with 64GB dual ranked memory, even with the 3600 and even when running at 1866mhz memory clock.

-The above CPU and memory are now in my FreeNAS system which now has an ASRock Phantom Gaming 4 motherboard (which has two 221100 m.2 real pcie4 x4 slots)

-During this adventure I canabalized my main desktop trying to find a winning combination. More or less my old PC is now my VMware host referenced above. So I bought a Ryzen 5 2600 because I got it for $100. I put my GSkill FlareX b-die memory in the ASRock rack board and put it in that box.i am still having memory issues. Windows doesn't see both my sticks of memory right now for some reason, but the BIOS does

-the IPMi is great, that is the only good thing I can say about this board. It is very discouraging that consumer motherboards are more stable, even a nearly 3 year old one,

Ryzen seems to work pretty well in FreeNAS. Perfoanfe is great. However, I may swap the Ryzen 5 2600 with the Ryzen 5 3600 I'm using in FreeNAS now because even on FreeNAS 11.2 beta 1 the GUI doesn't seem to be showing temperatures for 3rd gen Ryzen.

I hope this two week adventure and hole in my pocket saves someone some time

 

[Ericloewe]

Was that with the latest firmware?

 

[subhuman]

I did add more voltage to SoC which AMD recommends. Its sitting at 1.1V. RAM on the other hand is more heavily overclocked. I am using factory timings to get the 3600MHz and to increase stability I have increased a bunch of RAM voltages manually. I could list them here but it would be of no use I think. Overclocking RAM would only make RAM heat up more I don't think it would affect CPU temp drastically like that. Correct me if I am wrong.

Yeah, old post- I know.
Infinity Fabric. It runs at 1:1 with DRAM speeds up to and including 1800MHz (which is what you meant by that 3600MT/S). After which, it drops to a 1:2 ratio*. Anyway, you're at what is usually the limit, very few overclockers- even on liquid nitrogen- have managed to get the Infinity Fabric stable over 1800MHz. Bear in mind, you are running an OC memory speed- which isn't even listed as supported on that board! Unlike cores, memory clock doesn't vary. So it's always running at that speed.

*that's behavior for Zen and Zen+. Zen2 on a x570 chipset allows you to manually override it in BIOS