M.2 HBA card with 9 SATA ports, anyone play with this yet?

[TechUnsupport]

It's the one at the size of m.2 2280 with 9 SATA ports on it in a 3x3 grid. I have seen them around $30-$40 on Chinese site and Amazon. One FB user says it's using RTL9101 chip. The PCB have a marking PHDK PH519 on it. As far as I can tell, there seem to be only one chip under the heatsink, so it's likely not using port multiplier. And thus it's likely true that the chip maybe Realtek RTL9101. While there is no info on this chip, we also know that JMB and ASM HBA chip do not goes up to 9 SATA ports.

 

[louie1961]

All the ones I see on Ali Express that look similar say they are PCI 3.0x2 lanes. That means you have a max throughput of 1.8-2.0 GB/second throughput. Nine SATA drives operating at full speed could theoretically reach approximately 5.0-5.4 GB/second of bandwidth, meaning you would be bandwidth constrained with this card. However, since most spinning disks seem to top out at 200-250MB/sec in real life transfer speeds, you might be OK in real life use. The card will likely saturate in heavy sequential tasks (large copies, rebuilds) if many drives are active. But for more moderate or mixed loads (small files, random I/O, fewer drives actively streaming), you’ll see less limitation. Either way this is not a card I would choose, except maybe for a backup array.

That assumes this card doesn't have heat issues and doesn't thermally throttle.

 

[EasyRhino]

a lot of those goofball SATA adapters also use asmedia controllers as well.

(no personal experience.)

 

[p1415]

It's the one at the size of m.2 2280 with 9 SATA ports on it in a 3x3 grid. I have seen them around $30-$40 on Chinese site and Amazon. One FB user says it's using RTL9101 chip. The PCB have a marking PHDK PH519 on it. As far as I can tell, there seem to be only one chip under the heatsink, so it's likely not using port multiplier. And thus it's likely true that the chip maybe Realtek RTL9101. While there is no info on this chip, we also know that JMB and ASM HBA chip do not goes up to 9 SATA ports.

I tried PHDK PH519. Connected 8 SATA HDDs. Unfortunately, running `fdisk -l` couldn't even detect the file systems without ATA read errors. Several minutes later whole controller fell of the PCIe bus and could no longer be visible. The controller exposes something like 30 SATA endpoints even though max 9 are physically connected. With that amount of attention to detail it's not surprising that the device does not work properly.

I however had good experience with PH516 that has 6 ports and ASM1166 controller. Tens of TBs written and read without issues. My server case has very little airflow when CPU is not under load and the heatsink on the controller managed extended disk activity without problems. In my personal experience PH516 is perfect for home lab use case - acceptable bandwidth, low power usage, scalable - PCIe x16 can support SATA 24 disks with 4 x4 bifurcation.

It's worth mentioning that PH516 was designed with a backplate to reduce bending while inserting and removing SATA cables. PH519, at least the one I bought, didn't. The PCB on PH519 is flimsy, so even though I was careful inserting the cables into the card, I could have broken something. The card was not attached to the motherboard during the manipulation, I would have definitely broken the card in half during cable insertion.

 

[celemine1gig]

Very interesting indeed, as the Chipset is mentioned to be "RTL9101", but neither does Realtek list any data, nor is there any entry for Linux, mentioning support for it. So my guess is, that the ACHI driver under Linux picks up on it, but would probably need some workarounds for it to work correctly. And as there are none, this would explain what p1415 saw. Maybe sometime in the future, this will be supported sufficiently. Currenly, as an avid Linux user, I would not try it.

 

[beisser]

i also have the ph519, but from a different seller. the cards are all identical. im using it with proxmox passed through to truenas and have not noticed any kind of issues with it yet. so either my pc is more compatible with the card or p1415 had a faulty card.
i am currently utilizing 6 of the 9 ports.

the 30 sata endpoints are exposed because the chipset (just like the asm1166) supports port extenders/multipliers, which can be connected to each of the sata ports.

i can confirm that the pcb of the card is extremely fragile. i recommend removing the card when plugging in sata cable so that you dont break the pcb.

 

[celemine1gig]

@beisser: Very interesting. What kernel version did you use under Proxmox, that it worked perfectly with?

@p1415: Basically the same question.
What Linux kernel version did you have problems with?

Might be good to know, when this probably began to work error-free. Assuming that it was not a HW issue with p1415, but rather a SW-topic.

 

[beisser]

i dont use it in proxmox directly. it does get detected fine on kernel 6.17.2-1-pve though. im giving it via pcie-passthrough to a truenas 25.10.0.1 VM and its working there just fine.
im not having a super high load on it though. truenas is used as shared nfs-storage for my proxmox nodes and just hosting a couple of windows/linux-VM's

 

[beisser]

today i even installed it in an oculink to m.2 adapter and it continues working just fine. looks like this now:



you can see the rtl9101 card in the m.2 adapter laying on the right side next to the pc.
havent noticed any issues so far. certainly no disks dropping off or pcie devices disappearing.
performance seems to be similar to the asm1166 card is used before.

 

[nilfisk_urd]

I have a few questions about the sata card:
does it support aspm? how is the power consumption? does it run on pcie 3.0 x4 or x2?

 

[beisser]

this is the output if lspci -vv for the card:

01:00.0 SATA controller: Realtek Semiconductor Co., Ltd. Device 9100 (rev 02) (prog-if 01 [AHCI 1.0])
        Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B- DisINTx+
        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
        Latency: 0, Cache Line Size: 64 bytes
        Interrupt: pin A routed to IRQ 46
        IOMMU group: 16
        Region 5: Memory at dce10000 (32-bit, non-prefetchable) [size=64K]
        Expansion ROM at dce00000 [disabled] [size=64K]
        Capabilities: [40] Power Management version 3
                Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=375mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
                Status: D0 NoSoftRst+ PME-Enable- DSel=0 DScale=0 PME-
        Capabilities: [50] MSI: Enable+ Count=1/1 Maskable- 64bit+
                Address: 00000000fee00000  Data: 0000
        Capabilities: [70] Express (v2) Endpoint, IntMsgNum 0
                DevCap: MaxPayload 512 bytes, PhantFunc 0, Latency L0s unlimited, L1 unlimited
                        ExtTag- AttnBtn- AttnInd- PwrInd- RBE+ FLReset- SlotPowerLimit 75W TEE-IO-
                DevCtl: CorrErr+ NonFatalErr+ FatalErr+ UnsupReq+
                        RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop-
                        MaxPayload 256 bytes, MaxReadReq 512 bytes
                DevSta: CorrErr+ NonFatalErr- FatalErr- UnsupReq+ AuxPwr+ TransPend-
                LnkCap: Port #0, Speed 8GT/s, Width x2, ASPM L0s L1, Exit Latency L0s unlimited, L1 <64us
                        ClockPM+ Surprise- LLActRep- BwNot- ASPMOptComp+
                LnkCtl: ASPM Disabled; RCB 64 bytes, LnkDisable- CommClk+
                        ExtSynch- ClockPM+ AutWidDis- BWInt- AutBWInt-
                LnkSta: Speed 8GT/s, Width x2
                        TrErr- Train- SlotClk+ DLActive- BWMgmt- ABWMgmt-
                DevCap2: Completion Timeout: Range ABCD, TimeoutDis+ NROPrPrP- LTR+
                         10BitTagComp+ 10BitTagReq- OBFF Via message/WAKE#, ExtFmt- EETLPPrefix-
                         EmergencyPowerReduction Not Supported, EmergencyPowerReductionInit-
                         FRS- TPHComp- ExtTPHComp-
                         AtomicOpsCap: 32bit- 64bit- 128bitCAS-
                DevCtl2: Completion Timeout: 50us to 50ms, TimeoutDis-
                         AtomicOpsCtl: ReqEn-
                         IDOReq- IDOCompl- LTR+ EmergencyPowerReductionReq-
                         10BitTagReq- OBFF Disabled, EETLPPrefixBlk-
                LnkCap2: Supported Link Speeds: 2.5-8GT/s, Crosslink- Retimer+ 2Retimers+ DRS-
                LnkCtl2: Target Link Speed: 8GT/s, EnterCompliance- SpeedDis-
                         Transmit Margin: Normal Operating Range, EnterModifiedCompliance- ComplianceSOS-
                         Compliance Preset/De-emphasis: -6dB de-emphasis, 0dB preshoot
                LnkSta2: Current De-emphasis Level: -3.5dB, EqualizationComplete+ EqualizationPhase1+
                         EqualizationPhase2+ EqualizationPhase3+ LinkEqualizationRequest-
                         Retimer- 2Retimers- CrosslinkRes: Upstream Port

the card supports ASPM, but its disabled on my host because the host doesnt have a bios-option to enable it.
the card is running on pcie 3.0 x2

 

[nilfisk_urd]

thank you! a shame that all these cards are bandwith constrained

 

[beisser]

quite honestly these controller are fine in my eyes. 2GB/sec bandwidth should be anough unless you are using 25Gbit ethernet or faster and if you run that, you dont care about the powerusage of a beefier lsi controller anymore

i am quite happy with its speed and expandability. 9 additional drives at the cost of 18 euros and 1 m.2 slot (my minipc actually has 2 slots that have only 2 lanes to begin with) is a good tradeoff for me.

 

[ptf]

I just ordered one to see if it plays any nicer with my PEX8749 based splitter board. It can't be worse than what I've tried so far.

Bit puzzled why they went with 2 lanes of PCIe 3.0 as the M.2 format allows 4 lanes, I could imagine that it is challenging to route 4 PCIe lanes and 9 sets of SATA on a 2280 card but less clear why the chip would limit itself to just two lanes. I guess, as mentioned, it is based on the presumption that it is about right for mechanical drives.

There seem to be cards with 2x SFF 8087 ports as well - as described here

The actual RTL9101 seems to have been announced as far back as 2021



Source 컴퓨텍스 타이베이 2024 종합기사

 

[starshipeleven]

This "new" Realtek sata controller intrigues me, we are finally at a point where a single low power sata controller can kinda replace a HBA in a NAS-like environment, avoiding sata port multiplier bs or having to mount multiple controllers on a bifurcated slot (which is a huge waste).

The m.2 card with sff 8087 ports is neat, too bad even on ali it's twice as expensive as the one with sata ports while having one less "port" (each sff 8087 connects only 4 drives).

I did some searching on ali with the realtek chipset name and it seems there are also pcie cards like this one, that seem to utilize 2 pcie lanes (see the 2 pairs of caps near the pcie slot), while using 2 sff 8087 ports + a sata port at the top, exposing all the 9 sata connections.
The large heatsink does not allow to confirm this is actually a realtek controller and not the usual clowncar of asmedia controller + port multiplier, so it's not "guaranteed" to be actually what the seller claims it to be

Bit puzzled why they went with 2 lanes of PCIe 3.0 as the M.2 format allows 4 lanes, I could imagine that it is challenging to route 4 PCIe lanes and 9 sets of SATA on a 2280 card but less clear why the chip would limit itself to just two lanes. I

The chip package (the physical black chip where the controller silicon is put in) is standardized so the amount of pins available is limited by the package they have chosen. Although I'm sure they would have enough pins for 2 additional lanes if they dropped one Sata port so it becomes a x4 pcie and a 8 sata port controller.

But afaik no sata controller chip ever used more than x2 (even marvell ones are x2 at best if I remember correctly) and it's somewhat rare even for usb controllers to use more than x2, they upgraded to x4 only when they really really had to for usb 20gbit. So maybe it's significantly more expensive to include a pcie controller that can do x4 lanes and they decided it's not important enough for a card that will 99.999% of the times be used with mechanical drives anyway