Yeah, the usual Unraid data suggests typical ASMedia saturation of the 3.0 x1 lane. Given the minimal heatsink Iâm inclined to view the ASM1064âs ~900 MB/s limit as probably more of a feature than a limitation relative to the ASM1164âs 3.0 x2 interface.
Iâve this ASM1166 variant in test at the moment and, with good overall front to back airflow but without a fan directly onto the card, it gets plenty warm just from basic DiskSpd characterization at its ~1.8 GB/s max.
I ordered one of the ASM1166 6-Port from ALI, backed by the findings in this thread - thank you 
I also read here that the cards are super flimsy, BUT, I think the manufacturers realized that and my model came with a small thingy screwed to the back of the PCB.
I still wouldnât recommend plugging the SATA cables in when itâs installed, but thereâs basically no chance youâll snap it by accident.
Howâs the implementation? The inboard screw placement doesnât seem like itâd do all that much for downward deflection as SATA connectors are plugged in (or reinforce the connectors themselves) but if itâs also glued/epoxied/resined on then itâd be more effective.
As an aside, reinforced ASM1166 M.2s have been widely available for some time. I think itâs more that a lot of the parts listings donât emphasize reinforcement and many people donât look closely at the pictures, though sometimes itâs quite hard to tell.
https://www.aliexpress.us/item/3256803452021751.html?gatewayAdapt=glo2usa4itemAdapt
These kinds of things have been popping up as well. Anyone have an idea how theyâre implemented? Are there actually 2 ASM1116 on it, or is it a port multiplier? It claims itâs x4 so might be usable bandwidth, and I prefer it to an LSI HBA due to power usage. Thinking about picking one up.
8-port ASM1166 SFF-8087s are on Amazon and such as well but none of the listings Iâve encountered answer your questions. Absent more information, my working assumption would be itâs an 8-port ASM1064/1166 + JMB575 port multiplier variant.
Since the JMB575 is a 5x multiplier, in the ASM1166 case what Iâd hope for is one SFF-8087 is direct through the ASM1166 and the other is off the JMB575. Only real advantage Iâm seeing to that over a six port ASM1166 card with 6x SATA connectors is itâs probably easier to buy 1 m cables.
Thanks! Is an ASM1166 natively an x4 then? Iâve heard bad things about SATA multiplexers so thinking I should maybe just avoid.
Plus JMB585 doesnât support ASPM anyways so not sure how that would turn out.
Is anyone ever going to get around to releasing new adapters that take advantage of new pcie gens and/or use a lane count that motherboards actually support natively? Or are we going to continue to not fully utilize our precious lane allocation by plugging gen 3 x2 devices into gen4 (or 5 even now) x4 slots?
x6. A quick look through ASMediaâs PCIe to SATA and JMicronâs SATA to SATA product linesâll probably be helpful context. The general guidance to avoid port multipliers I suspect is mostly based on the JMB575 and, previously, JMB394 and 393. However, attribution back to the actual port multiplier used is rare. I believe itâs plausible the ASM1092 (and conceivably JMB572) could be better behaved but I donât know of anything that uses it.
Reliability issues and drive drops aside, even 2x SATA multiplication is marginal for current 3.5s as 550 MB/s split two ways is maybe 250 MB/s per drive after multiplier overhead. Even just on bandwidth itâs probably preferable to put in a second ASM1166 card as just an x1 slot still gives ~300MB/s to three drives concurrently.
Depends on what the motherboard has for slots but a common way to ladder it out would be
This is synchronized spin up, so watch the PSUâs minor rails as well as +12 with 3.5s. Some drivesâll peak at 1.5 A each from +5 and plenty of PSUs are 20 A max on 5 V, so even eight drives potentially gets pretty iffy after allowing for other motherboard demands and USB +5 delivery.
Be careful of confusing the JMB585 and 575. Iâm unsure its apparent lack of ASPM is the JMB585âs biggest problem but very likely it doesnât help.
Sure not looking like it, though I agree a 4.0 x1 version of the ASM1166 would be useful. 4.0 x4 I suspectâs too niche as thatâs 12+ SSDs.
Darn yeah fair point, not gonna be enough bandwidth for even x8 spinning rust. Maybe I should just give up on my low power consumption dreams and go with a LSI HBA like everyone else. Similarly priced to this solution and the cable management is cleaner.
Something like this seems solid, and I get the full 8 ports at 6gbps: IBM M5110 8-Port 6Gbps PCI-e IT MODE FW:P20 S2308 9207-8 I+2*SFF8087 SATA US | eBay
I wish I did more research before getting a JMB585 M.2 adapter, cause its giving me more trouble than its worth.
@lemma Do you have any opinion vs a LSI HBA and just going with a generic PCIe ASM1166 solution. I have 4 ports on my motherboard, only need 9 total. But it seems like maybe I should lean towards getting 8x full bandwidth ports over the ~250MB/s from the ASM1166.
Trying to decide between a 9207-8i and something like https://www.aliexpress.us/item/3256805261191720.html
For spinning rust, ASM1166 as long as the card assembly and thermals check out. I agree with aBav Broadcomâs pretty janky and, to be maybe mildly controversial, itâs my view the LSI cards are old enough to be done and itâs just the 9500-8i, -16i, and large OEM equivalents that are of interest for new builds. Also agree with aBav that significant-sustained workloads merit fan on card.
If itâs a server build with an open x8 slot and power no real object then maybe the 9500s are of interest, particularly as the -16i is rather required if doing Storinator type things. If itâs m(ATX) then a motherboard chosen with storage in mind + ASM1166 3.0 x2 is just about full bandwidth to 12-14 hard drives, which more than maxes every case I know of besides Define/Meshify XL. Thereâs not much actual application software (as opposed to stress/benchmarks like DiskSpd) capable of saturating an ASM1166 at its ~1.7 GB/s (~280 GB/s per drive at six drives, which is about best case outer edge sequential on current gen 3.5s, or ~340 with five).
What would push me to 9500-8i in your case is the combination where more than seven of the nine drives are SSDs, thereâs a regular use case for pulling more than 1.7 GB/s through the HBA that canât be met by the chipset ports, those workloads run long enough the difference between ~1.7 and ~2.2 or ~2.8 GB/s is important, and itâs something which isnât solved by using the x8 slot for a couple 4 TB NVMes instead. Thatâs⌠probably pretty niche. Maybe niche enough for an 8TB SN850X, similar consumer NVMe, or enterprise SSDs to make sense.
I have an ITX board, so was trying to decide between a PEX8747 based switch + stacking a bunch of ASM1166âs, cause ideally I would like ASPM support.
But then I ran upon this: Broadcom eHBA 9600-24i 24 Int. Ports PCIe SAS/SATA/NVMe Tri-Mode Storage Adapter | eBay for $25??
Seems highly questionable pricing wise, but gonna give it a shot.
That link shows as $320.89 for me.
That ASM board seems really cool for a low cost NAS build, since most SBCs do have an NVMe slot where this could be slotted.
Easy 6 SATA ports with a really smol computer for simple data storage and low power consumption. I might grab one of those to slap into my Orange Pi 5 with a bunch of 2.5" SSDs.
One thing that Iâm wondering is, when using just 2.5" drives and no regular ATX PSU, how can I power the drives? Maybe try to find some USB to SATA-power only adapters (since they only need 5v) along with one of those desktop USB charging stations? 
Yes, USB 5 V power for 2.5" drives (regular SATA SSDs, not SAS or NVMe enterprise drives) works fine 99 % of the time.
I tested installing Windows 10 in âWin-to-Goâ mode in 2018/2019 on a regular SATA SSD in an USB 3.0 5 Gb/s-to-2.5"-SATA adapter enclosure and it worked fine for years. Had to stop the test since in this Win-to-Go configuration Windows doesnât do the large feature upgrades.
If youâre just using 5 V power for the SSDs alone it should absolutely be okay since you donât have to additionally power the separate USB-to-SATA bridge chipset from an enclosure.
I agree wholeheartedly with your assessment on the tri-mode HBA from Broadcom. I just bought oneâŚliterally and the 4 drives attached, canât be found in Windows. This is however my first experience with this type of setup with raid and all; but it seems the only way itâs going to be possible is to have the OS installed on the HBA drives. Iâm still painfully working through data salvage right now so I have yet to see whether or not there is some workaround without the HBA being the boot device.
As far as SATA becoming obsolete in the near future. I donât foresee that happening yet. Only hardcore gamers (from what I can tell) are using M.2âs in this fashion. SSDâs just donât offer enough for the average budget in comparison to HDDâs. The only other reason anyone would have to change right now, is exactly what Iâm currently faced withâŚthe robustness of HDDâs seems to have gone down considerably over the last century. Drive just donât last the way they used to and probably by design so big tech can sell ore drivesâŚer ummâŚforce folks to buy more drives at the expense of data loss.
Turns out it was a pricing error lol, makes sense. The seller honored it for me though so Iâm pretty happy. My alternative was a PCIe gen 3 switch + a bunch of the ASM1166 M.2 adapters.
Be careful to read about TRIM and this board, i regret buying it
Iâm sorry to hear about your similarly bad Broadcom experiences.
My history with them is:
LSI RAID controller 9361-8i with external SAS Expanders, everything okay (that controller had been designed before Broadcom got involved with LSI/AVAGO)
Broadcom HBA 9400-8i8e (started out okay-ish but with issues, some issues have been solved, other catastrophic ones were introduced and wonât be fixed since Broadcom decided to declare it EOL in the meantime)
Broadcom HBA 9500-16i: Had the same issues as the 9400, but these were finally fixed with firmware 28 last year
Broadcom P411W-32P: Great on paper and performance-wise, horrible driver and firmware quality. Broadcom removed advertised features during the productâs life cycle (you canât attach NVMe SSDs directly anymore, you have to go through an active Tri-Mode backplane even though this HBA is a pure PCIe Switch, meaning you can only use it for NVMe, not SAS or SATA SSDs)
Broadcom sucks!
This is my first ever âraidâ setup. There are basically two reasons I will try to avoid Broadcom in the future (assuming there are feasible options). The first of those reasons is pizz poor software support. I find that itâs a shame that the writers seem to assume that everyone already knows and understands what theyâre doing and has experience with networking somehow. At every turn, when I search for the homepage of the software for some sort of help/support, Iâm redirected to the Broadcom page which also offers no direct support for a particular question. One must go on a search and destroy mission to NOT find an answer only to have to resort to reaching out to tech support. I remember a time when there was a bit of help already written into the software and it didnât require needing to be online.
The second reason is (I donât know if itâs normal or not), as itâs already been stated, I canât tell what drives are being used and how. I just now created the array and figured out it requires the Virtual Disks to have an disks at all in Windows. Even with that, I canât tell what the distribution of that data is. To be more exact, at the given raid level, how much of the actual physical disks are being used in relation to the virtual disks. Is it really being stored according to the given raid? It would be nice to have at least a good graphical representation somehow.
If Broadcom does in fact improve on customer and software support, theyâd definitely have a grreat product⌠TBF, the techs have been good about getting back to me. Yet, for some of my questions, it shouldnât have required writing (not calling) the techs for answers that should have be put in place to be readily found by beginners.
