MP All Models PCIe SSDs - NVMe & AHCI

[w1z]

The industry standard name is U.2, some drives are not even 2,5” and all use the new SAS U.2 cable/connector.

I know what U.2 is and you might have meant the 2.5" form factor drive uses U.2 connectivity thus requiring a U.2 NVMe PCIe adaptor. This doesn't make the product solely a U.2 product.

Samsung didn't even use U.2 in any of their product briefs/datasheets for the PM983.

 

[tsialex]

I know what U.2 is and you might have meant the 2.5" form factor drive uses U.2 connectivity thus requiring a U.2 NVMe PCIe adaptor. This doesn't make the product solely a U.2 product.

Samsung didn't even use U.2 in any of their product briefs/datasheets for the PM983.

It’s a SFF-8639 device? If it is, the industry standard name is U.2.

 

[w1z]

It’s a SFF-8639 device? If it is, the industry standard name is U.2.

Oh my. I know SF-8639 is the former name of U.2

Please read on the device more before making blanket statements. I already told you it comes in 2 form factors M.2 and 2.5".. ONLY the 2.5" model requires U.2 NVMe PCIe adaptor. M.2 is not U.2 and 2.5" is not solely U.2 as there's SAS and Sata drives that come in 2.5" form factor.

U.2 is the name of the connector not a form factor. As far as I know, U.2 NVMe SSDs only come in 2.5" form factor and there aren't M.2 NVMe SSDs that are U.2

I hope this is clear for you now.

 

[tsialex]

Oh my. I know SF-8639 is the former name of U.2

Please read on the device more before making blanket statements. I already told you it comes in 2 form factors M.2 and 2.5".. ONLY the 2.5" model requires U.2 NVMe PCIe adaptor. M.2 is not U.2 and 2.5" is not solely U.2 as there's SAS and Sata drives that come in 2.5" form factor.

U.2 is the name of the connector not a form factor. As far as I know, U.2 NVMe SSDs only come in 2.5" form factor and there aren't M.2 NVMe SSDs that are U.2

I hope this is clear for you now.

I'm talking specifically about the 2,5" variant.

SFF-8639 devices were all renamed U.2 for marketing purposes back in 2015 by the SFFWG consortium.

If a drive is made to the SFF-8639 spec, uses SFF-8639 cable and connector, why it's not a SFF-8639 device?

 

[w1z]

SFF-8639 devices were all renamed U.2 for marketing purposes back in 2015 by the SFFWG consortium.

If a drive is SFF-8639, uses SFF-8639 cable and connector, why it's not a SFF-8639 device?

Please reread all the posts I made on the PM983 in this thread again, including your responses, and you'll understand then where the disconnect is.

 

[tsialex]

Please reread all the posts I made on the PM983 in this thread again, including your responses, and you'll understand then where the disconnect is.

SFF-8639 is the broad Enterprise SSD Form Factor Version 1.0a defined by SSD Form Factor Working Group, that include 2,5", 3,5", SATA, SAS, NVMe drives that use the SFF-8639 mechanical connector.

PM983 2,5" version is not a drive made to the 2,5" subset of the SFF-8639 spec?
[doublepost=1558055792][/doublepost]Btw, Samsung uses the U.2 nomenclature in the LEARN MORE link from https://www.samsung.com/semiconductor/ssd/enterprise-ssd/MZQLB3T8HALS/

[doublepost=1558056341][/doublepost]One thing for sure, Samsung never uses the correct standard name in the full data sheet and they even removed it from the connector drawings. Why?

 

[joevt]

SFF-8639 is the broad Enterprise SSD Form Factor Version 1.0a defined by SSD Form Factor Working Group, that include 2,5", 3,5", SATA, SAS, NVMe drives that use the SFF-8639 mechanical connector.

PM983 2,5" version is not a drive made to the 2,5" subset of the SFF-8639 spec?

Btw, Samsung uses the U.2 nomenclature in the LEARN MORE link from https://www.samsung.com/semiconductor/ssd/enterprise-ssd/MZQLB3T8HALS/

One thing for sure, Samsung never uses the correct standard name in the full data sheet and they even removed it from the connector drawings. Why?

At least they include the connector drawings and the name/function of all the pins.
[doublepost=1558085192][/doublepost]

Please reread all the posts I made on the PM983 in this thread again, including your responses, and you'll understand then where the disconnect is.

I don't think there's a disconnect. You're both talking about the PM983. It has one interface: PCIe Gen3 x4. It has two form factors: 2.5" (uses a U.2 cable) and M.2 (connects to a M.2 slot). Other devices having a 2.5" form factor may connect with SATA or SAS, but the only option for the 2.5" PM983 is U.2.

I think you were mistakenly thinking that tsialex was saying that the PM983 was only U.2, when in fact he was only clarifying that the 2.5" variant uses U.2.

 

[w1z]

I think you were mistakenly thinking that tsialex was saying that the PM983 was only U.2, when in fact he was only clarifying that the 2.5" variant uses U.2.

You're thinking is wrong as I was trying to correct him on the above blanket statements. He edited his other posts but you can still read the unedited versions in my quotes.

Anyway, I'm done with this!

 

[joevt]

You're thinking is wrong as I was trying to correct him on the above blanket statements.

What exactly is wrong with each of those statements? Ok, let me try again to figure this out:

First, the statement "U.2 NVMe": this refers to the 2.5" form factor that you mentioned. The only problem I see here is that the form factor is omitted which is necessary to complete the description of this variant of the PM983: a 2.5" NVMe U.2 drive. Well, maybe just NVMe 2.5" was sufficient in the first place since U.2 is the only known connector for such a device?

Second statement: "The industry standard name is U.2, some drives are not even 2,5"". This again refers to the 2.5" form factor of the PM983. He is taking exception to calling the drive 2.5" instead of U.2. I see a couple problems: stating the fact that some U.2 drives are not 2.5" is unnecessary because we are talking about the PM983 which only comes in 2.5" and M.2 form factors. Calling the drive U.2 instead of 2.5" is insufficient because, as he says, some U.2 drives are not 2.5". So again, you need to call it a NVMe 2.5" U.2 drive or just NVMe 2.5" if you assume all NVMe 2.5" drives use U.2.

He edited his other posts but you can still read the unedited versions in my quotes.

If his posts were edited, they would say "Last edited". Your quotes and screenshots show the current contents of those posts.

Anyway, I'm done with this!

If you go away, I will be confused forever!

 

[flooglehorn dinglebop]

I've done a search through this thread and can't find the answer so please excuse me if I've missed something, but on the wiki it says
"Apple support the 1.3 NVMe standard. Any blade that need a special NVMe module/driver won't be supported"
But then further down it goes on to say that both the PM961, SM961 are good: aren't those NVMe 1.2?

Also has anyone tried the Adata XPG SX8200 Pro? It seems to be getting some good reviews but it's not in the wiki list. It is NVMe 1.3. The controller is SMI SM2262EN which looks similar to the HP listed as good in the wiki

I'm looking at upgrading my boot SSD in this case — currently using an older AHCI model, which is a bit small, and a 960 EVO in another slot for files (too small too) — I want to consolidate onto one larger and faster device in one slot and get a slot back for a network card and ideally putting a few NVMe drives in something like the Highpoint 7101A.

While I'm here if anyone could point me in the direction of a thread for updating the Samsung SSD firmware … last time I looked it was only doable on Windows and I don't know whether it's worth installing Windows (I have no desire to do so or reason otherwise) just for upgrading the firmware.

Thanks

Edit:
did some more searching and discovered

I just installed the Samsung 970 Pro in the Aqua kyro M2 I got yesterday. The rig runs great for an hour of
BM Disk Speed Test. The I/O Crest was the culprit and I'm a little nervous about trying another for running sessions. There was plenty of clearance with the Aqua card, the I/O Crest card, and the 580 pulse, especially for my purposes.

@mesaken I didn't quite follow your first post but I'm taking it from this follow up that the Adata blade did work for you (booting?) but the PCIe adapter was problematic?

 

[edgerider]

hi there!

does anyone have seen this :
http://amfeltec.com/pci-express-gen-3-carrier-board-for-6-m2-or-ngfss-nf1-pcie-ssd-modules

there is an interesting option at the bottom : « additional x16 uplink »

would that mean, that you could actually plug the card to 2 x16 gen2 pcie slot to span the gen3 6ssd over 2 gen 2 pcie x16 slot?

any how the more I deal with PCIEX16 expension with true pcie plex chip, the more I think intel and amd should just exit all pcie lane to a plex chip and let it dynamically manage pcie lanes on demand to pciex16 slot.

if a small company like anfeltec can do it on a smal pcie board i think apple could totally do a proprietary conector that exit all 40 lane of pcie gen 3 bandwith, and have an external box handle all pcie expension.
since i have my custom 10 slot pciex16gen2 expender, life is so easy :
shut down the computer, plug whatever card i want, turn the computer on and « voila » 2 or 3 gpu, 4 ssd, 10gbe, usbc, usb3, you name it it just work.
and it is just a one pciex16 gen2 link.
imagine a box with one cable, that takes all the pcie lane of the processor and put it on a expension box.

they could do a small one with like just 40 lane shared to x16x16x8 with no plex and a « pro » one with 2 connector for dual socket xeon who have 80 lane total.

 

[Pressure]

hi there!

does anyone have seen this :
http://amfeltec.com/pci-express-gen-3-carrier-board-for-6-m2-or-ngfss-nf1-pcie-ssd-modules

there is an interesting option at the bottom : « additional x16 uplink »

would that mean, that you could actually plug the card to 2 x16 gen2 pcie slot to span the gen3 6ssd over 2 gen 2 pcie x16 slot?

any how the more I deal with PCIEX16 expension with true pcie plex chip, the more I think intel and amd should just exit all pcie lane to a plex chip and let it dynamically manage pcie lanes on demand to pciex16 slot.

if a small company like anfeltec can do it on a smal pcie board i think apple could totally do a proprietary conector that exit all 40 lane of pcie gen 3 bandwith, and have an external box handle all pcie expension.
since i have my custom 10 slot pciex16gen2 expender, life is so easy :
shut down the computer, plug whatever card i want, turn the computer on and « voila » 2 or 3 gpu, 4 ssd, 10gbe, usbc, usb3, you name it it just work.
and it is just a one pciex16 gen2 link.
imagine a box with one cable, that takes all the pcie lane of the processor and put it on a expension box.

they could do a small one with like just 40 lane shared to x16x16x8 with no plex and a « pro » one with 2 connector for dual socket xeon who have 80 lane total.

You are reading it wrong.

The PCIe slots are not part of the PCB, so you can either connect the adapter with 8 or 16 PCIe lanes depending on the "uplink" adapter. You can see it clearly in the picture below.

 

[joevt]

does anyone have seen this :
http://amfeltec.com/pci-express-gen-3-carrier-board-for-6-m2-or-ngfss-nf1-pcie-ssd-modules

there is an interesting option at the bottom : « additional x16 uplink »

would that mean, that you could actually plug the card to 2 x16 gen2 pcie slot to span the gen3 6ssd over 2 gen 2 pcie x16 slot?

It means you can remove the x16 connector and replace it with an x8 connector with a screw driver.

any how the more I deal with PCIEX16 expension with true pcie plex chip, the more I think intel and amd should just exit all pcie lane to a plex chip and let it dynamically manage pcie lanes on demand to pciex16 slot.

if a small company like anfeltec can do it on a smal pcie board i think apple could totally do a proprietary conector that exit all 40 lane of pcie gen 3 bandwith, and have an external box handle all pcie expension.

imagine a box with one cable, that takes all the pcie lane of the processor and put it on a expension box.

they could do a small one with like just 40 lane shared to x16x16x8 with no plex and a « pro » one with 2 connector for dual socket xeon who have 80 lane total.

The switch chip used by Amfeltec is not made by Intel or AMD, and costs $$$ - Broadcom price gouging, like what happened with the EpiPen. Your idea of not using a switch chip might work for a modular Mac Pro but 3 slots would be too limiting. Current Mac Pros use x16x16x4x4 so maybe that should be the minimum but there may exist solutions that will let you use x8 when only one card is connected and x4x4 when two cards are connected. Much like in PCs where x16 lanes can be used by a single card or used by two cards as x8x8.

Imagine 40 lanes, providing up to 10 slots, divided into 3 groups like this:
[[[x16][x4]] [[x8][x4]]] [[[x16][x4]] [[x8][x4]]] [[x8][x4]]

The x16 groups can be used as x16, x8x8, or x4x4x4x4. Is there a chip that can do that?
The x8 group can be used as x8, or x4x4.
The x4 slots are interspersed among the longer slots to increase spacing between longer slots for cards that are wider.
You could go nuts and add x2 slots like this [x16 x2 x4 x2 x8 x2 x4 x2] to double the number of possible slots and to make the x8 slots quadruple wide and make the x4 slots double wide.

Are there examples of dual xeon PCs with 80 lanes (40 from each CPU)?

since i have my custom 10 slot pciex16gen2 expender, life is so easy :
shut down the computer, plug whatever card i want, turn the computer on and « voila » 2 or 3 gpu, 4 ssd, 10gbe, usbc, usb3, you name it it just work.
and it is just a one pciex16 gen2 link.

Is there a link to more photos of that monster? I want to see the backside. What board does it use? What PCIe switch chip? How are the lanes dividing among the 10 slots?
[doublepost=1558185240][/doublepost]

That new Sonnet card looks beautiful BUT I consider the biggest design flaw in all of them to be the fact that they won't support the width of a 1TB SSUBX - having owned the io-crest and the highpoint i find it aggravating that the spacing will only accomodate 1 from 2 slots on the io-crest and 2 from 4 slots on the Highpoint. It looks like the Sonnet card is no different to the Highpoint card in that respect.

What's a SSUBX? I could only find https://www.amazon.com/Odyson-SSUBX-Replacement-MacBook-2013-2015/dp/B06WGNVGX6 which doesn't look like an M.2 module at all. Are you using an adapter? There are M.2 to M.2 adapters which could let you move the SSUBX so that all M.2 slots are usable but I guess that could get messy.

No card is designed with 1TB SSBUX models in mind and never will. It's an Apple OEM blade that don't even have a standard M.2 connector.

does anyone have seen this :
http://amfeltec.com/pci-express-gen-3-carrier-board-for-6-m2-or-ngfss-nf1-pcie-ssd-modules

This new amfeltec allows for more and wider and longer M.2 devices than the Sonnet. Is 32 mm wide enough for a 1TB SSUBX? We never did get a response about what zedex meant about the SSUBX. I wonder what other things you can do with a 32mm x 110mm M.2 space.

 

[tsialex]

This new amfeltec allows for more and wider and longer M.2 devices than the Sonnet. Is 32 mm wide enough for a 1TB SSUBX? We never did get a response about what zedex meant about the SSUBX. I wonder what other things you can do with a 32mm x 110mm M.2 space.

1TB SSUBX is a Apple OEM drive used in MP6,1 and some rMBPs, it uses the 12+16 Apple connector. You can use an adaptor to convert it to the standard M.2 connector. The adaptor have a very bulky connector (11mm height x 22mm wide) and not very reliable, I have one here:




Using a pixel ruler with this photo from The Ultimate Guide to Apple’s Proprietary SSDs, total width is a little over 1,62 times the 12+16 connector (20mm), so the total width of a 1TB SSBUX is around or a little less than 33mm.

Anyone has the mechanical drawing for Apple proprietary 12+16 pin blades?

 

[zedex]

This is very cool conversation and exploration of potential performance enhancers.. i wanted to add these photos - they illustrate @tsialex 's point (no-one sets out to accomodate a SSUBX) and also my own. Whilst I think Alex is right, it's still hard not to love the SSUBX (1TB). Photos show.

IMG (1) shows two SSUBX installed on one HPT SSD7101 board - looks pretty crowded huh?



spot any other problems..?

the disks are TOO big and the BIG trade-off is you can't put the cover/heatsink back on the HPT SSD7101

this would be untenable/intolerable for some members. I'm more annoyed that the second SSUBX I acquired did not include a factory fitted heatsink - difference in resting/idle temp shows 27% vs 45%

 

[edgerider]

It means you can remove the x16 connector and replace it with an x8 connector with a screw driver.


The switch chip used by Amfeltec is not made by Intel or AMD, and costs $$$ - Broadcom price gouging, like what happened with the EpiPen. Your idea

Is there a link to more photos of that monster? I want to see the backside. What board does it use? What PCIe switch chip?

I built this from scratch with a g5 case and a cyclone microsytem pcie expender.

it as a 96 lane pcie switch gen 2
that do one gen2 16x host to 16x pcie cable to 16x slave hib to a 10 double width slot pcie backpane.
the slot are 8x/8x/8x/16x/16x/8x/8x/8x
because one slot is taken for the uplink card and one other slot is made for some kind of hardware monitoring card that i dont have.

so you only get a pciex16 gen2 link to the computer, but this is plenty enough to run all the card i need.

there is a varient with two uplink and two plx chip and each half of the board have it’s own 16x link.

they also have the same half board made for eatx with 16x/16x/x8/x8/8.

I also have one of those in my supermicro chassis with the 24 bay array true 24 port (no sas expender) areca 1880ix24.

the raid 60 on this is 2Gb/s all day long with 24 hsgt 4tb sas drive.

the challenge was to find a 20 slot pcie enclosure sloted grid and to do the custom foot to have the right height and cut the G5 chassis.

still have to do all the psu bracket and cable managment because it start to get crowded.

i bought a modular 600w psu and a 1600w mining psu to be able to use up to 4 titan x gpu.

it works good and everything is plug and play. it has worked since day one out of the box.

all my machine have the pcie x16 host card and I just plug the box to whatever computer need the included cards.

fun enoug it also work with the sonnet thunderbolt enclosure hooked to the macmini. it is slower and i only tried with the areca raid card, not gpu, but it works. i use it to initilize big raid array because the mac mini is enough to do this and like this my xserves and the macpro dont need to be blocked for 2 day to initialize a raid.

the expender in the supermicro chassis stays on there with the xserve, so my custom G5 enclosure is my swiss knife with all the pcie card i could need.

if i need one of the pcie card, i just shut off the machine were i need it , plug the pciex16 cable, reboot and there it is.

it is way easier than open evry machine each time to just plug one card for a couple of hour.

it cost me about 1000€ with parts from ebay.

[doublepost=1558201298][/doublepost]

You are reading it wrong.

The PCIe slots are not part of the PCB, so you can either connect the adapter with 8 or 16 PCIe lanes depending on the "uplink" adapter. You can see it clearly in the picture below.
View attachment 837715

yeah i know you can swap the connector on since gen3 squid from 8x to 16x
but on the bottom there is a mention of « extra x16 uplink and power supply »

but english is not my mother language so maybe i just don’t understand properly and it just mean you get both connector in the package x8 and x16 instead of just one or the other.

 

[joevt]

it as a 96 lane pcie switch gen 2

Very nice. Have you looked at any gen 3 solutions? I think they're much more expensive but you could use it to get full speed from a Thunderbolt 3 card (or any other PCIe 3.0 card with 8 or fewer lanes).

but english is not my mother language so maybe i just don’t understand properly and it just mean you get both connector in the package x8 and x16 instead of just one or the other.

Maybe it would be more clear if they called the extra adapter "Replacement" instead of "Additional".

 

[edgerider]

Very nice. Have you looked at any gen 3 solutions? I think they're much more expensive but you could use it to get full speed from a Thunderbolt 3 card.

yes I know!
I had a gen3 netstor pcie expander but as my mac was gen2 never noticed a difference.

my feelings are that 16xgen2 is still plenty enough for most of the things I have.

if the 2019 macpro have pciex16gen4 slot, I will still use those expender because i think only 4 nvme ssd or a very high end gpu can stress pciex16 gen3.

now I am happy with my setup, and the only thing that would help is a much faster and more modern cpu that support hvec and all codecs in hardware and way faster memory .

once you reach 5gb/s R/W with very low latency for cache and as long as your media are on a very powerful raid, in video editing I feel it is not anymore the storage that is holding you.

so far my machine is the fastest machine I have used with premiere on mac. the top of the line Imac pro is faster on 4k h264 media but on 5/6/8k prores 444 my machine crunches it.

but my friend maxed out Z8 with two p6000 make my mac look like what it really is : a 10 year old computer.

man that thing is fast... probably 5 time faster and absolutely no timeline render whatsoever, and exporting is like 10 time faster.

that is why the more it goes, the more i think pcie expender are the true modularity.

if the computer can be cheaper by just having the cpus the ram and basic I/O,
and then you have a chassis with all pcie lanes of both cpu on a switched board that translates pcie gen lanes that would be sweet.

you could put a gen 2 16x gpu on a slot but it would actually be seen as a 4x pcie Gen4 slot.

Pcie ssd is what have pushed the boundaries of bandwith in need for new highs.

so far now the cpu are the bottleneck in video editing because no software is written for 56 core software, therefore in most case a very fast and overclocked modern 6 core i9 machine is faster than a server grade monster in premiere or any adobe app because it is single core performance that count in their code.

I think very high end server cpu are not anymore built for general purposes but very oriented for VM.

I really hope that apple will bring back a form of XGRID because that was really a good way to increase speed back in the days.

with xgrid or when compressor / after effect where having a second machine to work on, you would litteraly see twice the speed.

with 10gbe the way i see it would be to have I servergrade main machine, and macminis as helper.

rendering on distributed computer with 4 6core macmini to help the xeon W would be blazing fast probably faster than a 50k$ server.

all you need is a 4 port 10gbe pcie card, and 4 x 2000$ mac mini and you are golden.

 

[startergo]

Very nice solution for NVME cooling:
https://www.glotrends-store.com/pro...-2280-internal-ssd-with-silicone-thermal-pad/

 

[tsialex]

Very nice solution for NVME cooling:
https://www.glotrends-store.com/pro...-2280-internal-ssd-with-silicone-thermal-pad/

Heat decompose rubber bands fast, I don’t really know if that material is better, but I really prefer heatsinks with metal clips that lock the heatsink over the blade

 

[bunnspecial]

Heat dissolve rubber bands fast

What sort of solvent might one find in "heat" to dissolve rubber bands?

 

[tsialex]

What sort of solvent might one find in "heat" to dissolve rubber bands?

Did you never saw a rubber band totally decomposed (better word perhaps) in a PC before? Heat and rubber bands don't work, after sometime they just melt and it's a real pain to clean it up.

 

[startergo]

Heat dissolve rubber bands fast, I don’t really know if that material is better, but I really prefer heatsinks with metal clips that lock the heatsink over the blade

Yes I was thinking about that too .
[doublepost=1558359571][/doublepost]Anyway I ordered a couple and will test them

 

[bunnspecial]

Did you never saw a rubber band totally decomposed (better word perhaps) in a PC before? Heat and rubber bands don't work, after sometime they just melt and it's a real pain to clean it up.

"Dissolving" something requires a solvent. They can "degrade" from heat, but they do not "dissolve."

 

[tsialex]

"Dissolving" something requires a solvent. They can "degrade" from heat, but they do not "dissolve."

Thx for the correction.