That is good but I got the board+card+cable for 800$+power supply=150$ Total 950 (and salvaged case)
and the Netstor is 2400$... Otherwise, it seems much more flexible...
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
External PCIe expansion (non TB3)
- Thread starter macguru9999
- Start date
- Sort by reaction score
and the Netstor is 2400$... Otherwise, it seems much more flexible...
we have macpro 4.1 or 5.1
so with cyclone adapter we are gettin a true 16x gen2 link with a 96 lane pcie switch you can have 2 x16 gen 2 and 3 x8 gen 2 card at full speed.
my anfeltec squid runs a 4000mo/s read write on the cyclone with titanx next to it.
pcie x16 gen 2 is still faster than pciex4 gen3
the netstor is different because the uplink is x16 gen 3 but the slot link on the mobo are x8x4x4x4 gen3
as i say the small cyclone is x16 gen2 to x16x16x8x8x8 gen 2
the big one is x16 gen2 to x8x8x8x16x16x8x8x8
I have both.
on a gen 2 machine the cyclone is faster, on a gen 3 machine the netstor is faster.
and any how only nvme ssd can saturate gen2x16, gpu show no diferrence being on gen 2 vs gen3 slot
cyclone also have 16x gen3 model but way more expensive.
[doublepost=1558794528][/doublepost]
you bought it from a seller in new zeeland right?
Last edited:
because i bought mine from him!
it was a bitcoin mining site, he release them slowly by slowly to keep the price up.
What's an mo/s?
The amfeltec (gen 3) works at full speed in the Cyclone because it has a gen 3 switch chip which makes it expensive. When speaking about gen 2 x16 RAID 0, the amfeltec will have the same speed in any of the Netstor's four slots (gen 3 x8), but only two of the Cyclone slots are gen 2 x16.
In the Netstor, you can get full speed from a gen 3 NVMe drive without a carrier card having a switch chip, unlike the Cyclone.
The Cyclone does not gain an advantage when hosted by a gen 2 machine because the Netstor can have gen 2 upstream with gen 3 downstream.
The Cyclone has these advantages:
1) price (a one-time advantage that doesn't exist after you buy a Netstor?)
2) number of slots (one or four more than the Netstor)
3) two or three full speed gen 2 x16 slots (all four of the Netstor slots are gen 3 x8). This is an advantage only if you have gen 1 or gen 2 x16 devices (old graphics cards? what's the opposite of future proofing?)
The Netstor has these advantages:
1) full build (power supply, case, cables, host and target cards, automatic-on when computer turns on) but this is a one-time advantage that doesn't exist after you've built a Cyclone machine.
2) four full speed gen 3 slots allowing full speed from gen 3 devices (up to x8) (not an advantage if you only have gen 2 devices but is future proofing)
3) more upstream bandwidth with a gen 3 x16 host (not an advantage for gen 2 hosts but is future proofing)
Last edited:
Thanks. Updated. That's the 5 slot system PCIe2-2707 using an 80 lane gen 2 switch x16x16x16x8x8.
@joevt
I have the squid gen 2 so there is no gain and the cycline is a gen 2 swich.
a gen3 gpu couldnt care less than this to be on a gen2 slot as long as it is 16 lanes.
I think you are trying to convince someone witch is beter.
they are different animal.
pcie work both way so in a gen 3 machine the netsor as a theorical bandwith of 16xgen 3 to the enclosure, but the cards themselves are conected via a x8 gen 3 link.
on the cyclone everything 3 of the slot gets x16 gen 2 bandwith.
even in a gen 3 machine the cyclone has a greater bandwith to each of those 3 16x slot because gen 3x8 is less than gen2x16
even the titan Xp cant saturate a gen2x16 link both way therefore the one card can be receiving while the second or 3 card can be transmiting.
so even on a gen 3 machine each of the 3 x16 gen 2 slot have more bandwidth.
the only case where the netstor would give a better bandwith is with gen 3 ssd because you could do exactly what a gen 3 squid do : pass the x4 gen 3 full bandwith on a slower but wider gen2 x16 slot or get full 4x4gen3 speed on a gen3 x 16 machine.
any how for gpu and in anycase the cyclone has a slower but wider bus to each of the 3 slot.
therefore it is recomended by blackmagic for gpu expension for davinci multi-gpu setup.
the netsor is a beter deal because you have a more modern switch, in a well rounded package, and the cyclone is way more expensive and require a more complex setup, but if you want to run gpu it is more versatile and you can run 3 gpu at full x16 gen2 bandwith.
I had the netstor it is a fantastic expension box but you cant run 3 high end 300w gpu, while the cyclone in the sc846 chassis with the 1200w redondant servergrade psu cant run 3 titan Xp all day without making breaking a sweat.
but you have a freaking jumbo-jet at take-off on your desk, this is why i built my custom g5 chassis with one 600w modular psu and one 1600w mining psu
witch have 8 double width slot and is dead silent.
see it as a custom netstor 255A on stéroid.
it hasn’t cost me much because I bought everything cheap and used on ebay, but if I had to pay for each component brand new in a true 20 slot redundant servergrade chassis I was looking north of 10k... and it actually cost me around 1000$.
i just sold my titan XP and i’m going to test 4 Rx580 because they are super cheap right now, and see if mac os can handle it. if it does support 4 gpu, i will go to 4 vega 7...
the custom 8 slot chassis allow me to have 4gpu , 1 areca raid card, 1 10gbe nic, 1 usb3 nic, and one blackmagic decklink card in one enclosure and to move it on the computer where I need it. see it more as a super USB hub.
the squid stay in slot 2 of the mac.
even if i upgrade to the future macpro and if it got gen3 or 4 x 16, I will keep it this way because it is so convenient to have everything in one external box.
same for my storage, I use to have promise fancy thunderbolt raid 8bay enclosure and now i have a xserve
hooked to the sc846 chassis via the small cyclone enclosure with 2 gtx780 6gb one sm951, one 10gbe nic and two areca raid card : one 1880xi24 for the 24 4Tb hsgt sas drive internaly and one 1880ix12 to two daisy chained netapp 4642 enclosure filled with 2 tb hd.
the xserve is in my basement and is my archival system and the macpro is on my desk with the custom 8 slot expender.
all is connected via 10gbe so i can work from my desk with the media being in the xserve if not to heavy, or just move them back and foward if I need to have it on local fast ssd.
very often when I work I send some export on the xserve even if it is slower, because then i can keep editing on the macpro.
but as soon as there will be a cheap and faster network nic, everything will go in the basement and I will just have the screens, keyboard and mouse on my desk and all the rest in the basement.
I do very long timelapse sometimes several years with 2-3 dslr shooting 1 25 megapixel raw file every minutes... so when i have to edit a 600k image timelapse i cant just do it on a brand new fancy machine. I need reliability of true servergrade hardware.
believe me or not when you are exporting a 8k sequence prores 444 from a media folder of 470k eos1dx2 raw files it takes 12 days, so you dont want it to crash because a you are using a single PSU machine.
dont ask me how I know it....♂️
other than that the netstor product are great and actually this is how I got in pcie expensions : i started with the na333b then had the na255, and got hooked to it so i sold the promise and the hackintosh I7 I had and went to cMP4.1 with twin x5690 and the full monty ... for actually super cheap.
[doublepost=1558814184][/doublepost]
[doublepost=1558814519][/doublepost]
this is the one i have in my custo chassis
[doublepost=1558814977][/doublepost]there is a big bro to this one with two x16 uplink and two 96 lane switch but i can find one. they usually run for well over 10k. it is used to have 8 gpu in linux machine.
I have the squid gen 2 so there is no gain and the cycline is a gen 2 swich.
a gen3 gpu couldnt care less than this to be on a gen2 slot as long as it is 16 lanes.
I think you are trying to convince someone witch is beter.
they are different animal.
pcie work both way so in a gen 3 machine the netsor as a theorical bandwith of 16xgen 3 to the enclosure, but the cards themselves are conected via a x8 gen 3 link.
on the cyclone everything 3 of the slot gets x16 gen 2 bandwith.
even in a gen 3 machine the cyclone has a greater bandwith to each of those 3 16x slot because gen 3x8 is less than gen2x16
even the titan Xp cant saturate a gen2x16 link both way therefore the one card can be receiving while the second or 3 card can be transmiting.
so even on a gen 3 machine each of the 3 x16 gen 2 slot have more bandwidth.
the only case where the netstor would give a better bandwith is with gen 3 ssd because you could do exactly what a gen 3 squid do : pass the x4 gen 3 full bandwith on a slower but wider gen2 x16 slot or get full 4x4gen3 speed on a gen3 x 16 machine.
any how for gpu and in anycase the cyclone has a slower but wider bus to each of the 3 slot.
therefore it is recomended by blackmagic for gpu expension for davinci multi-gpu setup.
the netsor is a beter deal because you have a more modern switch, in a well rounded package, and the cyclone is way more expensive and require a more complex setup, but if you want to run gpu it is more versatile and you can run 3 gpu at full x16 gen2 bandwith.
I had the netstor it is a fantastic expension box but you cant run 3 high end 300w gpu, while the cyclone in the sc846 chassis with the 1200w redondant servergrade psu cant run 3 titan Xp all day without making breaking a sweat.
but you have a freaking jumbo-jet at take-off on your desk, this is why i built my custom g5 chassis with one 600w modular psu and one 1600w mining psu
witch have 8 double width slot and is dead silent.
see it as a custom netstor 255A on stéroid.
it hasn’t cost me much because I bought everything cheap and used on ebay, but if I had to pay for each component brand new in a true 20 slot redundant servergrade chassis I was looking north of 10k... and it actually cost me around 1000$.
i just sold my titan XP and i’m going to test 4 Rx580 because they are super cheap right now, and see if mac os can handle it. if it does support 4 gpu, i will go to 4 vega 7...
the custom 8 slot chassis allow me to have 4gpu , 1 areca raid card, 1 10gbe nic, 1 usb3 nic, and one blackmagic decklink card in one enclosure and to move it on the computer where I need it. see it more as a super USB hub.
the squid stay in slot 2 of the mac.
even if i upgrade to the future macpro and if it got gen3 or 4 x 16, I will keep it this way because it is so convenient to have everything in one external box.
same for my storage, I use to have promise fancy thunderbolt raid 8bay enclosure and now i have a xserve
hooked to the sc846 chassis via the small cyclone enclosure with 2 gtx780 6gb one sm951, one 10gbe nic and two areca raid card : one 1880xi24 for the 24 4Tb hsgt sas drive internaly and one 1880ix12 to two daisy chained netapp 4642 enclosure filled with 2 tb hd.
the xserve is in my basement and is my archival system and the macpro is on my desk with the custom 8 slot expender.
all is connected via 10gbe so i can work from my desk with the media being in the xserve if not to heavy, or just move them back and foward if I need to have it on local fast ssd.
very often when I work I send some export on the xserve even if it is slower, because then i can keep editing on the macpro.
but as soon as there will be a cheap and faster network nic, everything will go in the basement and I will just have the screens, keyboard and mouse on my desk and all the rest in the basement.
I do very long timelapse sometimes several years with 2-3 dslr shooting 1 25 megapixel raw file every minutes... so when i have to edit a 600k image timelapse i cant just do it on a brand new fancy machine. I need reliability of true servergrade hardware.
believe me or not when you are exporting a 8k sequence prores 444 from a media folder of 470k eos1dx2 raw files it takes 12 days, so you dont want it to crash because a you are using a single PSU machine.
dont ask me how I know it....♂️
other than that the netstor product are great and actually this is how I got in pcie expensions : i started with the na333b then had the na255, and got hooked to it so i sold the promise and the hackintosh I7 I had and went to cMP4.1 with twin x5690 and the full monty ... for actually super cheap.
[doublepost=1558814184][/doublepost]
yep the 8 slot have a 96 lane switch
[doublepost=1558814519][/doublepost]
[doublepost=1558814977][/doublepost]there is a big bro to this one with two x16 uplink and two 96 lane switch but i can find one. they usually run for well over 10k. it is used to have 8 gpu in linux machine.
The only model I've seen on eBay these days:
https://www.ebay.fr/itm/offers-acpt-Cyclone-Microsystems-PCIe-x8-PCIe-414-PCIe-425-Adapter-Cable/192860214370?_trkparms=aid=111001&algo=REC.SEED&ao=1&asc=20160908105057&meid=79f9b4d23df44c8e988d856732af897b&pid=100675&rk=1&rkt=15&sd=192860214370&itm=192860214370&_trksid=p2481888.c100675.m4236&_trkparms=pageci:56dbf2db-7f35-11e9-9d3b-74dbd18017bb|parentrq:f0eda72d16a0ad483cf88c04ffe879f4|iid:1
It's a PCIE 1.0 x8X8X8 x4x4x4x4 with a PCIE 425 adapter.
https://www.ebay.fr/itm/offers-acpt-Cyclone-Microsystems-PCIe-x8-PCIe-414-PCIe-425-Adapter-Cable/192860214370?_trkparms=aid=111001&algo=REC.SEED&ao=1&asc=20160908105057&meid=79f9b4d23df44c8e988d856732af897b&pid=100675&rk=1&rkt=15&sd=192860214370&itm=192860214370&_trksid=p2481888.c100675.m4236&_trkparms=pageci:56dbf2db-7f35-11e9-9d3b-74dbd18017bb|parentrq:f0eda72d16a0ad483cf88c04ffe879f4|iid:1
It's a PCIE 1.0 x8X8X8 x4x4x4x4 with a PCIE 425 adapter.
Here you go complete system:
https://www.ebay.com/itm/PCIe2-2707...695396?hash=item340aa21b64:g:sSIAAOSw~pVc3Ess
Make an offer
squid gen 2 limits gen 3 NVMe devices to 1500 MB/s each. Your 4000 mo/s number comes from raiding 3 or more devices. A gen 3 squid can achieve that with fewer devices. For a Mac Pro host, a gen 3 squid can achieve the same max as a gen 2 squid in either a gen 2x16 or gen 3x8 slot, while a gen 2 squid can achieve max only in a gen 2x16 slot.
Right. A gen 3 gpu will have the same bandwidth between a gen 2 x16 slot and a gen 3 x8 slot. Older gen 2 or gen 1 GPUs will have more bandwidth in a gen 2 x16 slot.
I explained the differences. Someone can choose which differences they prefer.
The difference is negligible. gen2x16 = 64 Gbps. gen3x8 = 63.015384615384615 Gbps.
Any gen 3 device up to x8 will have better bandwidth in the Netstor. gen 3 x16 devices will have the same bandwidth as the Cyclone (but only in the Cyclone's gen 2x16 slots). Gen 3 NVMe, USB 3.1 gen 2 cards, x8 M.2 carrier cards, Thunderbolt 3 cards, ...
The cyclone is more expensive? gen 2 8 slot or 5 slot or both? Or do you mean a new Cyclone is expensive, while your used Cyclone is less expensive?
The Netstor can run 4 gen 3 x8 GPUs at the same speed as gen2 x16.
The Netstor has a 1200W PSU also. There is a Model NA255A-1500 with 1500W PSU recommended for something with four RTX 2080Ti. Otherwise, they say 1200W is enough for RTX 2070, GTX 1080Ti or Titan X (Pascal).
The Netstor seems quiet enough even with the cover off. Take-off is a couple seconds. The power supply can be replaced with something larger. There might be enough space for a second power supply.
Yup, used stuff can have good deals. Netstor can handle 4 Rx580 but not sure about vega 7 (you mean Radeon VII?) without PSU upgrade.
Yes, more slots = more stuff can be connected. You have host card in slot 1 of the Mac? Do you have any graphics in the Mac? If not, then you can't disconnect the expansion and still boot your Mac.
@joevt
I‘m not doing a who can be peeing the further contest here.
I have OWNED all the mentioned, and I HAVE the hardware to do realworld test.
I’ve been using pcie extension since 2014 so trust me when I say that in real word experience gen 2 x16 is faster than gen3x8 on a gen2 machine.
actually there is basicaly no diference in render time in most video software if the card is on gen2x8 or gen2x16, or gen3x8 or gen3x16.
in fact my macpro got faster render with 2 gpu than my hackintosh that was on gen3 with a Z77 mobo and 3770k.
so I agree the netstor is a good product, but as I had both, the cyclone is a very good alternative if you can find it for cheap on ebay.
would I be on the hunt for a pcie expension I would definitively go for a rackmount one with drives before going to just a GPU box.
if you want a normal gpu box the netsor
may be the way to go, but if you need a swiss knife like me then get a cyclone.
also worth noticing : with the netsor i could never use it in a thunderbolt chassis where i can plug both of the cyclone on my sonnet t2 chassis.
I also had some weird issues with the areca raid card on the netstor Na333 while I never ever had a issue with the cyclone.
there is a kind of logic when you compare a 10k 10 year old made in USA expender to a 5 year old 2500$ taiwan made one.
but either way PCIE expension rule, and i just pray that the new mac pro 2019 will have at least 1 x16 slot.
my macpro never boot without one of the chassis and if i have to for maintenance, i just slap a gt120 or my gtx780 6gb in.
the Xserve have an internal gt120 and dont require a gpu to boot.
I‘m not doing a who can be peeing the further contest here.
I have OWNED all the mentioned, and I HAVE the hardware to do realworld test.
I’ve been using pcie extension since 2014 so trust me when I say that in real word experience gen 2 x16 is faster than gen3x8 on a gen2 machine.
actually there is basicaly no diference in render time in most video software if the card is on gen2x8 or gen2x16, or gen3x8 or gen3x16.
in fact my macpro got faster render with 2 gpu than my hackintosh that was on gen3 with a Z77 mobo and 3770k.
so I agree the netstor is a good product, but as I had both, the cyclone is a very good alternative if you can find it for cheap on ebay.
would I be on the hunt for a pcie expension I would definitively go for a rackmount one with drives before going to just a GPU box.
if you want a normal gpu box the netsor
may be the way to go, but if you need a swiss knife like me then get a cyclone.
also worth noticing : with the netsor i could never use it in a thunderbolt chassis where i can plug both of the cyclone on my sonnet t2 chassis.
I also had some weird issues with the areca raid card on the netstor Na333 while I never ever had a issue with the cyclone.
there is a kind of logic when you compare a 10k 10 year old made in USA expender to a 5 year old 2500$ taiwan made one.
but either way PCIE expension rule, and i just pray that the new mac pro 2019 will have at least 1 x16 slot.
my macpro never boot without one of the chassis and if i have to for maintenance, i just slap a gt120 or my gtx780 6gb in.
the Xserve have an internal gt120 and dont require a gpu to boot.
No reason to go anatomical here
I'm not saying which is better, I'm only trying to get at the differences.Did you test the Netstor before or after the Mojave firmware update? Did you check the PCIe link status registers to make sure the upstream was connecting at gen 2 instead of gen 1 speed? The Netstor has many dip switches. Did you use the gen 2 mode dip switch setting for your Mac Pro, or did you use the gen 3 mode? Looking at the PCIe link status registers for the downstream devices will tell you the current link speed.
Even if it was faster, you don't say by how much. And then you say there's no difference in render times, which is contradictory (but I believe that part, - it's similar to people doing crypto currency using GPUs with x1 link widths).
I haven't tried that yet. Could be an issue that can be fixed with a dip switch change.
yep it has to be on gen 2 because areca 1880 dont play well with gen3 to gen2 on pcie swich everything has to be either gen 2 or gen 3.No reason to go anatomical here
Did you test the Netstor before or after the Mojave firmware update? Did you check the PCIe link status registers to make sure the upstream was connecting at gen 2 instead of gen 1 speed? The Netstor has many dip switches. Did you use the gen 2 mode dip switch setting for your Mac Pro, or did you use the gen 3 mode? Looking at the PCIe link status registers for the downstream devices will tell you the current link speed.
Even if it was faster, you don't say by how much. And then you say there's no difference in render times, which is contradictory (but I believe that part, - it's similar to people doing crypto currency using GPUs with x1 link widths).
I haven't tried that yet. Could be an issue that can be fixed with a dip switch change.
worth mentioning that all my thunderbolt machine are gen2 not gen 3 , never tryied T3 with pcie expension because gen3x4 is necessarly slower than gen 2x16
It's something maybe a person using any Mac that's not a classic Mac Pro might consider if they want more slots and already have a PCIe expansion box. If they don't, then they can just chain Thunderbolt 3 expansion boxes until they have the right number of slots (chaining is not supported for eGPUs though). They range from 1 to 3 slots. I don't recall any having more slots or large enough power supplies for multiple graphics cards, but that's beyond the scope of this thread. I think PCIe expansion hosted from a Thunderbolt expansion box is fair game here though.
You said it didn't work well with a sonnet t2 chassis. Maybe you had a t2 Mac? If the cure is a t3 chassis (not likely?) then it could be used with a t2 Mac using an Apple Thunderbolt 3 to Thunderbolt 2 adapter.
I've looked on this auction, this isn't the PCIE2-2707 model, but an old PCIE 1.0 7 slots x8 (x8/x4).
Kinda strange cause it is advertised as PCIe2-2707 andI've looked on this auction, this isn't the PCIE2-2707 model, but an old PCIE 1.0 7 slots x8 (x8/x4).
- Five Full Length PCI Express Ports
Hey joevt, hello edgerider,
nice to hear some good news from external pcie adapter stuff. I am also working since several years with the netsor 255 solution and nvidia cards especially for 3d rendering. But, when I upgraded to High Sierra with my cMP 5,1 my UI gets very sluggy with real peformance issues in displaying the OS windows. I never had this before on any mac... so I am now a little bit stuck on 10.12.6. Do have any solution or explanation about this???? I am running 1x1080 ti in the MacPro and two 1080ti in the netstor.
I never get more graficcards into the system than 3 ...? Your experience sound different.
Reworking on old G5 housings to bring them back into the cycle, sounds very interesting, I have still some G5s in the basement... Are there any howtodoes to be found in the www?
Thanks in advance for your help
Cheers from cologne
Martin
nice to hear some good news from external pcie adapter stuff. I am also working since several years with the netsor 255 solution and nvidia cards especially for 3d rendering. But, when I upgraded to High Sierra with my cMP 5,1 my UI gets very sluggy with real peformance issues in displaying the OS windows. I never had this before on any mac... so I am now a little bit stuck on 10.12.6. Do have any solution or explanation about this???? I am running 1x1080 ti in the MacPro and two 1080ti in the netstor.
I never get more graficcards into the system than 3 ...? Your experience sound different.
Reworking on old G5 housings to bring them back into the cycle, sounds very interesting, I have still some G5s in the basement... Are there any howtodoes to be found in the www?
Thanks in advance for your help
Cheers from cologne
Martin
@piznitram sounds like the nvidia driver is not corectly instaled, it can be bugy
-- How to "fix" web driver issues by force installing the older driver --
- https://forums.macrumors.com/threads/nvidia-webdriver-w-high-sierra-problem.2115396/#post-25989297
- https://forums.macrumors.com/threads/nvidia-driver-problems-beyond-10-3-0.2126322/#post-26218811
-- How to "fix" web driver issues by force installing the older driver --
- https://forums.macrumors.com/threads/nvidia-webdriver-w-high-sierra-problem.2115396/#post-25989297
- https://forums.macrumors.com/threads/nvidia-driver-problems-beyond-10-3-0.2126322/#post-26218811
Disable SIP
In high Sierra, open terminal and run
Code:
bash <(curl -s https://raw.githubusercontent.com/Benjamin-Dobell/nvidia-update/master/nvidia-update.sh) 378.10.10.10.15.114This may fix the multi GPU issue
With a lot of patience, I've found a Cyclone Microsystems PCIE2-427 backplane for less than 30$ , a Molex X16 PCIE cable (ridiculously short and rigid). I've just made a mistake to buy a PCIE2-426 card in low profile.
For less than 200$, it's a nice upgrade.
Do you know if I can buy just a full profile PCIE bracket, and a I/O shield for the 5 slots backplane?
, a Molex X16 PCIE cable (ridiculously short and rigid). I've just made a mistake to buy a PCIE2-426 card in low profile.For less than 200$, it's a nice upgrade.
Do you know if I can buy just a full profile PCIE bracket, and a I/O shield for the 5 slots backplane?
tu parles sans doute français?With a lot of patience, I've found a Cyclone Microsystems PCIE2-427 backplane for less than 30$
For less than 200$, it's a nice upgrade.
Do you know if I can buy just a full profile PCIE bracket, and a I/O shield for the 5 slots backplane?
je continue en anglais pour que tout le monde comprenne.
I never found any i/o shield as i thing those were proprietary to cylone.
the bracket are not easy to find , but my best bet would be to go in a computer brocker with the card, and tell them you need a bracket , then search the bin with all the old hba and buy whatever card were at least the screw from the bottom can fit.
as you see the cable is very stiff and you absolutely want a solid bracket , because the cable will tend to try to rip the card of the board.
regarding the chassis consider something with excellent ventilation capacity and space for one very big psu... because to power the board + 3 gpu, and two power hungry card such as one twin 10gbe and one raid card you will need 1200w minimum.
dont spare any expense on getting a trully silence chassis because you cant manage fan speed from the board.
Thanks for your answer edgerider.
My "eGpu" is in this PC case:
phanteks.com
I can add a mini-itx motherboard in it.
Actually, I use a Be Quiet 400W PSU, for the backplane and the 3 GPU, the two GTX 1070 (150 W each), are connected to the 8pin pcie cables, the RX 580 is on a vertical riser, with a 8pin pcie extender from the Mac Pro PSU. I've undervolted this GPU. I'll buy a good 1000W or 1200W Seasonic/Phanteks Revolt.
No shutdowns during benchmarks.
I've thought buying some NVME and two Highpoint cards 7101A, but it will be more expensive than making a new PC/Hackintosh and using the backplane for GPU.
I'll make some tests with the Kryo M2 in slot 3 in raid 0 with the dual Syba Nvme in slot 1 or in the Cyclone backplane.
Merci.
Guillaume
PS:for the PCIE426 low profile, I'll sell it, and buy a full profile one.
My "eGpu" is in this PC case:
Phanteks Innovative Computer Hardware Design
phanteks.com
I can add a mini-itx motherboard in it.
Actually, I use a Be Quiet 400W PSU, for the backplane and the 3 GPU, the two GTX 1070 (150 W each), are connected to the 8pin pcie cables, the RX 580 is on a vertical riser, with a 8pin pcie extender from the Mac Pro PSU. I've undervolted this GPU. I'll buy a good 1000W or 1200W Seasonic/Phanteks Revolt.
No shutdowns during benchmarks.
I've thought buying some NVME and two Highpoint cards 7101A, but it will be more expensive than making a new PC/Hackintosh and using the backplane for GPU.
I'll make some tests with the Kryo M2 in slot 3 in raid 0 with the dual Syba Nvme in slot 1 or in the Cyclone backplane.
Merci.
Guillaume
PS:for the PCIE426 low profile, I'll sell it, and buy a full profile one.