Mantiz Venus - different SSD performance using enclosure ports versus Macbook ports?  

Page 2 / 3 Prev Next
 

Trepach8225
(@trepach8225)
Active Member
Joined:11 months  ago
Posts: 9
July 23, 2017 8:01 pm  

Friends.  I would like to add an opinion and would agree with @Mymantiz_John regarding choice.  There are more and more options in the eGPU sphere which bodes well for all of us.  Every unit has its positive and negative aspects, and there are some great ideas for DIY and modifications.  However, I chose the Mantiz Venus precisely because I was looking for more than simply a eGPU box.   I desired an eGPU/docking station, knowing full well that could impact some GPU performance.  There is something elegant about coming into the office and plugging in just one cable.  If you desire only a box to house your GPU then there are many good option, yet, some of us are looking for more.  It’s early days in the eGPU/TB3 field, but we have come a long way when we consider what things were like only 24 months ago. Thanks to Mantiz and other companies that are forging ahead. Cheers!


ppm liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 24, 2017 4:46 am  

zwubel, thank you. We see the Mantiz hosts it’s ports using a 05e3:0617 Genesys 5Gbps USB 3.1 Gen1 hub.  Genesys’ range of USB 3.1 hubs is at:  http://www.genesyslogic.com/en/product_list.php?1st=3&2nd=10

USB 3.1 Gen1 is our 2008 USB 3.0 undergoing a name change:

https://www.msi.com/blog/usb-3-1-gen1-gen2-explained

USB 3.0 was launched in 2008, it provided an unprecedented increase in transfer speeds (up to 5Gbps) and added more power to connect more peripherals. In 2013 another major leap was made with the introduction of USB 3.1, doubling the speeds to as much as 10Gbps!

When USB 3.0 became USB 3.1 Gen 1

Recently however, the USB-IF organisation published an article where they announced that USB 3.0 connectors capable of 5Gbps (SuperSpeed) would now be classified as USB 3.1 Gen 1. While the new USB 3.1 connectors, capable of 10Gbps (SuperSpeed+) would now be classified as USB 3.1 Gen 2.

USB 3.0 is now called USB 3.1 Gen 1 (Speed up to 5Gbps)
USB 3.1 is now called USB 3.1 Gen 2 (Speed up to 10Gbps)

This tells us then:

– the total combined bandwidth for all Mantiz Venus ports (SATA+4xUSB 3.0+LAN) is 5Gbps
 


With this 5Gbps USB 3.1 Gen1 hub hosting the Mantiz Venus ports, what is the maximum combined bandwidth available to the enclosure ports?

5Gbps or 625MB/s

The Mantis Venus is pumping SATA, LAN and 5*USB 3.0 through the bandwidth of a system’s single USB 3.0 port!!  One thing that could have helped here is ..

 

Why didn’t the Mantiz Venus implement a 10Gbps USB 3.1 Gen2 interface for it’s ports?

Why wasn’t a bigger 10Gbps USB 3.1 Gen2 hub used to host the ports? Mymantiz_John alluded earlier on that a USB-C 3.1 port would be added to the next revision of the product. Will it be used there?

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 24, 2017 11:13 am  

To avoid Misleading & rumors around this forum.

Please check the below test data:

An upcoming Firmware:

Host: 2017  MBP 15″ 

Device: Mantiz Venus EGPU with MSI 1060 6G

CUDA Z:

 2017 07 24 下午6.41.52

Copy a 25G file from Device to Host ( USB 3.0  HDD ) , see CUDA Z Numbers here:

 2017 07 24 下午7.08.10

Copy 25G file from Host to Device ( USB 3.0 HDD ), See CUDA Z numbers here:

 2017 07 24 下午7.11.53

The New FIRMWARE is going to be the dynamic. 

 

 

Mantiz: ShopFacebookTwitter


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 24, 2017 12:05 pm  

Update In Game test

Host: 2017 MBP KabbyLake 15″

Device: Mantiz Venus + 1060 in Bootcamp

Game Far Cry Primal ( PCI bandwidth Heavy Game )

Mantiz FIRMWARE: Upcoming one

 2017 07 24 下午7.53.06

Min: 51 FPS

AVG: 63FPS

Max: 74FPS

Condition II:

Copy 25G from Host to Device USB 3.0 HDD 

See the picture

 2017 07 24 下午7.58.12

Result:

 2017 07 24 下午8.02.21

Min: 52

AVG:62

MAX:73

While Host to Device is sending 98-100MB data/ per sec to the device, In game Performance there’s ” NO DROP ” ” NO PERFORMANCE ISSUE”

 

Mantiz: ShopFacebookTwitter


ReplyQuote
PepeCZ
(@pepecz)
Eminent Member
Joined:8 months  ago
Posts: 36
July 24, 2017 5:43 pm  

Hi Johny, can I have novice question – what you mean that new firmware will be dynamical?

You mean for example: when I will have in my Venus for example Fury X and HDD, but in time when I will be using GPU for games and zero percent using hdd so Venus will work 100% for GPU. And when I will be for example in same time playing and copy files into hdd so speed for GPU will be little smaller .. and this will work dynamical? so when copy files will be finished so again GPU will have max performence

Edited: 7 months  ago

MY ULTIMATE DESKTOP BUILD
CPU AMD Ryzen Threadripper 1950X (16C/32T), CPU COOLER Corsair Cooling Hydro Series H115i, GPU Sapphire AMD Radeon RX Vega64 8GB HBM2, MOTHERBOARD ASUS ROG ZENITH EXTREME, RAM G.SKILL 16GB DDR4 3200MHz CL16 Trident Z RGB, SSD Kingston KC1000 480GB MLC, PSU Seasonic Prime SSR-1000TD, CASE PHANTEKS Enthoo Pro M Tempered Glass
MY NOTEBOOK
NOTEBOOK Acer Aspire A515-51G (i5-8250U 4C/8T, 8GB DDR4, 128GB SSD, 1TB HDD, NVIDIA GeForce MX150)
MY PERIPHERALS
DISPLAY Samsung C24FG73, KEYBOARD Logitech G810, MOUSE Logitech G703, MOUSEPAD Logitech G PowerPlay, WHEEL Thrustmaster T300RS
OTHER
ROUTER NETGEAR Nighthawk X4S, PHONE OnePlus 5 MidnightBlack, CHAIR DXRacer RV001/N


ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 24, 2017 6:20 pm  

@Mymantiz_john

1) I must say thank you for the support you give to your product! For sure we can see no other brand engineer here or in other places that hear customers and follows them this way. Great.

2) Impressive results for combined bandwidth. For the bandwidth, maybe it could be a problem with lower res (1080p) and higher fps? I will leave this to others more prepared than me on this topic. EDIT: Could I suppose also that the good results are also due to a transfer speed of only 100MB/s instead of 450MB/s?

3) Instead, for SSD speed, that you tested in a TB3 4x, 100MB/s to copy a file it seems unacceptable to me. It’s a very low speed for a SSD, even for a USB3.0 one. I suppose it has also the UASP protocol… Anyway, how do you explain such a low speed on a USB3.0 Gen1? As a comparison, I can get more speed from a USB3.1 (Gen2) card reader with a Lexar SD 1000x that goes @144MB/s… (yes the reader is Gen2 but the SD is lot lower than 5Gb/s, so we are still in the Gen1 comparison)!

Does speed grow making a copy without using the GPU?

 

 

 

Edited: 7 months  ago

BlackJack84 liked
ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 24, 2017 6:33 pm  
Posted by: PepeCZ

Hi Johny, can I have novice question – what you mean that new firmware will be dynamical?

You mean for example: when I will have in my Venus for example Fury X and HDD, but in time when I will be using GPU for games and zero percent using hdd so Venus will work 100% for GPU. And when I will be for example in same time playing and copy files into hdd so speed for GPU will be little smaller .. and this will work dynamical? so when copy files will be finished so again GPU will have max performence

An upcoming FW means: 

We had cooked the firmware and this firmware is waiting for Intel approval. Will be released @ any time.

Dynamic Means;
  If you don’t use any HDD or SSD through the IP ports, CUDA Z will stay @22xx MiB/S. But while you used, it will be reduce a bit to 21xx MiB/s. ( my USB 3.0 HDD occupied around 98MB/, if you use 3 USB HDD copying file, it will be reduce to 22xx Mib/S-300M=18xx Mib/s. and once you finished the tasks, it will be back to 22xx Mib/s

 

 

 

 

 

Edited: 7 months  ago

Mantiz: ShopFacebookTwitter


ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 24, 2017 6:36 pm  

Oh, sorry, I read now you had connected an HDD not a SSD, sorry… So results are good…

Can you test a SATA SSD?

Edited: 7 months  ago

ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 24, 2017 6:40 pm  
Posted by: franketto

@Mymantiz_john

1) I must say thank you for the support you give to your product! For sure we can see no other brand engineer here or in other places that hear customers and follows them this way. Great.

2) Impressive results for combined bandwidth. For the bandwidth, maybe it could be a problem with lower res (1080p) and higher fps? I will leave this to others more prepared than me on this topic. EDIT: Could I suppose also that the good results are also due to a transfer speed of only 100MB/s instead of 450MB/s?

3) Instead, for SSD speed, that you tested in a TB3 4x, 100MB/s to copy a file it seems unacceptable to me. It’s a very low speed for a SSD, even for a USB3.0 one. I suppose it has also the UASP protocol… Anyway, how do you explain such a low speed on a USB3.0 Gen1? As a comparison, I can get more speed from a USB3.1 (Gen2) card reader with a Lexar SD 1000x that goes @144MB/s… (yes the reader is Gen2 but the SD is lot lower than 5Gb/s, so we are still in the Gen1 comparison)!

Does speed grow making a copy without using the GPU?

 

 

 

1. Thank you, we need to clarify with some rumors especially for those Keyboard tester. We did test, we improve the disadvantage. and we provide the proof.

2. I’ll do more tests for High FPS, tomorrow.

3. I use my old U3 HDD to copy 25G file through Venus IO Ports, but I will make a comparison & crystal Disk benchmark tomorrow. Please wait for my update.

 

Mantiz: ShopFacebookTwitter


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 24, 2017 6:46 pm  

@franketto, he used a SATA HDD in a USB 3.0 enclosure attached to a Venus USB 3.0 port attached to a 4-lane TB3 notebook and then did the large file copies. 119MB/s D2H and 100MB/s H2D were the results.

The performance limits will be as explained earlier:

With this 5Gbps USB 3.1 Gen1 hub hosting the Mantiz Venus ports, what is the maximum combined bandwidth available to the enclosure ports?

5Gbps or 625MB/s

The Mantis Venus is pumping SATA, LAN and 5*USB 3.0 through the bandwidth of a single USB 3.0 port.

Due to USB interface overhead, the real-world figure will be less.  eg: Kisaten saw 247MB/s SSD peak transfer here on a 2-lane XPS 9350.

Mymantiz_John has shown the new, but not yet released to the public, firmware is dynamic so will see a decrease in eGPU bandwidth matching the bandwidth used by the Venus ports. The maximum decrease is a USB 3.1 Gen1 hub’s 5Gbps, or 15.6% of the TB3 enclosure total bandwidth.

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
goalque
(@goalque)
Honorable Member Admin
Joined:1 year  ago
Posts: 779
July 24, 2017 6:53 pm  
Posted by: franketto

Oh, sorry, I read now you had connected an HDD not a SSD, sorry… So results are good…

Can you test a SATA SSD?

Would be also interesting to see how Samsung portable SSD T1 (USB 3.0) and T3 (USB 3.1) perform. Should reach ~400MBps sequential write speeds.

automate-eGPU.shapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + Vega64@32Gbps-TB3 (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/RX580@32Gbps-TB3 (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 24, 2017 7:24 pm  

@nando4, yes, the interesting thing is he used only part of the 5Gb/s bandwidth and he got a decrease of 

1) Mantiz-Pc, 119MB/s D2H

from 2629.25 to 2498.4 that is a loss of 5%

1) Pc-Mantiz, 100MB/s H2D

 from 2217.92 to 2079.76 that is a loss of 6.2%

 

That’s a dynamically adaptation to the real use, not a fixed max 15.6%, so I think my previous question about using only the Gigabit (less than 100MB/s transfer), should result in less than this drop of 5% bandwidth. That’s good. Apart that I would use Gigabit transfer when not using GPU, so…

Of course, things would be better with a Gen2…


nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 25, 2017 9:23 am  
Posted by: franketto

Of course, things would be better with a Gen2…

Quite a bit better. It is not clear why Mantiz choose a 5Gbps USB 3.1 Gen1 hub (2008 technology originally called USB 3.0) to connect to the TB3 controller’s USB 3.1 Gen2 controller (2013 technology)?

venus usb hub

Above: Mantiz Venus  Genesys Logic 5Gbps USB 3.1 Gen1 hub (ID: 05e3:0617) [src: zwubel ]

 

As all the enclosure ports are hosted off this 5Gbps USB 3.1 hub equivalent to a USB 3.0 port’s bandwidth is available for 5xUSB 3.0, SATA and LAN combined. With such a bottleneck it then may be best to use those ports independently in a round-robin manner. Mantiz could help here . . .

 

Will Mantiz supply it’s customers an updated Venus daughterboard (contains the ports circuit) based on a 10Gbps USB 3.1 hub?

Mymantiz_John indicate earlier in this thread that a USB 3.1 port would feature in a future product.

Did Mantiz want to supply the daughterboard from that product which uses a 10Gbps USB 3.1 hub to it’s Mantiz Venus customers to lessen this bottleneck (5xUSB 3.0, 1xSATA, 1xLAN across the bandwidth of USB 3.0)?

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 25, 2017 11:35 am  

As asked here

Posted by: Mymantiz_John

Teach me, what is the bottleneck? 

Which has already been answered in this thread, along with a question:

Posted by: nando4

As all the enclosure ports are hosted off this 5Gbps USB 3.1 hub equivalent to a USB 3.0 port’s bandwidth is available for 5xUSB 3.0, SATA and LAN combined. With such a bottleneck it then may be best to use those ports independently in a round-robin manner. Mantiz could help here . . .

 

Will Mantiz supply it’s customers an updated Venus daughterboard (contains the ports circuit) based on a 10Gbps USB 3.1 hub?

Mymantiz_John indicate earlier in this thread that a USB 3.1 port would feature in a future product.

Did Mantiz want to supply the daughterboard from that product which uses a 10Gbps USB 3.1 hub to it’s Mantiz Venus customers to lessen this bottleneck (5xUSB 3.0, 1xSATA, 1xLAN across the bandwidth of USB 3.0)?

 

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 25, 2017 12:39 pm  
Posted by: Mymantiz_John

Teach me, what is the bottleneck? 

Well, it’s easy, I’ll make some examples:

1) let’s put a SATA SSD (capable of upto 500MB/s), you can’t use any other ports or Gigabit because that alone will saturate the bandwidth. Or Mantiz will get some bandwidth to Gigabit dropping the SATA bandwidth: this is a bottleneck.

2) Even if you don’t put a SATA SSD, put 2xUSB3.0 Gen1 enclosures SSD: each can be capable let’s say of 400MB/s. Well, each one alone will quite saturate your total bandwidht, so, like case 1), you get all speed from one and nothing from the other, or you drop speed from both: another bottleneck.

3) Let’s put all combined USB3.0 ports/SATA/Gigabit: don’t you imagine how low each device must go to adapt to the total 430MB/s? This is a bottleneck. Remember EACH ALONE 3.0 unit and SATA can saturate the actual Gen1 bandwidth into TB3. Of course with SSD, not with HDD, but today they are always more common.

It’s all a bottleneck: you must drop the speed of each unit because of the total Gen1 bandwidth of 430MB/s (theorical, in reality less than that…).

The only use without bottleneck is using mouse/scanner/printer and maybe 2xHDD 2,5”external + 1xHDD internal + Gigabit: ok, in this case you don’t have bottleneck. In all other cases, you will have.

In the more common situation, like the official photo of a SATA internal SSD, you just have a bottleneck… with only 1 unit connected!

 

With Gen2 you still will have a bottleneck at a certain point with more ports used, but it’s a lot better: it’s at least the double of actual total bandwidth… And anyway with 1SSD+Gigabit you’ll still go at full speed. Having this possibility today, I think it’s good to use it.

 

We are not saying you can’t use Mantiz actually, but if you want to make a top end product -and it deserves that- you should go with Gen2.

Remember that each SSD will give trouble speed actually, even only one.


nando4 liked
ReplyQuote
AngelRC
(@angelrc)
Eminent Member
Joined:10 months  ago
Posts: 27
July 25, 2017 6:06 pm  
Posted by: goalque
Posted by: franketto

Oh, sorry, I read now you had connected an HDD not a SSD, sorry… So results are good…

Can you test a SATA SSD?

Would be also interesting to see how Samsung portable SSD T1 (USB 3.0) and T3 (USB 3.1) perform. Should reach ~400MBps sequential write speeds.

As requested:

Test done with a Samsung T3 2TB SSD, Mantiz Venus and MBP 2016:

IMG 2230
T3

Connected to Venus (1st Picture) & Connected to MBP 2016 (2nd Picture)

T3 MBP
T3 Venus

 EDIT to clarify testing scenario: Test done with a Mantiz Venus without GPU and only T3 connected to it.

 

Edited: 7 months  ago

Freelancer IT Architect, Analyst & Developer
Macbook Pro 2016 15" 2,9GHz 16GB 1TB SSD Radeon Pro 460
Mantiz Venus with Sapphire RX Vega 64 8GB HBM2 Limited Edition + 3 x LG 27" 4K FreeSync displays + 1 x HTC Vive


theitsage and goalque liked
ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 25, 2017 6:35 pm  
Perfect.
So, now we can say the max speed on USB3.0 on Mantiz @TB3 4x with SSD is about rounded:
1) Mantiz-Pc, 365MB/s D2H
2) Pc-Mantiz, 350MB/s H2D.    EDIT (you uploaded a newly pic with higher speed…)
 
this confirms what we saw with HDD:
1) Mantiz-Pc, 120MB/s D2H
2) Pc-Mantiz, 100MB/s H2D

In D2H has more speed.

As we said earlier, the theoretically 430MB/s should be lower due to overhead. Here we have a 15% less due to overhead.
Maybe it could be something better with a SATA SSD, but anyway it will go on a USB3.0 connection, so maybe it’s the same speed.

 

EDIT:

Samsung T3 is 3.1 but Gen1… (max 450MB/s)

CrystalDisk bench will give some more consistent results than Blackmagic test…

Edited: 7 months  ago

ReplyQuote
kisaten
(@kisaten)
Active Member
Joined:11 months  ago
Posts: 15
July 27, 2017 3:36 am  

Been a long week for me, @JohnMymantiz_John and the support at Mantiz have been giving great support trying to get my system working, even sent a replacement board. I have also been contacting a bunch of dell reps (Japan and US) to try to figure out the deal with the 9350 and the weird behavior of this enclosure vs the Core (which had it’s own set of problems). 

Yesterday one of them told me that the stock wifi card can interfere with the thunderbolt controller (when PD is involved) so I had to lower the wifi power output to 50% (was at 75% and my other TB devices were fine but I guess I needed a bit more) and things started working properly (so far).

 

Here’s some tests from my system (maxed out xps13 9350) + Venus (500gb Samsung 850 evo SSD + MSI GTX 1060 6G Aero ITX OC ver.)

Crystal mark seems “normal” for now, I’ll try copying some stuff back and forth for real world estimates later tonight.

2017 07 26 19h47 32

(samsung 850 evo)

Rise of the Tomb Raider at 3200 x 1800 Max setting on my built in display benchmark running from the Venus SSD

2017 07 26 20h01 18

and now running on the built in display and laptop SSD (same settings)

2017 07 26 20h07 16

 

My TV is 1080p so here is Max settings at 1920 x 1080 on my built in display (laptop SSD)

2017 07 26 20h33 09

and same settings, 1920 x 1080 on my Sony 1080p TV (Laptop SSD + built in display disabled)

2017 07 26 20h39 46

 

Not bad so far, but no one else to compare with. I’ll be doing a few stress tests and wifi tests to make sure everything will continue to work, but these benchmarks look about right. Thanks @JohnMymantiz_John + team, I’ll be in contact with you over email in the next few days, and I’ll post some copy “tests” when I get a chance. 

Edited: 7 months  ago

ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 27, 2017 7:34 am  
Posted by: franketto
Perfect.
So, now we can say the max speed on USB3.0 on Mantiz @TB3 4x with SSD is about rounded:
1) Mantiz-Pc, 365MB/s D2H
2) Pc-Mantiz, 350MB/s H2D.    EDIT (you uploaded a newly pic with higher speed…)
this confirms what we saw with HDD:
1) Mantiz-Pc, 120MB/s D2H
2) Pc-Mantiz, 100MB/s H2D

In D2H has more speed.

So, with TB3 @2x the max speed on USB3.0 Gen1 is:

1) Mantiz-Pc, 350MB/s D2H

2) Pc-Mantiz, 320MB/s H2D.

A slightly less transfer speed, but quite near to equal. It depends also on the different benchmark prog…

So, this Nando4’s assumption :

    2-lane TB3: 5Gbps * 22/32 * 16/22 = 2.5Gbps or 313MB/s  [50% the bandwidth of a system’s USB 3.0 port!]

seems wrong. Even on TB3 @2x, still performs better than the 313MB/s theorical (to be diminished in practice). Instead we could say there is quite no loss between 2x and 4x.

Sadly the loss is in Gen1. 😉

Edited: 7 months  ago

ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 27, 2017 8:50 am  

@franketta, see the enclosure bandwidth below based on the TB3 enclosure being under no (max bandwidth) or complete eGPU load (min bandwidth). This will be the figures with the upcoming Mantiz Venus firmware that dynamically allocated USB hub bandwidth as opposed to the current firmware that statically allocated 10Gbps of bandwidth. Real-world throughput will be less than this due to protocol overhead.

 


 Q: If I want the best eGPU and storage performance, how should I host additional external storage?

 

 Anybody, especially the high performance 2016+ 15″ MBP owners wanting to maximize performance of external high performance storage AND eGPU devices is recommended to host their storage and eGPU from separate cables solutions to avoid TB3 link bandwidth competition.  That means your storage would be either:

  • a TB3 enclosure which separates eGPU and I/O traffic with 2 cables, eg: Asus ROG XG Station 2
  • a separate 10Gbps USB-C drive enclosure <--- recommended
  • a separate 5Gbps USB drive enclosure
  • a separate 22Gbps TB3 enclosure hosting your PCIe storage if can afford the $$ to do it

 

Example USB-C 2.5″ drive components

US$35 Satechi Aluminium 10Gbps USB-C 3.1 enclosure

Or US$20 Startech 10Gbps USB-C to SATA 2.5″ adapter

+ optional US$13 Cable Matters 3.3ft/100cm 10Gbps USB-C 3.1 cable

 

The above will ensure 100% of the TB3 enclosure bandwidth is available for the eGPU and your storage performance will be faster. Just be sure to check for DMI bottlenecking if using anything other than a 15″ MBP as anything more than a 10Gbps USB-C enclosure is a waste there. See here to understand Intel chipset architectural limitations.

 


Q: What is the eGPU bandwidth loss on enclosures with I/O ports (USB/SATA/Ethernet) ?

 

In this post we address what USB Generation hub TB3 enclosures with I/O ports use and how they distribute bandwidth between the eGPU and I/O ports. Conclusion being performance orientate users may wish to host high bandwidth storage on it’s own USB 3.1 Gen2 (10Gbps) port.

 

Firstly, let’s see what the enclosure marketting tells us about this: 

 TB3 enclosure marketting about I/O port usage effect on eGPU bandwidth

 

Mantiz Venus  Q: Will it reduce the bandwidth /FPS when I use the USB ports? A: No, it will not, FPS performance is the same with or without using the USB ports.

 

 

Nando4 >>  I/O ports use impinges on TB3 eGPU bandwidth. High bandwidth usage games will register a difference when I/O ports are active.  Seller confirms eGPU bandwidth differences here and here but argues there is no FPS difference between them.  Dubious marketting that opens the seller up for customer returns.

 

 

Razer Core V2 : The world’s first dual Thunderbolt 3 design provides dedicated lanes for both graphics and I/O ensuring fluid gameplay.

 

Nando4 >> Separate controllers that ultimately merge the PCIe lanes before sending the traffic down a single TB3 cable.

 


What is the I/O ports’ bus topology relative to the eGPU PCIe BUS?

 

Here  (src) we see TB3 enclosure has as PCIe BUS with two TB3 bridges, again with two PCIe Buses.

  •  the GTX960 eGPU is hosted off one PCIe BUS
  • a parallel PCIe BUS hosts the TB3 controller’s own USB 3.1 controller. In turn it hosts a enclosure USB 3.1 switch with I/O ports. That USB 3.1 switch can be USB 3.1 Gen2 (10Gbps) or, as we see in all I/O equipped enclosure ports, as USB 3.1 Gen1 (5Gbps).

 


Does the TB3 enclosure use static or dynamic allocation of I/O port bandwidth?

 

Example enclosure Electrical
link
static I/O port
bandwidth
available for
eGPU
TB3 cable “data”
bandwidth
Measured CUDA-Z
H2D
eGPU bandwidth

AKiTiO Node
(no I/O ports, full
eGPU bandwidth)

32Gbps none / 0% 32Gbps / 100% 22Gbps 22xxMiB/s   src
           
Mantiz Venus
(dynamic I/O bandwidth
allocation firmware)
32Gbps none / 0% 32Gbps / 100% 22Gbps 22xxMiB/s  src
           
Mantiz Venus
(static I/O bandwidth
allocation firmware)
32Gbps 10Gbps / 31.2% 22Gbps / 68.8% 22Gbps 15xxMiB/s   src

 

The TB3 enclosure’s firmware determines whether the onboard TB3’s controller’s USB port is using static or dynamic bandwidth allocation:

  • dynamic allocation with inactive I/O ports, provide full eGPU bandwidth. I/O ports can draw up to typically 5Gbps (15.6%) of available bandwidth.
  • static allocation of the typical 10Gbps (31.2%) for the TB3 USB 3.1 Gen2 (10Gbps) leaves a remaining 15xxMiB/s eGPU bandwidth.

 

We can guestimate the Razer Core V2, with a separate TB3 controller for I/O, will also apply dynamic I/O port allocation.

Oct-15-2017 alert>> Mantiz advises users to revert from dynamic (22xxMiB/s eGPU bandwidth) firmware to static (15xxMiB/s eGPU bandwidth) due to I/O ports being unusable under high eGPU load. 

 


I/O port bandwidth usage calculations

 

Below we calculate TB3 enclosure I/O bandwidth being under no (max bandwidth) or complete eGPU load (min bandwidth). Real-world measured I/O port throughput will be less than this due to protocol overhead.  

Hub type Min I/O ports bandwidth
(eGPU under 100% use)
Max I/O ports bandwidth
(eGPU not used)

 5Gbps USB 3.1 Gen1

4-lane TB3: 430MB/s
2-lane TB3: 313MB/s
625MB/s
     
10Gbps USB 3.1 Gen2 4-lane TB3:  860MB/s
2-lane TB3: 616MB/s
1350MB/s

 

A 10Gbps USB 3.1 Gen2 port can be found as another notebook USB-C port, or on Netstor HL23T, AKiTiO Node Pro/Lite enclosures

Min bandwidth calculation: 5Gbps/10Gbps is the USB controller input.  32 is the 32Gbps x4 3.0 TB3 enclosure controller. The 22 is the 22Gbps TB3 attenuatated ‘data’ link back to the notebook courtesy of Intel throttling. For a 16Gbps 2-lane TB3 notebook, a further 16/22 attenuation occurs on the notebook side.

  


How is eGPU bandwidth affected when a TB3 enclosure port is being used?

 

For a 5Gbps USB hub in the enclosure, eGPU bandwidth can be reduced by up to 15.6%. [5/32 * 100],  19xxMiB/s on CUDA-Z
For a 10Gbps USB hub in the enclosure, eGPU bandwidth can be reduced by up to 31.2% [10/32 * 100],  15xxMiB/s on CUDA-Z

 

 

Edited: 2 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


theitsage and cpauya liked
ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 27, 2017 11:28 am  

@kisaten

Could you try a Crystal test of SSD while you are stressing the eGPU?

So we could see the resulting loss.

 

 


ReplyQuote
kisaten
(@kisaten)
Active Member
Joined:11 months  ago
Posts: 15
July 27, 2017 12:17 pm  

Let Heaven benchmark run at extreme (just the window open rendering, not actually benchmarking) and got inconsistent results (probably dependent on what part of the bench was playing) So I upped the CDM passes to 3. Here’s the results (also incl. cuda-z)

2017 07 27 21h07 06
2017 07 27 21h03 26
2017 07 27 20h57 09
2017 07 27 20h53 39

 

Also did some copying, for “real world” with my tomb raider folder of mixed files kinda interesting:

First one is Venus to Internal (topped out at around 400, hovered in the 300s and spent the rest of the time around 200 ish)

2017 07 26 20h03 13

then Internal to Venus (pretty constant 250-260 <- the dip there is when started my screenshot program... or coincidence -though it went back up right after)

2017 07 27 21h08 55

Edited: 7 months  ago

ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 27, 2017 12:31 pm  

Yes, Crystal is better to test with 3/5 passes: it’s more accurate. So, let’s say, under eGPU stress, SSD goes:

Read 280MB/s

Write 290MB/s

This is a loss of about 18% on Read and 10% on write. So it seems Nando is quite right.

 

No wonder, real copy are lower speed than bench transfer: it’s normal, not a Mantiz limit.

But you should make a copy with only one very large file, not many files, one about 30GB or more. Only this way we could compare with the seq Crystal bench.

Edited: 7 months  ago

ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 27, 2017 12:44 pm  
Posted by: kisaten

Let Heaven benchmark run at extreme (just the window open rendering, not actually benchmarking) and got inconsistent results (probably dependent on what part of the bench was playing) So I upped the CM passes to 3. Here’s the results (also incl. cuda-z)

2017 07 27 21h07 06
2017 07 27 21h03 26
2017 07 27 20h57 09
2017 07 27 20h53 39

 

Also did some copying, for “real world” with my tomb raider folder of mixed files kinda interesting:

First one is Venus to Internal (topped out at around 400, hovered in the 300s and spent the rest of the time around 200 ish)

2017 07 26 20h03 13

then Internal to Venus (pretty constant 250-260 <- the dip there is when started my screenshot program... or coincidence -though it went back up right after)

2017 07 27 21h08 55

Do you satisfy with the result? 

Some PPL is asking why Mantiz Venus is not using the 3.1 GenII. allow me to reply in design point of view.

a. we start to develop the project since Sep 2016. For that Time being, there’s NO any PCI-USB Hub for 3.1. Gen II. No choice for that time being.

b. Asmedia there are few solution in Q1 2017, but one Hub is only supported for 2 ports, with the current 5 USB + 1Lan, we need to build 3 Asmedia PCI-USB Hub, it costs more than 18$ from the BOM Cost point of view. 

c. Look at the front USB cable. the latest technology for Real Gen II * real speed is around 7GB. it never can be that long. current one is 35CM, there’s no such thing ( around 7Gb ) appeared.

e. 2.4G wifi interfere issue. 3.1 GenII generate lots of 2.4G~2.5G noise, impact Laptop WIFI module a lot.

f. Considering all the phenomenon and timing, we still use the USB 3.1 Gen 1.

This is the best solution for the time being.

Edit:

Mantiz Venus is an eGPU, not design for the Thunderbolt PCIe SSD box or Thunderbolt PCIe-Fibernet lan card. We don’t need to have full H2D, In my various tests & INtel’s lab test result showed that more than 12xx Mib/s VS 22xx Mib/s is no significant difference.

But if we are belonged to the high speed storage or Fiber network adapter , it is really not suitable to add any other hub to share the 22GB ( Max Thunderbolt PCIe Bandwidth ). for the time being, it is a good & convenient IO ports for MBP 13″, 15″ and Windows Laptop users but also w/o GRAPHIC performance drop.

 

 

 

Edited: 7 months  ago

Mantiz: ShopFacebookTwitter


theitsage liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 27, 2017 12:45 pm  

@kisaten, does the Macbook allow Windows OS USB media boot fom a Mantiz USB 3.0 port? That’s important if you want to install Windows OS USB installation media. Otherwise would need to plug it into the USB-C (TB3) port directly.

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
goalque
(@goalque)
Honorable Member Admin
Joined:1 year  ago
Posts: 779
July 27, 2017 12:48 pm  

For me, they look pretty similar results, except in the last picture where Seq Q32T1 write speed has dropped down to 132 MB/s.

CrustalDiskMark does not yield consistent numbers. There is always variation. Valley/Heaven are not bandwidth hungry.

automate-eGPU.shapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + Vega64@32Gbps-TB3 (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/RX580@32Gbps-TB3 (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
kisaten
(@kisaten)
Active Member
Joined:11 months  ago
Posts: 15
July 27, 2017 1:13 pm  
Posted by: franketto

Yes, Crystal is better to test with 3/5 passes: it’s more accurate. So, let’s say, under eGPU stress, SSD goes:

Read 280MB/s

Write 290MB/s

This is a loss of about 18% on Read and 10% on write. So it seems Nando is quite right.

 

No wonder, real copy are lower speed than bench transfer: it’s normal, not a Mantiz limit.

But you should make a copy with only one very large file, not many files, one about 30GB or more. Only this way we could compare with the seq Crystal bench.

@franketto I copied a large folder of mixed files for “real world” numbers -how most people might experience it. Since the majority of my files arent over 10gb.

@nando4 I’m running an xps 9350, maybe you meant to ask @Mymantiz_John ?

 


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1580
July 27, 2017 1:19 pm  

@kisaten, same good question applies to your XPS 9350. Can you boot USB 3.0 Windows 10 installation media a Mantiz USB 3.0 port? We are seeing if the notebook UEFI firmware can communicate successfully with the enclosure’s Genesys Logic USB 3.0 controller to allow USB 3.0 media boot. This is before the OS has booted, so there are no drivers loaded yet.

Edited: 7 months  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
kisaten
(@kisaten)
Active Member
Joined:11 months  ago
Posts: 15
July 27, 2017 2:38 pm  

Bios is showing both the sata and usb drive connected on the Venus (allowing me to add them as boot options,) so I assume that that’s a yes?


nando4 liked
ReplyQuote
Rusty
(@rusty)
New Member
Joined:9 months  ago
Posts: 3
July 29, 2017 10:26 am  

Managed to get it going in the end, thank you 🙂

 

On another note, I am only getting around 140mb/s writes with a Samsung Pro 850 via the SATA port.

Via USB I get 225mb/s writes…. this doesn’t seem right to me?

When I had this installed in my iMac I was getting around 500mb/s write and reads.

Did someone say there is a firmware update to do with this?

If so when, and how would I update?

 

 


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined:12 months  ago
Posts: 425
July 30, 2017 11:41 am  
Posted by: Rusty

Managed to get it going in the end, thank you 🙂

 

On another note, I am only getting around 140mb/s writes with a Samsung Pro 850 via the SATA port.

Via USB I get 225mb/s writes…. this doesn’t seem right to me?

When I had this installed in my iMac I was getting around 500mb/s write and reads.

Did someone say there is a firmware update to do with this?

If so when, and how would I update?

 

 

Mantiz Venus Ports are under USB 3.1 Gen I bandwidth which is 5G. MBP is Gen II, 10G, which is larger than Venus 5G. 

 

Mantiz: ShopFacebookTwitter


wimpzilla liked
ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 30, 2017 4:16 pm  

This does not explain that behaviour!

He’s saying its SSD connected with SATA cable inside its Mac, is capable of 500MB/s.

 

When he connects this unit to the USB 3.1 5Gb of Mantiz he gets 225MB/s.

BUT

when he connects the SSD with the SATA port of the Mantiz he gets only 140 MB/s.

So, we know that SATA port of Mantiz isn’t a real SATA port bandwidth because it’s connected the same on the USB3.1 5Gb/s interface, but why SATA 5Gb/s is so much lower than USB 5Gb/s on the same board??

We are at a paradox that it’s better to connect an SSD enclosure on USB3.1 port instead of a direct SATA connection (even if it’s always a 5Gb USB bandwidth)…


ReplyQuote
gnu
 gnu
(@gnu)
Active Member
Joined:7 months  ago
Posts: 7
July 31, 2017 4:20 pm  
Posted by: Rusty

Managed to get it going in the end, thank you 🙂

 

On another note, I am only getting around 140mb/s writes with a Samsung Pro 850 via the SATA port.

Via USB I get 225mb/s writes…. this doesn’t seem right to me?

When I had this installed in my iMac I was getting around 500mb/s write and reads.

Did someone say there is a firmware update to do with this?

If so when, and how would I update?

 

 

I also have an EVO 850 and was disappointed by read/write performance using SATA port. My perfs are better than yours though, It goes up to 330mb/s for write and 370mb/s for read, but should be around 500mb/s as you said.

I don’t know if the bottleneck is hardware or software, but I don’t think there’s any firmware upgrade available for the Venus…


ReplyQuote
ikir
 ikir
(@ikir)
Honorable Member
Joined:1 year  ago
Posts: 588
July 31, 2017 4:43 pm  

I get higher numbers with Venus SATA port, last time I checked it was about 300MB/s

eGPU.it | LG 29" curved ultrawide display
MacBook Pro 2017 touchbar i7 3.5Ghz - 16GB RAM - 512GB PCIe SSD + Mantiz Venus with AMD Radeon RX 580


ReplyQuote
franketto
(@franketto)
Trusted Member
Joined:8 months  ago
Posts: 82
July 31, 2017 4:56 pm  

Well, 300MB/s is to be expected due to the architecture of Mantiz and TB3. That is a “normal” result for this configuration.

I suppose you, @gnu and @ikir are connected with a TB3 4x, not 2x.

 

Instead it seems @Rusty is connected with a TB3 @2x, so he has 225MB/s max speed. Either this is to be expected on 2x. BUT the 140MB/s on Sata, no, that does not make sense…

@Rusty:

1) are you connected on TB3 2x?

2) are you copying the same files on USB and Sata ports? To understand it better you should do a CrystalDisk Benchmark on both. Of course, all the test without using the eGpu…

Edited: 7 months  ago

ReplyQuote
Rycco
(@rycco)
Reputable Member
Joined:1 year  ago
Posts: 219
August 1, 2017 10:34 pm  
Posted by: franketto

Well, 300MB/s is to be expected due to the architecture of Mantiz and TB3. That is a “normal” result for this configuration.

I suppose you, @gnu and @ikir are connected with a TB3 4x, not 2x.

 

Instead it seems @Rusty is connected with a TB3 @2x, so he has 225MB/s max speed. Either this is to be expected on 2x. BUT the 140MB/s on Sata, no, that does not make sense…

@Rusty:

1) are you connected on TB3 2x?

2) are you copying the same files on USB and Sata ports? To understand it better you should do a CrystalDisk Benchmark on both. Of course, all the test without using the eGpu…

Which port on the Macbook Pro 2017 should I connect? The left one closer to the screen?

mid-2017 15" Macbook Pro RP555 + GTX1080i@32Gbps-TB3 (Mantiz Venus) + macOS & Win10


ReplyQuote
Page 2 / 3 Prev Next
  
Working

Please Login or Register