ALERT: half H2D performance issue on TI83 TB3 enclosures (Node, Venus, Devil Box, XG Station 2)  

Page 5 / 8 Prev Next
  RSS

valkylin000
(@valkylin000)
Active Member
Joined: 2 years ago
Posts: 9
April 7, 2017 12:57 am  

I am so  regret not buying AKiTiO Node.....ASUS has not responde my post on ROG forum dyet  🙁

Still got low performance in game with 1080 Ti...

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
Greenducky55
(@greenducky55)
Active Member
Joined: 2 years ago
Posts: 6
April 7, 2017 2:07 am  

 anyone else having errors trying to update the firmware on their node?

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
Greenducky55
(@greenducky55)
Active Member
Joined: 2 years ago
Posts: 6
April 7, 2017 3:17 am  
Posted by: Greenducky55

 

 anyone else having errors trying to update the firmware on their node?

   

Actually figured it out the firmware/driver for my thunderbolt controller was the problem. Just asus didn't offer an up to date one

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 7, 2017 3:50 am  
Posted by: valkylin000

 

I am so  regret not buying AKiTiO Node.....ASUS has not responde my post on ROG forum dyet  🙁

Still got low performance in game with 1080 Ti...

   

Asus ROG XG Station 2 updated H2D firmware fix on it's way - from the Asus ROG forum here

 

 

The new firmware with improved performance is in the middle of validation process, and should be release sometime soon.

 
Quote Originally Posted by joey.ken View Post
This is my setting of hardware

This is game FPS

This is someone same setting with GTX 1080

the game FPS difference is way too big...he 's cpu is i7 2700k

and you can see TI83-based enclosures VS TI82-based enclosures the host-to-device (H2D) bandwidth

http://imgur.com/DNn6Qyg

http://imgur.com/yX7weHr

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table    •    Several builds
2015 15" Dell Precision 7510 M1000M + macOS 10.14 & Win10


ReplyQuote
valkylin000
(@valkylin000)
Active Member
Joined: 2 years ago
Posts: 9
April 7, 2017 4:47 am  

Yes,that's my post on ROG forum~~ hope the FW can fix this problem~ 🙂

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


nando4 liked
ReplyQuote
VisViva
(@visviva)
Eminent Member
Joined: 2 years ago
Posts: 21
April 7, 2017 3:47 pm  

I just ordered an akitio node today, and it's expecting to come this weekend, and I'd love to see if this issue has been fixed.

 

I only have AMD graphics (RX470) as of now. Is there a way to test the throughout on AMD card? (i.e. no CUDA support)

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 7, 2017 3:58 pm  
Posted by: VisViva

 

I just ordered an akitio node today, and it's expecting to come this weekend, and I'd love to see if this issue has been fixed.

 

I only have AMD graphics (RX470) as of now. Is there a way to test the throughout on AMD card? (i.e. no CUDA support)

   

AMD's PCIeSpeedTest bandwidth testing utility worked under Win8 when I tried it. More details at https://egpu.io/forums/thunderbolt-enclosures/alert-akitio-node-half-h2d-bandwidth-issue/paged/2/#post-6142

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table    •    Several builds
2015 15" Dell Precision 7510 M1000M + macOS 10.14 & Win10


theitsage liked
ReplyQuote
theitsage
(@itsage)
Famed Member Admin
Joined: 2 years ago
Posts: 3207
April 7, 2017 4:00 pm  
Posted by: VisViva

I just ordered an akitio node today, and it's expecting to come this weekend, and I'd love to see if this issue has been fixed.

I only have AMD graphics (RX470) as of now. Is there a way to test the throughout on AMD card? (i.e. no CUDA support)

Run some game benchmarks as mentioned in this thread prior to the firmware update. Then run those same benchmarks after the firmware update to compare.

Best ultrabooks for eGPU use

eGPU enclosure buying guide

86 external GPU build guides


nando4 liked
ReplyQuote
win32asmguy
(@win32asmguy)
Active Member
Joined: 2 years ago
Posts: 8
April 7, 2017 7:10 pm  
Posted by: VisViva

 

I just ordered an akitio node today, and it's expecting to come this weekend, and I'd love to see if this issue has been fixed.

 

I only have AMD graphics (RX470) as of now. Is there a way to test the throughout on AMD card? (i.e. no CUDA support)

   

HWInfo64 has a "GPU Bus Load" sensor for my GTX 1070 which I have seen get as high as 20%, so I am guessing it correlates to a x16 PCIe 3.0 connection being 100%. If you have the same sensor on your RX470 its probably a good way to monitor throughput while benchmarking or gaming.

2016 15" HP ZBook Studio G3 + [email protected] (AKiTiO Node) + Win10


ReplyQuote
lalaboi
(@lalaboi)
Active Member
Joined: 2 years ago
Posts: 11
April 9, 2017 6:01 am  

Any idea how to update akitio node's firmware on a tb2 host?

Im on a 2015 15" m370x mbp and I cant update the node's firmware. Thunderbolt controller 156C

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
VisViva
(@visviva)
Eminent Member
Joined: 2 years ago
Posts: 21
April 10, 2017 2:56 am  

I just got the Akitio Node today, and i tested its transfer performance before and after firmware update.

 

I used the software called "BufferBandwidth.exe" under "C:\Users\<my user name>\AMD APP SDK\3.0\samples\opencl\bin\x86_64". It should be an executable that comes with AMD APP SDK 3.0
A little side notes, as for the previously posted "PCIe Speed Test v0.2",  i cannot get it working on my machine as it keeps complaining about a missing dll. (i tried using the dll with the same name under AMD APP SDK 3.0 directory, and it still returns an error)

I ran it on Windows 10 64bit, and my laptop is lenovo P50 (6700HQ and M1000M, TB3 via PCH), and i have a RX470 in Akitio Node. I ran it while using only external display (4K with FreeSync) via eGPU, which i later on considered a mistake. If anyone else is planning to do such test before and after firmware update, i recommend to not use display via eGPU so the bandwidth can be dedicated to speed test.

It is interesting to see the differences with "clEnqueueUnmapMemObject()"and "clEnqueueSVMUnmap()"

 

Below are the outputs before the firmware update.

Platform 0 : Intel(R) Corporation
Platform 1 : NVIDIA Corporation
Platform 2 : Advanced Micro Devices, Inc.
Platform 3 : Advanced Micro Devices, Inc.
Platform found : Advanced Micro Devices, Inc.

Selected Platform Vendor : Advanced Micro Devices, Inc.
Device 0 : Ellesmere Device ID is 00000273C11E4FF0
Build Options are : -cl-std=CL2.0 
Build:               _WINxx release
GPU work items:      32768
Buffer size:         33554432
CPU workers:         1
Timing loops:        20
Repeats:             1
Kernel loops:        20
inputBuffer:         CL_MEM_READ_ONLY 
outputBuffer:        CL_MEM_WRITE_ONLY 
inputSVMBuffer:      CL_MEM_READ_ONLY 
outputSVMBuffer:     CL_MEM_WRITE_ONLY 

Host baseline (naive):

Timer resolution     413.51  ns
Page fault           826.20  ns
Barrier speed        61.87   ns

CPU read             11.44 GB/s
memcpy()             8.88 GB/s
memset(,1,)          26.80 GB/s
memset(,0,)          27.12 GB/s


AVERAGES (over loops 2 - 19, use -l for complete log)
--------


1. Host mapped write to inputBuffer
 ---------------------------------------|---------------
 clEnqueueMapBuffer -- WRITE (GBPS)     | 5235.700
 ---------------------------------------|---------------
 memset() (GBPS)                        | 23.766
 ---------------------------------------|---------------
 clEnqueueUnmapMemObject() (GBPS)       | 1.161


2. GPU kernel read of inputBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.004

 Verification Passed!


3. GPU kernel write to outputBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.008


4. Host mapped read of outputBuffer
 ---------------------------------------|---------------
 clEnqueueMapBuffer -- READ (GBPS)      | 2.599
 ---------------------------------------|---------------
 CPU read (GBPS)                        | 10.929
 ---------------------------------------|---------------
 clEnqueueUnmapMemObject() (GBPS)       | 848.404

 Verification Passed!


5. Host mapped write to inputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueSVMMap -- WRITE (GBPS)        | 4915.834
 ---------------------------------------|---------------
 memset() (GBPS)                        | 22.472
 ---------------------------------------|---------------
 clEnqueueSVMUnmap() (GBPS)             | 1.172


6. GPU kernel execution using inputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.004

 Verification Passed!


7. GPU kernel write to outputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.008


8. Host mapped read of outputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueSVMMap -- READ (GBPS)         | 2.612
 ---------------------------------------|---------------
 CPU read (GBPS)                        | 11.237
 ---------------------------------------|---------------
 clEnqueueSVMUnmap() (GBPS)             | 779.217

 Verification Passed!


Passed!



 

 

Below are the outputs after the firmware update.

 

Platform 0 : Intel(R) Corporation
Platform 1 : NVIDIA Corporation
Platform 2 : Advanced Micro Devices, Inc.
Platform 3 : Advanced Micro Devices, Inc.
Platform found : Advanced Micro Devices, Inc.

Selected Platform Vendor : Advanced Micro Devices, Inc.
Device 0 : Ellesmere Device ID is 000001E11E241D70
Build Options are : -cl-std=CL2.0 
Build:               _WINxx release
GPU work items:      32768
Buffer size:         33554432
CPU workers:         1
Timing loops:        20
Repeats:             1
Kernel loops:        20
inputBuffer:         CL_MEM_READ_ONLY 
outputBuffer:        CL_MEM_WRITE_ONLY 
inputSVMBuffer:      CL_MEM_READ_ONLY 
outputSVMBuffer:     CL_MEM_WRITE_ONLY 

Host baseline (naive):

Timer resolution     402.90  ns
Page fault           1018.57 ns
Barrier speed        73.88   ns

CPU read             10.52 GB/s
memcpy()             7.70 GB/s
memset(,1,)          20.72 GB/s
memset(,0,)          19.82 GB/s


AVERAGES (over loops 2 - 19, use -l for complete log)
--------


1. Host mapped write to inputBuffer
 ---------------------------------------|---------------
 clEnqueueMapBuffer -- WRITE (GBPS)     | 1195.327
 ---------------------------------------|---------------
 memset() (GBPS)                        | 19.630
 ---------------------------------------|---------------
 clEnqueueUnmapMemObject() (GBPS)       | 2.384


2. GPU kernel read of inputBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.004

 Verification Passed!


3. GPU kernel write to outputBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.007


4. Host mapped read of outputBuffer
 ---------------------------------------|---------------
 clEnqueueMapBuffer -- READ (GBPS)      | 2.588
 ---------------------------------------|---------------
 CPU read (GBPS)                        | 9.091
 ---------------------------------------|---------------
 clEnqueueUnmapMemObject() (GBPS)       | 970.065

 Verification Passed!


5. Host mapped write to inputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueSVMMap -- WRITE (GBPS)        | 3309.140
 ---------------------------------------|---------------
 memset() (GBPS)                        | 19.060
 ---------------------------------------|---------------
 clEnqueueSVMUnmap() (GBPS)             | 2.432


6. GPU kernel execution using inputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.004

 Verification Passed!


7. GPU kernel write to outputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueNDRangeKernel() (secs)        | 0.007


8. Host mapped read of outputSVMBuffer
 ---------------------------------------|---------------
 clEnqueueSVMMap -- READ (GBPS)         | 2.619
 ---------------------------------------|---------------
 CPU read (GBPS)                        | 8.891
 ---------------------------------------|---------------
 clEnqueueSVMUnmap() (GBPS)             | 1021.940

 Verification Passed!


Passed!


 Edit: clarification about the PCIe Speed Test v0.2 

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


theitsage and nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 10, 2017 8:27 am  
Posted by: VisViva

 

I just got the Akitio Node today, and i tested its transfer performance before and after firmware update.

 

I used the software called "BufferBandwidth.exe" under "C:\Users\<my user name>\AMD APP SDK\3.0\samples\opencl\bin\x86_64". It should be an executable that comes with AMD APP SDK 3.0
A little side notes, i cannot get it working on my machine as it keeps complaining about a missing dll. (i tried using the dll with the same name under AMD APP SDK 3.0 directory, and it still returns an error)

I ran it on Windows 10 64bit, and my laptop is lenovo P50 (6700HQ and M1000M, TB3 via PCH), and i have a RX470 in Akitio Node. I ran it while using only external display (4K with FreeSync) via eGPU, which i later on considered a mistake. If anyone else is planning to do such test before and after firmware update, i recommend to not use display via eGPU so the bandwidth can be dedicated to speed test.

It is interesting to see the differences with "clEnqueueUnmapMemObject()"and "clEnqueueSVMUnmap()"

Below are the outputs before the firmware update.

1. Host mapped write to inputBuffer
 clEnqueueUnmapMemObject() (GBPS)       | 1.161

4. Host mapped read of outputBuffer
 clEnqueueMapBuffer -- READ (GBPS)      | 2.599

5. Host mapped write to inputSVMBuffer
 clEnqueueSVMUnmap() (GBPS)             | 1.172

8. Host mapped read of outputSVMBuffer
 clEnqueueSVMMap -- READ (GBPS)         | 2.612

Below are the outputs after the firmware update.

1. Host mapped write to inputBuffer
 clEnqueueUnmapMemObject() (GBPS)       | 2.384 ** doubled **

4. Host mapped read of outputBuffer
 clEnqueueMapBuffer -- READ (GBPS)      | 2.588

5. Host mapped write to inputSVMBuffer
 clEnqueueSVMUnmap() (GBPS)             | 2.432 ** doubled **

8. Host mapped read of outputSVMBuffer
 clEnqueueSVMMap -- READ (GBPS)         | 2.619

Thank you for posting your results confirming AMD RX470 is also affected by the H2D bandwidth bug. Here the OpenCL rather than CUDA interface is used. The latest AKiTiO Node firmware then giving you full bandwidth in both directions.

You also have a Lenovo P50 of which we have no existing eGPU implementation guide for. Would you like to submit one for our users to have confidence to proceed with a P50 TB3 eGPU implementation?

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table    •    Several builds
2015 15" Dell Precision 7510 M1000M + macOS 10.14 & Win10


Mymantiz_John and ikir liked
ReplyQuote
Mgear1981
(@mgear1981)
Eminent Member
Joined: 2 years ago
Posts: 36
April 10, 2017 9:05 am  
Posted by: lalaboi

 

Any idea how to update akitio node's firmware on a tb2 host?

Im on a 2015 15" m370x mbp and I cant update the node's firmware. Thunderbolt controller 156C

I have no idea if that is even possible. Perhabs try in bootcamp to install the thunderbolt software that would be compatible with your macbook thunderbolt controller. What you also can do is contact akitio support and ask them if it is possible to even update the firmware trough thunderbolt 2 or with the adapter ?

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
markbam
(@markbam)
Active Member
Joined: 2 years ago
Posts: 18
April 10, 2017 3:39 pm  
Posted by: Mgear1981

 

Posted by: lalaboi

 

Any idea how to update akitio node's firmware on a tb2 host?

Im on a 2015 15" m370x mbp and I cant update the node's firmware. Thunderbolt controller 156C

I have no idea if that is even possible. Perhabs try in bootcamp to install the thunderbolt software that would be compatible with your macbook thunderbolt controller. What you also can do is contact akitio support and ask them if it is possible to even update the firmware trough thunderbolt 2 or with the adapter ?

   

Reply from Akitio about firmware updates via TB2 and Adapter:

"The Node is compatible only with Thunderbolt 3 PCs running Windows 10, so the firmware utility will only be available for this kind of system, even in the future."

Mid-2017 MacBook Pro 15 inch Touchbar (Radeon Pro 560)
- Akitio Node (Radeon 580)
- Aorus Gaming Box (Geforce GTX 1080)


ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 10, 2017 7:35 pm  

Unfortunately we are not allowed to make a FW Update tool for MacOS.

This user was able to get it to work in BootCamp on his TB3 MBP. Maybe the same could work on a TB2 MBP.
https://egpu.io/forums/thunderbolt-enclosures/akitio-node-thunderbolt-3-egpu-enclosure/paged/6/#post-6718

Alias: KnightZero
AKiTiO Employee - Technical Specialist


ReplyQuote
Yukikaze
(@yukikaze)
Prominent Member Moderator
Joined: 2 years ago
Posts: 825
April 10, 2017 7:38 pm  

DanKnight, is it possible for you to develop a FW update tool for Linux? I would assume a Mac can be booted to Linux via a USB thumbdrive, and then such a tool can be used there to update the FW.

EDIT: While I am at it. How about Linux support in general? Are you involved with Intel on that front as well?

My eGPU Zoo - Link to my Implementations.
Want to output [email protected] out of an old system on the cheap? Read here.
"Always listen to experts. They'll tell you what can't be done, and why. Then do it."- Robert A. Heinlein, "Time Enough for Love."


ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 10, 2017 9:31 pm  

The only OS we are allowed to make a FW Update Tool for is Windows. The Node is certified only for Windows, so that is the only environment we can officially provide support.

Alias: KnightZero
AKiTiO Employee - Technical Specialist


ReplyQuote
VisViva
(@visviva)
Eminent Member
Joined: 2 years ago
Posts: 21
April 11, 2017 6:15 pm  
Posted by: nando4

 Thank you for posting your results confirming AMD RX470 is also affected by the H2D bandwidth bug. Here the OpenCL rather than CUDA interface is used. The latest AKiTiO Node firmware then giving you full bandwidth in both directions.

You also have a Lenovo P50 of which we have no existing eGPU implementation guide for. Would you like to submit one for our users to have confidence to proceed with a P50 TB3 eGPU implementation?

   

Thanks for the formatting of my post.

As for the P50 implementation, everything works out-of-box to me. I did not set any setting or DSDT in this process. I can try writing some later on once I gather more information in terms of benchmarks and in-game fps. 

 

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
taylorpoz
(@taylorpoz)
Active Member
Joined: 2 years ago
Posts: 5
April 13, 2017 2:57 pm  

Any updates on the Mantiz Venus FW update? The group buy closes tomorrow, and I would prefer not to purchase a bottlenecked device 😛

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 2 years ago
Posts: 489
April 13, 2017 5:36 pm  
Posted by: taylorpoz

 

Any updates on the Mantiz Venus FW update? The group buy closes tomorrow, and I would prefer not to purchase a bottlenecked device 😛

   

HI

Thanks for the comment, Intel didn't release the FW yet. It means there are still several things to be tested or modified. once it is released or allow to release, we will update the device.

 

Mantiz: ShopFacebookTwitter


ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 14, 2017 10:08 am  

How akitio did then? They already released it, you mean it is not with an official intel FW?

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 2 years ago
Posts: 489
April 14, 2017 10:16 am  

Intel clearly informed that the FW is still under testing and verification. It is not released yet. And I don't know the others but I do know Powecolor and Asus are also waiting for Intel's final release until now. 

Mantiz: ShopFacebookTwitter


ed_co liked
ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 14, 2017 7:29 pm  

Nice to know John. We bought your product because we trust you, mantiz' guys. And we know you will do your best to offer the best product out there. Because you listened your customers.

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ikir and theitsage liked
ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 14, 2017 10:44 pm  
Posted by: Yukikaze

 

DanKnight, is it possible for you to develop a FW update tool for Linux? I would assume a Mac can be booted to Linux via a USB thumbdrive, and then such a tool can be used there to update the FW.

EDIT: While I am at it. How about Linux support in general? Are you involved with Intel on that front as well?

   

Linux Support I've also asked quite a few times and it doesn't seem like they have any plans for it... yet (I'm hoping). If there will be any Linux support, it has to start from Intel first. But I'm also hoping for it as well, cause what's the point of a connector that does it all if it can't work on Linux, am I right? 😛

Alias: KnightZero
AKiTiO Employee - Technical Specialist


ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 15, 2017 12:14 am  

By the way, I just announced AKiTiO is doing a FW Update Service. More details here:  https://egpu.io/forums/mac-setup/akitio-firmware-update-service/

Alias: KnightZero
AKiTiO Employee - Technical Specialist


Medi47 liked
ReplyQuote
benr
 benr
(@benr)
Eminent Member
Joined: 2 years ago
Posts: 37
April 19, 2017 5:45 am  

I updated the Node firmware to v.25,1 via Bootcamp.

I'm running under macOS 10.12.4 on a 2016 15-inch MacBook Pro, with all latest Nvidia and CUDA drivers with Pascal support.

I'm seeing lower Host to Device and Device to Host numbers on macOS than people seem to be reporting under Windows. See screenshot. I feel like I'm usually seeing 2000MiB/s and above reported after the firmware update. 

Any thoughts? Is this even likely to affect GPU CUDA performance?

Any thoughts?

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
vava726
(@vava726)
Estimable Member
Joined: 2 years ago
Posts: 139
April 19, 2017 6:02 am  

These results compared with mine with a TI82 are very low.

Soon I will add my system & eGPU details or a build link to this my signature to give context to my posts


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 19, 2017 8:14 am  
Posted by: benr

 

I updated the Node firmware to v.25,1 via Bootcamp.

I'm running under macOS 10.12.4 on a 2016 15-inch MacBook Pro, with all latest Nvidia and CUDA drivers with Pascal support.

I'm seeing lower Host to Device and Device to Host numbers on macOS than people seem to be reporting under Windows. See screenshot. I feel like I'm usually seeing 2000MiB/s and above reported after the firmware update. 

Any thoughts? Is this even likely to affect GPU CUDA performance?

Any thoughts?

   

Pls run CUDA-Z under Windows and check H2D numbers. Have had a report of lower H2D numbers under macOS than on Windows on the same machine.

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table    •    Several builds
2015 15" Dell Precision 7510 M1000M + macOS 10.14 & Win10


ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 19, 2017 10:32 am  

At least we have it for the Akitio Node. I am just wondering what the other vendors are doing... Akitio was superfast fixing it...

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 2 years ago
Posts: 489
April 19, 2017 10:36 am  
Posted by: ed_co

 

At least we have it for the Akitio Node. I am just wondering what the other vendors are doing... Akitio was superfast fixing it...

   

Please count us in.
https://twitter.com/Mantiz_John/status/853981920253722624

 

 

Mantiz: ShopFacebookTwitter


ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 19, 2017 10:42 am  
Posted by: Mymantiz_John

 Please count us in.
https://twitter.com/Mantiz_John/status/853981920253722624

 

 

   

Sorry dude, I didn't know about it!! There is somewhere in a thread? These are awesome news!! You rock!!

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 19, 2017 11:13 am  

Please theitsage, update the FW in your device and publish it, and change the guide to TB3 in green color!!

Cheers.

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ReplyQuote
ed_co
(@ed_co)
Reputable Member
Joined: 2 years ago
Posts: 363
April 19, 2017 3:58 pm  

I have seen your update in the guide!! Thanks dude!! Just let us know when you have your FW updated and post the results...

mid-2017 Macbook Pro RP560 + MSI GTX1080Ti GAMING [email protected] (Mantiz Venus) + macOS10.13.5 & Win10


ReplyQuote
theitsage
(@itsage)
Famed Member Admin
Joined: 2 years ago
Posts: 3207
April 20, 2017 2:55 am  

I updated the AKiTiO Node with the latest firmware. The Mantiz Venus will soon be running the latest firmware to remedy H2D speed issue. We've seen comparisons between desktop PCIe vs. eGPU internal display vs. eGPU external display. What about dGPU vs. eGPU in a laptop? Here's your short answer. The performance is virtually the same.

 
  13R3 - GTX1060 dGPU XPS9365 - GTX1060 eGPU 13R3 - GTX1060 eGPU
       
Unigine Valley 2,368 2,067 2,408
Unigine Heaven 1,431 1,188 1,363
Unigine Superposition 2,085 2,164 2,073
3DMark Time Spy 3,537 3,507 3,677
3DMark Fire Strike 11,326 9,292 11,005

Best ultrabooks for eGPU use

eGPU enclosure buying guide

86 external GPU build guides


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 20, 2017 4:37 am  
Posted by: theitsage

 

I updated the AKiTiO Node with the latest firmware. The Mantiz Venus will soon be running the latest firmware to remedy H2D speed issue. We've seen comparisons between desktop PCIe vs. eGPU internal display vs. eGPU external display. What about dGPU vs. eGPU in a laptop? Here's your short answer. The performance is virtually the same.

 
  13R3 - GTX1060 dGPU XPS9365 - GTX1060 eGPU 13R3 - GTX1060 eGPU
       
Unigine Valley 2,368 2,067 2,408
Unigine Heaven 1,431 1,188 1,363
Unigine Superposition 2,085 2,164 2,073
3DMark Time Spy 3,537 3,507 3,677
3DMark Fire Strike 11,326 9,292 11,005

   

Those Alienware 13R3 GTX1060 dGPU versus GTX1060 TB3 eGPU results matching those posted at https://egpu.io/external-gpu-is-it-worth-it/ where a Mantiz Venus was used for testing. Were the eGPU results obtained after a H2D fixed firmware was applied?

Is the Alienware 13R3 a 4-lane TB3 device? A CUDA-Z bandwidth result screenshot obtained on the 13R3 would answer both questions.

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table    •    Several builds
2015 15" Dell Precision 7510 M1000M + macOS 10.14 & Win10


ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 20, 2017 7:30 am  
Posted by: theitsage

 

I updated the AKiTiO Node with the latest firmware. The Mantiz Venus will soon be running the latest firmware to remedy H2D speed issue. We've seen comparisons between desktop PCIe vs. eGPU internal display vs. eGPU external display. What about dGPU vs. eGPU in a laptop? Here's your short answer. The performance is virtually the same under the listed benchmaks:

 
  13R3 - GTX1060 dGPU XPS9365 - GTX1060 eGPU 13R3 - GTX1060 eGPU
       
Unigine Valley 2,368 2,067 2,408
Unigine Heaven 1,431 1,188 1,363
Unigine Superposition 2,085 2,164 2,073
3DMark Time Spy 3,537 3,507 3,677
3DMark Fire Strike 11,326 9,292 11,005

   

Edited for you.

You can't say dGPU have the same performances than the eGPU in general, this is misleading, a bench is not enough to be able to generalize, as i said more than one time now uselessly.

Plz since this forum is getting visibility, stop posting benchmark only and post instead as @goalque did, a game benchmark with.

I would gift you a game with a bench built bench in for the sake. Then post the benchmark and a couple of games.

  1. You bring more solid data.
  2. You remove any possibility of further wrong discussions.
  3. You feel more professional, since you seems take care of testing and reviewing the adapters.

 

Because it depend witch laptop/macbook you are testing, since not all the laptop use the same pci-e lanes from cpu or chipseet etc etc.

For me i would build a test bench and then test the max perfs, like with a TBE desktop and then on specific laptops with specific pci-e config.

If you aim is to asses the difference between eGPU and dGPU with the same gpu!

Now we finally get into the in game performance that is the main reason people pick up a new video card.

 From the review i was reading when writing this post, about new RX580 cards! From LANOC site!

2012 13-inch Dell Latitude E6320 + R9 [email protected] (EXP GDC 8.4) + Win10
E=Mc²


ReplyQuote
Page 5 / 8 Prev Next