2018 13" Dell XPS 13 9370 + GTX1070@32Gbps-TB3 (AKiTiO Node) + Win10 [dil950]  

 

dil950
(@dil950)
Active Member
Joined:10 months  ago
Posts: 7
February 3, 2018 8:02 am  

System specs  New Dell XPS 13 9370: 8th gen i7-8550U + 16 gb mem

 

eGPU hardware  Akitio Node running the June 7 2017 firmware+ Nvidia 1070 running 390.77 drivers + Thunderbolt drivers from Jan 8 2018

 

Installation steps (what did you do to get it all going?): Basically update the thunderbolt SW and firmware to the latest via the dell website.  Connect the Node to the power USB-C port.  Install Nvidia Drivers

 

Benchmarks : Timespy 4753

cuda

 

Comments (eg: how has the eGPU improved your workflow or gaming?)

haven’t tried gaming with it but Timespy was around 30 fps.  I previously had the same eGPU setup on a 9350 XPS 13.  The timespy score was around 3900 I think.  So a decent bump up going to the 9370.  

Setup was the same story where nothing worked until every component was updated to the latest rev.  The one thing that drove me bonkers was I had the node connected to the second USB-C port.  I always connected the laptop power supply to the first USB-C port just because it made sense and had the power icon drawn next to it.  For some reason the computer wouldn’t detect the Node when it was connected this way.  I had to switch the power supply to the second port and connect the node to the first.  Then windows started detecting it.  

Not sure if a timespy score bump of 800 justifies the cost of upgrading but it’s a decent bump so not complaining.

Edited: 2 weeks  ago

ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1578
February 3, 2018 8:10 am  

Thank you for your build.  Would you like to include bandwidth results so we can see if the XPS 9370 is now running a 4-lane TB3 controller and a GT4 OPI?

https://egpu.io/external-gpu-implementations-table/#perf

We would also love to see pictures of your gear.

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
dil950
(@dil950)
Active Member
Joined:10 months  ago
Posts: 7
February 3, 2018 4:11 pm  

Sure…here are the results from Cuda…i’ll run the other one later tonight

 

CUDA-Z Report
=============
Version: 0.10.251 64 bit http://cuda-z.sf.net/
OS Version: Windows x86 6.2.9200
Driver Version: 390.77
Driver Dll Version: 9.10 (23.21.13.9077)
Runtime Dll Version: 6.50

Core Information
—————-
Name: GeForce GTX 1070
Compute Capability: 6.1
Clock Rate: 1784.5 MHz
PCI Location: 0:60:0
Multiprocessors: 15
Threads Per Multiproc.: 2048
Warp Size: 32
Regs Per Block: 65536
Threads Per Block: 1024
Threads Dimensions: 1024 x 1024 x 64
Grid Dimensions: 2147483647 x 65535 x 65535
Watchdog Enabled: Yes
Integrated GPU: No
Concurrent Kernels: Yes
Compute Mode: Default
Stream Priorities: Yes

Memory Information
——————
Total Global: 8192 MiB
Bus Width: 256 bits
Clock Rate: 4004 MHz
Error Correction: No
L2 Cache Size: 48 KiB
Shared Per Block: 48 KiB
Pitch: 2048 MiB
Total Constant: 64 KiB
Texture Alignment: 512 B
Texture 1D Size: 131072
Texture 2D Size: 131072 x 65536
Texture 3D Size: 16384 x 16384 x 16384
GPU Overlap: Yes
Map Host Memory: Yes
Unified Addressing: Yes
Async Engine: Yes, Bidirectional

Performance Information
———————–
Memory Copy
Host Pinned to Device: 2188.61 MiB/s
Host Pageable to Device: 1778.07 MiB/s
Device to Host Pinned: 2596.54 MiB/s
Device to Host Pageable: 2087.86 MiB/s
Device to Device: 85.0072 GiB/s
GPU Core Performance
Single-precision Float: 7185.31 Gflop/s
Double-precision Float: 239.569 Gflop/s
64-bit Integer: 540.782 Giop/s
32-bit Integer: 2491.07 Giop/s
24-bit Integer: 1873.35 Giop/s

Generated: Sat Feb 03 08:09:36 2018

 


ReplyQuote
dil950
(@dil950)
Active Member
Joined:10 months  ago
Posts: 7
February 4, 2018 12:21 am  

Not sure if the Aida only applies to AMD cards..but here’s the result using AIDA64 Extreme.  The aida trial blocks out the memory write portion…I could only see it in the results txt..  So was Dell telling the truth, haha?

 

Benchmark Result Run Time Build Time
————————————————————-
Memory Read 2689 MB/s 1235 ms
– Pinned 2689 MB/s 11 ms
– Pageable 2199 MB/s 14 ms
Memory Write 2278 MB/s 1218 ms
– Pinned 2278 MB/s 13 ms
– Pageable 1812 MB/s 17 ms
Memory Copy 185907 MB/s 4797 ms
– 15 MB Block 109492 MB/s 0 ms
– 32 MB Block 138796 MB/s 0 ms
– 64 MB Block 160232 MB/s 1 ms
– 128 MB Block 172199 MB/s 1 ms
– 256 MB Block 179848 MB/s 3 ms
– 512 MB Block 183628 MB/s 6 ms
– 1024 MB Block 185907 MB/s 11 ms
Single-Precision FLOPS 7181 GFLOPS 6500 ms
– float1 6997 GFLOPS 629 ms 15 ms
– float2 7153 GFLOPS 615 ms 0 ms
– float4 7181 GFLOPS 612 ms 15 ms
– float8 7131 GFLOPS 617 ms 16 ms
– float16 7006 GFLOPS 628 ms 16 ms
Double-Precision FLOPS 237.3 GFLOPS 6187 ms
– double1 237.3 GFLOPS 579 ms 15 ms
– double2 237.2 GFLOPS 579 ms 16 ms
– double4 233.9 GFLOPS 588 ms 15 ms
– double8 227.4 GFLOPS 604 ms 15 ms
– double16 227.4 GFLOPS 604 ms 0 ms
24-bit Integer IOPS 2509 GIOPS 8640 ms
– int1 2503 GIOPS 878 ms 16 ms
– int2 2509 GIOPS 876 ms 16 ms
– int4 2473 GIOPS 889 ms 16 ms
– int8 2174 GIOPS 1011 ms 15 ms
– int16 2112 GIOPS 521 ms 16 ms
32-bit Integer IOPS 2526 GIOPS 7609 ms
– int1 2500 GIOPS 880 ms 15 ms
– int2 2526 GIOPS 871 ms 15 ms
– int4 2494 GIOPS 882 ms 16 ms
– int8 2177 GIOPS 505 ms 15 ms
– int16 2098 GIOPS 524 ms 0 ms
64-bit Integer IOPS 408.8 GIOPS 7875 ms
– long1 408.8 GIOPS 672 ms 16 ms
– long2 378.2 GIOPS 727 ms 16 ms
– long4 353.1 GIOPS 779 ms 16 ms
– long8 351.1 GIOPS 783 ms 16 ms
– long16 331.5 GIOPS 829 ms 15 ms
AES-256 20333 MB/s 33968 ms
– inline loop / 4 MB 19981 MB/s 820 ms 16 ms
– inline loop / 8 MB 20126 MB/s 814 ms
– inline loop / 16 MB 20125 MB/s 814 ms
– inline loop / 32 MB 20196 MB/s 811 ms
– inline loop / 64 MB 20202 MB/s 811 ms
– inline unroll / 4 MB 19890 MB/s 824 ms 0 ms
– inline unroll / 8 MB 20171 MB/s 812 ms
– inline unroll / 16 MB 20317 MB/s 806 ms
– inline unroll / 32 MB 20311 MB/s 807 ms
– inline unroll / 64 MB 20320 MB/s 806 ms
– define loop / 4 MB 19977 MB/s 820 ms 15 ms
– define loop / 8 MB 20123 MB/s 814 ms
– define loop / 16 MB 20182 MB/s 812 ms
– define loop / 32 MB 20176 MB/s 812 ms
– define loop / 64 MB 20204 MB/s 811 ms
– define unroll / 4 MB 19895 MB/s 824 ms 0 ms
– define unroll / 8 MB 20263 MB/s 809 ms
– define unroll / 16 MB 20328 MB/s 806 ms
– define unroll / 32 MB 20322 MB/s 806 ms
– define unroll / 64 MB 20333 MB/s 806 ms
SHA-1 Hash 39797 MB/s 11953 ms
– 512 blocks 7654 MB/s 523 ms 16 ms
– 1024 blocks 15243 MB/s 525 ms
– 2048 blocks 29980 MB/s 534 ms
– 4096 blocks 39797 MB/s 804 ms
– 8192 blocks 36490 MB/s 877 ms
– 16384 blocks 15917 MB/s 503 ms
– 32768 blocks 12267 MB/s 652 ms
– 65536 blocks 10874 MB/s 736 ms
– 131072 blocks 10725 MB/s 746 ms
Single-Precision Julia 1425 FPS 8391 ms
– float1 break 1390 FPS 737 ms 16 ms
– float1 stay / unroll 3 1425 FPS 718 ms 16 ms
– float2 stay / unroll 3 1201 FPS 852 ms 15 ms
– float1 stay / unroll 9 1304 FPS 785 ms 16 ms
– float2 stay / unroll 9 1080 FPS 948 ms 15 ms
Double-Precision Mandel 78.55 FPS 7563 ms
– double1 break 68.89 FPS 929 ms 15 ms
– double1 stay / unroll 3 78.55 FPS 815 ms 0 ms
– double2 stay / unroll 3 59.38 FPS 539 ms 16 ms
– double1 stay / unroll 9 77.25 FPS 829 ms 16 ms
– double2 stay / unroll 9 57.67 FPS 555 ms 0 ms

 

Here’s the Cuda Screenshot

cuda

Edited: 2 weeks  ago

nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined:1 year  ago
Posts: 1578
February 4, 2018 12:27 am  

Yes, 22xxMiB/s H2D CUDA-Z requires a 4-lane TB3 controller and GT4 OPI to obtain. That there is a maximum TB3 candidate ultrabook.

Edited: 2 weeks  ago

eGPU Port Bandwidth Reference TableeGPU Setup 1.35


ReplyQuote
dil950
(@dil950)
Active Member
Joined:10 months  ago
Posts: 7
February 4, 2018 1:11 am  

That’s good to know.  A nice silver lining to giving up all regular usb ports.  Feels weird having to carry around adapters to use usb peripherals haha.


ReplyQuote
  
Working

Please Login or Register