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

Page 4 / 8 Prev Next
  RSS

goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 3, 2017 8:52 am  

"Hitman's system requirements do specify that you need a quad-core CPU"

http://www.tomshardware.com/reviews/multi-core-cpu-scaling-directx-11,4768-3.html

It is a heavily CPU intensive game. I don't have experience of DirectX12 gaming generally. DirectX11 looked better in benchmark.

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 3, 2017 6:58 pm  

AKiTiO Update 4/3/2017:

We have a new firmware for the Node. Waiting on Intel approval. When it does drop, we'd really appreciate it, for everyone that does have a Node already, to give the new firmware a shot and posts your results. (Your data is the feedback. :P)

Alias: KnightZero
AKiTiO Employee - Technical Specialist


PFCBarefoot, nando4, ddqp and 2 people liked
ReplyQuote
jefniro
(@jefniro)
Estimable Member
Joined: 2 years ago
Posts: 138
April 3, 2017 7:21 pm  
Posted by: DanKnight

 

AKiTiO Update 4/3/2017:

We have a new firmware for the Node. Waiting on Intel approval. When it does drop, we'd really appreciate it, for everyone that does have a Node already, to give the new firmware a shot and posts your results. (Your data is the feedback. :P)

   

Sigh..... 

late-2016 13" MacBook Pro Touchbar + [email protected] (AKiTiO Node) + macOS & Win10


ReplyQuote
Lace
 Lace
(@lace)
Active Member
Joined: 2 years ago
Posts: 8
April 3, 2017 9:08 pm  

...and my product warranty is my parachute 🙂

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


ReplyQuote
Jaye
 Jaye
(@jaye)
Eminent Member
Joined: 2 years ago
Posts: 41

greg liked
ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 4, 2017 8:28 am  

Since you really begin to bench ingame, thing i would not though you spent time with. Since that i shall give you the basic rules on games performances.

  • Cpu is really important in games, to get basic good performances from Intel cpu's, you just need to have a true quad with four thread like an i5. Everything bellow will significantly produce lower fps in latest games. The main reason is that actual games are optimized generally for the use of 4 cores and 8 thread max, you noticed this with ryzen AMD cpu, where giving more cores see a gain only where the programs are coded for work with more than the 4/8 cpu config.
  • Also what matter for games is the amount of cache and the cpu clock speed. Even an old SB/YB perform decently ingame if you overclock them, as i already said, what allow the dual i7 to perform well is their high clock low TDP under turbo mode, but unfortunately clock can't replace also 2 cores 4 thread and 3MB of cache!
  • So like the cpu bench posted before, you clearly notice the basic fps line with 4/8 cpu same clocks. Whats above is a nice show of multithread usage, but in real daily ingame usage have a constant 60/100fps is enough! Unfortunately the 2/4 cpu produce only half of the framerate!
  • If you can't bench on the same CPU PLZ put them to the same IPC performances and disables cores if you need to compare different laptops. You have tools to do that, more than that, high end laptop are allowed to overclock/downclock. For the sake of keeping everything same, otherwise it is really difficult to compare.
  • DX11vsDX12, DX12 is better optimized on AMD side than on Nvidia side, AMD incorporate a hardware scheduler instead of the software scheduler on Nvidia side. DX12 allow an efficient management of computing task, freeing resources on cpu/gpu. So heavy DX11 games result lighter on DX12 api, with usually better gain on AMD gpu than on Nvidia ones, green lantern spend some time on DX12 drivers optimizations on pascal gpu, showing back good results! So DX12 = less charge on cpu/ram.
  • @nando4, beware mate to not take the pci-e bandwidth data from techpowerup and apply them straight in our case. In our case you have something in between the cpu/chipset and the gpu, the Ti82/83 driver. So the techpowerup data applies where there is nothing in between. The Ti82/83 silicon can't compete with cpu/gpu silicon, as same for the amount of data it could manage. So remove again 10%/25% from techpower data, imo.
  • Looking at the @goalque nice benchs, there is anyway something strange with the results, i can understand the cpu difference between the laptops, but still 81fps vs 60fps on 1080p is a lot. Plz specify the config when you are running test, cpu/memory speeds, etc, etc.
  • @goalque PLZ can you do again the test running a couple of bench in lower resolutions like 800*600, 1024*768 to fully remove the cpu bottleneck. For me the firmware have a link speed detection that detect the amount of data depending the resolution overhead. It could be nice to assay the fps output removing the cpu bottleneck and try to trick the firmware at the same time.
  • If you feel the game smooth it mean the frametime is good, so there is no lag on the transmitted data on Ti82/83 driver. As the lag is perceived as variation of the frametime, like 100fps = 9/12ms between frames, then a frame at 50/100ms would be perceived as a lag or stutter. Meaning the gpu handle easily the game producing internally a top constant i.e 100fps. What internally the gpu can compute and what goes on your screen is not the same.
  • What also bother me is the poor I/O bandwidth but the great internal device to device bus, meaning the controller could support higher speeds having a decent internal bus!
  • It is a fact that high end video adapter required a decent bandwidth to allow, like i said, to actually display the framerate they compute smoothly on screen, but they also need a decent hardware behind that follow the gpu computing, because a 1080 is hell fast, will be honest with you, any mac hardware atm couldn't possibly follow even with a high end TBE box.

Again if reading this, you do the bench correctly with another couples of games and there is no difference, especially on low fps/frametime, we need to find another way to explain this bandwidth difference. Meaning that anyway the eGPU firmware is locked to a fix value in all cases, imo. There is no way intel allow you, like i already said, to use the TBE controller for it's fully capacities, at least not into the eGPU consumer market.

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


jangoloti liked
ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 4, 2017 9:07 am  

I know that comparing a TB2 quad core against TB3 dual core may be misleading as for tracking down the reduced H2D speed impact on gaming.

Surprisingly, the Akitio Node performed better in this scenario with the RX 480, especially at UHD resolution.

Are there any games that are non-CPU intensive, and put stress on the PCIe bus?

Maybe one of these games?

https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 4, 2017 9:45 am  
Posted by: goalque

 Maybe one of these games?

https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/

   

Try these @FHD which had the biggest FPS improvement going from x4 1.1 -> x8 1.1:

  • Ryse - Son of Rome: +58.7% (28.1->44.6) here
  • WOW - Warlords of Draenor: +49.6 (106.3->159.0) here

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 4, 2017 9:50 am  

Just downscale the resolution and/or disable 2 cores, to remove the cpu bottleneck. Like i said i dunno nothing on mac software side, dunno if it is even possible. I try to help you here where i can, usually i do not post on mac thread, i do not have the knowledge for.  But i would assume you should find some command to do that on macos with admin right or some bench/test tools.

The games proposed in this review are a bit old, you should look at the techpowerup review Wizzard did for ryzen. There is good latest games you can test, you can also choose the ones less cpu bound, you will notice. Plz chose games with high basic 1080p framerate, so somehow light on cpu too. HERE  HERE

Yep i noticed that, with the RX480. I really dunno why the AMD seems perform so bad with the box, i mean i know why, but damn! A RX480 is not far from a GTX980 and should perform the same on UHD.

You could say the GTX980 is faster shining under 1080p, the RX480 shine under 1440p because it handle better the amount of data, you notice this on your eGPU setup, nvidia is faster on compute performing better with TBE adapter since required less bandwidth, instead AMD seems more bandwidth sensitive, maybe because of it's way to compute, with more slower but bulky data.

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


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 4, 2017 9:55 pm  
Posted by: wimpzilla

 

 

  • Looking at the @goalque nice benchs, there is anyway something strange with the results, i can understand the cpu difference between the laptops, but still 81fps vs 60fps on 1080p is a lot. Plz specify the config when you are running test, cpu/memory speeds, etc, etc.
  • @goalque PLZ can you do again the test running a couple of bench in lower resolutions like 800*600, 1024*768 to fully remove the cpu bottleneck. For me the firmware have a link speed detection that detect the amount of data depending the resolution overhead. It could be nice to assay the fps output removing the cpu bottleneck and try to trick the firmware at the same time.

 

Configuration:

2015 15" MBP (M370X) + Apple TB cable + Apple TB2-to-TB3 adapter + Akitio Node + DP-DP cable + Samsung 28" UHD Monitor (U28D590D)

Hitman:
Level of detail: Ultra
V-Sync off, 1920 x 1080 resolution

MSI afterburner’s averages (114 samples): 81 FPS (reference GTX 980) vs 54.6 FPS (reference RX 480)

The lowest available resolution in Hitman performance test was 1280x720 and it gave 63.8 FPS average to RX 480. Not a big jump. The GPU usage was not stable at all, why? Even more zigzag at 1280x720 resolution. Something is really wrong, we need another game before concluding anything. RX 480 cannot perform that poorly. I will do the same test with the 2016 13" MBP later.

RX 480:

GTX 980:

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 4, 2017 11:05 pm  

Dunno what happen with the RX480, there is a lot of reasons behind, maybe the game is not AMD side, drivers, TBE driver not AMD side, the box, the screen refresh, etc, etc, i couldn't give you an answer on that. Do you use the last relive drivers, the test was under dx11/12?

What i can say for sure is that the GTX980 perform really well with this configuration, this is a good example of nvidia card + TBE box + high end quad laptop. Either the framerate, frametime, gpu usage are really good on this game with this setup, nothing to say. This is a TB2 Ti82 16Gb/s on TB3 Ti83 box right?

On AMD side dunno what to say, it seems there is a lower cpu usage, better cpu usage worst gpu usage. The card seems bottleneck somewhere, the framerate is not sooo bad, but the frametime is a bit crappy.

Maybe give it a try with another game. And more than that give us your feeling about the game, you are the one that are performing these nice bench!

Here you have an overview of performances with almost updated drivers. 21 Feb 2017.

Try to do not overvolt too much the RX480, i got fairly better result at the end keeping my R9 270 under power pressure instead of feeding it with more voltage.

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


goalque liked
ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 4, 2017 11:47 pm  

More interesting results... only the hosting MBP and cable (Belkin 0,5m TB3 40Gbps) changed, the same settings:

Level of detail: Ultra
V-Sync off, 1920 x 1080 resolution
The latest 17.3.3 software from AMD and DX11
The latest Nvidia driver

RX 480:

GTX 980:

 

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 4, 2017 11:56 pm  

Clearly cpu/mb/ram bottleneck here, lower gpu t°, low gpu usage, high cpu usage, both frametime from 15/135ms but with different min/high fps.

So you can also conclude, unfortunately even the MPB 15 quad seems bottleneck the RX480 for unknown reason this time. The cpu is enough to deliver good computing performance, bandwidth, i.e with the GTX980. So the problem is elsewhere with the RX480, i think.

I can only assume on the MBP 16 TB3, the a dual i7/mb/ram is not powerful enough to manage the game and the TBE/eGPU enclose.

Now someone should test with a TB3 MBP with a true quad like the MPB 15 with the same enclosure. If you got the same result on both card/game it mean there is effectively maybe a performance problem on AMD side on this game at least. Also it mean that the bandwidth/cpu performances are enough to deliver a decent gaming experience with a GTX980/1060 on a MPB 15 quad TBE2 and TB3 enclosure.

Or you disable 2 cores, disable HT on your MBP 15, if you can.

I don't think the cable is the issue, i mean i hope. The cable length is quite short, the impedance should be ok. You can try to wrap it into aluminum foil, but hope it is not the issue. ^^

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


ReplyQuote
Poblopuablo
(@poblopuablo)
Eminent Member
Joined: 2 years ago
Posts: 27
April 5, 2017 2:48 am  

I don't have a MBP. But I just got my akitio node today. I have a strix 980 and a evga 970 ssc I can test with. 

My computer: 

Dell precision M3510 

Cpu: i7-6700HQ 

Ram: 32gb 18xx(forgot exact #)mhz

Ssd: 2x 256 nvme pcie ssd

Hdd: 512gb

TB3: yes.

 

However I don't have new games, and I don't have $ to buy games in the near future. 

I have a licence for titanfall 2 that I have not used yet, but that's about it.

Would you(the community) like me to get benchmarks or would this be more of a hindrance at this point?

 

 

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


ReplyQuote
Lace
 Lace
(@lace)
Active Member
Joined: 2 years ago
Posts: 8
April 5, 2017 5:18 am  

I have a MBP late 2016 13" i7 with an Akitio Node and a Gigabyte GTX 970 inside.

 

I like to do some CPU/GPU/Thunderbolt performance testing with MacOS , can you recommend tools for that?

I did some performance testing with games and I experienced frequent rendering "hangs", kind of stuttering in X-Plane 11, Bioshock Infinite and also in Benchmarks like Unigine Heaven. I am curious what the reason might be. I am also curious if the PCH-connected TB3 develops performance issues when the unquestionable High-End PCIe SSD is putting load on the chipset.

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


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314

ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 5, 2017 12:00 pm  

Well for help @goalque, since he is the one that is performing the bench in the right way, i can GIFT a couple of steam copy of HITMAN, to check on the same game with a MBP/High end laptop with a true quad and a TB3 coupled with an AKITIO NODE.

PM me if you have steam and wonder help to test the TB3/AKIO/QUAD CPU, if you have basic knowledge of gaming bench.

Again report graphs like  @goalque did, possibly with msi afterburner, min/max/average fps, frametime, gpu/cpu usage, t° etc.

This is exacly what we are trying to asses, @goalque already clearly show that TB2 + quad + GTX980 perform really well, the RX480 not so well, under HITMAN. Now we want asses the performances on TB3 + quad + GTX980/970 if possible. To finally answer this question and move on if problem, on a solution.

About the pci-e SSD M2.M, latest cpu/chipset have enough pci-e lanes to manage everything, but it is true that having a eGPU and a M2.M SSD put under pressure the chipset, if the TBE use chipset lanes. On a desktop it is not a problem because the pci-e lanes cames straight from the cpu.

@goalque: all U cpu are not intended for gaming, U for intel = low power, it mean the cpu have a lower TDP but same specs that the non U counterpart. All U cpu are rather used for battery lifetime than performance wise. Again they perform well because latest intel cpu have a far nice IPC, high clock 3.1ghz, low TDP allowing the cpu to boost without be downclocked!

=>We are far far from the performances of a i5-6300HQ unfortunately, Here. Intel is sometimes greedy since no competition on this consumer market segment!!!! !!! !! !

@lace: you can try ATTO Benchmark to bench your SSD, repeat the test a couple of times without eGPU, then with the eGPU connected. Check the relative read/write average speed and compare. ATTO Techpower. Lunch the tool as it is, default setting are ok.

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


ReplyQuote
Mgear1981
(@mgear1981)
Eminent Member
Joined: 2 years ago
Posts: 36
April 6, 2017 6:22 am  

I got myselves a macbook pro 2015 2.7 i7 quad and i got the akitio node and 980ti. More then happy to test wimpzilla

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


ReplyQuote
ddqp
 ddqp
(@ddqp)
Eminent Member
Joined: 2 years ago
Posts: 32
April 6, 2017 8:33 am  

updated:akitio node just update their FM to solve this issue, please double check it.

------------------------------------------------------------
AKiTiO
------------------------------------------------------------

AKiTiO Node

Firmware history:
==============================

Version: v1.0.0.4 B1-25+4.3.3
Date: 4/6/2017
-------------------------------------
Changed: Improved "Host to Device" memory copy speed 


Version: v1.0.0.4 B1-23+3.6.1
Date: 1/11/2017
-------------------------------------
Added: Support for Thunderbolt 3 hosts that do not support external GPUs

https://www.akitio.com/firmware/node-firmware

Late Macbook Pro 2016 13' touch bar + AKITIO node + GTX 1050TI 4G Windows


ReplyQuote
wimpzilla
(@wimpzilla)
Reputable Member
Joined: 2 years ago
Posts: 408
April 6, 2017 8:34 am  

@mgear: Send me by pm you steamid or log in on steam and add me. I would gift you HITMAN if you wonder do a couple of test with you Ti, since you have the +/- same raw power than a basic 1070.

I would need someone with a TB3 QUAND and a node to do the same.

Send me a pm.

Edit: Plz do some decent test before update the Node firmware, since if there is a gain and it bring fresh performance, we can clearly point out and compare both, before/after.

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


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 2 years ago
Posts: 489
April 6, 2017 8:44 am  
Posted by: wimpzilla

 

@mgear: Send me by pm you steamid or log in on steam and add me. I would gift you HITMAN if you wonder do a couple of test with you Ti, since you have the +/- same raw power than a basic 1070.

I would need someone with a TB3 QUAND and a node to do the same.

Send me a pm.

Edit: Plz do some decent test before update the Node firmware, since if there is a gain and it bring fresh performance, we can clearly point out and compare both, before/after.

   

Mantiz had the new FW too, expecting to see the complex bench that you guys did. we are also testing in parallel. 

Mantiz: ShopFacebookTwitter


ikir liked
ReplyQuote
benip2140
(@benip2140)
Active Member
Joined: 2 years ago
Posts: 14
April 6, 2017 10:44 am  

I found a solution for macbook users (maybe just for macbook 13" tb 2016) to update the firmware of akitio node

please follow this link

https://egpu.io/forums/thunderbolt-enclosures/akitio-node-thunderbolt-3-egpu-enclosure/paged/6/#post-6718

 

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


ikir and nando4 liked
ReplyQuote
ddqp
 ddqp
(@ddqp)
Eminent Member
Joined: 2 years ago
Posts: 32
April 6, 2017 11:19 am  

Here is the new Akitio node FM CUDA-Z result: it's fixed, and thanks Akitio

 

New FM

Old FM

Late Macbook Pro 2016 13' touch bar + AKITIO node + GTX 1050TI 4G Windows


ikir and nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 6, 2017 11:26 am  
Posted by: ddqp

 

Here is the new Akitio node FM CUDA-Z result: it's fixed, and thanks Akitio  

The big question is, has gaming FPS improved when do a before/after H2D firmware fix comparison?
Bandwidth dependent games are listed here and here. The H2D issue was the most visible indicator for investigation of the reported TB3 underperformance.

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


ReplyQuote
ddqp
 ddqp
(@ddqp)
Eminent Member
Joined: 2 years ago
Posts: 32
April 6, 2017 11:30 am  
Posted by: nando4

 The big question is, has gaming FPS improved when do a before/after H2D firmware fix comparison?
Bandwidth dependent games are listed here and here. The H2D issue was the most visible indicator for investigation of the reported TB3 underperformance.

Here you are, no difference with HEAVEN, other games I can make test in this weekend.

Late Macbook Pro 2016 13' touch bar + AKITIO node + GTX 1050TI 4G Windows


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 6, 2017 11:42 am  

Yep, there is no evidence yet that H2D TB3 bandwidth on the device side has an influence on gaming FPS or GPGPU performance.

I will retest.

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 6, 2017 11:52 am  
Posted by: goalque

 

Yep, there is no evidence yet that H2D TB3 bandwidth on the device side has an influence on gaming FPS or GPGPU performance.

I will retest.

   

Goalque, while gaming FPS may not be resolved here, at the very least, would you be willing to re-do the matlab tests? They were the most comprehensive study into the H2D issue.

If they are improved we can close this H2D thread as an area of study for TB3 underperformance.

A positive outcome being  CUDA apps benefitting from improved bandwidth.

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


ReplyQuote
win32asmguy
(@win32asmguy)
Active Member
Joined: 2 years ago
Posts: 8
April 6, 2017 4:15 pm  

I noticed a performance improvement with my setup. I have a HP Zbook Studio G3, connected to an Akitio Node + Nvidia GTX 1070 Founder Edition. I run five World of Warcraft client at the same time. Before the update, the background clients would drop to 1-0.5 FPS. Now, after the update I can get a steady 30FPS on each background client and maintain 60FPS on the foreground client, similar to how it performs in a desktop with the same specifications.

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


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 2 years ago
Posts: 1815
April 6, 2017 4:19 pm  
Posted by: win32asmguy

 

I noticed a performance improvement with my setup. I have a HP Zbook Studio G3, connected to an Akitio Node + Nvidia GTX 1070 Founder Edition. I run five World of Warcraft client at the same time. Before the update, the background clients would drop to 1-0.5 FPS. Now, after the update I can get a steady 30FPS on each background client and maintain 60FPS on the foreground client, similar to how it performs in a desktop with the same specifications.

Thank you for dropping by and advising the performance improvement. We do not as yet have a HP ZBook G3 eGPU implementation: https://egpu.io/external-gpu-implementations-table/?table_filter=%22zbook%22#search

Would you like to submit one along with your now full speed CUDA-Z  H2D goodness and some photos? The ZBooks are very nice systems with deserved eGPU representation.

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


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 2 years ago
Posts: 1314
April 6, 2017 4:50 pm  

@nando4: I will do that later. The first thing to do is to find another game (preferably available from Steam) as I want to find out the reason for underperforming RX 480 under Hitman. Maybe "virtual super resolution", "GPU scaling" or "scaling mode" in Radeon settings have some effect.

A quick test indicated that GTX 1060 couldn't reach the performance level of reference GTX 980 so this might be also GPU architecture related issue.

By coincidence, AMD released Crimson ReLive Edition 17.4.1 two days ago, and Nvidia released 381.65 today 🙂

automate-eGPU EFIapple_set_os.efi
--
late-2016 13" Macbook Pro nTB + [email protected] (Netstor HL23T) + macOS & Win10
late-2016 13" Macbook Pro nTB + GTX980/[email protected] (Netstor HL23T) + macOS10.13 & Win10


nando4 liked
ReplyQuote
Mgear1981
(@mgear1981)
Eminent Member
Joined: 2 years ago
Posts: 36
April 6, 2017 6:58 pm  

the new firmware for akitio node has been given free. I updated and flashed it. Will post results soon. Perhabs for everyone who owns the node to update the firmware and test it out!

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
Mgear1981
(@mgear1981)
Eminent Member
Joined: 2 years ago
Posts: 36
April 6, 2017 7:02 pm  
Posted by: wimpzilla

 

@mgear: Send me by pm you steamid or log in on steam and add me. I would gift you HITMAN if you wonder do a couple of test with you Ti, since you have the +/- same raw power than a basic 1070.

I would need someone with a TB3 QUAND and a node to do the same.

Send me a pm.

Edit: Plz do some decent test before update the Node firmware, since if there is a gain and it bring fresh performance, we can clearly point out and compare both, before/after.

   

I have sent you a message

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


ReplyQuote
Poblopuablo
(@poblopuablo)
Eminent Member
Joined: 2 years ago
Posts: 27
April 6, 2017 7:09 pm  

Can someone post a link of how to set up the akitio node on a desktop that has an integrated GPU(Iris) and a dedicated gpu(amd firepro) so I can actually utilize my akitio node

 

I have a m3510 Dell precision. I have it recognized and the driver for it my evga 970(in the node). My problem is that it isn't getting utilized(heaven bench mark is 5-6fps) 

Side note: I'm trying to use internal display. I don't have an external available to test. 

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


ReplyQuote
win32asmguy
(@win32asmguy)
Active Member
Joined: 2 years ago
Posts: 8
April 6, 2017 8:09 pm  
Posted by: nando4

Thank you for dropping by and advising the performance improvement. We do not as yet have a HP ZBook G3 eGPU implementation: https://egpu.io/external-gpu-implementations-table/?table_filter=%22zbook%22#search

Would you like to submit one along with your now full speed CUDA-Z  H2D goodness and some photos? The ZBooks are very nice systems with deserved eGPU representation.

   

Yes, I would like to submit a guide for it, possibly over the weekend. It has been a great experience so far. I have the version of the Studio with a 6700HQ and no Quadro graphics, and the eGPU has basically been plug and play after doing a few firmware updates. It appears to be an x4 PCIe link based on CUDA-Z and Firestrike benchmarks.

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


DanKnight and nando4 liked
ReplyQuote
Mgear1981
(@mgear1981)
Eminent Member
Joined: 2 years ago
Posts: 36
April 6, 2017 9:27 pm  

I noticed a good performance gain to on my 980ti and the akitio node. I do have the feeling tough Akitio can turn the thunderbolt 3 pipeline more open. They are on the good rood but they can push out more. It's funny they claim in their firmware update release that it shouldn't have a performance impact. I do get about 10 fps more in Company of heroes 2 and about 5 fps more in civilisation 6. Both are cpu intensive games. I am curious tough if you get more performance from these games in windows bootcamp setup. Anyone who can relate to graphic performance difference with egpu between mac os and bootcamp windows ?

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


nando4 and DanKnight liked
ReplyQuote
DanKnight
(@danknight)
AKiTiO Tech
Joined: 2 years ago
Posts: 80
April 7, 2017 12:20 am  

Intel and AKiTiO would like to thank everyone for their patience and for reporting this issue in the first place. We love the support we're getting from this community and excited to see the good feedback from the new FW so far.

Update for those people with pre-orders. They will be arriving soon. Containers are on-route to resellers and distributors that sell to resellers.

-cough- @jefniro -cough-

Alias: KnightZero
AKiTiO Employee - Technical Specialist


ikir, Dad, nando4 and 2 people liked
ReplyQuote
Page 4 / 8 Prev Next