Enclosure & Hardware Discussions
ALERT: half H2D performance issue on TI83 TB3 enclosures (Node, Venus, Devil Box...
 

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

 of  8
  RSS

Poblopuablo
(@poblopuablo)
Eminent Member
Joined: 3 years ago
 
Posted by: wimpzilla

If you can't change cpu, just put them on the same perfomance clocking them to the same IPC, so the 4700HQ to 4ghz and the 6700HQ to 3.7Ghz, you can use cinebench or cpu-z to evaluate the IPC performances.

   

Great suggestion! That would make things accurate. 🙂 

 

Also, I hope my comments didn't come off strong, I wasn't trying to attack/argue or anything, but rather just clarify. 🙂 Thanks for help and input. 🙂 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

@wimpzilla: Yep. I got your point, thanks 🙂 Ingame results are very important. Frame drops occur often in multiplayer fights. I am not a gamer unfortunately. Someone who has a bunch of games and TI82/TI83 boxes with ~500MiB/s-1000MiB/s H2D difference, please post gaming results! If there is still no difference, this is clearly something CPU-PCH-TB3 controller related…

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


ReplyQuote
Poblopuablo
(@poblopuablo)
Eminent Member
Joined: 3 years ago
 

I believe @Mymantiz_John has multiple hosts/cards/and I think he has TI82 &83. If he(John 😉 you can do it I believe in you) had the time for such an extensive review.Or if anybody else has these things feel free. 🙂 It would be a comprehensive assessment for the betterment of the community. 🙂 

 

Granted you have everything:

2-3 different gpus

1 TI82 GPU egpu(core or devil box)

1 TI83 GPU(venus or node) egpu dock

1 TB2 egpu dock(thunder 2 + psu)

1 TB3 -> TB2 adapter

1 TB2 -> TB3 adapter

1 laptop with TB2

1 laptop with TB3

Multiple games to bench mark with.

A ton of time. 🙂 

 

 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
XCVG
 XCVG
(@xcvg)
Trusted Member
Joined: 3 years ago
 

It occurred to me earlier that looking at minimum FPS might be more useful than looking at average FPS. I've heard sporadic complaints of stuttering and framerate dips, rather than overall reduced performance.

Late-2015 15" Dell XPS 9550 GTX960M + [email protected] (Mantiz Venus) + Win10


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 3 years ago
 
Posted by: Poblopuablo

 

I believe @Mymantiz_John has multiple hosts/cards/and I think he has TI82 &83. If he(John 😉 you can do it I believe in you) had the time for such an extensive review.Or if anybody else has these things feel free. 🙂 It would be a comprehensive assessment for the betterment of the community. 🙂 

 

Granted you have everything:

2-3 different gpus

1 TI82 GPU egpu(core or devil box)

1 TI83 GPU(venus or node) egpu dock

1 TB2 egpu dock(thunder 2 + psu)

1 TB3 -> TB2 adapter

1 TB2 -> TB3 adapter

1 laptop with TB2

1 laptop with TB3

Multiple games to bench mark with.

A ton of time. 🙂 

 

 

   

HI:

I would like to do the test ( after my national holiday from now till next Wed ) but unfortunately, I don't have TB2 Box. and I see no reason why I need to test the TB2 Box in TB2 Laptop. All TB2 laptop MBP 2015 CPU performance is better than the MBP 2016 even Quadcore too. it will only address the unreasonable result. 

and ingame performance wise, I had tested Far Cry Primal for many times, tested with different resolution and  video quality. the result shows the core (2200mb ) is same and  even a bit lose to Mantiz Venus ( 1100mb ). but the lose few FPS is in the allowance. 

PPL always like to see what they want to see, to believe what they want to believe ,  but my comparison is just the comparison. not intend to drive PPL where I want to be. if someone is really that believe in TB2, just go ahead.

to imagine if someday, I got a new FW, make every device to be 2200mb ( Actually I had that FW already ) but still got the 3Dmark, Valley, Heaven  benchmark the same, ingame benchmark the same, is it a real happy ending? I doubt, then another topic will be posted for example " why 1100mb can run to xxx benchmark then turn it to 2200mb how comes the score is the same??  it is really a waste of time.

 

 

Mantiz: ● ●


ReplyQuote
Jetcopter
(@jetcopter)
Active Member
Joined: 3 years ago
 

So I decided to go out and buy a Razer Core off Craigslist and ran some benchmarks.  I don't really know what to make of this...

Same card in both: Asus Nvidia GTX 1080

AkiTiO Node TI83

http://www.3dmark.com/fs/12177273

Razer Core TI82

http://www.3dmark.com/fs/12179972

They are almost exactly the same.  I'm starting to think I wasted a GTX 1080 putting it in a Thunderbolt enclosure. 

 

 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
wimpzilla
(@wimpzilla)
Honorable Member
Joined: 3 years ago
 
Posted by: Mymantiz_John

HI:

I would like to do the test ( after my national holiday from now till next Wed ) but unfortunately, I don't have TB2 Box. and I see no reason why I need to test the TB2 Box in TB2 Laptop. All TB2 laptop MBP 2015 CPU performance is better than the MBP 2016 even Quadcore too. it will only address the unreasonable result. 

and ingame performance wise, I had tested Far Cry Primal for many times, tested with different resolution and  video quality. the result shows the core (2200mb ) is same and  even a bit lose to Mantiz Venus ( 1100mb ). but the lose few FPS is in the allowance. 

PPL always like to see what they want to see, to believe what they want to believe ,  but my comparison is just the comparison. not intend to drive PPL where I want to be. if someone is really that believe in TB2, just go ahead.

to imagine if someday, I got a new FW, make every device to be 2200mb ( Actually I had that FW already ) but still got the 3Dmark, Valley, Heaven  benchmark the same, ingame benchmark the same, is it a real happy ending? I doubt, then another topic will be posted for example " why 1100mb can run to xxx benchmark then turn it to 2200mb how comes the score is the same??  it is really a waste of time.

My last post in this thread because i feel loosing my time.

Sir you are wrong, your test are wrong and badly done, pls refrain to continue to insist, again plz. You begin to be boring and begin to appear like a sand seller!

If you want continue on this way, not including real game benchmark before giving an opinion, i suggest you to change job.

IMO, atm i wouldn't buy your box, for 2 reasons, 1st you are not professional, 2nd you are not showing decent review of your own hardware.

It is really sad that manufacturers can't bench decently their own hardware and came here to find support thinking it would be easy, it is the opposite.

Good luck for the future, for sure i would not recommend your box until you performed decent bench.

See you! 🙂

Edit: @XCVG: It is what i pointed, lower fps and frametime! I would give some explanation why, being a power gamer (online BF1 mainly), you should care about games. I'm actually benching my eGPU setup with mafiaIII, why? Because it is an open world game, like GTA, it required resources from both cpu computing the word dynamics and gpu computing the graphics. Like GTA you can fully say theoretically your machine is used almost at 100% on overall. And the main reason it is optimized like s**t, so if i got decent perfs here i know every other game will run +40fps low easly.

Well if i only look on valley my setup give without AA 40fps average, nice was really happy to get 1750pts on valley with my R9 270. Lunched mafiaIII 1080p everything low, begin to shoot randomly 40fps, get a car, get in a crowded heavy game scene, fps drop to 22fps, unplayable.

Steep down resolution to 1680*X instead 1920*X, lower fps driving in heavy crowded scene 28fps, decently playable, on foot and shooting 30/40fps. Better than a console, with better graphics, infinite games etc etc et... .. .

 

 

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


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 3 years ago
 
Posted by: wimpzilla

My last post in this thread because i feel loosing my time.

Sir you are wrong, your test are wrong and badly done, pls refrain to continue to insist, again plz. You begin to be boring and begin to appear like a sand seller!

If you want continue on this way, not including real game benchmark before giving an opinion, i suggest you to change job.

IMO, atm i wouldn't buy your box, for 2 reasons, 1st you are not professional, 2nd you are not showing decent review of your own hardware.

It is really sad that manufacturers can't bench decently their own hardware and came here to find support thinking it would be easy, it is the opposite.

Good luck for the future, for sure i would not recommend your box until you performed decent bench.

See you! 🙂

   

thanks for your advices and see you.

Mantiz: ● ●


ReplyQuote
Mgear1981
(@mgear1981)
Eminent Member
Joined: 3 years ago
 

Well there does seem to be something wrong, my setup

macbook pro 2015 i5 2.9 GHz 

nvidia gtx 970

I bought the akitio node because i liked the full enclosure and beeing future proof.

I use the Apple Tb 3 to 2 adapter

there seems to be really loe performance on the games i tested like company of heroes 2 and civilisation 6. Far lower even then when i use the same gtx 970 in my old akitio 2. The performance drop when comparing them for the node is about an extra 25%. 

Those results and performance drop turn even worse when entering my pc rig gtx 980 ti, i Ger even lower results with the 980 ti in the akitio node then the 970 in the akitio 2 enclosure.

 

lets hope they fix iT soon otherwise there is no usage for the akitio node..

 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
XCVG
 XCVG
(@xcvg)
Trusted Member
Joined: 3 years ago
 
Posted by: Mymantiz_John

HI:

I would like to do the test ( after my national holiday from now till next Wed ) but unfortunately, I don't have TB2 Box. and I see no reason why I need to test the TB2 Box in TB2 Laptop. All TB2 laptop MBP 2015 CPU performance is better than the MBP 2016 even Quadcore too. it will only address the unreasonable result. 

and ingame performance wise, I had tested Far Cry Primal for many times, tested with different resolution and  video quality. the result shows the core (2200mb ) is same and  even a bit lose to Mantiz Venus ( 1100mb ). but the lose few FPS is in the allowance. 

PPL always like to see what they want to see, to believe what they want to believe ,  but my comparison is just the comparison. not intend to drive PPL where I want to be. if someone is really that believe in TB2, just go ahead.

to imagine if someday, I got a new FW, make every device to be 2200mb ( Actually I had that FW already ) but still got the 3Dmark, Valley, Heaven  benchmark the same, ingame benchmark the same, is it a real happy ending? I doubt, then another topic will be posted for example " why 1100mb can run to xxx benchmark then turn it to 2200mb how comes the score is the same??  it is really a waste of time.

   

It may be an issue in some games, though the issue seems to not be an overall performance loss but rather stuttering and dropouts. These are what I'd look for while testing, along with minimum FPS. Reduced bandwidth is also a potential issue for professional applications, and I'd be very interested to see how it affects Photoshop and Premiere in particular.

Late-2015 15" Dell XPS 9550 GTX960M + [email protected] (Mantiz Venus) + Win10


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 3 years ago
 
Posted by: XCVG  

It may be an issue in some games, though the issue seems to not be an overall performance loss but rather stuttering and dropouts. These are what I'd look for while testing, along with minimum FPS. Reduced bandwidth is also a potential issue for professional applications, and I'd be very interested to see how it effects Photoshop and Premiere in particular.

   

HI:

any suggestion to measure the performance in Photoshop & Premiere? Mantiz is the only one manufacturer to face this " issues " and following the guide and valuable suggestion to do the tests.  We are willing to take any productive suggestion & method to make things better. we are here  not just only  for the sales , but also we want to learn & listen to the voices to improve our built quality & enrich knowleges. thanks again for your point out, and let's figure out how we can simulate or finding the workaround to see there's a difference or not for the H2D issues. 

 

 

 

 

Mantiz: ● ●


ReplyQuote
Mymantiz_John
(@mymantiz_john)
Vendor
Joined: 3 years ago
 
Posted by: Mgear1981

 

Well there does seem to be something wrong, my setup

macbook pro 2015 i5 2.9 GHz 

nvidia gtx 970

I bought the akitio node because i liked the full enclosure and beeing future proof.

I use the Apple Tb 3 to 2 adapter

there seems to be really loe performance on the games i tested like company of heroes 2 and civilisation 6. Far lower even then when i use the same gtx 970 in my old akitio 2. The performance drop when comparing them for the node is about an extra 25%. 

Those results and performance drop turn even worse when entering my pc rig gtx 980 ti, i Ger even lower results with the 980 ti in the akitio node then the 970 in the akitio 2 enclosure.

 

lets hope they fix iT soon otherwise there is no usage for the akitio node..

 

   

I am quite interested in your case, can you please run few benchmark with these 2 boxes in your MBP?

Mantiz: ● ●


ReplyQuote
XCVG
 XCVG
(@xcvg)
Trusted Member
Joined: 3 years ago
 
Posted by: Mymantiz_John

 HI:

any suggestion to measure the performance in Photoshop & Premiere? Mantiz is the only one manufacturer to face this " issues " and following the guide and valuable suggestion to do the tests.  We are willing to take any productive suggestion & method to make things better. we are here  not just only  for the sales , but also we want to learn & listen to the voices to improve our built quality & enrich knowleges. thanks again for your point out, and let's figure out how we can simulate or finding the workaround to see there's a difference or not for the H2D issues. 

  

I went ahead and looked up benchmarks for professional applications. I use some of these apps but I'm by no means an expert in them, so perhaps someone with more knowledge in the field can jump in.

When first launched, Puget Systems created a script to benchmark Photoshop's GPU acceleration. It's old, but it should still run. Premiere Pro I can't find much on, but Techspot documented some of their procedures in a review. I think the ever-popular Cinebench has a GPU component. I looked up Blender benchmarks, and it looks like someone created a suite for it. I know AutoCAD is very popular but I'm not familiar with CAD so I don't have much to say on that one.

Does the reduced H2D bandwidth also affect USB devices?

Finally, I think a lot of people are upset by this development because even if the real-world performance difference is small, it's still a hit and it's still less than what we potentially could be getting. A lot of people here want to get the most out of an already limited solution, and getting half of what we're supposed to be getting is quite upsetting.

Late-2015 15" Dell XPS 9550 GTX960M + [email protected] (Mantiz Venus) + Win10


ReplyQuote
ddqp
 ddqp
(@ddqp)
Eminent Member
Joined: 3 years ago
 

Yes, same result is not the key point, key point is what we are supposed to get and what they promised to get.

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


nando4 liked
ReplyQuote
itsage
(@itsage)
Famed Member Admin
Joined: 3 years ago
 

Intel is the Thunderbolt god. Don’t blame these limitations on the enclosure manufacturers. Voice your opinions and findings in a constructive way so that insiders like DanKnight and Mymantiz_John can deliver your frustrations to Intel.

Best ultrabooks for eGPU use | eGPU enclosure buying guide


ddqp liked
ReplyQuote
ddqp
 ddqp
(@ddqp)
Eminent Member
Joined: 3 years ago
 
Posted by: itsage

 

Intel is the Thunderbolt god. Don't blame these limitations on the enclosure manufacturers. Voice your opinions and findings in a constructive way so that insiders like DanKnight and Mymantiz_John can deliver your frustrations to Intel.

   

We know it s not their fault, it s the Intel FW or policy issue. We are on the same side for now with enclosure manufacture.

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


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 

I postulate that if this H2D issue is fixed, a GTX1080 TB3 eGPU user will see a 13% increase in their GTA V FPS results here, more with other games:

What sort of gaming FPS improvement could be had if the half-H2D TI83 TB3 enclosure bandwidth issue was fixed?

This is a summary of the games seeing more significant FPS increase with a x4 1.1 -> x8 1.1 bandwidth increase (approx. equivalent to the H2D issue being fixed). Take from techpowerup’s GTX1080 PCIe Scaling test @ FHD

* Hitman: +46% (63.7->93.1)  here

* Far Cry Primal: 43% (51.4->73.9) here

* Just Cause 3: +42% (73.7->104.7)  here

* Total War Warhammer : +39% (43.2->58.6)  here

* Assassin’s Creed Syndicate: +29%  (55.1->70.9) here

* BF1: +26% (97.8->123.4) here‘s

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


rawweb liked
ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

@Nando, we can’t compare for example x4 1.1 vs x4 2.0 (both directions) to Thunderbolt protocol, especially as only the H2D DATA part is the problem. I emphasize the word DATA. That's what CUDA-Z does - a memory copy method that transfers only data.

USB-C alt mode allows DP, HDMI, MHL or Thunderbolt over the USB-C. These are all different protocols.

This mode enables 4 lanes of DP, and simultaneous USB 3.1 and DP. The following picture represents the alt mode on the device side, and we can see that DP0, DP1, DP2, DP3 are connected to TX1, RX1, TX2, and RX2 which are "4 high speed lanes up to 20Gbs each". In Thunderbolt mode, low-speed RX and TX lines (LSRX and LSTX) are reserved according to datatec.de.


"Thunderbolt uses them as low speed RX/TX lines".

Sources:

http://www.ti.com/lit/wp/slly021/slly021.pdf
http://literature.cdn.keysight.com/litweb/pdf/5992-1628EN.pdf
http://www.datatec.de/shop/artikelpdf/5992-1393en_d.pdf

 

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 
Posted by: goalque

 

@Nando, we can’t compare for example x4 1.1 vs x4 2.0 (both directions) to Thunderbolt protocol, especially as only the H2D DATA part is the problem. I emphasize the word DATA. That’s what CUDA-Z does – a memory copy method that transfers only data.

I used the comparison as it’s x4 1.1 is close to the measured half-H2D throughput, (8Gbps vs 9.22Gbps) x8 1.1 being close to what full H2D throughput would be (16Gbps vs 18.44Gbps). H2D is the predominant directional traffic in those gaming examples, hence disregarding the unaffected D2H flow.

We are leveraging techpowerup’s PCIe scaling testing to see how the x4 1.1 -> x8 1.1 bandwidth increase improve gaming FPS. It was as much as +46% so is significant.

* Hitman: +46% (63.7->93.1)  here

Your points about DATA:

* the affected TI83 TB3 enclosures don’t have DP ports so no competing “DISPLAY” Thunderbolt channel traffic there (unless the user is daisy chaining ports).

* Concurrent enclosure internal USB/SATA/LAN use while benchmarking will impinge on the problematic H2D bandwidth. We’ve measured this offline (enclosure internal SATA HDD).  Our users should unplug drives while benchmarking to enclosure no additional DATA overhead.

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

All the TB3 enclosures have the USB-C (Thunderbolt 3) connector and the TB3 controller that utilize those 4 lanes (TX1, RX1, TX2, RX2) but as low speed. Just guessing if Thunderbolt uses DP protocol lanes somehow simultaneously.

I wanted to point out that CUDA-Z transfer only data, but we don’t know how much data is transferred via H2D in gaming situation. I haven't seen any performance test that proves H2D has an influence on gaming FPS. 

I did not find any performance difference between the Node (TI83) and Devil Box (TI82) yet. Mantiz confirmed the same, no difference compared to the Razer Core. I will conduct a couple of gaming benchmarks if there is detectable and repeatable difference.

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 
Posted by: goalque
 

I did not find any performance difference between the Node (TI83) and Devil Box (TI82) yet. Mantiz confirmed the same, no difference compared to the Razer Core. I will conduct a couple of gaming benchmarks if there is detectable and repeatable difference.

   

Yes, we are looking for that. John Mantiz’ results showed a tiny FPS improvement to the TI82 Razer Core.

@goalque, maybe you have the tools to simulate the H2D problem on your 15″  MBP (M370x) by sending frames to the eGPU?  I believe your MBP‘s SSD is fast enough to do this test.  It also doesn’t have DMI bottlenecking.

– attach a 4K/5K LCD to your eGPU
– obtain a 4K video
– play it back with “5K Player” incrementing the variable speed playback rate until you start seeing stuttering whereup you’ve saturated the H2D link

Doing back to back testing to see if the TI82-enclosure allows higher speed playback compared to the TI83 one would prove the H2D path has more bandwidth now.

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


goalque liked
ReplyQuote
funkyscooter
(@funkyscooter)
Active Member
Joined: 3 years ago
 

Not sure if this helps but I was pointed to a bandwidth test in DaVinci Resolve when I first built my eGPU and was less than happy with the performace in Resolve.

Location on Mac OS  cd /Library/Application Support/Blackmagic Design/DaVinci Resolve

./bandwidthTest

Reports the following for a 2015 MBP TB2 to Highpoint RocketStor

[CUDA Bandwidth Test] - Starting...
Running on...

Device 0: GeForce GTX 980 Ti
Quick Mode

Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1290.4

Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1406.7

Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 241897.4

Result = PASS

 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
valkylin000
(@valkylin000)
Active Member
Joined: 3 years ago
 

i am use GS62vr 7RF with  Asus strix 1080ti in XG station 2,and i try 3Dmark test the performance  it looks normal,but play game (the witcher 3,rise of the tomb raider ) in 1080p

guess what? i got very low fps in both game (about 40-50 fps only...) 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


nando4 liked
ReplyQuote
valkylin000
(@valkylin000)
Active Member
Joined: 3 years ago
 

this is 1080ti  1080p test of rise of the tomb raider.....

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


ReplyQuote
valkylin000
(@valkylin000)
Active Member
Joined: 3 years ago
 

while running The witcher 3 in 1080p all max out (GS63VR 7RF  , XG Station2 ,Asus ROG strix 1080Ti O11G) the FPS is a joke............and the CUDA-Z

HTD only 738 

Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts


Jaye and nando4 liked
ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

Mid 2015 15" MBP (M370X) vs Late 2016 13" MBP (non-touch) - a comprehensive comparison

Windows 10, the latest drivers from Nvidia and AMD. Samsung 28" UHD Monitor (U28D590D) via DP-DP cable.

Tomb Raider 2013: Tests were done twice for each combination, and the average value was taken. Quality was set to Ultimate. V-Sync was off in 1920 x 1080 resolution.




No any significant difference between the Akitio Node (TI83) vs Devil Box (TI82).

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


kotlos and nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 

No any significant difference between the Akitio Node (TI83) vs Devil Box (TI82).   

The same can be said about the TB2 2015 vs TB3 2016 Node results with the sole exception being the RX480 FHD results (84 vs 86.3), which then has the peculiar GTX980 FHD result (102.5 vs 100.3).

Turns out Tomb raider is not a bandwidth sensitive game: https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/17.html . Unlike the following which did see a noticable FPS increase going from x4 1.1 to x8 1.1. As found in techpowerup’s GTX1080 PCIe Scaling test @ FHD:

* Hitman: +46% (63.7->93.1)  here

* Far Cry Primal: 43% (51.4->73.9) here

* Just Cause 3: +42% (73.7->104.7)  here

* Total War Warhammer : +39% (43.2->58.6)  here

* Assassin’s Creed Syndicate: +29%  (55.1->70.9) here

* BF1: +26% (97.8->123.4) here

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


goalque liked
ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

So the Hitman 2016 would be the best candidate. If I chose a couple of tests for the Hitman, what hardware combinations would you suggest? It is also worth to note that Hitman supports DirectX 12 whereas Tomb Raider 2013 doesn’t (2016 does).

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

Hitman is now downloaded through Steam. Will report the results 🙂

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


nando4 liked
ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 
Posted by: goalque

 

So the Hitman 2016 would be the best candidate. If I chose a couple of tests for the Hitman, what hardware combinations would you suggest? It is also worth to note that Hitman supports DirectX 12 whereas Tomb Raider 2013 doesn’t (2016 does).

   

Any of those games that shows market FPS differences bandwidths would be good to test. We are looking to correlate FPS against CUDA-Z H2D values, which I believe you have H2D as 1100MiB/s (Node), 1300MiB/s (Thunder2) and 1700MiB/s (Devil Box).  A +50% difference from lowest to highest.

We’d then be looking for half the FPS improvement in techpower’s PCIe  scaling from x4 1.1 to x8 1.1 (+100%).  So if Hitman on the Node is our control at 100%, we’d be looking for 123% on the Devil Box. That would be an unmissable improvement.

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 

Hitman 2016 : Mid 2015 15" MBP (M370X) vs Late 2016 13" MBP (non-touch)

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

DirectX11:


DirectX12 (visual artifacts, frame drops):


Now there is a clear difference between Akitio Node & Devil Box, but not what we expected. The CPU makes the biggest difference, not Thunderbolt 3.

This convinces me to sell my 2016 13" MBP.

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


nando4, enjoy and kotlos liked
ReplyQuote
XCVG
 XCVG
(@xcvg)
Trusted Member
Joined: 3 years ago
 

What about minimum framerates? Were there any noticeable drops?

Late-2015 15" Dell XPS 9550 GTX960M + [email protected] (Mantiz Venus) + Win10


ReplyQuote
goalque
(@goalque)
Noble Member Admin
Joined: 3 years ago
 
Posted by: XCVG

 

What about minimum framerates? Were there any noticeable drops?

   

Significant drops occasionally, fluctuated a lot. We should not look at the average only. DirectX12 went down to 1 FPS or 2 FPS.

Of course this doesn't describe the real in-game situation.

GTX 980 was superb. I was amazed about that overall smoothness.

 

automate-eGPU EFIapple_set_os.efi
--
2018 13" MacBook Pro + Radeon [email protected] + Win10 1809


ReplyQuote
wimpzilla
(@wimpzilla)
Honorable Member
Joined: 3 years ago
 

Nice ingame bench!

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


ReplyQuote
nando4
(@nando4)
Noble Member Admin
Joined: 3 years ago
 
Posted by: goalque

 Now there is a clear difference between Akitio Node & Devil Box, but not what we expected. The CPU makes the biggest difference, not Thunderbolt 3.

This convinces me to sell my 2016 13″ MBP.

   

GTX980 gives us a pretty clear result here. AT FHD, the 2015 15″ MBPs quad-core CPU and ‘adequate’ TB2 bandwidth sees it shine (66 vs 52). At UHD, TB2’s bandwidth is constricting performance and the 2016’s TB3 bandwidth sees it deliver the better performance (30 vs 39).

Yep, the CUDA-Z H2D bandwidth report for your Node and Devil Box is not being reflected in FPS. The question remains, why is there a difference of 1100, 1700, 2200 MiB/s CUDA-Z H2D for a AKiTiO Node, Devil Box and Razer Core respectively on a 2016 13″ MBP?

The new discovery from your results is the large difference in RX480 FPS between the Node an Devil Box, especially at UHD (42 vs 28). Why is that occurring?

eGPU Setup 1.35    •    eGPU Port Bandwidth Reference Table

15" Dell Precision 7510 (Q M1000M) (6th,4C,H) + GTX 1080 Ti @ 32Gbps-M2 (ADT-Link R43SG) + macOS 10.13.6 & Win10


ReplyQuote
XCVG
 XCVG
(@xcvg)
Trusted Member
Joined: 3 years ago
 
Posted by: goalque

 

Posted by: XCVG

 

What about minimum framerates? Were there any noticeable drops?

   

Significant drops occasionally, fluctuated a lot. We should not look at the average only. DirectX12 went down to 1 FPS or 2 FPS.

Of course this doesn't describe the real in-game situation.

GTX 980 was superb. I was amazed about that overall smoothness.

 

   

Better in DX11 than DX12? Is this the case only with eGPU or more generally as well?

Late-2015 15" Dell XPS 9550 GTX960M + [email protected] (Mantiz Venus) + Win10


ReplyQuote
 of  8