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

@Nando, the XPS 9365's numbers were done through the Mantiz with older firmware. I will re-run these with the new firmware once I have it. The Alienware 13 R3 has 4x lane. Dell document on PCIe lanes on its Thunderbolt 3 laptops confirms this.

@wimpzilla, I greatly appreciate your input and knowledge. These benchmarks only tell part of the story, that I agree. eGPU is a hobby of mine and between a full-time job and a Mac repair business, it's hard to do elaborated testings in my spare time. I'd welcome suggestions and games to run more benchmarks. The only game I have and play regularly is PES 2017. 😀

Since you are putting effort in, and again be rewarded by great visibility for this forum, i only suggest you to start from now to get the good habit to bench, even it is from your free time.

Since one day when you will show data like these, i hope for you and this forum, to more specialized people like companies, you will get teared down for proving nothing solid.

And it will end everyone giving his opinion randomly without any real hardware proof, without a solid conclusion that allow to move further, like already happen in this thread.

To assess to mantiz and akitio there were a problem into the bandwidth, we spend dunno how many pages of nonsense. And for this i will take the example made by mantiz, it is not because it show 2kMb/s transfer, that is really what happen, number could be falsified by the firmware, FPS output NOT!

This is the only thing i can notice and will continue to repeat until i saw decent performances bench that put the hardware under stress and give decent overview of the hardware performances.

And again, performances on a bench do not reflect the real ingame performances, because not all games are coded to same way, are optimized the same way for AMD/Nvidia gpu, Intel/AMD cpu, etc.

I can gift you a couple of games with built in benchmark on steam, just run the bench and post them along your usual benchmarks.

Like this you propose an extended review, because the best example is the video posted along, on laptop is fluid, the other stutter like hell, even if the superposition bench give the same result! Pity!

This is the main difference from review where companies generally send you their product to test, you can't be too bad on the review and the one you purchase yourself the product and give a real review without any pressure from the company!

@winmpzilla for real games bench you can see my threads -  https://egpu.io/forums/builds/2012-15-mbpr-gt650m-gtx106010gbps-tb1-akitio-thunder2-win10-enjoy/ and  https://egpu.io/forums/builds/2013-15-mbpr-iris-only-gtx106016gbps-tb2-akitio-thunder2-win10-internal-display-optimus-enjoy/

I have videos etc.... 

You must know that eGPU is not only for gaming, most of the people here wants to use Pro Apps for work etc... 

Iam a gamer and use my eGPU and my Macbook for that. The bench are the best ways to test the eGPU power like i did with my thread for the performance drop. In modern games we have bad ports from Xbos or PS4 to Windows and we must wait for 1000 patches to get solid FPS even on Desktop PC. From my bench i know that eGPU + external display (Monitor) all GPUs have 20% performance drop for FULL HD resolution. On the Internal Display we have 20-50% performance dropbased of Thunderbolt (1 2 or 3) and the GPU!

@itsage is great person and he makes the best news, reviews etc and you can see that he makes that for free (no banners in the forum), he is not a gamer like me to post gaming bench. I can give you some channels in YouTube where gamers tests their eGPU performance like these users

• https://www.youtube.com/channel/UCyOYYXm5u0JoMVxXFJSKsNg/videos
• https://www.youtube.com/channel/UCzaxv22N1CAIJ_OXSlbBMig/videos?shelf_id=0&sort=dd&view=0
• https://www.youtube.com/channel/UChbOIU1a7DWzIQxYyjQ0zAA/videos?view=0&shelf_id=0&sort=dd

I would also be interested to see some additional results for FPS in popular games, Overwatch, Witcher or whatever is trendy among review sites.

For someone looking at buying a new portable notebook and eGPU, this would be a big help determining if lower CPU power (like on the Dell XPS 9365) or half TB3 bandwidth (like 9560) are going to cause significant dropoffs in real world gaming performance.  I know the gaming benchmarks are much more repeatable, so both results have value.  Be known for the best reviews of TB3 enabled notebooks and the most comprehensive eGPU testing and you will suck away gaming and GPU compute enthusiasts from the big sites as this technology gains recognition:-)

Thank you all for the suggestions. @wimpzilla please PM me the game codes on Steam. I'll run those in-game benchmarks and update my blog posts.

Sure I send you a pm, just add me on steam, i will gift you a couple of games with built in benchmark.

Be known for the best reviews of TB3 enabled notebooks and the most comprehensive eGPU testing and you will suck away gaming and GPU compute enthusiasts from the big sites as this technology gains recognition:-)

That's what i pointed, unfortunately into the eGPU world it need the double of the work. Because you need to know what are the performances of a gpu and then what it deliver under a eGPU solution. A lot of misleading information in between for the basic consumer. Since there is no specialized site about at the moment, it would be nice to be the 1st and the best at as you said! And begin to ask companies to send you product to review! It is how forum built and grow.

Thank you again @wimpzilla. I was able to run benchmarks through 3 games tonight. Click the links for the screen caps with additional information. I got GTA V installed but was struggling to get to the benchmark menu to show up.

I will check for the GTAV benching issue, usually people that do review build an automated process, that can be repeated, to ensure a good data output, X times.

It's pretty clean the performances, even with a drop, are really good. Give us your feeling about, you are the one who is reviewing, if you feel the bench smooth and fluid everything is ok.

So these bench showing good result, hoping you will continue on that, and a nice video like you already did, is almost perfect having a good visual and informative impact.

Nevertheless if you feel the frametime not smooth, the bench stutter, plz spend a bit of time to check with MSI Afterburner the usual FPS/frametime/CPU/GPU utilization.

Quote from GTAForum

1.the games does a save point ,after prologue

2.after mission 1,try to benchmark

i did succes benchmark after above.1 & 2 steps.

If GTAV give the same overall result you are ready to go, since it is heavy on both cpu and eGPU.

@enjoy: I know your threads is nice and informative, with in game pics and bench, it is a fairly enjoyable thread visually, especially those in game high res pics. ^^

But here was for review and testing purposes, again to try to help how to show results easily.

@itsage

thanks for these infos! 🙂

what egpu box are you using for these tests?

i'm really curious about benchmarks in a laptop with a quadcore cpu i7-6700hq or i7-7700hq and a normal thunderbolt3 x4 port ( tb3 port is tb3/pcie in alienware) like hp zbook g3 / ux501vw /msi gs63 

I had more time to install these same games on the Dell XPS 9365. CPU makes a difference for sure but not as much as I thought. Tom Clancy's The Division actually popped up a warning saying this XPS 9365 is not capable to play this game. I ignored and proceeded with Ultra settings to match that of the Alienware 13R3. I ran these tests through the AKiTiO Node with the latest firmware which has the H2D fix.

One thing I can conclude from all these tests is that I will not consider a bulky gaming laptop such as this Alienware 13 R3 as my only computer. The cooling system is too loud. The XPS 9365 with the Kaby Lake-Y CPU showed its weakness in gaming benchmarks. That wasn't its selling point though. Therefore I'd take the XPS 9365 over the Alienware 13 R3 any day. Then pair it with a GTX 1070 eGPU to compensate external transmission loss.

Posted by: itsage

 

I had more time to install these same games on the Dell XPS 9365. CPU makes a difference for sure but not as much as I thought. Tom Clancy's The Division actually popped up a warning saying this XPS 9365 is not capable to play this game. I ignored and proceeded with Ultra settings to match that of the Alienware 13R3. I ran these tests through the AKiTiO Node with the latest firmware which has the H2D fix.

One thing I can conclude from all these tests is that I will not consider a bulky gaming laptop such as this Alienware 13 R3 as my only computer. The cooling system is too loud. The XPS 9365 with the Kaby Lake-Y CPU showed its weakness in gaming benchmarks. That wasn't its selling point though. Therefore I'd take the XPS 9365 over the Alienware 13 R3 any day. Then pair it with a GTX 1070 eGPU to compensate external transmission loss.

	13R3 - GTX1060 dGPU	13R3 - GTX1060 eGPU	XPS9365 - GTX1060 eGPU
Unigine Valley	2,368	2,408	2,067
Unigine Heaven	1,431	1,363	1,188
Unigine Superposition	2,085	2,073	2,164
3DMark Time Spy	3,537	3,677	3,507
3DMark Fire Strike	11,326	11,005	9,292
Metro Last Night Redux	83 FPS	71 FPS	58
Rise of the Tomb Raider	54.28 FPS	47.83 FPS	43.13
Tom Clancy's The Division	49.9 FPS	40.8 FPS	39.1

   

thanks! so there is only a loss of fps between 20%(tom clancy/ rise of) and 30% (metro last) for that dell 9365. i think it is a good result:)

Then pair it with a GTX 1070 eGPU to compensate external transmission loss.

Unfortunatly this is the trend of the forum, i acknowledge that. But is is only a trick since FPS OUTPUT = GPU+CPU computing.

If you remove the botleneck on the GPU, you will still bottleneck with the CPU, so bottleneck more the 1070.

Like i already said before, there is no laptop, imo, even the top high end you tried, that deliver enough performances for match a 1070, even less a 1080, no words on 1080ti and up. Try to understand when the bandwidth problem is the issue, when the hardware is the other.

Between eGPU/dGPU  1060, this is the difference between bandwidth. Between eGPU 1060 on both different laptops, the hardware is the issue, especially if the specifications are different but share the same TB3 config. It is easy to confuse both things here.

I should mention the 13R3 was plugged in at all time due to fear the battery would run out during game download/installation/benchmark. The XPS9365 was running on battery in the meanwhile.

A shoutout to AKiTiO Support for their participation in our community and responsiveness with issues/concerns. Without the involvement of Thunderbolt 3 enclosure manufacturers, we would not have received the quick turnaround for this H2D bandwidth issue as we had.

I also want to thank @wimpzilla for calling me out because my initial benchmarks were not comprehensive. I appreciate your input to make eGPU.io better.

Posted by: itsage

 

I should mention the 13R3 was plugged in at all time due to fear the battery would run out during game download/installation/benchmark. The XPS9365 was running on battery in the meanwhile.

A shoutout to AKiTiO Support for their participation in our community and responsiveness with issues/concerns. Without the involvement of Thunderbolt 3 enclosure manufacturers, we would not have received the quick turnaround for this H2D bandwidth issue as we had.

I also want to thank @wimpzilla for calling me out because my initial benchmarks were not comprehensive. I appreciate your input to make eGPU.io better.

   

max load for that ultrabook is 24.3watt
https://www.notebookcheck.net/Dell-XPS-13-9365-2-in-1-Convertible-Review.193704.0.html

You should bench with the AC power supply, since on battery you are not using the performance power management profile.

So you get more or less half performances from the cpu. Try again with the power supply.

Generally you can bench with the AC brick without battery, to feed all the current to the motherboard VRM along the performance power management plan.

Here you are not assessing the power consumption, so it is ok to test with the AC brick, since you are in a setup where the laptop is on a desk. The eGPU is not powered by battery at then end. But you can assess that too if interested. I expect the result in line with the others with the AC brick, the results are already really good. Again if you did not notice big frames drops or stutters you are ready to go.

To confirm the cpu limits, run GTAV bench, as i'm benching on MafiaIII, to stress the cpu + eGPU. In my case i clearly saw the TDP lock effect and my cpu performances limiting my eGPU performances.

I can gift you a last game, like i did to @mgear, Hitman 2016 with the ingame bench, then i finished to sell all the free keys i holded on steam. ^^

ASUS XG Station 2 just released V23P firmware

Here's my benchmark result using 3D mark Time Spy test http://www.3dmark.com/3dm/19479798?

But sadly, I forgot run the test before update firmware.

Just played WoW a while, FPS  has significant improment, from 50+ to 80+, It seems Asus ROG XG Station 2 has solved H2D problem

Posted by: jasl

 

ASUS XG Station 2 just released V23P firmware

Here's my benchmark result using 3D mark Time Spy test http://www.3dmark.com/3dm/19479798?

But sadly, I forgot run the test before update firmware.

Just played WoW a while, FPS  has significant improment, from 50+ to 80+, It seems Asus ROG XG Station 2 has solved H2D problem

   

My score is 6937 on the same NUC, running with the Titan X Pascal.

Can you help me to understand these Data ? 38GB/s Is it strange ?

I guess that's the GPU Memory bandwidth, i.e. between VRAM and GPU.
Mem clock * Bus width = Bandwidth
1200Mhz * 256bit = 38400 MB/s
Nothing to do with TB3 bandwidth, either H2D or D2H.

Posted by: itsage

 

I had more time to install these same games on the Dell XPS 9365. CPU makes a difference for sure but not as much as I thought. Tom Clancy's The Division actually popped up a warning saying this XPS 9365 is not capable to play this game. I ignored and proceeded with Ultra settings to match that of the Alienware 13R3. I ran these tests through the AKiTiO Node with the latest firmware which has the H2D fix.

One thing I can conclude from all these tests is that I will not consider a bulky gaming laptop such as this Alienware 13 R3 as my only computer. The cooling system is too loud. The XPS 9365 with the Kaby Lake-Y CPU showed its weakness in gaming benchmarks. That wasn't its selling point though. Therefore I'd take the XPS 9365 over the Alienware 13 R3 any day. Then pair it with a GTX 1070 eGPU to compensate external transmission loss.

	13R3 - GTX1060 dGPU	13R3 - GTX1060 eGPU	XPS9365 - GTX1060 eGPU
Unigine Valley	2,368	2,408	2,067
Unigine Heaven	1,431	1,363	1,188
Unigine Superposition	2,085	2,073	2,164
3DMark Time Spy	3,537	3,677	3,507
3DMark Fire Strike	11,326	11,005	9,292
Metro Last Night Redux	83 FPS	71 FPS	58
Rise of the Tomb Raider	54.28 FPS	47.83 FPS	43.13
Tom Clancy's The Division	49.9 FPS	40.8 FPS	39.1

   

Nice

Can you please post the CPU clocks while running the games? 

Also, I still remember the rush for a quad core Q6600 back in the day, 11 years ago. Everybody saying that you need it for gaming, if not now then in 2-3 years. Intel made quite a few bucks on gamers. Still does with all the recent I7s with 4C/8T. 

Well, nice one anyway

Where would I be able to contact Powercolor about this issue? It seems that mine has this issue as well.

I finally have all the equipment to saturate the Thunderbolt 3 Data Limit of 22Gbps. This is the result of Samsung 960 EVO NVMe flash storage installed in the AKiTiO Node with the latest firmware.

Nice Idea, this thing is damn fast!

Try with ATTO Benchmark, give you a better overview.

This is from a 960Evo review, but you can find plenty to get a basic starting point. You can try also other tool suggested in the review and compare.

I ran this Samsung 960 Evo NVMe M.2 SSD through all Thunderbolt systems I have atm. Here are the results:

1. X99 Motherboard M.2 slot

2. X99 Motherboard Thunderbolt 3 port - AKiTiO Node

3. Late 2016 15" MacBook Pro - CPU -  x4 Thunderbolt 3 - AKiTiO Node

4. Razer Blade Stealth - PCH - x4 Thunderbolt 3 - AKiTiO Node

5. Dell XPS 12 9250 - PCH - x2 Thunderbolt 3 - AKiTiO Node

6. Late 2013 Mac Pro -  Thunderbolt 2 - AKiTiO Node

@itsage, you are getting less than USB 3.1 (Gen2) of  960MB/s performance levels from your notebooks' TB3 link for the writes (H2D).

Why? Is the Node only certified for eGPU use with Intel throttling it's use for other PCIe products? Are they segmenting markets,  so partners can charge a premium for a storage enclosure with unlocked 22Gbps performance? We know Intel are providing an eGPU and PCIe TB3 enclosure firmware.

Posted by: nando4

@itsage, you are getting less than USB 3.1 (Gen2) of  960MB/s performance levels from your notebooks' TB3 link for the writes (H2D).

Why? Is the Node only certified for eGPU use with Intel throttling it's use for other PCIe products? Are they segmenting markets,  so partners can charge a premium for a storage enclosure with unlocked 22Gbps performance? We know Intel are providing an eGPU and PCIe TB3 enclosure firmware.

 

 

 

The results of Late 2016 15" MacBook Pro - CPU -  x4 Thunderbolt 3 - AKiTiO Node shows that the differences between CPU and  PCH tb controllers are inappreciable.... if you don't have a system that overload all the pch lines.

@vava, the PCH vs CPU comparison cannot be made given the SSD H2D throughput isn't anywhere near the 22Gbps TB3 limit. If it was then we'd expect to see it's throughput constricted due to the PCH-attached devices like SSDs, wifi, lan sharing the 32Gbps DMI  bandwidth.

Nando you are right, but the performance of this experiment are really strange because, MB controller is  very slow. It's not a fair experiment for DMI saturation and TB controllers loss of efficency but I think that the pcs  used during the experiment sessions were using part of DMI  for lan and other connected devices.

My skepitical question is: "Can we saturate DMI during a standard game session, affecting the performance of a PCH controller? Or the advantage of a CPU TB controller is visible only when you use a system for professsional purposes as multi-sources rendering, brute force algorithm etc etc?

Interesting results, dunno if the controller noticed that is an SSD or used it own eGPU general profile.

• Desktop vs Desktop TB3: The TB controller seems to allow only a peak of 1000MB/s write, 2400MB/s read. It is 500MB/s less on write and 700MB/s less on read, when compared to a basic cpu 4x pci-e. On overall the result are good, the r/w speed are consistent along the different size files. The other thing you notice is the TB controller seems to have a neat write/read max value set, not always read speeds cap this value, but the write speeds definitely does. This cut-off seems to be applied when the file size is bigger than 8KB.

• Desktop TB3 vs MBP TB3: Well dunno if the hardware performance difference alone could explain the additional 180MB/s r/w loss compared to the desktop TB. I dunno how to explain it. It seems the controller also set a fixed cut-off value. It seems there is also a different setting depending the file size. I suppose the max r/w speeds are different if the TB controllers is hooked on a desktop or laptop motherboard. Really dunno here if it is a normal hardware behavior limitation and/or a different TB3 config.

• PCH TB3 vs CPU TB3: As @Vava726 said, the difference between both is not huge, there is at max 60MB/s write difference, read speeds are the same,  i would had expected more. The transfer speeds are less uniform, even if there is a cap the max value is not always reach imo, this fit well with the fact that the DMI is used/busy for other things also. When the system is under stress it could have an impact on performances, maybe a bit bigger than what we saw here.

• PCH 2X TB3/PCH TB2 vs PCH TB3: Same trend as before about the write speeds, the main difference here is the read speed, ranging from 700MB/s less to 900MB/s less for the TB2. This explain the performances differences users saw between TB2/2X TB3 and a the TB3. Note also that the performances difference is due to the read speed in this cases, it also explain why the TB2 still perform well when not saturated.

I'd like to know if the sata controller of Venus can be more efficient.

I love the fact someone from Akitio is an active member on this forum... looking forward the release of the Node lite.

Posted by: Eightarmedpet

I love the fact someone from Akitio is an active member on this forum... looking forward the release of the Node lite.

We have active participation from AMD XConnect, AKiTiO, Mantiz, Sonnet, and others on our forums.

Brilliant to hear. Can I make a suggestion, I'm not sure if it's just me but your sites IA doesn't make complete sense to me (maybe I just need dig around more) I've had to bookmark a couple of things because I struggle to find them... feels like a visual browse page under nav pages might be good? The mini egpu recommendations page is a good example of buried content... happy to lend a hand if needed.

Posted by: Eightarmedpet

Brilliant to hear. Can I make a suggestion, I'm not sure if it's just me but your sites IA doesn't make complete sense to me (maybe I just need dig around more) I've had to bookmark a couple of things because I struggle to find them... feels like a visual browse page under nav pages might be good? The mini egpu recommendations page is a good example of buried content... happy to lend a hand if needed.

Please post under Feedback & Suggestions forum so that we can further discuss. We'd greatly appreciate your help. Mockups would be wonderful.

Will do!

Just a quick heads-up: Gigabyte's Aorus Gaming Box has the same problem. I've contacted Gigabyte but have received no reply.