Alienware Graphics amp question

If you want to connect an external monitor, do you connect the external monitor cable to the laptop or to the graphics amp itself or to the video card in the graphics amp?

 

Amp. Also don't get high end GPU's they will get bottlenecked.

 

what about for multiple monitors? i.e. if someone wants to connect using TB3, mDP, and HDMI

 

I mean the card will simply be handled as an extra card. All cards you plug to the AMP card will be handled by the AMP card, when you connect the screen to the notebook then it will handled by the notebooks graphics solution (iGP or dGPU), however note:

If you plug the card in your notebook instead of your AMP and use the AMP card, then the card will lose even more performance since it will have to route everything from the external card through the notebook to the monitor, instead of direcly to the monitor.

 

I see a typical graphics card has 3 mini display ports and one HDMI so thanks for the info about having to connect the monitors only to the GPU in the graphics amp to get the best performance and not connect one to the TB3 port on the laptop

imagehosting

 

Just make sure you don't mix anything crazy. The higher end the card the more the nerf.



The performance is generally really unstable on external cards. The average seems decent, but the lows are really a problem. So be careful with that. I wouldn't really recommend using anything above a 1060 for this kind of setup. If you have a 1080 and just want to play for fun on a notebook and got a 1080 arond, go for it, but would never recommend buying a 1080 specifically for eGPU stuff.

 

Wow that's crazy man, so Alienware still didn't fix the initial issues plaguing the graphics amp to this date? And there I was thinking one could buy an Alienware 17 R5 now and get the graphics amp for when the next gen GPUs are out to be more future proof

 

You are showing TB3 results. Which runs over PCH and thus causing more degradation in performance. I am running a heavily OC-ed 1070 (2100mhz) in an AGA for 2 years and recently I use my AGA on the newer AW15R3 which also has a 1070 built in. Both OC-ed and they perform pretty much identical which just a very very very slight edge for the 1070GTX in the AGA. What you say is basically incorrect.

The AGA is nothing more than an external PCI express port, the cable is just sending the PCI express signal back to a mini pci express slot within the laptop which is connected directly to the CPU. You cannot compare this to the TB3 setups which run through the PCH and benchmarks have shown multiple times the AGA is far more consistent in terms of performance.

His data shown TB3 results which is not comparable to the AGA. The AGA has yet to show any bottlenecks in my setup

TB3 solutions run on the PCH, so it has to share resources with all other devices hooked onto the PCH. Those screenshots are from Jayz2cents channel where he tested a MSI laptop with a TB3 external dock.

 

The newest interface has a max bandwitch of 30GB/s, TB 3 has 32GB/s. You cannot run a 1070 not nerfed, unless they made a 2018 version that magicially runs with much higher bandwitdh, which I'm not aware off.

EDIT: nvm seems like the AGA uses PCI-E 4x which is around 40GB/s?

Do me a favor and make a screenshot of GPU-Z.

Also your AGA also goes through the PCH, Dell lied to you. If you want proof, go open HWinfo64, check your PCH tab and u'll see the card there.

Also a notebook GTX 1070 (especially in a smaller notebook like your R3) runs significantly slower than a desktop 1070. If you have the same speeds on a desktop GTX 1070 wth a AW 3 1070, then you already confirmed, that there is a severe bottleneck.





So yeah, no way you'll run a GTX 1070 stock let alone OCed at full speeds on PCI-E 4x.

@Phoenix:
Maybe one day there will be a PCI-E 16x connection which would allow to run at full speed, lets hope and pray =D

 

I cant comment on the for example older 13inch versions with ULV cpu's if on those the AGA is running on PCH or not But in the AW15R2/AW17R3, AW15R3/AW17R4 it does not run grom the PCH.

Its marked in HWinfo as connected to one of the PCI busses as you can see in my screenshot. Not to the PCH slots which are filled except for one by my SSD's and Wifi card.

Also as you can see from the GPU screenshot it is running at 4X speeds. same as what HWinfo says for that port. The AGA works by fully disabling the internal Geforce GPU and it is a reroute to an external port. That is why you also cannot see both GPU's at the same time.

4X is still not a huge bottleneck for any current GPU out there as proven by various benchmarks. https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080_PCI_Express_Scaling/

As you can see in the techpowerup article. Running the card through a PCH-PCI express slot (chipset results) matters.



Both my 1070's are overclocked by the way thats why they score almost identical

1070gtx in the AGA @ 2050, 7700HQ https://www.3dmark.com/fs/15122509
laptop 1070 running at 1950 (but has more cuda cores) https://www.3dmark.com/fs/15120340


So no TB3 and the AGA are not comparable at all. The AGA performs almost with no performance degradation (or lets say very minor). TB3 however is unfortunately handicapped by the fact that it is connected to the PCH and also has to deal with protocol overhead over TB3.

Why I still use the AGA you might ask? Its a nice monitor, keyboard dock and the laptop runs completely with the fans turned off under load. that huge heatsink can cool an i7 7700HQ without spinning up while gaming.

 

It is a huge bottleneck to people like me. Lets take your benchmark as an excample. You card which runs at 2150mhz would run like this in a benchmark:
https://www.3dmark.com/fs/15241456

That's still a 15% bottleneck in a benchmark that utilizes the CPU very well. Games tend to be much more unforgiving, especially when you're playing a game that uses the bandwitdh more.

https://egpu.io/alienware-graphics-amplifier-review-faster-than-thunderbolt-3/

 

My bad it was 2050 that run and that guy ran with much higher memory clock speeds. I did not overclock the memory. Also I compare apples to apples. You are showing a run where a user used a 8700K and even though the graphics score is mostly based on the GPU alone, the CPU still is a factor.

Again I have shown you numbers of an internal 1070GTX which is overclocked, 100mhz less but with more cuda cores vs an external GPU with a bit less cuda cores but 100mhz more clockspeed and you can clearly see they almost score identical. Also I have posted a link where PCI express scaling is not a factor yet. Your link, and i have seen that one before is bottlenecked by an Intel ULV cpu which might run on the PCH? Not sure, that is the only source that I saw that claimed that the AGA was running through the PCH. But as shown in my setup this is not the case.

The 1070GTX that I use was actually part of my deskop setup, which now is somewhere discarded in my closet because there where no performance differences at all.

 

The AGA does not significantly bottleneck. Stop spreading disinformation. My 13r3, 7700HQ with a 1080ti FE smokes 17r4s with the 7820K and 1080. It doesn’t hold up to most desktop processors with a 1080 ti, but matches the numbers from the few examples running older desktop or newer mobile cpus.


Sent from my iPhone using Tapatalk

 

CPU has nothing to do with the GPU score on firestrike. It's not a factor in firestrike unless you take the combined score into consideration which we don't do. Also again, a GTX 1070 notebook version is significally worse compared to a desktop GTX 1070. Saying that your desktop 1070 runs the same as a notebook 1070 is confirming a bottleneck.

@Ihtfp06 Where is your benchmark? Where is the numbers showing your performance on the 1080TI.

 

Thats why there are 1070s with equal clockspeeds in the 3dmark database that perform lower on lower specced CPU's......

Even though the difference is small, its there. Other users claim that tested it around 3 to 5% https://linustechtips.com/main/topi...-a-slightly-higher-firestrike-graphics-score/

Which is similar to the score difference between your link and my link (even though that guy also has a heavy memory OC next to a 100mhz higher boost).

https://www.3dmark.com/fs/10010898
This is run I did when my 1070GTX was still in my desktop, slightly higher clocked, higher memory clocks as well, and again you can see that graphics score are in exactly the same range. It cannot be more apples to apples, I provided you with 3 Firestrike runs, 2x the same 1070GTX used in a desktop and AGA and an AGA vs internal 1070 comparison on the same laptop. There is no bottleneck.

In DX11 Nvidia is offloading quite a bit onto the CPU, not as much as AMD (thats why in general AMD tends to perform a tad better on DX12).

Also you conveniently ignore the rest of my posts.

-No, the AGA is not connected through the PCH (for all Quadcore Alienwares).
-Performance is near identical with similar clocked internal and external 1070's to prove the bottleneck you portray is not there.
-Techpowerup and other sources clearly state PCI express bandwidth is not a big factor yet between 4x, 8x, 16x.

 

25K on Firestrike using the laptop display.


Sent from my iPhone using Tapatalk

 

I got the equipment in my sig, if benchmarks needed, let me know.

Everything is running stock speeds, only repasted with AS5 the CPU/GPU and eGPU. I have already replaced a motherboard so I will not be overclocking right now.

 

Simple answer @Phoenix -

AGA hooks up directly to the PCIe lanes/CPU. So you get full PCIe 3.0 x4 bandwidth. Still bottlenecks higher end GPU's GTX 1080+

TB3 based solutions go through the PCH for conversion to PCIe signal and then to CPU, so that adds conversion overhead along with adding a decent amount of latency.

 

When using my external 4K monitor on my Titan X in the graphics amp (AW13 i7 7700HQ), the performance was very close to regular desktop cards.

When rendering onto the internal display the performance was definitely less.

If you play GPU bound games and use the external monitor(s) to game with the amp + GPU you should be fine.

 

One thing to note with 4k is that, there are other factors that start to make a difference. At 1440p/1080p is where the bottlenecks are actually perceivable.

 

The AGA and AW laptop with external monitor is a great combo for at home gaming as opposed to a tb3 connected box or stand alone internal gpu. I’ve had very good performance with 1080ti and Titan xp in AGA and my aw 17r4 and aw13r3....

 

Just ran it on an external display. 27.5K. Tell me again how it’s bottlenecked?


Sent from my iPhone using Tapatalk

 

Would be nice if we could combine pci express lanes, like using two Thunderbolts for 5 pcie

 

Lol you will never fill up 4x PCIe 3.0 lanes with Bandwidth. The GPU's own memory bandwidth, as well as internal CPU latency and RAM latency would come into play long before you fill up 4x PCIe lanes.

Proof:
https://www.pugetsystems.com/labs/a...-E-Speed-on-Gaming-Performance-518/#4kResults

https://www.techpowerup.com/reviews/Nvidia/GTX_980_PCI-Express_Scaling/22.html

I second Rinneh's opinion.

Stop spreading false info. The AGA is currently the fastest external GPU implementation.

https://egpu.io/alienware-graphics-amplifier-review-faster-than-thunderbolt-3/

Any difference in Benchmarks between a desktop card and an AGA will be a result of the Laptop CPU and it's supporting Northbridge chipset and RAM.