XPS 15 9570 + RTX 2080 = Extreme performance drop, any solutions?

Ok, I finally got my hands on RTX 2080 and Razer Core X, and while I was really excited about the boosting performance, I am very disappointed by the fact that I am facing close to 35% performance drop.

XPS 15 9570

1.41 bios
i7 - 8750H
32GB 2400mhz ram
1TB 960 evo

RTX 2080 with Razer Core X

In theory in Firestrike, RTX 2080 should get 27000 on graphics score, but I am getting 18000 ish only. Sure I was expecting performance drop, but nothing like this. I assumed that 10 ~ 20%, but close to 35% is quite surprising. Am I the only person having this much performance drop or is this normal for XPS users.

 

Well, there are some benchmarks at eGPU, but not perfectly comparable as this is a new card.

Razer blade 15 (running Asus with 2080). He has some Firestrike scores

https://egpu.io/forums/builds/2018-...tb3-asus-xg-station-pro-win10-1803-theitsage/

Well there is one guy with some 9570 benchmarks (running Aorus with 1080):

https://egpu.io/forums/builds/dell-xps-9570-15-aorus-gaming-box-1080-win10/

You can search here for other builds or look later when more people have the RTX 2080. . .

https://egpu.io/build-guides/#perf

 

The higher the FPS on a normal system the greater the performance loss on an egpu. There is too much latency in signal conversion. If you run benchmarks bottlenecked by the GPU you will see closer performance parity

 

I assume your video signal is doing a round trip "laptop -> egpu -> laptop screen" in which case your probably hitting a bandwidth issue down the wire.

Connecting the eGpu to an external monitor will free up > 50% of the bandwidth as its one way traffic, hence the dramatic improvements when setup that way.

Setting vsync to 60Hz (or whatever the built in monitor refresh is) may help as pumping more frames than that down the wire would be a waste of bandwidth as they ultimately get thrown away anyway?

 

I am using an external monitor, so I don't think that's the issue here.

Thank you, it seems like the performance drop is normal D:

Ty for the info. It seems like these kinds of performance drop is normal. But still, it is way more than i expected.

 

9570 + 2080 is

May I ask what you expect to achieve with that config? With the CPU throttling going on in the 9570, I would imagine that the 2080 would always be bottlenecked but could be wrong

Kudos for trying this out though, respect new thinking and people trying the unexpected

 

Well, considering that I've undervolted, changed the thermal compound as well as applied thermal pads on VRM, I was expecting 15 ~ 20% performance drop, not 35%. Especially with external gpu, my cpu stays 80 ~ 82c with full 3.9hz on all 6 cores.

 

3.9 GHhz on all cores would mean zero thermal and power limit throttling, is that really possible with the XPS15 unless you happen to use Davids famous MacBook Pro cooling solution?

 

@improwise it's absolutely possible. A game is not a stress test

 

Stress tests like Prime95 get PL1'd to 45W, around 3.1 GHz, after short turbo time expires. But in Cinebench, higher clocks are reported, maybe the CPU isn't loaded continuously (load spikes are visible in graphs).

 

It wasn't me that claimed "full 3.9hz on all 6 cores" AFAIK that is max for that CPU which made me curious how it was possible as noone else seem to have achieved that other than short times for the 9570.

 

If a 1080 already gets a 20% performance decrease over TB3 due to overhead and what not, it only makes sense that a 2080 will experience a bigger bottleneck (assuming all other factors are the same; screen resolution, testing app, cpu, etc.). The issue with eGPU is that the bandwidth is limited, and a higher framerate increases this bandwidth limitation. The lower the framerate, the lower that bandwidth limitation is going to hurt, as the GPU isn`t trying to push through as many frames as possible.

It is then highly likely that your performance bottleneck will be much lower if you try to game on 4K settings with higher graphical settings. It may very well be that the 2080 might be able to game at 60fps on 4K Ultra, while it can only handle 120fps on 1080p Ultra for example (where you would expect a much higher framerate given the capabilities of the 2080). These are random numbers, but they illustrate the point.

In other words: diminishing returns.

 

Odd thing is, according to Dell Community in Korea, Vega 64 egpu get's 20000+ with egpu

 

Thanks for putting up your findings on this and the other thread. Its solidified my decision not to pursue eGPU and maintain a gaming rig instead.

 

If you have the ability to have a desktop, I would stick with it for now. TB3 is fast, but just not fast enough to give the desktop performance for the price. Maybe the nest gen TB port can deliver it though. I prefer Egpu because my primary device is a laptop, but the performance isn't strong enough to play modern titles.

 

Maybe when the TB controller is integrated into the CPU as Intel has been hinting for a few years, we will see a bump in performance.

Today, Alienware Graphics Amplifier can be faster in some cases:

https://egpu.io/forums/thunderbolt-...s-amplifier-review-faster-than-thunderbolt-3/

 

While I love Alienware in general, I don't really see the point of external gpu for their laptop. Isn't the whole idea of egpu "make your ultrabook super powerful?". Then why bother buying egpu for alienware, when it already has a decent graphics card for the laptops?

 

Not always true, doing small FTFs in Prime95 after 78w turbo boost mine will PL1 to 56w (PL1 is 56w). Interestingly, If i lift the laptop up to increase airflow it will throttle the power. I am not sure exactly why yet, but reverting Intel DPTF to revision A00 removes this behaviour.

 

Interesting.

If you look at the 9550 & 9560 thermal threads, you will see a significant boost in thermal performance of those laptops when lifting well above the desk (obviously to increase airflow to the fans as you note).

In fact, I saw good thermal performance boost on the 9550 & 9560 by propping up the back just a bit. Notes:

-Back foot is a long piece of rubber which helps separate hot exhaust and cool intake air (thanks @GoNz0 ). So ideally you have an equally long foot to prop up the computer positioned in the same place as the XPS' rubber foot.

- Keyboard position might not be very egronomic so may cause hand and arm issues over time.

 

Yeah, I'm relatively lucky in that my desktop has a i7-4790 so that's enough CPU to not bottleneck too bad for at least another year or two. After that its $$$ for the entire mobo + RAM + CPU upgrade path which is hard to justify except the sunk cost of the Steam library (vs buying all console games again, ugh). Hope in 2 years time when it comes time to upgrade the 9570, eGPU technology will have progressed again.

The 4790 probably edges out the 9570 hex-core in many real life scenarios as it can maintain 3.6Ghz indefinitely on all cores even with the humble CM212 cooler. Even with undervolt I'm staying at 3.0Ghz on all cores on the i7-8750H

 

They also sell an ultrabook type I believe, like the Razer stealth

 

Unless you are talking about M15, there's nothing like it, and even M15 is too powerful to take the benefit of eGPU.

 

I also seen about a 4 degree drop in average CPU temperature in DTFP A00 with the rear lifted, and maintaining 56w PL1.

The following quote confirms my suspicion about DTFP A01, thankfully we have the option, though there could be drawbacks due to other changes.

" This confirms my assumption. With increased airflow DTFP lowers performance to lower the laptops surface temperatures because it believes it is on your lap,



" The Dynamic Power Mode Policy can therefore, adapt to the notebook's usage. If the system is docked, the policy can set a higher temperature limit. Similarly, if the system is hand-held or worn, the policy can set a lower limit in real-time. Thus, maximum performance within the dynamically adjusted power limit can be provided. "#



https://www.notebookcheck.net/Dell-...-policies-across-its-XPS-lineup.247462.0.html "

 

@J99 Very interesting info.
Where do you monitor the status of DTFP? how do you know if you are in A00 or A01?

 

You can see which version you have installed in device manager.

You can download both versions from Dell Support,

A00 https://www.dell.com/support/home/d...productCode=xps-15-9570-laptop&driverId=JPXGW
A01 https://www.dell.com/support/home/d...productCode=xps-15-9570-laptop&driverId=591DK

 

Hey, I am not that surprised about the performance drop. I tried an external 1070 with my 9570 and the speed was barely above the internal 1050 ti. I ended up sending it back. I also never understood where the bottleneck was, but the egpu 1070 was never much faster than on my old 9550, where I also tried it.