Best Desktop Graphics for your Laptop - Proprietary vs. Thunderbolt!

[AlexTheGreatish]

The ultimate showdown between the Razer Core and the Alienware Graphics Amplifier, but which external GPU solution will come out on top?

 

Buy the Graphics Amplifier on Amazon: http://geni.us/aa1C

Buy on Newegg: http://geni.us/BWU9v

 

 

[Str_Mape]

No first posts for people this time >.>

[l0bst3r]

Well, that's what I call a non-clickbait-thumbnail

[shadowbyte]

this thumbnail is beautiful

[sNowCat]

By chance was Fallout 4 with the High Resolution texture pack bench marked? Always wonder if the 50GB of extra textures would saturate and bottleneck the PCIe 4x connection.

[Coffey]

He should look at old way of hooking up an external graphics card to a laptop, with express port and mini pcie. It would be intresting to see how much a pcie 2.0 1x would bottleneck a 1080ti

[Drak3]

The Alienware solution has the advantage for this reason: It's a dedicated GPU box. It's PCIe 3.0 x4 connection is dedicated to the GPU.

The Razer Core, on the other hand, allocates its PCIe bandwidth to support a Gigabit NIC, 5Gbps USB controller, and the GPU. Thunderbolt 3 spec tells us that Thunderbolt 3 only carries a PCIe 3.0 x4.

[O8h7w]

There is a slight but important error in this video: the bottleneck actually IS in the laptop. The Thunderbolt 3 cable and interface as such is not the problem, the problem is that the Thunderbolt 3 controller - on the motherboard! - is attached to the PCIe lanes coming through the chipset rather than those coming directly from the CPU. This was mentioned in several reviews of the Skulltrail NUC.

 

As far as I know this is not required, so an ambitious motherboard manufacturer could conceivably attach a Thunderbolt 3 interface to the PCIe lanes that come directly from the CPU. But I have yet to see one, which might mean there are technical or legal obstacles.

Edited to add disclaimer: I do not know these things for sure, but I remember having read and heard this from reputable sources discussing external GPUs on previous products - including on the Razer products. I have no information on the Alienware system outside the video. I might look it all up a bit better in a couple of days.

[User Name]

It's crazy the hate that you see in the YouTube comments. I have a graphics amp and love it. My case for getting it is traveling for work I can use the laptop for work then game in the hotel get home and use the graphics amp with my 980ti. Lol my laptop runs 10hr days like 7 days a week, surprisingly it is like a 5% fps loss over a tower so not too bad IMO. Found one online for like $150 USD so it was worth it. Glad to see the video because I was looking into razer at one time. Glad I chose alienware.

[Cazzzz]

 

@O8h7w is right on. The performance difference between the two would be equalized if the test was able to be performed on a laptop with PCIe lanes direct to the CPU, or, perhaps even just lanes that could be dedicated to the device instead of shared. So I have some questions that come out of this:

 

1) If anyone was going to try to design a Thunderbolt controller that used CPU-based PCIe lanes, wouldn't Razer want to do that? So if I had access to Razer I would want to ask them if that is something they are working on, and if not, why? (@O8h7w, you're probably right that there is some legal or technical challenge.)

 

2) If the goal is truly to get desktop gaming performance on a laptop, wouldn't the goal also be to dedicate 16 lanes to graphics the same way a desktop card would (or at least 8)? Both these solutions use 4.

 

3) I'd be interested to see some more testing to see performance differences between CPU-based PCIe lanes of x16, x8 and x4 compared against shared PCH-based lanes of whatever number are available. I know that the x8 versus x16 PCIe lane test has been done before, but in light of the fact that the PCH bottleneck is a factor, I'd be interested to see all of those numbers side by side. I think this could be done with a desktop motherboard and the same video card.

 

 

[Drak3]

1 minute ago, Cazzzz said:

If the goal is truly to get desktop gaming performance on a laptop, wouldn't the goal also be to dedicate 16 lanes to graphics the same way a desktop card would (or at least 8)? Both these solutions use 4.

Asides for the top end, The Titan XP, Titan Xp, and 1080Ti, 3.0 x4 isn't a bottleneck, and for those top end cards, its barely a bottleneck.

 

2 minutes ago, Cazzzz said:

The performance difference between the two would be equalized if the test was able to be performed on a laptop with PCIe lanes direct to the CPU, or, perhaps even just lanes that could be dedicated to the device instead of shared.

No, it wouldn't. The performance loss on the Core is due to the need to allocate for the extra IO on the core. DMI 3.0 runs at the same speed as PCIe 3.0, 8GT/s, and has 4 lanes. The only impact that DMI would cause is slightly longer load times in games, which wouldn't show up in benchmarks like what LTT used, and would barely be noticeable. The only thing the Southbridge would be doing is accessing storage occasionally. If it were a desktop system, the likelihood of the DMI causing a bottleneck would go up, but only because desktops typically have more IO that need access through the Southbridge and DMI.

 

On the Razer system, there would be little using the DMI, as most of what the southbridge is responsible for that requires the DMI isn't there, or uses too little of the available bandwidth to result in a noticeable effect. On the other hand, the Core needs to have PCIe resources allocated for the NIC and USB controller, which takes enough bandwidth at all times to produce some bottlenecking.

[dragoon20005]

MSI solution on the Stealth Laptop was directly connect the GPU to the full x16 lanes to the CPU but using the big bulky dock which Linus previously complain of noise issues

 

 

not the most pretty but you get full x16 lanes vs x4 on these two docks which JaysTwoCent shown it bottlenecks high end GPU as well as mid range cards like the GTX1060

 

eGPU makes sense when the onboard GPU of the laptop is only Intel offering like the MSI Stealth

 

but the price of the laptop + dock - the GPU puts me and a lot of my friends off

 

 

[Drak3]

6 minutes ago, dragoon20005 said:

MSI solution on the Stealth Laptop was directly connect the GPU to the full x16 lanes to the CPU but using the big bulky dock which Linus previously complain of noise issues

MSI released a slim version of the dock a while ago, about the size of a 'gamer' laptop cooling pad. It also limits the GPU to x8 lanes, the other x8 were split between storage, additional IO, and a sound solution.

 

6 minutes ago, dragoon20005 said:

x4 on these two docks which JaysTwoCent shown it bottlenecks high end GPU as well as mid range cards like the GTX1060

JayzTwoCents shown that some eGPU docks bottleneck some cards. Running a straight PCIe 3.0 x4 won't actually be a bottleneck for the heaviest gaming until you're using the Titan XP, Titan Xp, or the 1080Ti, and even then, it's only a slight bottleneck.

[dragoon20005]

Just now, Drak3 said:

MSI released a slim version of the dock a while ago, about the size of a 'gamer' laptop cooling pad. It also limits the GPU to x8 lanes, the other x8 were split between storage, additional IO, and a sound solution.

 

JayzTwoCents shown that some eGPU docks bottleneck some cards. Running a straight PCIe 3.0 x4 won't actually be a bottleneck for the heaviest gaming until you're using the Titan XP, Titan Xp, or the 1080Ti, and even then, it's only a slight bottleneck.

yes i saw the 2nd gen version and hopefully Linus or JaysTwoCent could review it and see how it performs

 

you need to see the video again, the eGPU dock was the Razer Core and with a GTX 1060 installed.

 

it performed way worse than the internal GTX1060

 

putting a GTX1080 make matter worse and it only gain like 10-20fps across the games he tested

[Cazzzz]

@Drak3, I get what you're saying-- the bottleneck isn't the DMI, but instead the fact that the PCH-based lanes are being shared between the GPU (through Thunderbolt) as well as the NIC, storage and USB. This is bogging down the GPU whenever there are other requests from those devices. I guess what I don't understand is why having even four CPU-based lanes dedicated to a GPU-specific Thunderbolt port (if this is even possible) and the rest of the I/O shared between the other PCH-based lanes wouldn't solve the issue by isolating it from the other busy devices.

 

Point taken, though. I was assuming a bottleneck from DMI also.

[Drak3]

Just now, dragoon20005 said:

you need to see the video again, the eGPU dock was the Razer Core and with a GTX 1060 installed.

 

it performed way worse than the internal GTX1060

 

putting a GTX1080 make matter worse and it only gain like 10-20fps across the games he tested

The Razer Core is prone to multiple bottlenecks. Thunderbolt has higher overhead (although that won't cause nearly that much issue), and the Core splits its x4 connection between the GPU, the NIC, and the USB controller.

 

Running a GTX 1080 on a desktop, with only a 3.0 x4 connection, and it performs the same as if it on a 3.0 x8 connection.

[dragoon20005]

Just now, Drak3 said:

The Razer Core is prone to multiple bottlenecks. Thunderbolt has higher overhead (although that won't cause nearly that much issue), and the Core splits its x4 connection between the GPU, the NIC, and the USB controller.

 

Running a GTX 1080 on a desktop, with only a 3.0 x4 connection, and it performs the same as if it on a 3.0 x8 connection.

PCI-E from the chipset vs the CPU lanes are vastly difference

 

like you said , there will be overheads and latency

 

older GPU will have no issues with X4 PCI-E 3.0, but it will affect more recent GPUs.

 

PCI-E x4 3.0 is like PCI-E x8 2.0 it did affect it greatly especially SLi and CF setup

[Drak3]

1 minute ago, dragoon20005 said:

older GPU will have no issues with X4 PCI-E 3.0, but it will affect more recent GPUs.

In heavy gaming, only the Titan XP, Titan Xp, and 1080Ti, and it's a small bottleneck.

 

2 minutes ago, dragoon20005 said:

PCI-E x4 3.0 is like PCI-E x8 2.0 it did affect it greatly especially SLi and CF setup

No, it doesn't. Until recently, PCIe 2.0 x8 was more than enough to drive a GPU in gaming, and it's still enough to drive all but the most extreme GPUs.

 

5 minutes ago, dragoon20005 said:

PCI-E from the chipset vs the CPU lanes are vastly difference

Unless you introduce a multitude of PLX chips to introduce more partial lanes, they really aren't. The big difference between the two is that the southbridge acts as a middle man, which does add some latency, but not enough to cause the bottlenecking we see with the Razer Core.

[NumLock21]

The Alienware amplifier is now on sale at Amazon. $170

https://www.amazon.com/Alienware-9R7XN-Graphics-Amplifier/dp/B00PCJXN0I

[Fma96580]

couldn't this solve the issue?

 

https://arstechnica.co.uk/gadgets/2017/05/intel-thunderbolt-3-royalty-free/

 

(read the article not just the url)

[ribbitd]

LTT crew I know you guys have a lot of video and idea line ups i get it but can you take a look at some thing that is comparable to this solution for better graphics to your laptop. The idea and product has been out for a long time but seemingly a lot of people like this idea and its very cost effective compared to this solution when its only proprietary. Here i will provide links to ebay and alternate sites if you want to take a look. All require a power source like a laptop charger and or a power supply some are different but all in all spending about $100 USD or less might be a better alternative for extending the life of you laptop because of the video card upgrades and prices of laptops. Make sure to pick the right cord because there are 3 kinds (NGFF,Mini PCI E, Express Card) Also as a added benefit you can use old laptops to bring gaming life back to them with this and other alternatives because of the device. I know you guys are always busy so I hope you guys take a look at this idea. http://www.ebay.com/itm/V8-0-EXP-GDC-Beast-Laptop-External-Independent-Video-Card-Dock-To-PCI-E-16X-/112301900457?var=&hash=item1a25b702a9:m:ma1v5_26pJRfY3U1NzhpPhQ https://www.banggood.com/Mini-PCI-E-Version-V8_0-EXP-GDC-Beast-Laptop-External-Independent-Video-Card-Dock-p-1011222.html?rmmds=search http://www.gearbest.com/laptop-accessories/pp_229101.html https://www.aliexpress.com/store/product/EXP-GDC-Laptop-External-Independent-Video-Card-Dock-with-PCI-E-Interface-Black/1048722_32456172794.html https://www.amazon.com/Laptop-External-PCI-Graphics-Card/dp/B00Q4VMLF6

[dragoon20005]

8 hours ago, Drak3 said:

In heavy gaming, only the Titan XP, Titan Xp, and 1080Ti, and it's a small bottleneck.

 

No, it doesn't. Until recently, PCIe 2.0 x8 was more than enough to drive a GPU in gaming, and it's still enough to drive all but the most extreme GPUs.

 

Unless you introduce a multitude of PLX chips to introduce more partial lanes, they really aren't. The big difference between the two is that the southbridge acts as a middle man, which does add some latency, but not enough to cause the bottlenecking we see with the Razer Core.

i went into more digging and this recent review shows something

 

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080_PCI_Express_Scaling/24.html

 

they did test PCI-E 3.0 x4 from CPU and chipset which is similar to the Razer Core

 

with a GTX1080 being tested here

 

you will get up to 8% performance hit vs the PCI-E 3.0 x8 and x16 modes

[King_of_Oz]

I haven't seen the video, but I'm putting my chips on the razor core, because you can use it with any device as long as it has thunderbolt.

[koobear]

On 5/24/2017 at 2:38 PM, Coffey said:

He should look at old way of hooking up an external graphics card to a laptop, with express port and mini pcie. It would be intresting to see how much a pcie 2.0 1x would bottleneck a 1080ti

Yo, I have (very old) benchmarks for this!

 

http://www.3dmark.com/3dm/4629516?

 

Proof (this is an older picture--I typically use an external monitor since using the laptop's display results in a drastic drop in performance): 

 

[silentdoh]

I currently use an older laptop (Ivy Bridge Clevo Rebrand) hooked up to an MPCI-E EGPU as my 'office' "gaming" machine. (It had a 660M that died on it, so now it gets a GTX 760). I'd love to see the cheaper options like these tested against the "High End" ones like the Razer Core. There unfortunately has just not been enough research done into how these configurations perform, despite how 'cool' they seem, especially at first.