Why didn't Alienware hang the Graphics Amplifier off the CPU's PCIe lanes?

In Razer's case I can understand them wanting to stick to standard TB3, but given that the Graphics Amplifier is 100% proprietary, why didn't Alienware split the CPU's PCIe lanes into a x8/x8 configuration, use one x8 for the internal GPU, and use the other x8 for the Graphics Amplifier?

Heck, they could even add an option in the BIOS to disable the Graphics Amplifier port and dedicate all 16 PCIe lanes to the internal GPU if they don't want those without the Graphics Amplifier to feel like they're losing out. Every desktop mobo that supports SLI/CF has a similar option for single GPU setups, so it's definitely technically possible.

Hanging the Graphics Amplifier off of the chipset's 4 PCIe lanes is just such a weird choice in comparison, and that's not even taking latency into account...

 

First of all it hardly has any effect as in turning it into a bottleneck. Recent cards shown hardly any performance decrease when running into 4x mode. Only a few frames. The performance loss is mostly only present when using the loop back function to the internal laptop screen.

Second, there are simply not enough lanes available in laptop chipsets to begin with. The Alienware laptops also have 2 pci express nvme slots, TB3 port that need their lanes.

So no its technically not possible unless Intel redesigns their chipsets.

 

Err, the whole point of using the CPU's PCIe lanes is to bypass the chipset (and the limitations you mentioned) entirely.

After all, the CPU and chipset each have their own PCIe controller, so if something is directly attached to the CPU's PCIe controller, it's not subject to DMI 3.0's bandwidth limitations.

 

Again there arent enough lanes from the cpu AND chipset. Pretty much all laptops start out with pci express 8x on the GPU to begin with. Which is tied directly to the CPU.

As you can see there are from the CPU only 16 available https://ark.intel.com/products/88967/Intel-Core-i7-6700HQ-Processor-6M-Cache-up-to-3_50-GHz

GPU takes 8 lanes, 2x 4x nvme slots. 16 are taken already for the CPU. The chipset handles 4x for the AGA (which is still sufficient for any GPU out there) and 4x for TB3 and cannot go beyond 4x lanes as stated by Intel here https://ark.intel.com/products/90584/Mobile-Intel-HM170-Chipset

THere is simply not more available.

 

Because they want you to buy next Dellienware. If you have GA you will be stuck with dellienwares only.

 

Read the above. this is like spreading fake news.

Frank Azor already stated they could update the bandwidth simply with a cable when higher bandwidth options become available. > http://www.pcworld.com/article/3083...cs-card-tech-beats-thunderbolt-3-options.html

 

Think the amp still runs on dmi port, latency problem...


@D2 Ultima

 

AGA passes through the DMI and is thus limited to a PCI/e 3.0 x4 connection speed. Thunderbolt 3 is too; it is incapable of achieving its 40Gb/s... only 32Gb/s of it is usable. The technology is simply "rated" at this. No idea if AM4 can bypass this.

The problem with TB3 eGPUs I believe is the conversion to PCI/e from TB3. You can ask @bloodhawk for more information. A diy solution with M.2 NGFF will achieve the same perf as the AGA and beat TB3 as well, but afaik nVidia being anti-consumer locked higher end cards from using those with such diy solutions.

Sent from my OnePlus 1 using a coconut

 

Neah they can still be used, just need a bit if BIOS level tweaking.

M.2 = AGA (i think) > TB3.

 

While that may be the case for cards today, how about 3-5 years in the future when you'd like to give your laptop an upgrade with a new graphics card? Are you confident that a PCIe 3.0 x4 connection will not bottleneck cards designed for PCIe 4.0 x16?

Well, if Alienware is hanging SSDs off of the CPU on gaming laptop, then I suppose planned obsolescence is the only rational explanation...

 

Just the reference design of what Intel offers. But I am not sure if all PCI express ssd ports are in use directly on the CPU. Fact is that 16 lanes is not enough for current day configurations. TB3, SSD slots etc all want those lanes.

But regarding the bandwidth. I am not that worried about it. It only will start to become a bottleneck when GPU's need more CPU cycles but that trend isnt happening right now. DX12 & Vulkan are all made with less CPU cycles in mind to eliminate overhead.