eGPUs still bottlenecked by thunderbolt?

Hey guys,
I’ve been interested in buying an eGPU soultion for a while now but last I checked 1-2 years ago eGPUs were severely bottlenecked by thunderbolt bandwidth capacity and were effectively functioning at 60% of max potential and one had to crack open their notebook and connect the GPU to the internal PCMCIA slot if they wanted to obtain the full bandwidth. I am just curious if anything has changed now with thunderbold 3 etc? Does eGPU make more sense now? Thanks.

yeah, 3060ti or 6700 would be the max I’d put in a egpu probably.

2 Likes

Since Thunderbolt has not gotten any faster the answer is that is is indeed still bottlenecked.

Arent PCMCIA slots those credit card-like adapters that existed during the 90s?

The servere bottleneck reports were mostly from people trying to use the eGPU to render on the main laptop screen.

I have an RTX 3060 single fan from Asus. Bought a NUC and gpu didn’t fit because it was too large. While looking for a new case I’ve used TB from a Staturn Mantiz Pro. Got a new case and moved to the PCI-E slot and difference was barely noticeable. I was expecting a 10-20% performance improvement from tb3 to pci-e but barely got 3-5%.

So use an external monitor plugged into the external GPU. Or don’t use e-GPU at all. Apple make it popular and then killed it. This is coming from someone who has buyers remorse from getting an enclosure.

Sorry I got that wrong. It’s mPCIe or NGFF/M.2.

Apparently those are better and more accessible than Thunderbolt. I don’t have an opinion on GPUs connected that way tho. All I’ve seen are videos with guts opened and extended cables connected to raw gpus supported by some skeleton enclosure at best.

In any case most of the TB bottleneck complains probably came from people who use MacOS and use the eGPU to render things in the main laptop screen. Communicating both ways through a single cable will probably bottleneck.

But I still wound’t recommend Thunderbolt connected GPUs. Its a niche and you will surely run into bugs and weird quirks. In my case the external gpu only became available after logging into the OS.

My dream has become a reality! Asus’ new notebook ROG Flow X series now has an eGPU port which is a direct PCIe 3.0 x8 connectivity (63GBps total gpu bandwidth). This might be a desktop killer for me. So happy. Supports up to Radeon RX6850M GPU or nvidia 3080.

64Gbit/s is 50% more than thunderbolt can offer, but still way less (75% less to be precise) of PCIe 4.0 x16.

I heard that PCIe 4.0 x8 only is a couple of % loss in performance on high-end cards…but PCie 3.0 x8 still is like putting a 3080 in a 6 year old board. And comes with a proprietary connector with all its troubles .

1 Like

Well I think it has to come with a proprietary connector since there are no official direct PCIe port implementations (this is still a mystery to me). I know there are issues with this implementation. For starter the eGPU enclosure doesn’t need to be so small, hot and loud - after all isn’t the whole point of this model that this enclosure isn’t portable? If you needed it to be portable then you’d just get a Zephyrus G14 or something like that with a built-in GPU and you don’t care about battery life or having an ultrabook. No one in the eGPU community is concerned with GPU portability, they just want to be able to use their notebooks as notebooks when they don’t need a GPU and they don’t want to have both a desktop and a notebook to achieve this. I am sure plenty of people will be still happy if the eGPU brick is 2x the size.
As for your second point, I am not sure why they decided to go with x3, maybe they will upgrade to x4 soon.