Thunderbolting your video card | Hacker News Reader

Thunderbolting your video card

blog.codinghorror.com

120 pointsplg33 comments

If anyone is interested in eGPU for TB2, I recently wrote an article on my hassle- and hack-free Macbook build. Not nearly as powerful as this one, but still 4x as fast as my integrated 750M!

https://news.ycombinator.com/item?id=13303912

I thought I had read about a TB2 build here in the past. Thanks for the link! I've been considering a mid-level Mac mini to get back into the macOS game, and something like this would allow me to do away with my gaming rig (I don't do AAA games, but the ones I do play tend to require more than Intel graphics).

Caveat: the 1050 Ti is still an entry level card, but at least I can play modern games now. :)

When I upgrade again, I might opt for a molex-powered SFF card and hack up a Dell DA-2 adaptor like some people do.

The 1050 Ti would be more than enough for my needs. I currently run a GTX 960 and I have yet to find its bottleneck among the games I play, apart from Rust which is so poorly optimized it lags even on a 1080. It's my understanding that the 1050 Ti matches or beats the 960 on nearly every metric, and does it at 75 watts compared to the 960's 120 watts.

Caution for the OP considering using it with a Mac Mini: there are no nvidia OS X drivers for the 1050Ti. You're stuck with 9xx or earlier cards.

And TBH, if you want this solution to boost a desktop computer, you're better off just building a Hackintosh. It may make some sense for a laptop, but i think it's still cheaper to build a separate gaming computer.

For the record, the build I describe in my article is strictly for use with BootCamp! I wanted to avoid having to install any 3rd party software which AFAIK is only possible in Windows. (And besides, OSX gaming performance is pretty awful.)

> Caution for the OP considering using it with a Mac Mini: there are no nvidia OS X drivers for the 1050Ti. You're stuck with 9xx or earlier cards.

Yep I'm aware, and most of my games don't have native macOS ports. The ones that do would run fine on the Mac's Iris graphics. I would be dual-booting via BootCamp for heavy lifting.

> And TBH, if you want this solution to boost a desktop computer, you're better off just building a Hackintosh. It may make some sense for a laptop, but i think it's still cheaper to build a separate gaming computer.

I already have A Windows 10 gaming desktop, my goal would be to consolidate into macOS for workstation duties and Windows 10 for gaming, and free up space on my desk. I'm not keen on building a Hackintosh or turning my current gaming rig into one; I've been there done that and it's just not worth the hassle. I also would prefer to run macOS on a real Mac just for the support side of things.

I would then be able to give my 16 year old niece the "old" gaming rig (it's a Skylake based system I just built a few months ago) as a massive upgrade from her Core 2 Quad. She's getting into computer animation and 3D modeling and needs something better.

I don't know about you, but I used to dual boot and i still have the Windows installation on my Hackintosh (which is beefy enough to handle gaming. Hint: don't buy the gaming hardware and try to make OS X run on it, buy hardware that works well in a hackintosh and is good enough for games instead.). In time though, I simply stopped playing the games that required me to reboot and now I only get stuff that runs on OS X.

Of course, it's easier if you have a very strong dislike for the current AAA games. I stopped touching anything from EA/Ubisoft long before i stopped rebooting for gaming.

What keeps the price high on these enclosures? In the past the assumption had been that it's because Intel is charging a lot for the TB3 controllers, but unless I'm reading it wrong, the prices on those look to be very reasonable (sub $10) at this point: https://ark.intel.com/products/codename/56890/Alpine-Ridge.

Does anyone have a BOM estimate of this (or any of the competing) Thunderbolt 3 enclosures?

It's probably too niche of a market to see price drops.

There are GPU enclosures with a lower price, for example the AKiTio Node for 270 dollar[1]. Since the new Macbook Pro, new GPU enclosures have been popping up everywhere.

[1] https://www.bhphotovideo.com/c/product/1303819-REG/Akitio_AK...

Good to know but doesn't really answer the question. The BoM of just the signal adapter is what I really want to know. Is it $5 or $80? The price of a power supply is a known factor, and boxes are cheap.

Add the PSU, box and other parts + assembly all at low volume and factor in fixed cost such as design - I can see them staying above 200$ if they suddenly don't start selling a lot more.

But that's the question. How much is fixed costs? I'm not surprised at a high cost for a low volume part, but I want to know if it's going to eventually drop to $40 or $140.

Most of the price is probably caused by low volume. Once a major manufacturer start selling it, I expect the price would half if not more.

Expensive is putting it mildly. $500 is pretty much the budget for an entire desktop system minus GPU. So I could buy a laptop with TB3, then drop another $1000 on eGPU, or take that $1000 and by a whole nother gaming system.

Maybe the next step is putting the graphics processor in the display, and extending the processor's I/O bus out to the display with Thunderbolt/PCIe.

I don't see many usecases for that - besides the cable length problem mentioned in the article, it would have the effect of reducing choice for the consumer.

I could perhaps see displays including a very basic GPU and advertising it as a plug & play display, no need to check if your video card has enough ports, but I believe (citation needed) multiple display outputs have been standard for years now.

If that's the usecase you're after, I'd also suggest that you're better served by USB - there have been basic USB external GPUs available for years now that would do the trick.

This doesn't seem to be a popular idea but I think it makes a lot of sense. The issue with the idea is that you're now halfway there to an all-in-one, and I think manufacturers will just keep adding components until they get there because theres way more consumer demand for that.

I am fully on this thunderbolt 3 GPU bandwagon as well. I pre-ordered an Akitio node, and I'm quite bummed out that it hasn't been delivered yet. I'm very much looking forward to playing some games on my MacBook Pro with it!

Though if the Mantiz Venus starts arriving before my order is shipped, I may need to switch.

Ps: If anyone else is interested in doing an e-GPU with their mac, https://egpu.io/setup-guide-external-graphics-card-mac/ is a great place to start.

There is a missing product in the market right now. These high-end GPUs are all designed to fit into decade old computer cases. The pcb and chips are tiny and thin, you should be able to throw the actual guts of a GPU like a frisbee. GPU manufacturers should be designing external GPUs, instead of consumers having to purchase enclosures and power supplies for GPUs still designed to go into pc cases.

The current modularity is important, because the market thrives on it. TB3 enclosures are expensive; to duplicate that cost across every graphics card sold would be expensive.

Yes, it'd be smaller. But the increase in cost would not be worth it for the current niche market. Until TB3 penetration across regular consumers (as compared to the techie niche) goes up, there's no point in making slick consumer-market pre-builts.

Modularity is a fast horse. The current state of external GPUs is like phone blocks, and I'm waiting for the iPhone. A gpu enclosure with power costs about $300 USD. An Nvidia 1060 6gb gpu also costs about $300 USD. From my perspective I don't understand why I wouldn't just cough up a bit more money and build an entire pc. An Ncase M1 pc case is practically just as portable as any of these gpu enclosures.

It's a product people don't know they want. Currently, the fast horses are fast enough. When enough people are using "fast horses" where cars would be more appropriate and where volume pricing would favor cars, we'll see cars on the market.

The market just isn't ready for the car. And you mention wanting an iPhone. The problem is that you'll more than likely get a Newton MessagePad.

edit: also, I've come to the same conclusion wrt PC cases. I'm considering building a "portable desktop" that I can take with me, that mirrors video over a network connection. Then I can use a laptop to interface with it.

I'm not in the market for a Mac Pro, but I keep thinking that something like this is the future for high-end computers—a capable base system with the ability expand virtually infinitely by attaching another Thunderbolt 3 component.

What's wrong with eSATA/SATA/PCIe/USB/other existing technologies? I don't fault Intel for wanting to try their chops in the data transfer tech market but it's saturated enough as it is and real high end computers will make use of eSATA/SATA.

Thunderbolt - all of the versions - can be summarized as "External PCIe". It's just that ePCIe isn't nearly as marketable a name as Thunderbolt.

It's not. Thunderbolt is a very very closed standard by Intel. With standardized protocols, like regular PCIe, to start building a device, getting started is pretty much a case of ordering a few chips, reading the datasheets, and starting up in a design program. Thunderbolt, on the other hand, in order to even start looking at the datasheets, you have to apply to the developer program, and wait an unknown amount of time for intel to decide if you can even start to develop. Once your device is ready, you have to wait again for intel to approve it, and there is a chance they won't -- the first Thunderbolt GPUs were promised 5+ years ago, but Intel wouldn't approve for them being sold.

>Thunderbolt, on the other hand, in order to even start looking at the datasheets, you have to apply to the developer program, and wait an unknown amount of time for intel to decide if you can even start to develop.

That sounds excellent. Fewer amateur hour broken, incomplete and potentially computer damaging implementations!

>What's wrong with eSATA/SATA/PCIe/USB/other existing technologies?

That they are all multiple specifications to bother with (you mentioned 4+others), and that they don't offer the full capabilities, speed and power draw TB3 does. Not to mention the ugly cabling.

>and real high end computers will make use of eSATA/SATA.

There's nothing unreal about a pro wanting an expandable portable compute to function as high end.

I would like to "thunderbolt" my CPU, memory and GPU. I don't like to be stuck with the low power laptop CPU when sitting at my desk. Nor do I want to pay for and carry the workstation class laptop with lousy battery life.

My interest in this is adding a 10 GbE SPF NIC to my 2016 MBP running macOS (you can find drivers for 10 GbE, but tradically not yet for 25 GbE cards). I don't know if has been tried yet.

This stuff is the future, most people are going to be happy with a T470 or MBP13 hooked up to an Atikio Node + AMD RX480 8GB.

Reply on news.ycombinator.com