AMD wants to standardize the external GPU
arstechnica.com
arstechnica.com
Sounds like yet another flash in the pan to me. VR could be a killer feature, of course, and price would determine the success, but unless they use a standard connector and don't require all manufacturers to add in yet another display/bus to the thin laptops, I don't see this going anywhere.
They are trying so hard to compete with phones they aren't doing anything to improve performance. Back in the day you could buy a new PC every 2 years and get a big performance boost, today that is not so. Get a cheap netbook and put in an SSD and you'll get something that feels more responsive than an i7 machine saddled with a mechanical drive.
No wonder people are not buying PCs. I remember when Tom's hardware used to be interesting but for a long time they have been colonized by this "power efficiency" thing which is really bogus because you can work really hard to power optimize your machine and then some stupid flash ad makes your CPU spin at 100% and it is all for nought. People buy things because of product attributes they can feel, not because of attributes which require careful measurement under controlled conditions
I prefer quiet.
A few years ago I built a pretty awesome-looking, water-cooling machine, and just the noise of the water pump was louder than anything I had before. The fans I needed to the radiator also were just as loud as my previous GPU fan.
If Microsoft, at some point, said that you couldn't put a windows 10 or 11 sticker on a machine that has a mechanical C: drive, everybody would be saying that this version of Windows is awesomely faster than anything they've seen before.
Study after study have shown that screen real estate and system performance improve our work output, let us use more of our working memory for task-critical state, etc.
If you need portability, then sure, laptops are necessary trade-off so that you can get any work done at all. However, that's all they are: a trade-off.
It's a shame if we're only optimizing for use cases that are ultimately regressive.
The only thing that comes close right now is a Thinkpad P70. Can get it with 64GB ECC RAM, Xeon E3-1505m, 4K 17" screen, 2 SSDs + 1 HDD, and a Quadro M5000M. The only problem is that it can't run OSX in some supported fashion, and the battery life likely isn't great. Sad that Apple has been falling short in the professional market for a while now and they expect everyone to be happy with thinner & lighter.
If your project has tendrils into the local system, that's the real problem to fix, because it's going to introduce pointless inefficiency at every level: new developer setup, production deployment, support for enterprise-installation, and ... inability to use the ideal hardware. available.
Performance translates directly into working efficiency. There's no such thing as "fast enough", there's only "too expensive".
Everyone can do eGPUs now via USB-C that uses TB3 via alt mod protocol. Razor was the first one to support this a few months ago with their Blade ultrabook. Sadly, they are behind with releasing the eGPU case but their case can by used on any laptop that has TB3 and as long as the manufacturer has not disabled some stuff for TB3 via the bios.
If Asus have it ready, then where is it?
AMD have been fairly good at providing open source software and freely licensed standards, certainly much better than Nvidia. Their adaptive frame rate technology is royalty-free, they have an MIT-licensed middleware platform for games and GPGPU programming, and they are the main backer of Vulkan.
Thunderbolt 3 external GPU enclosures are still rare, but products such as the Razer Core are on the way and should work with any laptop that supports Thunderbolt 3, of which there are a growing number.
There should be nothing stopping AMD from supporting eGPU enclosures via Thunderbolt 3. I'm honestly perplexed why they think we need yet another competing standard.
So possibly a long play on VR taking off, wanting to provide support to Macs and stopping dominance by Nvidia.
There has been already a standard for about a decade and it's called PCI Express :) I had an external GPU setup about 8 years ago using the ExpressCard slot which worked at 1.6Gbit/s at the time (about 1/4th of what PCIe(1) X16 offered, but still more than enough for even upper mid-range GPU's of the day).
Since ThunberBolt came out you could run external GPU's without any issues or even performance bottlenecks.
The PCIe(3.0) x4 on TB3 these days offers more bandwidth than PCIe(2.0) x16 does and it's more than enough to saturate any high end GPU on the market today or in the foreseeable future.
Intel with IRIS and other IGP's and MSFT through DX12 and WDDM improvements support pretty much GPU hotswap which make this a plug and play setup (historically if you wanted to use an external GPU you had to do a full system reboot to initiate it).
But this the main problem for AMD atm, Thunderbolt and pretty much the entire eco-system for external GPU's today is based on Intel techonlogy, AMD has yet to implement TB or managed to create a competitor to it and it has tried and failed twice already - LightingBolt which was later re-branded and redesigned as DockPort.
So if AMD want to support external GPU's all they need to do is bite the bullet and implement TB at the cost of their pride and some royalty fees, but as it seems now all they want to do is get some PR and then abandon the project complaining about market monopoly like they've done quite a few times in the past.
Shame what have became of AMD, I loved them in the Athlon x64 days, too bad they tried to play dirty early on and then got fixated on the same tech they had without capitalizing on their market success to push it forward, they thought they can play the same game as Intel and they failed, poorly.
AMD's playbook is leading them to a dead end, at this point i hope they'll spin off their GPU department into a separate company that might get back on the right path that ATI has been going towards (which oddly enough also hit a dead end because they were too "cool" to play ball after their meteoric rise during the R300 vs GeforeFX days).
But hey Ars seems to like rooting for the "underdog" even when it doesn't deserve it, but at least the amount of fanboyism in comment section made me laugh a bit.
It's not just a matter of license fees. AFAICT, Intel won't even let anybody else create chips for cables or peripherals, let alone host chips.
All they need to do is have a TB controller and have a free PCIe interconnect, from looking at say the Promontory chipset (AMD's answer to Intel's PCH) you can just drop it in without sweating.
Sounds like Intel is guilty of "illegal tying". https://en.wikipedia.org/wiki/Tying_(commerce)
On top of that, the DMCA has interoperability clauses which may need to be updated https://en.wikipedia.org/wiki/Reverse_engineering
> On the other hand, AMD has tried to standardise the external GPU before—and failed miserably. In 2008 AMD launched its XGH external graphics standard, which essentially just took the pins from PCIe slot and passed them through to an external connector (a solution copied by Alienware for its Graphics Amplifier). This was a smart idea at the time, particularly as USB 2.0 and ExpressCard didn't offer anywhere near the amount of bandwidth required for a high-end GPU. Unfortunately, very few laptop manufacturers used XGH, with Fujitsu Siemens' Graphics Booster being one of the rare commercial examples.