I would argue construction is probably the least digitised. The Economist had an article recently that construction is the only sector in manufacturing in the UK that declined in productivity over the past two decades. I'm not really in the topic but the construction I see in the city (London) looks not like they adapted any processes developed over the past 1-2 decades. Maybe that's where the next revolution happens.
After reading this, it is clear how much can still be done. Holland has in the past 3 - 4 years consistently been innovating and producing much better yield in all sort of agriculture products. The first two years most US companies has been saying those prices aren't sustainable, and they will come down soon, it is now clear those better then expected yields are not mere coincidence.
This makes returns on investment into new software, trainings and lost temporary productivity diminishing. Without government changing standards on engineering output, the market for flexible projects is not that large (which is a bit of a chicken and an egg problem).
It also can not be ignored that until the Switch came out, they were making the SoCs for every current generation console.
While I agree, my only criticism would be that the "old" data center market is effectively consolidating to commoditized hardware. In other words - we have less idle CPUs in data centers and the diminishing returns of Moore's Law is hitting. So while the new wave is positive, the underlying revenue stream may not be consistently holding.
Cisco is still one of the more remarkable stock stories I've seen in the last two decades. Something like ten straight quarters of accelerating revenue growth, quarter over quarter, leading up to 68% growth for the third quarter of 2000. Three quarters later their quarter over quarter growth had evaporated down toward zero.
That same shock deceleration basically killed Lucent, Nortel and Sun Microsystems. The three were worth around $600 billion together during the bubble. Sun of course struggled on as their business melted, but they were a shadow of their former self by 2009. Oracle nearly got caught in the same trap (also having a crazy ~$200 billion valuation during the bubble), instead they went on a competitor acquisition spree and created as much of a moat as they could. 17 years later, their stock is finally back to where it was during the bubble (in nominal terms).
Navi, the next gen modular GPU are going back to TSMC 7nm. But that is professional offering appeared to by the end of 2018, and consumer offering in 2019.
i.e There is a capacity limitation of how well AMD can do within the next 24 months. That is excluding AMD's APU launching soon which I think is going to be a big hit.
P.S - That is GF's Fab 8 20% capacity increase included. It is clear AMD is not the only one using Fab8, I wonder who are these other customer using 14nm in GF.
Does it?
I'm genuinely interested in this as Vega appears to have high power usage requirements and less than spectacular mining results. I mean they're fine, but not the multiple of current cards that was rumoured before release. I owned a Vega 64 briefly, but sold it on as I didn't really want the noise or heat. And it still doesn't look like any real partner cards have been released, the lineup is all reference cards.
Have you seen any sales figures that would show the mining popularity of vega?
Similar for the steam statistics on linux, AMD just doesn't compete on anything that pushes the GPU on linux. Even things like vlc, mplayer, plex, and friends often get the AMD GPUs in a wonky mode.
The problems I most often see is new windows fail to map, I just get a black window. Presumably something to do with memory allocations of the backing store. It's also frequently munges the console so I can't get to text mode to unwedge the AMD GPU.
So generally if you what a stable linux box I'd recommend nvidia or intel GPUs. Things just work, you can easily have month long uptimes, and things don't freak just because you are playing 1080P video full screen. No pixel droppings, no failing to map, console always works, etc.
Not sure what you looked at, but I see quite a rapid progress of bug fixes: https://www.gamingonlinux.com/wiki/Mesa_Broken
Never had problems with vlc or mpv on AMD. It works just fine with vaapi. If anything munges the console, it's Nvidia blob which doesn't even support framebuffer.
Also, it sounds like you aren't using Mesa, but use AMD closed stack instead. Don't do that. In short, if you want a stable experience - AMD is way ahead of Nvidia.
This is true, but needs a caveat: a lot of work has gone into improving Mesa and the rest of the Linux graphics stack recently and it's now pretty much caught up (performance-wise) to the proprietary AMD drivers. However, that's only happened in the past 6 months or so, so if you're not using a rolling-release distro like Arch or Gentoo you might still be better off with the proprietary drivers until your distro catches up.
Vega is loud and power-hungry, and expensive The last generation of AMD card was snapped up by miners, I've not heard of it happening with Vega.
I am genuinely interested in whether miners are really buying these. Have you any reason to think they are?
I couldn't find much when I looked, beyond speculation about how amazing they might be (but turned out not to be). AFAICT for ethereum you're better off with nVidia 1070s which can cost about 75% as much, have about 75% of the hash rate, and use under 50% of the power.
I have recently switched from an NVIDIA Quadro card to an AMD FirePro card. I use GNOME/Wayland and the difference is quite big, with NVIDIA on nouveau, there was regularly flickering (at random moments), artifacts and other problems. The FirePro with the amdgpu driver on the other hand works completely flawless on my machine.
[1] https://www.phoronix.com/scan.php?page=news_item&px=XDC2016-...
nouveau was always pretty awful, but the nvidia binary drivers always Just Worked, across many models for many years, whereas with AMD it's always been a crapshoot whether a given card would work properly or not.
The binary drivers always kept up with the important things (indeed they were well ahead of AMD in terms of doing accelerated video decoding on linux via Xvmc and later vaapi, their xinerama integration was always better...). Sure, they don't support Wayland, but that's a solution in search of a problem; if and when there's actual value to be had by using it I have confidence that nvidia will support it.
X is also insecure from the ground up, that alone is enough incentive to move away from it.
Not
- if you don't want any GUI application to be able to read keystrokes/mouse events/screen grabs of any other GUI application,
- if you want per-display scaling factors,
- if you want a tear-free desktop experience.
If AMD is worse, they're failing to clear a pretty low bar.
The one chink recently was the new display code, which was a requirement for Vega (but not for earlier generations), and which was a genuine open sourcing of an existing, Windows-focused code base. That process took longer than one might hope for, but we're clearly over the hump now with it going upstream.
AMD had slideshows describing the feature on the new Vegas, but decided to disable it before release. This alone stopped me from building a threadripper + vega machine. Now instead I'm waiting for Intel to add more cores, then I will use their integrated graphics between multiple VMs.
It depends what you value. I've consistently bought Nvidia hardware because of the proprietary drivers.
But more to the point - sure, Linux gamers are quite excited about all the work that's going now into the open graphics stack.
I just checked the last steam hardware survey, linux is at 0.60%.
The linux steam hardware has pretty much been a flop. Ubuntu is ditching unity.
I sit in front of linux desktops at home and work. I do occasionally game some under linux. I have a linux phone (which has a linux kernel). But I fully realize it's a niche and feel lucky when any game comes to linux. I don't really see any reason for any real optimism.
Further I would argue it is irrelevant to the initial argument.
What has Canonical to do with it? Neither "Linux", "Linux Desktop", "Linux Gaming" or "AMD graphics under Linux" are in any way tied solely to Canonical or what Canonical does. Again, am I missing something?
Fair point.
> What has Canonical to do with it? Neither "Linux", "Linux Desktop", "Linux Gaming" or "AMD graphics under Linux" are in any way tied solely to Canonical or what Canonical does. Again, am I missing something?
Ubuntu is the only official supported distribution by Steam and GOG (they also support Linux Mint, but it's based on Ubuntu).
My point is, those who work on Mesa (OpenGL / Vulkan) and Linux graphics stack have way more direct impact of Linux gaming (they are fixing bugs that affect games, improve performance, add new functionality and so on).
You didn't pay attention either. The growth was well described here: http://boilingsteam.com/linux-gaming-in-2016-the-good-the-ba...
And that was last year, since then it only improved. And please, stop using Steam hardware survey for any Linux market estimation. It's useless.
That actually says (if you follow the links to try to find the actual claims): The Linux Market share on Steam is about 1%, Mac is about 4%. They are relatively stable around that. But obviously more people join Steam every day.[1]
So... about 1%, using Steam to estimate (not steam hardware though).
[1] https://boilingsteam.com/our-fifth-podcast-with-feral-intera...
There is no evidence presented anywhere which disproves that. Your claim was that it showed evidence of growth, but if you read it then it turns out it is using the steam survey, and shows no growth in percentage terms at all.
And there is actual sales data that comes from developers. It's very different from those survey numbers.
Also, since there is no info on methodology of that survey, you can't know even what it means. I've heard from many Linux users, that they never got one while using Steam on Linux, while they got it while using it on Windows for example. It even never comes up in Valve's own SteamOS, so it's clearly not something Valve put a lot of thought in.
So, I'll stand by what I said. Data from that survey is useless as is and should not be applied for any market evaluation.
Show me a quote. There is a larger number of titles being published on Linux, but there is nothing about sales or usage.
And there is actual sales data that comes from developers. It's very different from those survey numbers.
In that article or elsewhere? Can you point to it please?
There were several articles with numbers from developers on GOL: https://www.gamingonlinux.com
Search for sales statistics or something of that sort there.
The OP's comment "The several dozen people playing AAA games on linux must be thrilled." seems pretty realistic.
Increased number of games is simply because it is close to zero-effort for most game engines to press the button and deploy to Linux Steam. That doesn't mean anyone is actually playing them.
I had a quick look at the GOL site and I didn't see anything obvious claiming growth. Ironically (given the topic of this HN story) I did see this:
Get ready to become a neural detective as 'Observer' is now on Linux, AMD not supported.... I spoke with Aspyr Media, who confirmed to me the team has "currently no plans to support AMD at this time for Observer".[1]
Yet, clearly gaming on Linux will take off any day now... (And this is from someone who runs a Linux desktop computer)
[1] https://www.gamingonlinux.com/articles/get-ready-to-become-a...
It's far from zero effort. Besides, engines are making Linux support easier, is also driven by demand. But hey, legacy execs would rather talk about how Linux gamers don't buy games, instead of actually making games for Linux.
I find it odd that AMD can get the drivers right for gaming but that's about it.
If you want good performance using open-source drivers, AMD is the way to go with Linux [2].
[1]: A old comparison (Jan/2017): https://www.phoronix.com/scan.php?page=article&item=mesa171-.... Nowadays the performance is even better.
[2]: Open-source drivers is important not only because FSF and all, but because the open-source drivers follows the advancements in Linux graphics stack much closer. Things like KMS and Wayland works much better with open-source drivers than NVIDIA proprietary drivers. Even things simple like Xrandr is ramdomly broken in NVIDIA drivers.
NVIDIA binary drivers really only win in performance. Even the features in NVIDIA drivers tends to be more bugged: i.e. I use compton as my composite manager. It needs a very old OpenGL version (like 1.1 or 2.0 capability), and NVIDIA drivers still mess up and I need to activates some workarounds in compton to get usable desktop. MESA (used by open-source drivers) has a much better OpenGL implementation, and I can basically get a glitch free desktop without workarounds. I also remember getting random glitches in Chrome and Gnome Shell, just to cite more examples.
So it is not an artificial comparison. Want performance and CUDA? Yeah, go to NVIDIA. Want just a stable, modern desktop (Wayland and composite glitch-free)? Open-source drivers is the way to go.
https://www.phoronix.com/scan.php?page=article&item=radeonsi...
The AMD open source driver is just about on-par with their closed source Windows driver. One of the benefits of Open Source is also being utilized now. Game companies like Feral Interactive, who develop many Linux ports, have been improving the driver's performance for their games.
But I think you are missing the point here and that is: if you want to buy a GPU now for a Linux machine and you don't want to use a big proprietary blob, AMD is by far the best choice, because the amdgpu driver outperforms the nouveau driver.
No doubt, I said so explicitly elsewhere in this thread. I.e. that on Linux Nvidia will lose its current dominance.
Disclaimer: I own both NVIDIA and AMD.