Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Intel, tilting at windmills again.
Always have to like articles that provide a morning chuckle.
https://www.marketwatch.com/story/intel-makes-it-a-three-way-race-with-amd-and-nvidia-on-graphics-chips-2018-06-12
Throwing money down the rabbit hole is not going to thrill investors.
https://www.marketwatch.com/story/intel-makes-it-a-three-way-race-with-amd-and-nvidia-on-graphics-chips-2018-06-12
Throwing money down the rabbit hole is not going to thrill investors.
Comments
Current game: Pillars of Eternity
Played: UO, AC, Eve, Fallen Earth, Aion, GW, GW2
Tried: WOW, Rift, SWTOR, ESO
Future: Camelot Unchained? Crowfall? Bless?
거북이는 목을 내밀 때 안 움직입니다
EQ1, EQ2, SWG, SWTOR, GW, GW2 CoH, CoV, FFXI, WoW, CO, War,TSW and a slew of free trials and beta tests
Furthermore, the commonly used APIs for GPU programming work well and make portability between architectures (whether the same vendor or different vendors) easy to do. (Well, except for CUDA, which has killing portability as its primary reason to still exist, but games generally don't use it.) If you want to make architecture-specific optimizations, that can add a little bit of work, but you don't really have to do that. Other than that, all that you really have to worry about is driver bugs.
EQ1, EQ2, SWG, SWTOR, GW, GW2 CoH, CoV, FFXI, WoW, CO, War,TSW and a slew of free trials and beta tests
Right now we are within spitting distance of iGPUs being able to handle 95%+ of what gamers and developers want. And that number is growing. Sure, you can say 4K/Raytracing/VR... and sure, those things need more power than an iGPU can probably deliver in the next 5 years, but they are also items that only the top 5% or so are really using right now too, and everything is getting more powerful or offloading processes smartly that even that is within reach sooner than later.
So entering now with a big push into discrete GPU - the purpose certainly isn't gaming. Gaming happens to get some nice marketing buzz going though, and Intel certainly loves their marketing, so I don't doubt such a product would have some gaming variant, probably with it's own branding and big marketing campaign. But it certainly isn't a gaming-first endeavor.
I think AI has a good bit to do with it. That market has exploded recently. I also think mining has a good deal to do with it - I don't think mining as a hardware driving force has a lot of legs left to it, but related tech (particularly block chain encryption) certainly has potential.
The other interesting part is the AMD factor. Intel and AMD were very hush hush about their partnership on an iGPU, and KabyG it does exist, it's been remarkably low key for what otherwise (at least on paper) should be a breakthrough product. Products that contain KabyG have been expensive, poorly energy optimized, and not all that attractive. I wouldn't necessarily point the finger at either Intel or AMD, I think more of it is optimization than any glaring hardware deficiency, but when you do mashups like this optimization is a tricky support topic to iron out.
And the other side of that coin - even though Intel had AMD signed up as a partner, they sniped out Raja Koduri, the former head of the RTG group and was heavily involved with Vega/Navi.
There have been rumors that Vega was primarily designed to Apple specifications (iMac Pro primarily, which may be true), and discrete cards just happened to come as a byproduct of that... and Navi is in a similar situation except the customer is Sony (for presumably the PS5). Maybe that's why Intel saw the need to go their own path - AMD wasn't catering to Intel's needs, and Intel wanted something to battle nVidia, not to go courting Apple or Microsoft with, and that isn't the direction that AMD seems to be focused on.
Intel is desperately trying to find new markets. The x86 market is not growing and hasn't been for a long while. They missed the boat entirely on mobile. They are ok on communications, but they aren't a market leader. A lot of their other attempts to branch out have fizzled (not because they didn't have good ideas, but because their support didn't seem to be behind any of them - they just kinda threw a lot of random ideas at the wall halfheartedly and acted surprised that none of it stuck), and now their biggest advantage - their fabrication, appears to be struggling.
Just speculation and random comments, but it's fun to randomly comment. If I were to bet at a company that will look radically different in 10 years time, my vote would entirely go to Intel. All it will take is Apple pulling the plug, and the rest of the PC manufacturers will fall in line like they always do - and the entire consumer PC marketplace will change almost overnight. Intel will be left holding just server business, and look an awful lot like IBM does today... In fact, I wouldn't be surprised if those two look at merging or one buying out the other if/when that does occur, as they would be natural bedfellows, or at least one would drive the other out of existence.
TL;DR I wouldn't expect a great gaming GPU out of Intel no matter what... no, especially (considering Raja is over there) because Intel is saying that one is come. Intel is a gasping fish out of water and is grasping for anything they can to try to remain relevant to shareholders.
PC nVidia (Kepler, Maxwell, Pascal)
PC AMD (GCN)
PC Intel (iGPU)
PS4 & PS4 Pro
XB1 & XB1X
Switch
iOS A8/A9/A10/A11
Android (and it's 10 bajillion different hardware configurations)
Developers are already supporting a lot more than just 2 hardware configurations, even if you just look at Red/Green. APIs and good drivers are what make that possible. It doesn't hurt that Fortnite is built on Unreal, and Unreal supports a wide array of hardware with it's API, making it a lot less painful for a developer to move pretty quickly across different platforms -- accomodating all the various input form factors becomes a bigger hurdle than graphics at that point.
Anyway... Nobody ever won by doing nothing... more or less.
This have been a good conversation
Intel is already building an ASIC for AI, aren't they? If you build an ASIC for an algorithm and run the same algorithm on a general purpose chip with the same resources dedicated to both, the ASIC will win, and usually by a lot. General purpose chips have their uses for algorithms not important enough to have built an ASIC for them (which includes nearly everything), but once you've got an ASIC, the general purpose chips aren't still needed for that particular algorithm.
I can't see cryptocurrencies as driving this, either. Other than building an ASIC, the best way to build chips with cryptocurrencies in mind is an FPGA with a suitable memory controller. Intel conveniently just bought Altera, so they're now one of the top to FPGA vendors.
It's possible that Intel wants the game console market. They can't get there without a respectable GPU, which is something that they don't have right now.
거북이는 목을 내밀 때 안 움직입니다
The problem is one of physics. It's cheaper to move data a millimeter within a chip than a thousand miles across the Internet. That's what they have to "fix" in order for game streaming to be viable as anything other than a very low end option.
If game streaming as a way to play games (as opposed to watching someone else play where the latency doesn't matter) at high graphical settings catches on, ISPs will revolt and find some way to sabotage it. The entire ISP business model completely collapses if your average customer is trying to use 10 TB/month of bandwidth and your network only has enough capacity for them for 100 GB/month per customer.
거북이는 목을 내밀 때 안 움직입니다
To give you some idea of the relative bandwidths involved:
Typical broadband Internet connection: 0.01 GB/sec
High end (currently gigabit) Internet connection: 0.125 GB/sec
Monitor cable for mainstream monitor (currently 1080p at 60 frames per second): 0.5 GB/sec
PCI Express bandwidth for relatively good video card: 10 GB/sec
Global memory bandwidth for video card: 300 GB/sec
Local memory bandwidth for video card: 10000 GB/sec
Register bandwidth for video card: 100000 GB/sec
All of those numbers will go tend to go up in the future, but the ratio between them will likely stay similar to what it is today. One thing that should be striking is that some of the numbers are massively larger than the others. Using one of the higher things on the list when you should be using one of the lower ones can kill performance in a hurry. The goal of streaming games is to do exactly that.
To make things worse, all of those except for the Internet connections are things that you can readily use all day long. With the Internet connection, if you try to do that all day every day, your ISP will probably try to shut you down one way or another, as that would be more than 10 TB/day on a gigabit Internet connection.
Computations are cheap. Internet bandwidth is expensive. That's dictated by physics, and the laws of physics aren't going to change just because someone doesn't like them.
The only real advantage of game streaming is that you can skip the time it takes to download a game. But the amount of data that goes through your monitor cable in one minute is more than the size to download most games. If you've got an Internet connection that can deliver that kind of bandwidth, then downloading games is pretty trivial and streaming becomes pointless.
거북이는 목을 내밀 때 안 움직입니다
You are limited on bandwidth - local bandwidth will always outstrip remote bandwidth.
There's probably smart ways to split up that workload rather than just say it all has to be one side or the other. Just to say Steaming will never work I don't think is entirely accurate - maybe we don't do 100% streaming assets, we have a good deal still accomplished locally, but some portion of that is calculated and generated remotely and streamed.
How exactly that looks i don't know, and I'm just looking at things with my head in the clouds.. but seems there is some potential there.
Both rendering capacity and bandwidth capacity are going up all the time... hard to say how it will look in 5-10 years time... (Except in the US, where everything will be Comcast or AT&T and will still be vastly inferior to the rest of the developed world)
For streaming a movie, whoever made the movie can look at the whole product and compress it across time. You can't do that if you're trying to play a game in real time. You can still try to compress single frames in isolation, but that might mean that a given image quality takes ten times as much space as if you could compress across time.
Well, you could compress an image across time for games, but the latency would be so awful that it would be useless. If you press a button and it doesn't seem to register until a full second later, that would be regarded as unplayable by the standards of 40 years ago.
With Netflix, there's also the issue that rendering the movie locally isn't an option. If game streaming were the only way that it's possible to play games, then people would do it and it might seem okay if we didn't know of any other way. But if you can get a massively better experience for less money by rendering the game locally on an integrated GPU, why stream it? Not only is that the case today, but the gap is probably going to widen in coming years.
Bandwidth isn't the only problem. Latency is a huge problem, too, and probably more intractable than bandwidth. It takes time to send your inputs to the remote server that is running the game. It also takes time to send the completed image back. It takes time to compress and decompress the image. It also effectively backs you up by a frame for the time it takes to transmit that frame. That could easily add an extra 100 ms of latency.
For comparison, the difference between a high end gaming rig and a low end one at the same graphical settings might be about 50 ms. Unless the frame rate is low enough that your brain doesn't interpret it as motion, it's better to think of the effects of higher frame rates in terms of latency than raw frame rates.
거북이는 목을 내밀 때 안 움직입니다
¯\_(ツ)_/¯
And - as commented above - Intel is a very big company. What they lack in discrete graphic experience they make up for in the sheer amount of cash they could decide to invest.
And if they do create a solid discrete graphics chip - and presumably market it under brand name X I would be very surprised if this wasn't followed by a cpu featuring X. So this could be a strategy that downstream could boost cpu sales.
Furure consoles - possibly; I would also add TVs. And then - as mentioned - there is also the whole AI.
My argument isn't that there will never be any streaming of games at all. Especially for purely turn-based games, the latency doesn't kill the game entirely. My argument is that gamers will never accept it if someone tries to make streaming into the primary way to play a game. Make streaming into the only way that a game can be played at all and you greatly diminish the market for that particular game.
It's kind of like saying that if someone today tried to launch a new game console to compete with PS4 Pro and Xbox One X, but it offered less performance than the original Xbox One (non-X), cost $700, and didn't have any notable advantages to compensate, people wouldn't buy it.
A bit like fixed broadband connections and mobile broadband connections are both being developed. The world has enough markets for multiple different competing technologies, each of which has their own advantages and disadvantages. Developing one thing doesn't mean abandoning the other.
거북이는 목을 내밀 때 안 움직입니다