Slashdot Mirror
Ask Slashdot: What Is the Latest and Greatest In Computer Graphics Research?
OpenSourceAllTheWay writes: In the world of 2D and 3D Visual Content Creation, new tricks that ship with commercial 2D or 3D software are almost always advertised as "fantastically innovative". But when you do some digging as to who precisely invented the new "trick" or "method" and when, you often find that it was first pioneered many many years ago by some little known computer graphics researcher(s) at a university somewhere. Case in point, a flashy new 3D VR software that was released in 2018 was actually based around a 3D calculation method first patented almost 10 years ago. Sometimes you even find that the latest computer graphics software tricks go back to little-known computer graphics research papers published anywhere from 15 to 25 years ago. So the question: What, in mid-2018, is the latest and greatest in 2D or 3D computer graphics research? And which academic/scientific publications or journals should one follow to keep abreast of the latest in computer graphics research?
If you're looking for the great classics in computer graphics, many not so little-known graphics papers are in SIGGRAPH proceedings.
https://en.wikipedia.org/wiki/...
Starting in 2003, all SIGGRAPH papers are published in ACM TOG
https://en.wikipedia.org/wiki/...
It seems to me the biggest trends this year involve accurate 3D scanning tools which eliminate tedious low polygon approximations. The latest engines are fantastic but without proper modeling and texturing stock, there isn't much point.
Ray driving is similar to ray marching except high performance sports cars are used instead of slow moving army dudes.
Every time I see a thread about the latest/greatest in computers or graphics, I always think of this Dilbert from 1995.
It must have been something you assimilated. . . .
Here http://kesen.realtimerendering... This truly helpful fellow collects nigh every paper from every conference covering such in an easily browsable site. The only 2 things not covered are here http://gdcvault.com/ and here http://advances.realtimerender...
Ha ha ha ha... wow, that is... wow, so wrong.
GPUs have increased many-fold in performance since 10 years ago. Not even the fastest video card from back then could power a VR headset today, or support modern gaming on a 4K monitor. CPUs have made less of an increase in raw clock speeds, but have made huge jumps in core count and instructions per clock (especially in specialized areas, like vector units). RAM capacities have gone through the roof. Drive technology has made the jump from HDD to SSD, and then from SATA-based SSDs to PCI-Express.
Yes, from one generation to another is usually a relatively small difference - but with generational changes every 18 to 24 months, over the course of a decade you are looking at much bigger improvements than your comment stated. And this is all without talking about things like using GPUs for general-purpose computation, which has vastly improved performance in many areas of computing.
By the way, this is intended more as to refute the parent comment - not as a direct answer to the subject of the main post.
William George
If all you want is to be able to browse the latest graphics research papers in a convenient fashion, the #1 site to go to is Ke-Sen Huang's page. Every paper released at every major conference from the past 10 to 20 years is there, with links to everything you'd want: ACM reference page, free access preprint if available, website for the paper if available, etc. It's an amazing resource and something you just have to have bookmarked.
If you want something more curated, it becomes trickier, but a fun way of doing it is to look for the "technical papers preview" videos online for SIGGRAPH. A fairly long-standing tradition of that particular conference is to kick off the whole thing with a very short, usually humorous blurb of every technical paper being presented that year, done by the authors of each paper, in one giant marathon session on the first day. Each paper gets like 30 seconds to pitch its idea and show it off visually, and while you can't find the full 2-3 hour presentation that contains all of them, there's usually a shortened version online with some interesting/promising examples.
Nope. There are no longer generational changes every 18 to 24 months. The computer you had in 2008 is essentially the same as you have now in 2018. Not sure what you are talking about regarding RAM capacities for example. How have they "gone through the roof" in 10 years?
Haven't you heard of Qbit Blockchain Deep-Learning Microservice Serverless 4D.js Rendering?
Table-ized A.I.
How about you just Google this?
Who cares about this little niche called GPUs?
GPU ram has.
Any advance in hardware is taken up by mining, leaving research for graphics a hardware generation behind.
A "Big" whoosh to you for not understanding what TFA is
about and how 110010001000's response was appropriate.
CAP === 'governor'
Just check the Latest Siggraph papers.
Takes a year or two to be assimilated
Thats true. That is because when you hit a performance brick wall due to physics you can add RAM and cores to keep the performance train going. There hasn't been any "generational changes" in the last 10 years though.
I was talking about actual research, not hardware specs. But even if he is talking about specs it hasn't been changing much. Because, Physics.
Windows NT is a tad bit outdated, m8
does mining cryptocurrency count?
I think OP was talking more in the mainstream. There just isn't as much going on.
Case in point: I don't game (disclaimer), but my primary computer today is a late 2013 Macbook Pro (15 inch Retina). The i7 It shipped with is till perfectly adequate today. The 8GB RAM it shipped with still feels adequate, and is still in line with most new laptops out there. The 256 GB SSD has been close to full for a while, but that's till the standard SSD size.... But unlike a lot of newer models, I can upgrade my SSD at will.
In all likelihood, that computer will remain my primary computer into next year. I'll probably replace the battery soon. And will take that opportunity to replace the SSD.
This is the first time in all my years of computing that a machine has remained not just in service, but perfectly usable even as I continue to update the OS and Applications, over this span of time.
10 years is probably a little stretch even with the slowed pace of advancement in hardware, but not much of one. My current desktop was built with price/$ in mind 10 years ago, and still runs modern software fine, although a few programs want more than 6gigs of ram in the last year. The graphic card was upgraded once and is only 6 years old but runs everything I've thrown at it fine. Obviously my 10 year old moniter isn't 4K. I only intended to keep the system about 5 years, but hardware advancement has slowed so much it has lasted twice as long as intended. I'll probably replace sometime in the next year, and exspect to get 10-15 years this time. Using 1998 hardware for modern software in 2008 would have been laughable. Mid range 2008 hardware in 2018 doesn't even require lowering settings for games.
Lame.
If your standard is "essentially the same", then why stop there? Your computer in 2008 is essentially the same as your computer in 1998, which is essentially the same as your computer in 1988, which is essentially the same as your computer in 1978. I'll stop there, although if I was willing to push the point I could keep going.
If you had said that Dennard Scaling has slowed down you would have been on solid ground. However my computer of today is much improved upon the 2008 model, and thank goodness you aren't responsible for purchasing my computers.
Distance fields as primary objects may change how we do a lot of computing both in and out of the graphics domain.
http://iquilezles.org/www/articles/raymarchingdf/raymarchingdf.htm
The top-end desktop CPUs in 2008 from Intel were the first generation of the Core i7 series. The maximum amount of RAM that CPU supported was 24GB. Today, the top-end Intel and AMD desktop processors support 128GB of memory (Core X series and Threadripper both) - and if you go over to the single-socket Xeon W you can get 512GB. Dual-socket CPUs support more now, and supported more back then, but we are looking at ~5 times the RAM capacities today that were available then. 2-4GB was typical for an average desktop back then, if you don't want to look at the high-end, and now 16GB is easy to get while many systems have even more.
On the GPU side, as an AC also noted below, memory capacity has gone up substantially. The top-end desktop GPUs in 2008 had 1GB of video memory, while top-end cards today are at 11 or 12GB (depending on whether you consider the Titan series to still be desktop cards, or if you consider that to max-out at the 1080 Ti). Workstation cards go even higher, with capacities double what you find on consumer cards (24GB on the Quadro P6000, for example).
And to debunk your statement that "The computer you had in 2008 is essentially the same as you have now in 2018":
In 2008, I had a dual-core CPU running ~2.5GHz with 8GB of memory, a GPU with 512MB of memory, and a brand-new 80GB Intel SSD that could push about 250MB/s read and write speeds.
Today I have a quad-core CPU running at ~3.4GHz with 32GB of memory, a GPU with 8GB of memory, and two SSDs (500GB at ~550MB/s and 400GB at over 1500MB/s).
So my CPU has doubled in core count, increased over 50% in clock speed, and is several times faster overall thanks to a myriad of other improvements. My GPU is on the order of 30 times faster with 16 times the amount of memory. My SSDs have over 10 times the total capacity and are almost 2 to 6 times faster (and I don't have cutting-edge SSDs, myself).
Oh, and generational changes? They do still average 18-24 months from Intel on the CPU side and NVIDIA on the GPU side. Sometimes a little longer, sometimes shorter. From 2008 to 2018 we've had ~8 generations in the Core series from Intel. NVIDIA has gone from Tesla to Fermi, then to Kepler, Maxwell, Pascal, and now Volta (though that has only shown up on GPU-compute oriented cards thus far). So 8 generations over 10 years from Intel breaks down to about 1.25 years per generation, or 15 months. 6 generations over 10 years for NVIDIA is an average of 1.66666... years, or 20 months.
William George
Wow. Doubled in core count and 50% faster in 10 years? That is pretty good. My mistake.
Windows 10 will soon have the GPU ability to play solitaire.
As I noted, my comment was just in regards to the sweeping - and very incorrect - statement made by 110010001000. I work with computers, but I do not personally perform research to design new hardware or software approaches to graphics. The original question is also very wide-open, so much so that I do not feel I can directly answer it... but I didn't want to leave such a disparaging comment about computer technology unchallenged.
Could you perhaps enlighten me as to what you think I missed regarding the original "article" (really just a user-submitted question)?
William George
I am sure you are an IT guy who reads Intels marketing thoroughly, but haven't noticed that your new computer isn't much different from your old one.
That is just my personal system, which is actually running a CPU from ~ 4 generations ago (so about 5 years old). I'm on the cusp of an upgrade (likely this year) to a 6-core at over 4GHz, and if I were into any applications that benefited from higher core counts I could get a 16-core AMD or 18-core Intel processor. It all comes down to what an individual user needs, wants, and of course can afford.
If your point was more along the lines of "basic Internet and office application usage isn't any more complex today than it was 10 years ago, so a computer with similar specs will still do the trick" - then you'd have actually been fairly correctly. But many areas of computer use can and do use far more powerful hardware today: gaming, video editing, 3D rendering, photogrammetry, machine learning, and more. Since the original question was about graphics advancements, it isn't really fair to come back with an answer that 'nothing much has changed' (a rough paraphrase of your comment).
William George
I'm curious as to why no one mentioned anything about voxels, volumetric video, light-fields, or photogrammetry.
I'm not an academic researcher, but from what I can tell, you can basically thank Lytro, 8i, and Depthkit for the latest in "3D camera" tech: http://www.depthkit.tv/
Lytro was bought by google recently, so who knows where that's headed... As for voxels: https://www.atomontage.com/
I think that the next frontier of graphics will be taking the whole planet and sticking it in a graphics engine using volumetric capture, and represented by voxels instead of polygons. Polygons are hitting their upper limit of usefulness, as the amount of complexity involved with creating more realistic images is very high, and increases in GPU power are yielding smaller and smaller improvements in graphical quality over time. Its possible that voxels will be the best way to overcome this. But I'm sure you all will tell me how I'm wrong :)
See http://www.connellybarnes.com/work/ and the references therein.
See http://www.connellybarnes.com/... and the references therein.
LOL - nice try. Check out the plethora of articles I write on a regular basis, looking specifically at performance of different computer hardware in high-end applications:
https://www.pugetsystems.com/a...
(any written by William George are mine, in case that wasn't obvious from my /. username)
William George
Oh I see. You sell computer systems. No wonder.
I guess going from 4 cores to 18 cores is in 10 years is pretty good. I hope they will add another core this year. That would be awesome. By the way, Intel "generations" are just marketing speak. Meaningless.
Oh for Pete's sake... so does that mean nothing NVIDIA or Intel says can be true, just because they sell GPUs / CPUs?
Sure, a company selling something certainly might stretch the truth to try and get folks to buy things - but that doesn't mean that all the stuff said (or written) by anyone selling a product is automatically incorrect. Where I work we actually don't do much advertising, or make crazy claims: we actually run real-life software to see how it performs, and then publish the results publicly. We could keep the info to ourselves, if we wanted to only benefit our company and our customers, but we don't. We could also make wild claims without backing them up, but again we talk about what we test, how we test it, and then let the results speak for themselves.
I'm not even going to plug the name of the company or link to our website again, since that isn't even the point of the conversation we were having. I didn't bring it up until you tried to accuse me of just being "an IT guy who reads Intel's marketing" (effectively claiming I don't really know what I am talking about, and that I am not qualified to comment on this stuff). Linking to my extensive writings and research was the fastest way I knew of to prove you wrong.
Now, can we all just go back to being civilized? Computer hardware has made huge advancements, but you have a fair point that for an average, basic computer user it hasn't be a hugely dramatic shift in the last 10 years. I would compare that to cars: we've got huge advancements in electric and hybrid tech (especially batteries) going on, as well as the beginnings of smart / self-driving cars, but the vast majority of drivers are still using gas-powered vehicles that aren't all that different from the cars of 10 years ago. So car tech *is* constantly being researched and improved, but that trickles down to most folks on a much slower timetable. Is that sort of the idea you were trying to get across?
William George
Does this also cover what goes into movies?
C88 is always like this, don't take it personally.
Sounds like a kid. performance improvement is not "generational change" heck. go and study for your school exam instead, kiddo
If you're more into math and art than optimization tricks, check out Bridges.
(I (re)?discovered math art about 3 years ago, and it sort of reminded me of the early 90s demoscene, except this time it's for grownups. I got into Bridges as soon as I heard of it, and it's my third year taking part in some way; there's also an art exhibition and a short film festival for those of us who'd rather just show off what they do instead of giving lectures.)
Escher was the first MC and Giger invented the HR department.
Delaunay triangulation was invented before the digital computer... He wrote the original paper in french while living in Russia. Or maybe the other way around. One thing that is weird is that delaunay of dimension n is associated with a hull of dimension n minus 1 .. For the same set of points. That is probably useful in some way we havent found out yet... So in thw matter of a few weeks this useless fact will become cutting edge simply because of human society and culture. In that sense we have no idea what is truly truly new...
...no, this is almost entirely Intel being so terrified of cannibalizing their high-profit-margin "Server CPU" business that they've refused until recently to offer more than a quad-core processor to home consumers for over ten years now.
- WolfWings, too lazy to login to /. in way too long, and too lazy to explain more verbosely what even a brief google search can explain better.
"Case in point, a flashy new 3D VR software that was released in 2018 was actually based around a 3D calculation method first patented almost 10 years ago"
Where's the case you are making your point with so the rest of us can see what you are talking about?
Why is this not closed yet like that moronic Stack Exchange-like servers as being too broad and out of scope or whatever fucking crap their dickwad mods conjure up
as soon as i can afford to upgrade my own card to something made in the last five years.
Photogrammetry software is very cool, I dod not know this type of algorithms are possible.
How about NVIDIA Ray Tracing?
Are pretending to be retarded or...?
On the Oregon Cost born and raised, On the beach is where I spent most of my days
Well... i'll bite
Graphics subsystems existed in 2000 which would operate the latest VR headsets if some work was involved. The InfiniteReality4 from SGI was on par with a modern day GTX1060 or maybe 1070. The key difference was that it took racks of equipment to do what one chip that fits in a laptop does today.
Now that said, to run a VR headset requires a graphics card capable of feeding 50 or more frames per second on a synchronized output timer. 10 years ago, NVidia was selling Quadro solutions specifically targeting the Broadcast SDI market. While I wouldn't sing from the roof about the number of triangles or pixel rendered per frame, the cards were certainly able to meet the frame buffer requirements of modern VR headsets.
24GB of RAM or 128GB of RAM is no real difference. It's mostly about the availability and cost of RAM that matters. Judging based on :
https://jcmit.net/memoryprice.htm is a historical RAM website. If I follow the trend and go to NewEgg and find a 16GB DIMM, $165 is the price which seems to represent the cheapest memory you would actually intentionally purchase today. That works out to $0.01007080078125 a meg... which places us at the average price of memory in 2009.
Of course, this doesn't account for inflation. Using https://westegg.com/inflation/ to adjust the prices, $165 today would be worth about $145 back then, So we can say you can buy a bit more RAM for the same comparable money, But certainly not even close to double.
The reason for this is partially economic, alternatively... we simply didn't need more than 16GB in most computers and the market isn't being driven by greater RAM demand. As such, the computer we had in 2008 is still perfectly suitable from that regard today.
I have tested performance on a 2600K, 3770K, 4790K, a wide variety of modern notebook and server processors as well. I don't have any desktop processors newer than the 4790K because the performance improvement is simply negligible. Just not worth the cost. And if I need more cores instead of faster cores, I personally own a 128 core Xeon E5 cluster plus another 70+ cores in the same rack. I pretty much never turn them on... and they are modern.
I still use the 2600K regularly since I love the case it's in and it's usually around as fast as my Surface Book 2 (top model)
There have been many changes... I will gladly submit that NVidia and AMD have made huge strides in GPU, and I simply wanted to correct the "Not even the fastest video card from back then could power a VR headset today" purely for technical accuracy. I would simply have said "graphics cards from that time look like 8 bit graphics compared to today's cards" for example as it makes an obvious exaggeration to prove a legitimate point. With regards to the rest, you mention maximum capacities but don't mention realistic configurations.
And BTW I'm a RAM whore. I have terabytes of it in my cluster. I keep a 20 Core E5 with 384GB of RAM for fun.
does that mean nothing NVIDIA or Intel says can be true, just because they sell GPUs / CPUs?
Moving away from a boolean value of True or False, the correct answer is not completely True. That is guaranteed because of marketing BS.
There are hard numbers that claim revolutionary speed and quality increases. These are also guaranteed to also be not completely True because of marketing BS. Core counts, clock speed and memory size do not scale linearly with results, even though that is implied by marketing comparisons.
So called "real world" tests are also deliberately misleading. As they say, YMMV. It's always a cherry picking contest driven by, you guessed it, marketing.
Is it better then it was 10 year ago? Yes, of course it is. How much? Somewhere between 1.5 and 5 times better, for some vague value of "better". Moore's Law, an exponential size over time relationship, maps very indirectly into user land.
Why is Snark Required?
> In the world of 2D and 3D Visual Content Creation, new tricks that ship with commercial 2D or 3D software are almost always advertised as "fantastically innovative".
That's the marketing department. It's what they do. For a living.
Pro tip: If you're interested in the real thing, stop reading press releases and turn to academic papers (although some of them are press releases too, so navigate carefully).
A lot of more recent Graphics papers (mostly Image Processing, actually) are using Convolutional Neural networks to do various things. There has been a lot of low hanging fruit in the areas of denoising, and various image manipulation techniques, so results in those areas have been transformed in the last few years.
One such "hot" area that has application in the broader area of computer graphics, is the denoising of path traced images. Path tracing uses stochastic light bouncing techniques to produce a highly accurate image (in terms of lighting effects), but these images are noisy (due to the stochastic nature of the rendering process), requiring a large amount of samples to "average away" the noise, and hence being slow to render. Neural networks can learn to remove the noise from such images, potentially allowing for photorealistic images to be created extremely rapidly, perhaps even in realtime. In my view, this is the most exciting "game changing" area in graphics at the moment.
The most destructive creatures in the universe. Spent all their time inventing new ways to destroy things and generally make a nuisance of themselves.
It is probably a good thing they all drowned when the ice caps melted after the self inflicted global warning.
If we make a graph of performance per CPU core, you will sere that in the last 10 years we have been reaching a plateau. Do the same for the 90s and you'll see a steep difference.
Now, I hear "but multiple cores!". No, no, stop right there. The average computer user and the typical programs he uses is unable to get much gains from more than 4 cores. We already had 4 cores CPUs available 10 years ago... So there you go.
Paint it anyway you like it. CPU performance increase have been low in the last 10 years, and it seems it will only get slower. We need a breakthrough and the industry haven't been capable of pulling this off.
The most significant jump in graphical improvement was from a Voodoo card. Since then, everything has been seemingly incremental in comparison.
It's a perfect time for being wasted.
A perfect time to watch the stars.
- Burden Brothers, "Beautiful Night"
SIGGRAPH is the ACM computer graphics research conference. You won't find anything more cutting edge. Each year they produce a video "SIGGRAPH $YEAR : Technical Papers Preview Trailer". This is exactly what the OP was looking for. Here's 2017's video:
https://www.youtube.com/watch?...
Oh thats easy.
Follow John Carmack on Twitter !
Hivemind harvest in progress..
The really fundamental advances take a long time to be fully explored. There is little significant that doesn't build upon earlier work.
Check out Conformal Geometric Algebra, which is the basis for the company Geomerics' Enighten software for real-time global radiosity lighting for games. (Now part of ARM / Silicon Studio).
See the lectures linked from the first link, in particular lecture 7 on CGA. These are by Chris Doran, one of the founders of Geomerics, a member of the Cambridge GA group. Also see Leo Dorst's GAviewer CGA tutorial for interactive visualization and a better idea how this can be used in computer graphics. GA is also a lingua franca for physics and simulation that subsumes vector algebra, imaginary numbers, homogeneous coordinates, quaternions and a zillion ad-hoc hacks that have made graphics far more complicated than it needs to be. The papers in the field are nearly all written to be understood and require little background knowledge.
"Is life so dear, or peace so sweet, as to be purchased at the price of chains and slavery?" - Patrick Henry
> GPUs have increased many-fold in performance since 10 years ago.
So what? The OP was talking about stuff like innovative new mathematical/logical approaches to performing graphics rendering, not what hardware you're doing it with.
Do you not understand that just doing the same old shit on a bigger chip every year is actually not scientific progress?.
They are doing a lot of interesting things at NVIDIA...
https://www.nvidia.com/en-us/research/computer-graphics/
Iâ(TM)m very fond of their recent paper at I3D with Lucasfilm and Unity on using low sample-count ray tracing with denotising to get rid of speckly artifacts and combining it with traditionally lit objects in a clever way.
Itâ(TM)s really slick.
http://research.nvidia.com/publication/2018-05_Combining-Analytic-Direct
As I mentioned, my reply was not so much directed at the original question as at the comment above mine (the original "None" comment).
Regarding the original Ask Slashdot question, I am sure if I am qualified to answer... but it seems to me that the question is pretty wide-ranging and not very focused. The query includes mention of 2D and 3D visual content creation, consumer software, VR, graphics research, patents, and more. I would think that checking out conferences like SIGGRAPH and GTC would showcase a lot of the ongoing research and development in these fields, but I don't follow any publications (which is what the question ended with a request for).
William George
Actually he's being rational. The retarded one is William who keeps throwing out meaningless numbers.
I'm doing the same shit on my computer today that I did 10 years ago. Not much has changed.
Only difference is code bloat, which is the reason we see these small improvements.
Every 10 years nvidia talks about ray tracing like it will be a game changer. Hint: it wont. It will die and in 10 years nvidia will put out an press release about new techniques in ray tracing.
I wondered why he was working so hard to bend and twist and stretch every single spec to fit his narrative. Now it makes sense -- he's a computer salesman.
Q: What's the difference between a car salesman and a computer salesman?
A: The car salesman knows when he's lying.
> I'm doing the same shit on my computer today that I did 10 years ago. Not much has changed.
Same here, except nowadays my computer can't even keep up with my typing - it regulary freezes up mid-sentence or shoves some app in my face when it needs attention.
My very first Mac, a lowly 100MHz PPC 601, always kept up with my typing and NEVER let anything interrupt me. That was 1995.
it used to be that the actual "graphics" was the limiting factor in "peoples minds" now with the textures and engines the realism has got to the point the limitation is the way in which things move
for example https://www.youtube.com/watch?v=uFJvRYtjQ4c
https://github.com/sebastianstarke/AI4Animation
regards
John Jones