Slashdot Mirror
GPUs Keep Getting Faster, But Your Eyes Can't Tell
itwbennett writes "This brings to mind an earlier Slashdot discussion about whether we've hit the limit on screen resolution improvements on handheld devices. But this time, the question revolves around ever-faster graphics processing units (GPUs) and the resolution limits of desktop monitors. ITworld's Andy Patrizio frames the problem like this: 'Desktop monitors (I'm not talking laptops except for the high-end laptops) tend to vary in size from 20 to 24 inches for mainstream/standard monitors, and 27 to 30 inches for the high end. One thing they all have in common is the resolution. They have pretty much standardized on 1920x1080. That's because 1920x1080 is the resolution for HDTV, and it fits 20 to 24-inch monitors well. Here's the thing: at that resolution, these new GPUs are so powerful you get no major, appreciable gain over the older generation.' Or as Chris Angelini, editorial director for Tom's Hardware Guide, put it, 'The current high-end of GPUs gives you as much as you'd need for an enjoyable experience. Beyond that and it's not like you will get nothing, it's just that you will notice less benefit.'"
they need to handle more stuff happening on the screen.
The Kruger Dunning explains most post on
Make it draw less power!
Mr. America walk on by your schools that do not teach Mr. America walk on by the minds that won't be reached
-Multimonitor gaming
-3D gaming (120 Hz refresh rate or higher)
-4K gaming
keep em coming, and keep em affordable!
One thing they all have in common is the resolution.
So 2560x1440 and 2560x1600 27"s only exist in my imagination?
... can certainly tell. The more onscreen objects there are the more slowdown there is. This is why I like sites like HardOCP that look at MIN and MAX framerates during a gameplay session. No one cares that a basic non-interactive timedemo gets 100's of frames a second, they are concerned with the framerate floor during actually playing the game.
In cutting edge games, FPS still suffers even at low resolutions.
Many users are going to multi-monitor setups to increase their visualization and even cutting edge graphics cards cannot handle gaming at 1920x1080 x 3 display setups on taxing games or applications (e.g. Crysis).
For games, GPU's have to process 3D geometry, light, shadows, etc. Number of pixels is not the only factor. This is so lame.
Sig? Heil
8K resolution, 120hz. Nuff said.
That statement makes the rash assumption that GPUs will somehow continue to grow in speed and complexity while everything around them remains static. What about stereoscopic displays which would double the required number of pixels to be rendered for the equivalent of a 2d image? What about HMDs like the forthcoming Oculus Rift, which over time will need to continue pushing the boundaries of higher resolution displays? Who on earth is thinking that the display industry is thinking "whelp, that's it! we've hit 1080p! we can all go home now, there's nothing left to do!" ? 1080p on a 24 inch display is nowhere close to the maximum PPI we can perceive at a normal desktop viewing distance, why is that the boundary? Why are 24" displays the end? Yes, improving technology has diminishing returns. That's nothing groundbreaking, and using that to somehow suggest that we have peaked in terms of usable GPU performance is just downright silly.
If you're talking 2D desktop-type computing (surfing, emails, writing documents etc) the point of this article has already been true for at least a decade.
If you're talking 3D hardware rendering (most usually gaming), there is no such thing as enough GPU power, as its also about consistently achieving the highest framerates your monitor can handle, while having every eye-candy setting maxxed out on the latest AAA games, which are mostly already developed to get the most out of the current and next generation hardware. Its a moving goalpost on purpose.
A GPU is no longer a Graphics Processing Unit, it's a general purpose DSP usable for tasks that have simple logic that must done in a massively parallel fashion. No, I'm actually not talking about mining bitcoins or specialty stuff, I'm talking about things like physics engines.
On the other hand, they are still WAY behind the curve measured by the "My screen isn't 4xAA RAYTRACED yet" crowd.
They ARE out to get you simply because They are in it for themselves and they don't care about you.
Now maybe we can have gameplay and originality again.
"There is considerable debate over what is the limit of the human eye when it comes to frame rate; some say 24, others say 30,"
That's what is studied and discussed as as the lower limit to trick people into thinking it is in motion. I believe there are other studies where they have used pilots as test subjects where they could spot an object between 1/270 a second and 1/300 a second. In addition, there's another study that our brain (and perhaps eyes) can be trained by watching movies/tv to be more relaxed and accept lower frame rates such as 24 as fluid, or higher. Different careers can have an impact as we are exposed to different things visually.
Additionally frame latency can continue to be driven down (with diminishing returns) with higher performing cards even if the frame rate stays constant.
These are often marketed as GPGPU products, nVidia's Tesla for example, rather than taking a bunch of Geforces and putting them together.
Aren't there are other areas of science that a faster GPU benefits namely structural biology and the modeling proteins?
Even ignoring that, the guy is a fucking idiot.
He seems to be confused about the function of a GPU- they are doing far more than simply pushing pixels onto the screen. Wake up buddy, this isn't a VGA card from the 90's. A modern GPU is doing a holy shitload of processing and post-processing on the rendered scene before it ever gets around to showing the results to the user. Seriously man, there's a reason why our games don't look as awesomely smooth and detailed and complex as a big budget animated film- it's because in order to get that level of detail, yes on that SAME resolution display, you need a farm of servers crunching the scene data for hours, days, etc. Until I can get that level of quality out of my desktop GPU, there will always be room for VERY noticeable performance improvement.
There is absolutely no reason to have 1080p as a "standard" max resolution. 5 years ago I got a nice Princeton 24", 1920 x 1200 monitor at a good price. And I expected resolution to keep going up from there, as it always had before. Imaging my surprise when 1920 x 1200 monitors became harder to find, as manufacturers settled on the lower "standard" of 1920 x 1080 and seemed to be refusing to budge.
It's great and all that a 1080p monitor will handle 1080p video. BUT... when it does, there is no room for video controls, or anything else, because it's in "full screen" mode, which has limitations. I can play the same video on my monitor, using VLC, and still have room for the controls and other information, always on-screen.
Now certain forces seem to want us to "upgrade" to 4k, which uses an outrageous amount of memory and hard drive space, super high bandwidth cables, and is more resolution than the eye can discern anyway unless the screen is absolutely huge AND around 10 feet away.
Whatever happened to the gradual, step-wise progress we used to see? I would not in the least mind having a 26" or 27", 2560 x 1440 monitor on my desk. That should have been the next reasonable step up in monitor resolution... but try to find one from a major manufacturer! There are some on Ebay, mostly from no-names, and most of them are far more expensive than they should be. They should be cheaper than my 24" monitor from 5 years ago. But they aren't. Everything else in the computer field is still getting better and cheaper at the same time. But not monitors. Why?
Not to mention that the world hasn't standardized on 1920x1080. I've got half a dozen computers / tablets and the only one that is 1080p is the Surface Pro. The MacBook Pro with Retina Display is 2880x1880. Both of my 27" monitors are 2560x1440. I don't have any idea what this dipshit is thinking, but his assumptions are completely wrong.
I wouldn't let a 1920x1080 monitor grace my cheap Ikea desk.
I should use this sig to advertise my book ISBN-13 : 978-1501515132.
Until, when I look at a video game on my screen and look at a live action TV show and can't tell the difference, there is room for improvement. Perhaps the improvement needs to come from the game developers, but there is still room and I do not believe we have hit the pinacle of GPU performance.
By the way, 4K will replace 1080p very soon, so the article is doubly moot.
Thats why I have my old 1920x1200 panel on an arm off the wall. TECHNICALLY it isn't touching the desk.
16:10 FTW!
Your ad here.
I agree, the guy must live in the ghetto, using recycled year 2000 equipment.
For his crapstation maybe he wouldn't get any benefit from a faster GPU.
Monitors are getting bigger all the time, and the real estate is welcome when editing mountains of text, (as is a monitor that can swivel to portrait). 1920x1080 is not sufficient for a large monitor. I don't have any that are that limited. People aren't limited to one monitor either.
Sig Battery depleted. Reverting to safe mode.
Aren't there are other areas of science that a faster GPU benefits namely structural biology and the modeling proteins?
Absolutely, I run complete atomistic molecular dynamics simulations of viruses that cause disease in humans (enterovirus simulations around the 3-4 million atom mark). Five years ago I had to use a supercomputer to model 1/12 of a virus particle which barely scraped into the nanosecond range. I'm now able to run complete virus simulations on my desktop computer (Tesla C2070 and Quadro 5000) and I get 0.1ns/day or on my 2U rack (4x Tesla M2090) with 2 viruses running simultaneously at almost 0.2ns/day. That's using the last generation of nVidia cards (Fermi), I should in theory be able to almost double that with the new Kepler cards. I will be VERY interested to see how the next ?Maxwell architecture pans out in the future. I can see a time in the not to distant future when I can model multiple instances of virus-drug interactions on-site here in the lab and get results overnight that I can compare with our "wet lab" results. I use NAMD for the simulations which works well with the CUDA cards.
You don't need a GPU at all. A screen is 2Mpixels. Refreshing that about 60 times per second is enough to create the illusion of fluid motion for most humans. So that's only 120Mpixels per second. Any modern CPU can do that!
Why do you have a GPU? Because it's not enough to just refresh the pixels. You need (for some applications, e.g. gaming) complex 3D calculations to determine which pixels go where. And in complex scenes, it is not known in advance what objects will be visible and which ones (or part) will be obscured by other objecs. So instead of doing the complex calculations to determine what part of what object is visible, it has been shown to be faster to just draw all objects, but to check on drawing each pixel which object is closer, the already drawn object or the currently being drawn object.
Err, wrong.
24-30FPS is enough *with proper motion blur*
Without motion blur, you need about 3x that.
We only get good enough framerate at 1920x1200 (the One True Resolution) because of a lot of shortcuts. Improved computing power could allow games to make the transition to better lighting models (whatever they call the new ray0tracing stuff) that are both easier for artists/world builders and look better and more natural. It would also be nice to stop thinking of everything in polygons, but there's so much tooling there beyond the GPU (and if you push the poly-count high enough it doesn't matter visually).
Socialism: a lie told by totalitarians and believed by fools.
Seriously man, there's a reason why our games don't look as awesomely smooth and detailed and complex as a big budget animated film- it's because in order to get that level of detail, yes on that SAME resolution display, you need a farm of servers crunching the scene data for hours, days, etc.
That reminds me of the Final Fantasy movie from 2001, I remember watching that and being struck by the realism of the characters, especially the individual strands of hair of the female lead. Apparently she had 60,000 strands of hair that were individually animated and rendered, and her model had 400,000 polygons. The Wikipedia article has some interesting details:
Square accumulated four SGI Origin 2000 series servers, four Onyx2 systems, and 167 Octane workstations for the film's production. The basic film was rendered at a home-made render farm created by Square in Hawaii. It housed 960 Pentium III-933 MHz workstations. Animation was filmed using motion capture technology. 1,327 scenes in total needed to be filmed to animate the digital characters. The film consists of 141,964 frames, with each frame taking an average of 90 minutes to render. By the end of production Square had a total of 15 terabytes of artwork for the film. It is estimated that over the film's four-year production, approximately 200 people working on it put in a combined 120 years of work.
To your point, this bears repeating:
with each frame taking an average of 90 minutes to render
This isn't exactly a GPU pumping out 40 frames per second where it can afford to make several mistakes in each frame. Also to your point, here's another interesting detail:
Surprisingly for a film loosely based on a video game series, there were never any plans for a game adaptation of the film itself. Sakaguchi indicated the reason for this was lack of powerful gaming hardware at the time, feeling the graphics in any game adaptation would be far too much of a step down from the graphics in the film itself.
"Our two-party system is like a bowl of shit looking at itself in a mirror." - Lewis Black
Aren't there are other areas of science that a faster GPU benefits namely structural biology and the modeling proteins?
Even ignoring that, the guy is a fucking idiot.
Hear hear. I sorely miss the old ars, where new CPU architectures were explained - in detail - by someone who actually knew what they were talking about. Nowadays we have op-ed political ramblings, coverage of every. single. thing. to do with Apple and some guy drinking the latest fad food and giving us all a daily update on what colour his shit is. (Seriously, that happened.)
If God forks the Universe every time you roll a die, he'd better have a damned good memory.
But everyone here is missing the REAL advantagse of new GPUs....power consumption and price. My HD4850 uses over 110w and reaches nearly 190c under load, the HD7770 I'm getting for my BDay? It is nearly 60% faster while uses less than HALF of the power and creating less heat by nearly half. And while my HD4850 new cost nearly $300 the HD7770? Can be had for less than $100.
Now that CPUs have pretty much maxed out the big gains will be in GPUs, both on power and price. Sure the top 'o the line will be a giant wallet raper but those are really ePeens more than anything and the biggest sales will be the $75-$150 range and that is where we are seeing some really sweet cards.
ACs don't waste your time replying, your posts are never seen by me.
I do some work on the side for a hardware raytracing company and you're mostly right. Shameless plug: http://caustic.com./ And speaking as a VFX artist ray tracing is way easier. When you aren't cheating everything it becomes much simpler to get to "realistic". Global Illumination also goes a long way to help. I can take a game asset with textures and geometry and normal maps etc and render it with a raytracing engine and it looks dramatically better.
The problem with current technology is that there is something of a divide in performance. Present ray tracing technology is about 5x too slow to match a good rasterized game. You could deliver 10-20 fps at decent resolution with ray tracing but wouldn't get any noticeable benefit. To really get that silky smooth GI you need another 20-30x faster or so (even with a dedicated ray tracing chip).
The challenge then is to improve ray tracing chips fast enough to catch up to GPUs. I think in 3-4 years you'll see a number of games which deliver exceptional ray traced images. Rivaling film renders in real-time. But 3-4 years in spite of this author's nonsense is a long time in GPU technology. In the last 3-4 years we've seen tessellation, the first instances of GI and dynamic light reflections. These make a huge difference. They're total hacks but game developers can't sit still and as much of a pain as they are--they work. It would be more of a paint to rewrite their engines from scratch to take advantage of a whole new rendering pipeline.
The other challenge is that the reason many films look so good is because of 2D cheats in the composite. If you look at a raw render out of Arnold, Brazil, Renderman or Vray it's not really like what shows up on screen. There is a lot of sweetening, a lot of one-off lighting tricks in post and in the render which only look great from the one angle. Games have to look good from every angle. I don't know that they'll ever achieve that. They'll look more photographic but the ultra polish of a film comes from lighting TDs, cinematographers and compositors all working in tandem to polish a shot for days or weeks. If things just looked good from every direction all the time--the VFX work on a feature film would be dramatically reduced. So in that regard game developers are going to have it way harder than film people. Not only does it have to render at 60 fps... but you can't cheat detail. You have to make it look good from 300 yards as you drive your car down main street... all the way to jumping out and walking up to 2" away and reading the headline.
First off you're so wrong it hurts. Until very recently graphical framerates in the average FPS were relatively insulated from "physics framerates", in the days of TFC for example 100fps didn't make your rockets any more accurate because the SERVER calculated its trajectory.
Secondly it's been proven time and time again that humans are perfectly capable of detecting framerates well into the hundreds. Fighter pilots can not only detect a SINGLE frame from somewhere around 1/200th of a second but even tell you what enemy jet it was. Gamers are similar, until consolization forced a lower standard onto everyone and covered it up with a lower FOV filled with massive amounts of bloom and blur performance was judged by the gold standard of a solid 60fps minimum, 30 was choppy and 100 was idea.
A bullet may have your name on it but splash damage is addressed "To whom it may concern."
Mine sure does.
Using two none-SLI GPUs, a GTX670 and an older GTS460. I use the 460 for physx and a couple of older 20" monitors, and a 27" WQHD as main monitor on the GTX670. I've also hooked the 670 up to the TV, running 1080p for when I'm in couch potato mode. I can see a dramatic difference in speed changing between TV and main monitor.
The difference between 1080p and WQHD is remarkable in certain games, like eg. Civilization 5. Most games with an isometric view benefit from it.
Actually this is borderline hilarious/obtuse of an article in the first place. This is implying that increasing a GPU's power is about resolution - when more and more demanding games require those same increases of a GPU's power.The article could not be further from any truth whatsoever.
So can eyes tell when GPU's get faster? Absolutely. You just need to put in a context people can understand. In this article they have a call of duty video of the xbox one side by side with PS4, and then a video with PC side by side with PS4 for the same game. So resolutions are: 720p -> 1080p playtation -> 1080p (pc, fully upgraded). Anyone can easily see the differences made, and it's not "hard to tell".
I hope this shows people that the folks at Toms Hardware Guide are sometimes correct and sometimes completely idiotic.
Additionally, you will be able to see a difference if your game stutters at 30FPS vs 60FPS. So there as well, the need for better performing GPU's still exists.
Silly rabbit. GPUs are for mining bitcoins.