|
jisforjosh posted:For reference, the 2080ti with DLSS and the same settings was hitting 42 FPS average drat that is a huge jump.
|
#
?
Sep 14, 2020 19:02
|
|
|
|
| # ? Jun 5, 2024 21:55 |
|
|
bonus: it's a rifle firing a whole pistol cartridge
|
|
#
?
Sep 14, 2020 19:10
|
|
|
It's 65% more bullet, per bullet!
|
|
#
?
Sep 14, 2020 19:13
|
|
|
Zedsdeadbaby posted:
What *is* raytracing anyway/what does it add to a rendered image? I've never been really clear on that other than "it looks better"
|
|
#
?
Sep 14, 2020 19:21
|
|
|
axeil posted:What *is* raytracing anyway/what does it add to a rendered image? I've never been really clear on that other than "it looks better" https://www.youtube.com/watch?v=oCsgTrGLDiI This explains it well enough for surface level
|
|
#
?
Sep 14, 2020 19:24
|
|
|
axeil posted:What *is* raytracing anyway/what does it add to a rendered image? I've never been really clear on that other than "it looks better" digital foundry always does well produced explainers https://www.youtube.com/watch?v=moKV5_BpxjM
|
|
#
?
Sep 14, 2020 19:27
|
|
|
axeil posted:What *is* raytracing anyway/what does it add to a rendered image? I've never been really clear on that other than "it looks better" It helps to understand that traditional lighting in games is a hodgepodge of different effects that when assembled well appear to mimic light, but they aren't actually simulating real light. There's a bespoke system for shadows, another system for global illumination, a system for reflections etc., and they all have to be carefully tailored to look realistic and not conflict with one another. Raytracing can replace any or all of those effects by simulating actual rays of light, which takes a lot of compute power, but can lead to photorealistic accuracy without requiring a ton of artist labor. Raytracing can make shadows blur at their distant edges, it can get totally accurate reflections, and it can do photorealistic bounce lighting in indirectly lit scenes.
|
|
#
?
Sep 14, 2020 19:34
|
|
|
raytracing makes reflections stop doing this poo poo https://www.youtube.com/watch?v=CKFmilV8dJs https://www.youtube.com/watch?v=he-I2mPCuLI repiv fucked around with this message at 19:46 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 19:41
|
|
|
Bigsteve posted:Benchmark seems impressive but dlss on makes it less so in terms of brute strength. It's not like AMD is gonna drop anything that's more powerful in terms of brute strength than the 3080/3090, and their RT implementation is going to be first gen. Soooo, yeah. It's a huge jump over what's possible right now, and as has been mentioned, it's unlikely we'll see too many games coming out with RTX but not DLSS, so this is a pretty reasonable demonstration of the idea that, yeah, Ampere can actually do real nice RTX effects at viable framerates now.
|
|
#
?
Sep 14, 2020 19:44
|
|
|
repiv posted:raytracing makes reflections stop doing this poo poo I don't even have an RTX capable card but since seeing RTX reflections I instinctively just pan the camera up/down when seeing reflections in games to see this effect. It's like soft-edged shadows, once you see them enough times when you don't have them they stand out massively, whereas when they were first on the scene and produced a massive performance impact they were seen as largely frivolous.
|
|
#
?
Sep 14, 2020 19:54
|
|
|
my biggest pet peeve with current lighting technology that ray tracing should solve is how flat and dull textures suddenly look when you enter a shadowy area
|
|
#
?
Sep 14, 2020 19:58
|
|
|
I still hate random soft-edge shadows. Doom 3 style stencil shadows are to me the most artistically appealing shadows have been, and if you're going to go soft shadows to chase photorealism, the softness better scale properly with distance.
|
|
#
?
Sep 14, 2020 19:59
|
|
|
Rinkles posted:my biggest pet peeve with current lighting technology that ray tracing should solve is how flat and dull textures suddenly look when you enter a shadowy area Unfortunately that is global illumination which fixes that. And that is legit the most performance expensive effect with RT. Though rtxgi which uses raytracing to calculate lightprobes in real time rather than per pixel seems to be rather performant if their ue4 stream is trustworthy
|
|
#
?
Sep 14, 2020 20:03
|
|
|
RTXGI also has the huge advantage of simulating infinite bounces (one additional bounce per frame) basically for free, while per-pixel RTGI is barely practical even when limited to a single bounce It's probably the way forwards, perhaps with an optional ultra mode that does per-pixel GI for the first bounce then samples RTXGI probes for deeper bounces
|
|
#
?
Sep 14, 2020 20:06
|
|
|
unfortunately you do lose that near perfect Ambient occlusion proper RTGI generates, Also indirect shadows are going to be way less accurate, but we have gone so long without indirect shadows even existing that I can see most people being willing to sacrifice that. The infinite bounces thing is just a recusion loop using the previous frames results, and can be done with RTGI, but it introduces lag in light changes. Heck Lumen for UE5 which is Raymarched has indefinite light bounces.
|
|
#
?
Sep 14, 2020 20:12
|
|
|
I really wish EVGA would post the part numbers/specs on their website so I can start looking at them. I've never followed a launch before, so I assume this is normal. Just weird that NewEgg has some listed already and they don't have them listed on their own site, outside of the announcement ad.
|
|
#
?
Sep 14, 2020 20:22
|
|
|
Anime Schoolgirl posted:bonus: it's a rifle firing a whole pistol cartridge That's fine, there's nothing wrong with that. They're called PCCs, pistol-caliber carbines. Extending the length of your barrel gives the powder more time to burn, expand, and build pressure behind the bullet before it leaves the muzzle, increasing muzzle velocity, and impact energy. Up to a point, of course, a barrel that's too long means you lose energy to friction and your barrel pressure turning into a barrel vacuum. It's what John Wick was using in the 3rd movie in the lobby shootout scene. It's just that I can't not ever think of the old Portal fake commercials whenever I see something like that.
|
|
#
?
Sep 14, 2020 20:22
|
|
|
Twibbit posted:unfortunately you do lose that near perfect Ambient occlusion proper RTGI generates, Also indirect shadows are going to be way less accurate, but we have gone so long without indirect shadows even existing that I can see most people being willing to sacrifice that. I don't see how you can do cheap recursive bounces with per-pixel RTGI, you'd end up re-introducing screenspace artifacts because the data available in the previous frame is view dependent The closest I think you could get would be to have an Ultra GI setting that does per-pixel GI for the first bounce then falls back to RTXGI for deep bounces, and just use the RTXGI + SSAO/SSGI on lower settings AFAIK Lumen uses a probe structure similar to RTXGI to do their recursive bounces, though there's still not much information on that repiv fucked around with this message at 20:44 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 20:26
|
|
|
SwissArmyDruid posted:That's fine, there's nothing wrong with that. They're called PCCs, pistol-caliber carbines. Extending the length of your barrel gives the powder more time to burn, expand, and build pressure behind the bullet before it leaves the muzzle, increasing muzzle velocity, and impact energy. Up to a point, of course, a barrel that's too long means you lose energy to friction and your barrel pressure turning into a barrel vacuum. It's what John Wick was using in the 3rd movie in the lobby shootout scene. The problem is what is shown is not the bullet, it is the entire round. 
|
|
#
?
Sep 14, 2020 20:28
|
|
|
Counter point: It's the future It's a video game It looks cooler with the reflections.
|
|
#
?
Sep 14, 2020 20:29
|
|
|
SwissArmyDruid posted:That's fine, there's nothing wrong with that. They're called PCCs, pistol-caliber carbines. Extending the length of your barrel gives the powder more time to burn, expand, and build pressure behind the bullet before it leaves the muzzle, increasing muzzle velocity, and impact energy. Up to a point, of course, a barrel that's too long means you lose energy to friction and your barrel pressure turning into a barrel vacuum. It's what John Wick was using in the 3rd movie in the lobby shootout scene. what about a carbine that shoots pistols? the pistols can shoot knifeguns also
|
|
#
?
Sep 14, 2020 20:31
|
|
|
When I say brute force I mean it would be interesting to see how it does without the secret sauce of dlss. I'm not saying dlss is not the future its just that its using magic rather than grunt. Would be a direct showing of how much the actual improvement is.
|
|
#
?
Sep 14, 2020 20:35
|
|
|
Seeing the RT-enabled performance without DLSS is also an optimistic preview for how raytracing will perform on RDNA2 
|
|
#
?
Sep 14, 2020 20:41
|
|
|
Bigsteve posted:When I say brute force I mean it would be interesting to see how it does without the secret sauce of dlss. I'm not saying dlss is not the future its just that its using magic rather than grunt. Would be a direct showing of how much the actual improvement is. There were benchmarks done without DLSS, I think it was like 80% faster in Doom Eternal lol Not at all, but it isn't supported in every game so you can't count on it by default vvv mobby_6kl fucked around with this message at 20:47 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 20:41
|
|
|
Bigsteve posted:When I say brute force I mean it would be interesting to see how it does without the secret sauce of dlss. I'm not saying dlss is not the future its just that its using magic rather than grunt. Would be a direct showing of how much the actual improvement is. I don't understand the people saying this. DLSS is part of the "actual improvement". They've added extra silicon specifically for the calculations behind DLSS - more, better Tensor Cores. I get that not every game supports it, but not every game supports RTX either. Do people feel that DLSS is somehow "cheating" in the way that it works or something?
|
|
#
?
Sep 14, 2020 20:46
|
|
|
AirRaid posted:Do people feel that DLSS is somehow "cheating" in the way that it works or something? People do indeed seem to feel that. I think it’s mostly predicated on an outdated mental model of what “real resolution” is, but a lot of people seem to want to exclude DLSS from performance measurements and it can’t all be trying to keep AMD from getting crushed savagely. I don’t recall, were people like this about shaders when they first appeared? “Sure you can do these effects in real time now, but games don’t support it yet!”
|
|
#
?
Sep 14, 2020 20:48
|
|
|
.
|
|
#
?
Sep 14, 2020 20:51
|
|
|
I don't think that's really it. DLSS support is going to continue to be extremely limited for some time, so why put emphasis on how much it has improved? Sure it is a good metric for improvements from RTX2000's, and the technology will be helpful, but right now direct comparisons without it is more valuable.
|
|
#
?
Sep 14, 2020 20:52
|
|
|
People were luddites about deferred shading, I remember nerds spitting feathers over post process tricks like SMAA and the like. Just shout them down and reminisce fondly about them in a few years time
|
|
#
?
Sep 14, 2020 20:53
|
|
|
AirRaid posted:Do people feel that DLSS is somehow "cheating" in the way that it works or something? I can understand the issue: if you're measuring pure "brute strength" on a more or less level playing field, you can be reasonably confident that you'll get the same performance improvement regardless of what game you throw at it. Relying on DLSS or looking at RTX speeds doesn't necessarily tell you a whole lot about what the performance is going to be on games that don't utilize them. Now, we're getting to the point where these cards are so fast, that they're going to be more than what you need to push games without RTX enabled in the first place, rendering that issue somewhat moot except in the edge cases like FS2020 or MSFS. Still, people play those games, and it's reasonable that they'd be interested in seeing how much improvement they can expect when locked in engines that won't be supporting any fancy technologies any time soon. But, yeah, you're missing the point if you're complaining about DLSS charts because you plan on playing a game that supports it but want to run with it disabled for some reason. Voxx posted:I don't think that's really it. DLSS support is going to continue to be extremely limited for some time, so why put emphasis on how much it has improved? Sure it is a good metric for improvements from RTX2000's, and the technology will be helpful, but right now direct comparisons without it is more valuable. Uhh, are you kidding? Because it lets NVidia reasonably slap their name on charts showing them being enormously faster than anything AMD could possibly hope to put out, as well as enormously faster than anything NVidia has previously put out, which is a hell of an encouragement for people to upgrade off their 9- and 10-series cards. I mean, it's fine enough to say that a 3080 is 50% faster than a 2080 or whatever. To say that it'll be 150% faster than a 1080 (or whatever the actual number is) in some of the biggest name games coming out in the near future is a HUGE difference. DrDork fucked around with this message at 20:57 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 20:53
|
|
|
To me, I think just asking "hmm what does it do with DLSS off?" just sets up the mental framework to compare to AMD, since it's something that the next-gen AMD cards simply won't have. Essentially, "how will this stack up against the competition, ASSUMING that the proprietary thing isn't involved?" (Which, at the moment, is a reasonable assumption for a lot of games). Granted, I don't really even have a horse in this race - I'm not interested in anything non-EVGA until/unless someone gives me a compelling reason to be, so it's not like I'm going to use that matchup for any sort of real-world decisions. It just helps me understand where the two companies' products are, relative to each other. I mean, yeah, I want to see benchmarks utilizing DLSS for games that support it, because that's gonna tell me what my expected performance will be. I just wanna see both numbers.
|
|
#
?
Sep 14, 2020 21:01
|
|
|
some people do feel it's cheating (and will until AMD implements it, much like the arguments that RTX was pointless and a complete waste up until it came out that AMD was putting it into RDNA2) but there's also the question of adoption. At the end of the day it still is only like 7 games (although now that it actually improves quality and now that it doesn't need to be trained per-game we can probably expect that to rise significantly), and people do want to get an idea of how it's going to perform on legacy titles/etc. at this point we basically need both imo, DLSS performance is very interesting for the future and pure raster is important for legacy. (to some extent though there isn't just "one" DLSS. you can set the internal resolution to basically arbitrary levels, not just the fixed 1440p->4K upscaling or whatever. iirc at one point someone was poking around with this with Control via nvidia inspector or something and they basically found that they could still get a pretty usable 4K image even at loving 240p, they were literally rendering the game at 320x240 and scaling it all the way up. how do you bench that? basically you can get near-arbitrary amounts of performance depending on how low you are willing to go, obviously a game has one "normal" setting but that normal setting could vary by game, maybe one game renders at 720p internally and another renders at 1080p, so card vs card scaling could vary dramatically between games. Or you could have "framerate targeting" where you say "I want this to run at 140 fps for my 144 hz monitor" and it plays around with the resolution target in realtime, so the framerate is fixed across cards and the quality varies. So there is also an aspect of "this is a giant loving mess and reviewers haven't really hammered out a coherent protocol to address it, because games don't even know what they're doing with it yet".) "cheating" is an interesting topic though because basically all of computer graphics is cheating by one perspective or another. Checkerboard rendering is sub-native. Effects or physics are often rendered at sub-native (or at a slower framerate). Non-raytraced lighting is a giant steaming pile of cheating. We just have come to accept certain forms of cheating as being normal and good. DLSS is another way for game engines to cheat and get a bunch of "free" performance, it's not really the cards cheating (and AMD will eventually implement it, via DirectML or whatever else, they just won't have dedicated hardware this generation probably.) We've actually already seen the first title that basically requires DLSS btw. Control is basically designed to use DLSS and runs like rear end without it. Can you play it without it, yeah, but it's like sub 50 fps at 1080p with a 2060, and sub 40 fps with RTX on medium. At 1440p a 2080 is 54 fps with RTX off, and 42 fps on medium. Wirth's Law in action, even without RTX the games are immediately soaking up all the extra compute power and just running like rear end anyway. Paul MaudDib fucked around with this message at 21:16 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 21:03
|
|
|
Zarin posted:I mean, yeah, I want to see benchmarks utilizing DLSS for games that support it, because that's gonna tell me what my expected performance will be. I just wanna see both numbers. We'll get those numbers from the actual reviews. But it's silly to think that NVidia wouldn't be pushing the tech and games that make them look good. I mean, AMD kept touting AoC framerates for the same reason: everyone wants to make their stuff look the best that they can so people will buy it.
|
|
#
?
Sep 14, 2020 21:04
|
|
|
SwissArmyDruid posted:That's fine, there's nothing wrong with that. They're called PCCs, pistol-caliber carbines. Extending the length of your barrel gives the powder more time to burn, expand, and build pressure behind the bullet before it leaves the muzzle, increasing muzzle velocity, and impact energy. Up to a point, of course, a barrel that's too long means you lose energy to friction and your barrel pressure turning into a barrel vacuum. It's what John Wick was using in the 3rd movie in the lobby shootout scene.
|
|
#
?
Sep 14, 2020 21:39
|
|
|
CaptainSarcastic posted:The problem is what is shown is not the bullet, it is the entire round. It's a feature, not a bug.
|
|
#
?
Sep 14, 2020 21:45
|
|
|
Not sure what to make of that FLIR video posted earlier. Am wondering whether I should just try going for an AIB card with triple fans (like that Asus TUF, 2.7 slot thick). The FE reviews are going to release tomorrow, right?
|
|
#
?
Sep 14, 2020 21:54
|
|
|
Combat Pretzel posted:Not sure what to make of that FLIR video posted earlier. Am wondering whether I should just try going for an AIB card with triple fans. The FE reviews are going to release tomorrow, right? How to feel about the FLIR video: The parts of the graphics card designed to get hot and dissipate said heat, get hot! That's kind of it. There's no real data around it to really form any other opinion from it.
|
|
#
?
Sep 14, 2020 21:55
|
|
|
AirRaid posted:How to feel about the FLIR video: The parts of the graphics card designed to get hot and dissipate said heat, get hot! The other salient point is that the whole thing gets hot, unlike a lot of normal AIB cooling solutions. This may mean people have to be more careful with cable routing to ensure they're not touching the card instead of being able to basically drape poo poo wherever. The point temps being in-band with existing cards suggest that, whatever the TDP of that card actually ends up being in practice, the FE cooler does a good job getting it out of the die. Also looks like it'll be real toasty for any card directly below it, regardless of what those fans are doing, but we pretty much already knew that was going to be the case.
|
|
#
?
Sep 14, 2020 21:58
|
|
|
I don't know. I've been looking at the TUF card gallery, editing the URLs you can get high resolution pictures, and the cooler looks to me making large contact to all important parts, including VRMs. And it's pretty oversize and has three fans blowing on the heatsink instead of just two. I hope some of the reviewers has reviews for both the FE and the TUF online tomorrow, to see if there's a considerable thermal difference. I keep mentioning the TUF because it looks like one of the visually least offensive AIBs. --edit: I guess one might argue that the AIBs typically blow the hot air back onto the PCB considering the heatsink layouts, and that's also not desirable. Combat Pretzel fucked around with this message at 22:03 on Sep 14, 2020 |
|
#
?
Sep 14, 2020 22:01
|
|
|
|
| # ? Jun 5, 2024 21:55 |
|
|
the new embargo for FE is the 16th, AIB is still the 17th - not sure what you're expecting tomorrow
|
|
#
?
Sep 14, 2020 22:05
|
|
