Yeah, these guys want to make money. Or as others put it, are "greedy". So the only way RT is shoved in our throats is if Nvidia pays them enough money to do so.

i am yet to see an actual game implementation (seen gameplay video comparisions on youtube thats it nothing eles)that makes me go AW $HiT I NEED RTX CARDS now Physx made me go wow but we all know how that went. I know i know RT is in its in

infancy and what we are seeing is basic tweaks to allow some sort of RT Implementation ONLY Dynamic Shadows Or GLobal Illumination Or Reflections. purely eye candy that made me go meh i'll not notice it 5 mins later and my gaming experience

hasnt changed at all. Physx did affect it to certain extent when used superbly as it was in Assasins creed Black flag and Batman Arkham series even with metro last light and 2033 but thats about it. Never really used to its true potential

now sure Realistic Graphics are always gonna blow our mind but at the cost of significant frame rate? while not really looking that different so till we reach something that blows my mind i aint running out to get a RTX card. heck if anyone

that would be a day when we have crazy computing power at our disposal like in 4k 60-90fps high or ultra on a entry level potato machine cause thats the level of computing power needed to run full RT at 1080p 60 fps low-medium

at the end of the day whats easier to sell Ray Traced Realistic shadows / reflections or 8k 120FPS just look at PS5 ive heard 8k more than ray tracing being said by the Ps5 Architect.

8k is just the display output bandwidth, it's not going to be for games or maybe a couple small games

See the problem is that low-end GPUs will still have ray tracing cores, which means that instead of more cuda cores there are ray tracing cores. And ray tracing cores are Application Specific Integrate Circuits - ASICs, have very limited usage.

On top of that Ray Tracing requires the GPU to render the entire environment around you and store it into memory all the time for all the lighting, shadows, reflections, refractions, shading and so on to work, so not only is the method more demanding on its own, but it is also a lot more demanding due to rendering a ton more for it to work.

And visuals don't make a game ground-breaking, good gameplay, mechanics, level design, story with good amount of content and good quality to it do. Literally Crysis is a great example of that. Crysis 1 was loved not because of its graphics, but because it was genuinly very good, it had a unique and innovative gameplay, great level design, great gunplay, great missions, it was long and in general it was amazing. Both Crysis 2 and especially Crysis 3 had much better graphics but they weren't nearly as good as Crysis 1 so they sold less and they were not even close to Crysis 1 in terms of quality and quantity.

this is a dumb question more like a statement as it pure economics and no one would buy it due to the pricing and utility value. but wont a Standard Graphic Card design MORE CUDA COREs at faster clocks over the larger die as the RTX die is

Crazy huge and then have a separate card with RT CORES on it and the AI cores as well or maybe skip them cause upscaling really? and then essentially give the ability to Have it run in a DUal Graphic Card mode? and as its only RT CORES even

more cores and much better performance. heck make the thing backward compatible and let me plug it up with my GTX 1080 or let IT go compatible with AMD CARDS and Hijack their cards like they Hijacked AMD FREESYNC? sure like i said a card li

like that would be crazy expensive and no one would really buy it due to the little games that support it and its exorbitant price. but then Nvidia could i theory have a GPU that has 4K 60-100 FPS capable Card and a RTX card capable of prov

providing some killer Visuals!! heck i know many Animation Studios would Dro tons of Money for RAY TRACING ONLY CARDS imagine how much it would lower the cost and time of movie production. Ive seen a SIngle 2K frames render for hours if it

Dude you went all over the place. XD

Ok from what I'm getting, you are asking if ditching the RT cores and Tensor cores for more Cuda cores would result in better performance and scaling? The answer is yes and no. It will be much better general purpose GPU performance(I say general purpose GPU performance as GPUs are ASICs too, but much less than ever before) and would be much better than having RT and Tensor cores.

The thing is though Cuda cores do cunsume a bit more power than RT cores and especially Tensor cores, so a 754mm^2 die like the rtx 2080Ti would consume more power and heat, probably a 350W TDP card which in the high limit of standard GPU air cooling. But a 650mm^2 die GPU at 12nm should be 250-275W TDP.

And the problem with dedicated RTX cards for companies is that Nvidia are absolute Arseholes to work with and unlike normal GPUs Tensor Cores and Ray Tracing Cores don't have such wide and vague patents as the regular GPUs and on top of that Nvidia is not the first to have Ray Tracing and Neaural Network ASICs, so they can't monopolize them, so many other companies can make their own AI and Ray Tracing ASICs and many have.

Biggest example being Tesla. Nvidia tried to milk them with their AI chips and the folks at Tesla were no fools and they ditched Nvidia all together, made their own AI chips that are much better than Nvidia's and are now using them.

Otherwise for regular consumers dedicated Ray Tracing cards are a better solution, but we don't need ray tracing at all.

Designing Ray Tracing ASICs and ASICs in general is much simpler and cheaper and faster than designing GPUs or CPUs. Just like bitcoin ASICs. Ray Tracing has very generic instructions and very few at that, so it's not like it's expensive to develop. AI is harder, but again the math and logic behind it is not even of the complexity of the GPUs which are much simpler than CPUs.

Back on GPUs(My turn to go all ove the place). A 650mm^2 on 12nm (example name)gtx 2080Ti with ONLY Cuda cores(SMs to be more accurate as there are more things than just cuda cores), would be at least 40% faster than a gtx 1080Ti at 417mm^2, considering the gtx 2080Ti wouldn't clock as high as the gtx 1080Ti

Though sadly due to amdahl's law, going above 6000 cores for GPUs starts to have a big diminishing return as even graphics and graphics cards are not 100% parallel due to overhead, latencies, stalling and in general graphics not being truly 100% parallel.

That's why AMD went for higher transistor count Stream processors(CUs) for Navi and thus higher IPC, instead of putting more cores and that's why Nvidia went for higher IPC with maxwell and higher clock speeds with pascal and turing instead of putting more cores.

Also when it comes to ray tracing I remember the 5120core voltage being about as good as the rtx 2080Ti in ray tracing games, so you can definitely brute force through ray tracing if needed. XD

interesting I was always under the impression that Graphics Scalabilty was much rather simple number of Cores+clock speeds sure i knew that certain types of cores are more jack of all trades and certain are much more specialized and super f

Manufacturing nm improvements but this is news is that why MULTI GPU sucks that much? i was always under the impression it was simply the optimization sure you cant get 100% 2nd gpu performance simply down to communication between both the

cards but at 80% performance? ive seen Multi GPU performance to really be meaningless and as you add more cards the performance gets even crappier Ive seen a max 4 GPU system and as he added more the performance for the additional gpu start

started going even lower that being said this was about 2012 I guess i think AMD had the 7000 series at the time i wanted to go multi gpu the shopkeeper demoed it on his system to say why we should avoid MULTI GPUS and i ended up picking th

HD 7950 which was later rebranded to rx 280x (I Guess) is this why we keep seeing so many Architectural changes in GPUS so often as opposed to CPUS? to try and get an IPC boost?

The Next GPU I consider I would want would have to have 1080p at 120fps ultra stable no issues sure a dip to 100 is realistic but thats it. then again with Next Gen consoles about to register a massive push i think Game Engines will make th

Zen 3 is rumored to have 4 threads per core? can it be possible with Zen 4 or Zen 5 that they improve the quality of cores? or will we simply have to wait for a whole new Architecture then? at this point I am quite significantly out of my

depth in CPU core knowledge but i am guessing what your saying is that Intel cores are so good because they are huge? I am guessing they are huge because of their Desktop Architecture philosophy of More faster Processing rather than More Si

Simultaneous Processing which would make sense as to why their performance can change so dramatically with clock speed changes and their Processor size is a effect of that and not a cause. NO doubt they have tried to make multi core cpus on

singular dies. now if I understood you correctly then If AMD eventually move to a smaller manufacturing process 5nm? would it be possible that amd rather than increasing the number of Cores like they did from their jump From 14nm to 7nm by

going from 4 cores per chiplet to 8 cores per chiplet that they would then have the ability to maintain the 8 cores but try and improve their Single COre Performance? THat being Said isnt AMD's current SIngle core IPC comparable to that Of

It's not just bigger cores. And IPC is NOT just a single metric unit, it's a combination of how many of each x86-64 instructions the CPU can do per cycle and for cache bandwidth and latency.

If for Zen1 there is as good of an optimization in compilers, frameworks, engines and no on as for Intel's architecture, it would have higher IPC overall than Coffeelake, excluding AVX 512, encryption and L1 cache bandwidth.

And intel's cores have less transistors than Zen, they are just bigger due to intel's 14nm+++ being bigger than TSMC's 7nm. When AMD was at Glofo's 12/14nm Zen1 and Zen+ had about as big cores as Coffeelake because it lacked 256 bit FPUs and used more 128 bit FPUs.

More of the same execution units + optimization of the architecture to compensate for the extra execution units(branch predictors, schedulers, out of order execution, etc, etc).
But then there is the other part, adding new instructions, which again increases core sizes, but not necessarily general purpose performance as it takes at least 3-5 years for new instructions or even improved versions of previous instructions to be utilized by compilers, engines and developers as whole.

For example both AVX 256 bit and AVX 2 512 bit were unused for a number of years outside of synthetics and some custom vendor software.

Overall bigger cores(as in higher transistor count cores) = better cores when done right and when not introducing new instructions.

AMD's clock speeds are entirely limited by the process node(Global Foundry's 12/14nm and TSMC's 7nm) the transistors of those process nodes are just too slow.
People may bash intel's 14nm+++ but it's only 20% bigger than TSMC's 7nm and only 15-25% less efficient, while being faster and scaling better.

AMD currently has a big decoder bottleneck in their Zen1 and Zen2 CPUs, but instead of solving it and getting 33-50% higher integer IPC, they are just making their cores retain the same Transistor Count, but shrink the die size through new process nodes, but doubling that L3 cache was a dramatic increase in transistor count and it's primarily for server use.

Zen2 is for all intents and purposes a server CPU first and a desktop/regular consumer CPU second.

And Zen 3 having 4 threads per 1 core just further solidates that AMD cares for the companies(for servers basically) and not the regular consumer and I'm not surprised that's where the majority of the money for CPUs are and soon for GPUs too...

As companies don't need single core performance nearly as much, if at all compared to regular consumers.

And I hope that Zen4 will have at least solved the decoder bottleneck, but I doubt it, that's why they are most-likely designing Zen3 to have 4x threads per 1x core.

And what sucks is that AMD's SMT is fake and castrated. For example IBM's SMT for each thread you get 90-100% extra performance as they have true SMT with true superscalar pipelines that work in parallel with advanced scheduling to fill the empty pipeline stages.

What AMD and Intel have are just the advanced scheduling that fill the empty pipeline stages and instead of making it hardware level, they have a simpler way where the bios and OS kernel see the extra schedulers and scheduling as extra threads and AMD's (fake)SMT gives up to 30% higher performance per core, while intel's Hyper-threading(they didn't call it SMT as they knew it's not) gives up to 25% higher performance per core.

And in general single-core performance should be tested with the (fake)SMT/HT included.

But again we have no hardware level parallelism yet(IBM is working on it for example and some 3rd party engineering companies that are not fortunate enough to make their own CPUs, but do such designs that other companies license or even straight up borrow), so software optimization is needed for those extra threads to be used as if they were extra cores, which means that Amdahl's law is much heavier.

Amdahl's law when applied to CPUs and GPUs and other ICs expects perfect hardware level parallelism, so for CPUs especially adding the software optimization factor means that it's even worse than the deminishing return is even worse than it should be.

thanks mate this was very educational I wasnt aware about Real And Fake MULTI THREADING. but how come Intel after coming up with Hyper Threading quite sometime ago havent come up with TRUE SMT yet or somthing rather near it or rather the Ne

next step of Hyper Threading or are they only aiming for incremental updates with generations to get closer to it?

Hyper-Threading and AMD's (fake)SMT is just half of the scheduling of real superscalar SMT.
And I doubt we will any time soon though and even if they do, we first need hardware level parallelism. And then again above 16 threads is pointles for 99.9% of the software, games absolutely. Amdahl's law can calculate it. And even 16 threads won't give 100% scaling.

lol

... limited amount of raytracing games do today, and you really need to drop to 720p with 2060 for it to be viable. And only 1080p 60FPS raytracing card they have is 2080Ti and I would dare to say that you must be insane to buy 2080Ti to do 1080p. Thirdly, even developers don't like it, because they still have to do quite a bit of work, it is not that "it just works" thing,...

... there already was one game that canceled raytracing implementation, but I just can't find which one it was... Google is flooding my searches with Minecraft results. Anyway, what developers are waiting is more open implementation they can use. Not this heavily proprietary stuff from company that likes to push people around. And when we get that, I doubt current RTX cards will run it well.

The value is in the eyes of the beholder. If you don't like it - you won't find value in it and you will look at it from the eyes of that person (which is what you're doing). If you like it - you'll see the value.

Now greed is a big part of it. But if games costed 90-120$ since 2012-2013 then there would be serious backlash for micro-transactions.

And Fortnite didn't have micro-transactions until it went free to play with the battle royale game mode. And those micro-transactions are sadly normal for F2P games, but the battlepass system in fortnite is the least terrible micro-transaction system till date. You can literally earn it back by playing the game and just pay once for it and get all the future ones for free with enough game time. Even CoD is implementing it now.

Otherwise, as game dev costs increase and the price stays at static 60$ they will have to cut up the game for DLCs, introduce micro-transactions and loot boxes.

For example, Square Enix lost money from Rise of the Tomb raider and Shadow of the Tomb Raider, even though both sold a few million copies in their first year. They barely broke even with Tomb Raider 2013. Keep in mind that the average price for a AAA game in it's first year is 45-50$, not 60$ as not everybody gets it on launch or the first couple of months and unlike steam, game prices on consoles start going down after the second month.

Sales beyond the first 6 months don't matter much either. Rise of the Tomb Raider as of late 2018 has sold only 3.5 million copies for a 3 year duration...that's just not enough cover a game made by 300+ people with a minimal(bottom 20%) average developer/employee cost of 10 000 usd a month(120 000 usd a year) per person for a 2.5 year development cycle.
And then have some money for the next more expensive game.

Otherwise I used to care about the graphics too, at the bottom of my priority list, but still I did, until I found out that thanks to them games cost hundreds of millions of dollars...
I used to think that just like GPUs and CPUs graphics get better and remain at the same price or even lower price as time went on, because of all the improvements in the graphics tools, I was wrong.

And games are meant to be played, not watched, that's what movies are for. We as gamers NEED for them to scale down game budgets back to the mid 2000s or we will keep getting simpler, blander, more watered-down games with each sequel or new IP.

The majority of games have barely, if at all improved since the mid to late 2000s be it sequels or just games similar to those back in the day. Only the graphics, but on the other hand the quality and quantity of content has suffered. And as I said above the gameplay and mechanics have barely improved. There are exceptions of course, but in general they haven't.

All for what? Graphics? Nah... but again people prioritize graphics for some reason, as if graphics make a game better. :/

Games can certainly be jaw-droppingly gorgeous without being cutting edge just because of the sheer amount of time, effort, and money spent on its artistic aspects. And the reverse can be true as well.

Higher quality graphics take more time to be drawn/scetched or whatever in the 3d modeling softwares, animated and integrated and it takes more advanced and complex engines to run them. It's like drawing an anime character vs drawing a realistic fully detailed human, except it's even harder and slower.

And many people don't seem to pay attention to how much visual objects there are in modern games compared to older games. Nowadays there would be hundreds of objects on the screen at a time, back in the day it was basically a tunnel, or if it was opened world it was flat or almost flat ground with textures and textured rectangles, unless it was a more important building then there was som depth to it

Nowadays modeling a can of gas would take longer than modeling a building 15 years ago

And sadly the tools for making graphics haven't gotten much better, or should i say they have gotten much better, but graphics have outpaced them tenfold.

As I said nowadays there are hundreds of artists and animatiors and motion movement engineers and cutscene artists, whereas back in the day 10-30 were enough for an entire AAA game.

And I agree, I personally prefer good art style and graphical design with much more unrealsitic and lower quality and technologically worse graphics to realistic graphics even if they get the realistic graphics to be indistinguishable from real life. If anything realism shouldn't be the golden standard in graphics, creative and unique design and artstyle should be.

But that said the general public cares only for the technologically advanced graphics

And ray tracing won't make it faster to light up a place at all, if anything it will be slower, until some automated system based on AI or algorithms is made, but that can be said about the current fake methods of lighting, reflections, shading, etc, etc.

Nvidia's demo of just plop the light source in and it's done is just not realistic to anyone who wants to make good looking environments. You can do that in source engine too and yeah it won't be in real time, it's source engine after all, but it will render it and boom, light source with reflections and shading, but it won't look good. Same will be with Ray Tracing, if not harder as it's more complex and it actually makes surface properties of textures matter much more and much more complex and objects become light sources of their own instead of just bouncing off them and what-not. I haven't used ray tracing and I don't know the details in depth but it's far more complex than what we currently use.

I don't know the details indeed, but Nvidia's demonstration of just putting a light source in and being done with it is unrealistic.

It really isn't once you know how it works. Presentation is enveloped in marketing speak but it's not that far removed from the actual use Real time RT it can provide.