Calling all low and mid GPU owners - shall we swap RT for more performance or lower prices?

[Aquinus]

I think the poll sort of says everything. A nice whopping 11% actually care about RT whereas there is an event split for the rest wanting more performance or more cost effectiveness. To be honest, that doesn't surprise me. I couldn't give to craps about RT. All I want is a cost effective upgrade from my Vega 64 should I ever be in a position to build a new machine. That means a reasonable price and more performance than what I have now. If it comes with RT, fine, but it's not a requirement by any stretch of the imagination. This difference probably becomes even more stark when you start talking about mobile GPUs.

All in all, I still think RT is a niche market and nVidia is really trying to change that. I can't say it's succeeding though.

 

[Onasi]

Ultra settings don't make as much sense at 4K as they do at 1080p or even 1440p. You don't need that much antialiasing for example. With DLSS/FSR/XeSS advancement, it's also no more mandatory to play at native 4K as it occasionally looks worse than DLSS (and I expect it to become widespread by late 2020s), and yeah, performance gains are much more visible than supersampling artifacts.

Antialiasing is the least of problems, actually. We are no longer using forward renderers and MSAA is dead. Modern temporal methods are almost free performance wise, but you do pay in ghosting and artifacts, depending on implementation. Mostly the expensive settings are now stuff like volumetrics, global illumination, shadows, complex AO and so on. Whether those look better in 4K and lowering them makes or does not make sense is subjective.

 

[AusWolf]

As long as it takes. What we are fighting for is the future of "gaming" design, not AI designing disguised as gaming "advancement". If we are not lied to, we might support it.

The only choice you have, is where you spend your money, unfortunately. If you disagree with a GPU design, or philosophy (although I'd argue not to buy PC hardware based on philosophy), don't buy it. Although, this won't stop the masses from spending a month's salary on the latest Ultra Super RT GPU just because its model number is one higher than the one they currently have.

 

[Dr. Dro]

It seems to me that you are very good at constructing long texts devoid of any argument other than saying that everyone who thinks for themselves and questions is a faboy and absolutely everything Nvidia does is right, and the future undeniable, unquestionable. A suggestion: you might as well replace the text with a photo of you kissing a Huang statue, achieving the same effect!

RT proves inefficient, consuming die space, energy, and resources, particularly in the specified context. With manufacturing processes becoming increasingly expensive and cache no longer substantially shrinking, proportional advancements are unlikely in the foreseeable future. Given the current stage, as we're beyond the beginning of the century, the imminent challenges make it apparent that there's insufficient margin for genuine RT viability in average consumer GPUs.There is no counter argument other than generic texts because no one refutes reality.

Thanks for proving my point. It's just clubist resentment after all. Mind you, I'm the same guy who openly throws the fact that the 4090 itself is a low quality die, but nah, I'm just very good at long winded texts to defend a corporation lmao

 

[AusWolf]

The title poses a bit of a trick question. The reason it is, is because it needs to be understood why silicon is designed the way it is - both from a technical and from a marketing and manufacturing perspective.

When a company like NVIDIA approaches silicon design, things like RT cores are not being considered 'extra' or 'bonus' or 'optional'. These things are in the very basic building blocks of a unified and scalable architecture. The outlier being GTX 1660 and the transition period between 10 series and 20 series.

Typically, when making smaller dies, the question isn't if RT core real estate can be replaced with more traditional CUDA cores, but typically what size silicon and what performance output you want out of a product - since the building blocks already exist and include those RT cores. Not using them or designing special small cores that do not include them might in fact be very cost inefficient. You will have to reroute and redesign the architechture to not leave any wasted space. The only decision left for the silicon design team is when to press the go-ahead and implement these kinds of cores as a part of your building blocks - a decision made by NVIDIA earlier, and by AMD a bit later with RDNA.

As a result, the entire stack is ought to get RT cores, regardless of how viable the performance is when we talk about more entry level products such as the RTX 3050 or even the RTX 4060. The idea behind persisting with having all the none-traditional type of compute transistors in the silicon is that as the technology matures, and as using the same building blocks help shaping products, what didn't use to be that viable will start being somewhat viable. There's quite a difference between how an RTX 2060 deals with Raytracing and how an RTX 4060 does in terms of end-result.

So yes, it sucks that my 200mm^2 GPU can't really handle Raytracing to a satisfying level (to me) and instead of more theoretical CUDA cores i get silicon space dedicated to RT. But, that's how a master-design works with scalable silicon, and that's NVIDIA's (or AMD's decision, if we were to talk about RDNA/2/3) in order to be able to properly work their design teams and their rented fab nodes.

Eventually, more and more none-gaming applications are also starting to use these silicon real estate parts, like modeling, architecture (houses and apartments in this context) and many other types of CAD, simulation and engineering tasks. By now, everybody is implementing RT-intended hardware in their silicon including ARM based mobile SoCs. Its the current natural progression of GPUs.

It's all about costs and benefits. If it would cost Nvidia or AMD 50% more to redesign an architecture to cram 20% more shader cores into a chip instead of RT or AI, they won't do it. The GTX 16 series is an outlier. Somehow I doubt that Nvidia made truckloads of money on it (although personally, I think the 1660 Ti was quite good, and even the 1650 was a decent upgrade over a 1050 Ti).

 

[Beginner Macro Device]

Whether those look better in 4K and lowering them makes or does not make sense is subjective.

In my case, these settings at High mostly look better than Ultra at 1080p (didn't own a 1440p display though) and it prevents about 10 to 30 percent of GPU punishment depending on a game and location. Considering I can't quite tell the difference between native and FSR: Quality at 4K (if we don't count buggy implementations like that in the CP2077 game for example) it's another 30 to 40 percent performance boost. This way 30 FPS become 50. And at Balanced, it's still far ahead native 1080p Ultra in terms of picture quality, and most importantly, I'm having this 60 FPS experience. Going UW1440p would be better but that 4K display was $150 and honestly, I never felt like it's possible for such a cheap device to be so good. Not a single regret.

Although, this won't stop the masses from spending a month's salary on the latest Ultra Super RT GPU just because its model number is one higher than the one they currently have.

I just know the GPU I wanna upgrade to doesn't yet exist so I'm waiting for future generations. Yet I'm about to buy an RX 6900 XT not because I want to upgrade my main rig but because I want to have a fast backup GPU (RX 480 → RX 6700 XT). And I also wanna butcher the said 6900 XT just for the memes.

 

[AusWolf]

And even cutting out both RT and Tensor cores would not bring performance boost to 30%. Price is a whole different discussion though, especially in that market segment.

We wouldn't save on price, either, as the money saved on manufacturing a smaller die would first have to be spent on designing it.

 

[Dr. Dro]

Sure its the future. But define 'future'. Might easily take another 10 years.

It took similarly long for DirectX 11 to be widely adopted. Budget concerns are what slows down progress at this point, making high end video games with advanced graphics has become as, if not more expensive than making movies and few studios can afford these multimillion budgets, and at the same time, having a great game with an alternative art direction or simpler graphics is quite feasible, so there you have it.

 

[AusWolf]

I want to have a fast backup GPU (RX 480 → RX 6700 XT)

I know that feeling far too well. I have a shelf full of backup hardware. Some of them are actual backup stuff (like my 6500 XT), but others are mainly of sentimental value, like my half-height, half-length, LP GT 710.

 

[Vayra86]

It took similarly long for DirectX 11 to be widely adopted. Budget concerns are what slows down progress at this point, making high end video games with advanced graphics has become as, if not more expensive than making movies and few studios can afford these multimillion budgets, and at the same time, having a great game with an alternative art direction or simpler graphics is quite feasible, so there you have it.

Its not just budget.

Look at the die size of a 4090. This is what Nvidia can do at this point. We rely almost exclusively on meagre shrinks now to move forward. That stagnation isn't a budgettary issue nor a game issue. Its even doubtful the hardware can provide. RT adoption isn't business 'as usual'.

Also, its definitely not a given that because a tech gets industry support and 'momentum', that it succeeds and is here to stay. Look at AR and VR, they're in that space too, but its not going places. We've had quite a few of these brainfarts in the days gone past. And most of the time the conclusion wrt these failures is 'It was too early / the tech wasn't ready / the market wasn't ready'. Its all more of the same though: some shit just doesn't stick.

 

[Onasi]

Thanks for proving my point. It's just clubist resentment after all. Mind you, I'm the same guy who openly throws the fact that the 4090 itself is a low quality die, but nah, I'm just very good at long winded texts to defend a corporation lmao

I am still finding it hilarious that AD102 has never been implemented as a full fat die into anything. Even insanely expensive professional cards don’t max it out. I assume the yields are gutter trash.

 

[Beginner Macro Device]

I know that feeling far too well. I have a shelf full of backup hardware. Some of them are actual backup stuff (like my 6500 XT), but others are mainly of sentimental value, like my half-height, half-length, LPP GT 710.

You're way more extreme than me in this department. I only have a one full backup system at once. Usually even less. Like today for example: Z490 Vision D, a couple decent PSUs, RX 480, a FullHD monitor, a couple damaged (but still acceptable for basic tasks) keyboards, and a little cheeky 250 GB HDD is all I have for my backup as of now. Gotta buy an SSD, a CPU, a mouse, and RAM to be 100% backed up.

I wonder would it make any sense to create dedicated RT devices. Like, you have your CPU running its CPU things, your GPU doing its raster and a little ray tracing here and there, and another PCI-e slot inhabitant that strictly enhances RT performance. Like ASICs but made for RT and not for mining, and mounted inside a gaming PC.

Or... might injecting RT cores instead of E-cores make LGA1700-alike CPUs more interesting..?

 

[Dr. Dro]

Its not just budget.

Look at the die size of a 4090. This is what Nvidia can do at this point. We rely almost exclusively on meagre shrinks now to move forward. That stagnation isn't a budgettary issue nor a game issue. Its even doubtful the hardware can provide. RT adoption isn't business 'as usual'.

The 4090 being severely cut down (almost as a much as a 6800 relative to the 6900 XT), power budgets thrown sky high to compensate should tell you a thing or two about yield at this scale, particularly considering not even enterprise has a maxed AD102 yet.

Budget concerns are quite real, RT adoption is actually supposed to simplify development and reduce budgets over time as well. There's even market concerns, not everyone has the latest hardware, and limiting your customer base is very unwise in a business sense.

Ada is the third wave of RT ready hardware from Nv, but Turing is similar to Evergreen that it pioneered the "generational leap tech" (with the ATI HD5000 cards it was DirectX 11), so normalizing for AMD's lag, let's say that the RT era truly started with RDNA 2 and Ampere roughly 3 years ago.

By the time the HD 5870 turned 3 in 2012, the largest majority of games still targeted DX9 on Windows XP, including major AAAs like Borderlands 2, while we were still in awe of quad SLI builds pushing Metro 2033 with DX11 tessellation.

If anything we are doing quite fine so far.

 

[Onasi]

I wonder would it make any sense to create dedicated RT devices. Like, you have your CPU running its CPU things, your GPU doing its raster and a little ray tracing here and there, and another PCI-e slot inhabitant that strictly enhances RT performance. Like ASICs but made for RT and not for mining, and mounted inside a gaming PC.

This is the same idea that was already tried with dedicated physics acceleration by Ageia before being bought out by NVidia. It’s not practical. The trend in computing for years has been heterogeneous hardware that can do all or most things by itself. Dedicated anything has gone the way of the dodo unless we are talking about having acceleration for certain tasks on the SoC like Apple and now Intel and AMD does or (surprise) RT and AI acceleration as part of GPGPU.

 

[AusWolf]

You're way more extreme than me in this department. I only have a one full backup system at once. Usually even less. Like today for example: Z490 Vision D, a couple decent PSUs, RX 480, a FullHD monitor, a couple damaged (but still acceptable for basic tasks) keyboards, and a little cheeky 250 GB HDD is all I have for my backup as of now. Gotta buy an SSD, a CPU, a mouse, and RAM to be 100% backed up.

I'm way too extreme to recommend others the same. I've got two HTPCs (one of them has an 11700 and now a 2070 in it), and about 2-3 backup CPUs and GPUs from each vendor, also several backup SSDs.

I wonder would it make any sense to create dedicated RT devices. Like, you have your CPU running its CPU things, your GPU doing its raster and a little ray tracing here and there, and another PCI-e slot inhabitant that strictly enhances RT performance. Like ASICs but made for RT and not for mining, and mounted inside a gaming PC.

I think the main issue would be PCI-e bandwidth and latency... although it might finally be a solid case for gen 5 speeds (unlike those ridiculous SSDs with slapped-on CPU coolers).

 

[BMfan80]

RT is the first thing i disable (even when i had my 4080) as its waay too subtle.
Maybe its my 43 year old eys but i can BARELY notice RT visually.. the performance hit i can notice though.

I'm in the same boat as you.

All the games I have played have been fine without RT, I would rather have a game play smooth with good visuals than a game that
plays like manure, but I can see a billboard "perfectly" on a window.

 

[Vayra86]

The 4090 being severely cut down (almost as a much as a 6800 relative to the 6900 XT), power budgets thrown sky high to compensate should tell you a thing or two about yield at this scale, particularly considering not even enterprise has a maxed AD102 yet.

Budget concerns are quite real, RT adoption is actually supposed to simplify development and reduce budgets over time as well. There's even market concerns, not everyone has the latest hardware, and limiting your customer base is very unwise in a business sense.

Ada is the third wave of RT ready hardware from Nv, but Turing is similar to Evergreen that it pioneered the "generational leap tech" (with the ATI HD5000 cards it was DirectX 11), so normalizing for AMD's lag, let's say that the RT era truly started with RDNA 2 and Ampere roughly 3 years ago.

By the time the HD 5870 turned 3 in 2012, the largest majority of games still targeted DX9 on Windows XP, including major AAAs like Borderlands 2, while we were still in awe of quad SLI builds pushing Metro 2033 with DX11 tessellation.

If anything we are doing quite fine so far.

I totally respect your view on it; I'm just a lot less certain of progress myself, but then again, I'm a pessimist by nature A big part of that view is also based on the state of the world and the pressure on chips for the last... well, many years really. Things are getting worse - the economical conditions too. Budget concerns won't get lower.

I am still finding it hilarious that AD102 has never been implemented as a full fat die into anything. Even insanely expensive professional cards don’t max it out. I assume the yields are gutter trash.

Yep, Nvidia is definitely going to have to adopt chiplet in gaming segment sooner or later...

 

[Deleted member 57642]

RTX is pure marketing for 2xxx series and low-mid range 3xxx and 4xxx. It's part of the card and you paid/pay for that feature - but can't actually use it (most games having this feature - are unplayable if Ray Tracing is activated on this card).

And yes - they people who design this cards "knew what they were doing". They knew this cards suck big time (again, practically useless) - but didn't even take the moral approach into account (to keep this feature for high end cards - or at least cards where this feature actually makes sense and can be used). Instead, they use this feature as "a paid demo" for high end or future models. As in, even with a low end RTX card - you might be able to activate Ray Tracing at the lowest resolution and get a 7 FPS Ray Tracing demo... where only the Full Card (cause low-mid range - are more like Sample Cards) - can use this feature at playable rates.

Last but not least - how many products have useless features - which make no sense, but they're still there? Clever design? You could put it like that - if you admire machiavellianism. Or why not, if you admire this guy:

... since it's the same thing.

 

[R0H1T]

making high end video games with advanced graphics has become as, if not more expensive than making movies and few studios can afford these multimillion budgets

Is it really? Why do end users still do beta testing when the game's released?

How many Marvel movies with crappy CGI do well?

 

[ty_ger]

Where is the option "Yes, I think RT is a gimmick."?

 

[jesdals]

The pursuit of incorporating Ray Tracing (RT) into games, started by Nvidia, comes at the cost of allocating space on its die exclusively for RT. This valuable space could otherwise be utilized for additional shaders, thereby enhancing overall performance. This trade-off is becoming increasingly apparent in the escalating prices of GPUs, driven by the soaring costs of chip production and development, and a divergence from advancements in density.

Achieving higher performance in Ray Tracing (RT) entails augmenting the dedicated hardware on the die for this purpose. There's no magic involved, and as a consequence, GPUs are likely to become so expensive that what is currently considered low-end will become the equivalent of today's mid-range pricing. Nvidia isn't dismayed by this situation because its profit is directly tied to a percentage of the price; the greater the price, the larger its share.

It is evident that low and mid-tier GPUs lack the necessary power to effectively handle Ray Tracing, except when employed in an extremely limited manner. I won't even delve into mentioning Path Tracing (PT) due to its resource-intensive nature. So, taking all this into account, answer the question below.

- Would you be open to sacrificing the capability to run Ray Tracing on mid and low-end(under U$500) GPUs in exchange for either a 30% boost in performance or a 30% reduction in price, while maintaining the same level of performance as the current lineup?

The very premise is a misunderstanding of market mechanism - competition give lower prices not restrictions like that. Just see how AMDs Ryzen have changed cpu prize/performance ratio - no one wants 4 core cpus today (even though they seem to reapear). So if you want to have more performance with lower price - skip the upgrade cycle or buy used, or buy a Intel GPU.

 

[Denver]

Thanks for proving my point. It's just clubist resentment after all. Mind you, I'm the same guy who openly throws the fact that the 4090 itself is a low quality die, but nah, I'm just very good at long winded texts to defend a corporation lmao

Yeah, you complain about Nvidia and still throw money into Jensen's pocket, great critic. it has everything to do with the topic at hand...

In essence, your argument boils down to the assertion that personal preferences shape opinions, with statements like "You're Y, so you dislike RT, and I'm X, so I favor RT." However, the focus remains on individuals rather than the underlying ideas. I've previously explained... and the reason why I find RT unfeasible and useless in the context presented is also in the initial post. It's clear that most seem to agree that more performance or cheaper GPUs are preferable to the ability to run RT on GPUs in this segment.

The very premise is a misunderstanding of market mechanism - competition give lower prices not restrictions like that. Just see how AMDs Ryzen have changed cpu prize/performance ratio - no one wants 4 core cpus today (even though they seem to reapear). So if you want to have more performance with lower price - skip the upgrade cycle or buy used, or buy a Intel GPU.

I agree that we need competition, but on the manufacturing process side, TSMC has been swimming alone for half a decade. When production and development costs escalate exponentially with each generation, and the density provided does not proportionally align with the price, companies operate based on the anticipation of requiring increasingly more funds.

 

[Beginner Macro Device]

you complain about Nvidia and still throw money into Jensen's pocket

Maybe, just maybe, because AMD GPUs are way worse?

 

[Vayra86]

Yeah, you complain about Nvidia and still throw money into Jensen's pocket, great critic. it has everything to do with the topic at hand...

In essence, your argument boils down to the assertion that personal preferences shape opinions, with statements like "You're Y, so you dislike RT, and I'm X, so I favor RT." However, the focus remains on individuals rather than the underlying ideas. I've previously explained... and the reason why I find RT unfeasible and useless in the context presented is also in the initial post. It's clear that most seem to agree that more performance or cheaper GPUs are preferable to the ability to run RT on GPUs in this segment.

That is true but until consumers vote with wallets that shit aint happening, or not a lot. Polls do not sell GPUs and people are hypocrites - or put differently, commerce is stronk.

Maybe, just maybe, because AMD GPUs are way worse?

Except they aint the only drawback is lower RT perf. FSR is catching up. You get more raster perf per $ and more VRAM. Im still problem free with RDNA3. No regrets after over 10 years on team green. Nvidia served me well, but Ill be damned before I start serving Nvidia, which is inherent to purchasing their product carrying the RTX sticker. They can stick their planned obscolesense right back in Jensens shit oven. I vote with my wallet.

Chiplet is the future of GPU progress and cheaper silicon. Not monolithic chips loaded with software to make them competitive. Nvidia wants to create a performance blackbox with proprietary bullshit. F.U.C.K. That.

 

[evernessince]

It's 100% appropriate to compare each card's performance relative to the games that people are playing when it came out. Because that's want people are going to do: buy a current-gen GPU and play current games. Someone who buys a 7600 or 4060 will have a much better gaming experience even with the most difficult games today than someone with a 950 in 2015, because these cards are a tier up. Unfortunately they cost it too.

That's not the problem, the problem is you are trying to compare two entirely different test suites. I stated the reasons why in my last comment, I will not restate them.

That's what I responded to and continue to disagree with as their performance at 1080p Ultra is well above 60fps. By this measure then the 950 was one of the most useless releases ever with it's 45 fps, so I fail to understand how it can be used as a good example with the others so easily outperforming it in their respective times.

You are completely ignoring the fact that perceptions on how much FPS you should get out of a GPU have completely changed since 9 years ago. A 950 hitting 45 FPS at 1080p would not be nearly as bad as a modern graphics card hitting 45 FPS 1080p. This is why relative performance to the flagship is simply better, it provides context of the value you are actually getting.