The only spec of this card that I appreciate, is the 24 GB memory, but that's for machine learning and, according to NVIDIA, that's not even for the target audience.
I'm sorry you lost your sense of excitement and wonder, but this 4090 really looks quite great.
Granted, ray tracing is optional (I can't even distinguish between RT and non-RT in games like CP:2077), and DLSS is a game-changer for 4k gaming. But there are plenty of games that I've been playing would be served better by a 4080 over my 3080.
Mostly because fake reflections in triangle-based rendering is convincing enough unless you're looking for it or have just gotten good at spotting it.
It's like bad kerning. If you don't know what to look for, you probably don't notice it or pay it any mind. But once you do, you see it EVERYWHERE. [0]
But I see artifacts of fake reflections in games everywhere. The problem is that most games rely on screen-space reflections which don't work if the reflected object isn't on the screen. So imagine being over water, looking over mountains in the distance, and you see the mountains reflected in the water. But as the camera pans down and the peaks go off screen, suddenly the mountain peaks get cut off in the reflection.
The only way to render a mirror accurately in triangle-based rendering is to effectively use a technique similar to Portal, but then your transform engine has to would twice as hard, since it has to project the scene twice. Even then, you're limited to flat mirrors without taking shortcuts.
Curved objects can't be reflective in triangle-based rendering at all. They fake it well enough by using environment mapping of textures, but then they won't accurately reflect things around them.
[0] Relevant XKCD: https://xkcd.com/1015/
It's great for a movie displaying on a wall, not for a game on a desktop monitor.
If your 4k pixel count is based on 3-5 screen driving/flight simulator on big screens, then sure that's great to have
I build a new machine about once every 4 years for gaming/development, and I just don't really see all that much difference between the capabilities of the machine I built two generations ago (now ~9 years old) and the one I specced out last year.
I can still load up modern AAA titles on the 9 year old machine and they run just fine at 1080p 60fps with some minor tweaks to the graphics. Turns out the GTX 970 from 2014 is still more than capable.
Basically - these top end cards are a complete waste of cash for the average consumer. Hard to even claim you're future-proofing the machine with it - you're just plain wasting money.
Before the 30 series, everyone knew RTX Titan was not a gaming card, but rather for compute/professional work.
Nvidia rebranded their Titan cards as 3090/4090 and people lap that up for some reason.
It was literally double the price of a 3080, but was at most 20% faster, and usually only 10-15% faster.
nVidia's claim that "It's for 4K gaming!" was marketing bullshit. With a such a small performance premium over the 3080, I'd argue that if a 3080 can't handle a game in 4K, the 3090 probably can't either.
The "1%" who have unlimited budget for consumer purchases like this, are millions of people.
Two reasons people are gobbling these really expensive cards:
* Inventory constraints of the 3xxx series pushed people to buy any available card and it just so happens the 3090 was there waiting for them as the only available card (or within parts bundles).
* Streamers normalize overpriced setups. It used to be cool to be thrifty with pc builds, now its just all about "performance" as games become less and less optimized on launch.
Namely, there are 2 big issues: - Virtualization: Getting vGPU functionality out of them (most of the time, there's no point in having a full 1080 Ti for only 1 job, we want to have it split on more than 1 VM). - Kubernetes workloads: I'm running a simple workload that uses the GPU (for testing for instance) and k8s wants it all for only one pod.
You'll need to make sure to not overload gpu mem, otherwise you'll get oom errors. You can use k8s patch functionality to add custom resources, essentially use that to represent gpu mem and include the amount of gpu the pod uses in its definition, that way you don't go over the limit.
Then don't buy it?
These are flagship cards. It's like the $100,000 Corvette in Chevy's lineup. It's fun to watch the reviews and see on the showroom floor, but most people aren't in that enthusiast market segment. They're going to the dealership for the $40,000 Camaro or the $30,000 Colorado truck because those get the job done for them at a reasonable price.
If you're an enthusiast and this is your thing, you save up the money and buy the flagship. It's a small audience, but the product is there for those who want it. If you're a casual gamer, you get something like a 3060 and it works great as long as you're not maxing out settings and trying to play everything at 4K.
I honestly don't understand all of the criticism for these flagship parts. When auto makers release a ridiculous flagship car, everyone cheers it on and understands the purpose. When hardware makers release a flagship part, it seems everyone wants to criticize it for existing.
Besides all that, releasing "consumer" cards with such overly inefficient power consumption is just being tone-deaf at a time where half of the Western world is facing rapidly rising electricity bills.
https://tpucdn.com/review/nvidia-geforce-rtx-4090-founders-e...
https://tpucdn.com/review/nvidia-geforce-rtx-4090-founders-e...
in contrast to the breathless rumors, those power numbers look incredibly reasonable... it's less than the 3090 pulled last gen, and it's 1.7x the perf/w of a 3080 or 2x the efficiency of a 3090. 3090 was always known to be pushed a little far, but, there is an easy 70% uplift in efficiency here, and none of the power-creep stuff ended up actually happening.
4090 Ti will eventually happen, and will pull about the same as 3090 Ti. Which, if it gets 1.7-2x the perf/w, sounds fine.
I think you're right. I imagine a lot of people don't like the feeling that something faster exists outside of their budget.
Ironically, I suspect these people would be happier if the faster cards simply didn't exist and the GPU lineup topped out at something like a 3060. It's funny that people get angry that faster parts simply exist.
These new GPUs and even CPUs are entirely about power and performance. The upfront cost is one thing, but the long term cost of the both the energy consumption and central air system having to handle the massive thermal load is getting out of hand, particularly in a hotter region.
It's severely overlooked or just lightly complained about as "this thing needs cooling" without really elaborating what that entails or means. It's one thing to cool down a large studio area with a commercial HVAC system for doing youtube videos, but it's another to try and cool a bedroom in a house with the normal central air system.
I just wish these PC component manufactures would have some more focus on making their products more energy efficient, do more with less. You can always do more by consuming more, that's less impressive. And to go back to cars. Take a huge engine, and one half its size, both output the same power, but one uses a third as much of the same fuel. Which one is more impressive?
I understand flagship is the extreme edge. But when the power consumption of the flagships keep increasing, it normalizes the lower and mid ranges to also keep increasing. I just want to see a focus on efficiency, stop doing more with more. Do more with less instead.
Efficiency should be better than previous gens, at least this was the case for the CPUs I looked at (looking at their power efficiency ratings at full TDP they're hard to tell).
Marginal efficiency is lower when the power envelope is expanded to make these higher performance numbers achievable.
I'd wager that performance matched, these "power hungry" current gen GPUs/CPUs would show better efficiency than previous gens.
Flagship cars generally do get significantly worse gas mileage than mainstream cars.
But you're exaggerating the inefficiency of a 4090. In terms of performance-per-watt, it's not that far off from their lower level cards.
If low power is a primary driver, it's very easy to move a slider and reduce the power usage for any of these cards.
> I just wish these PC component manufactures would have some more focus on making their products more energy efficient,
They do! Efficiency and performance are one in the same. They wouldn't be able to squeeze out this much performance if they didn't optimize the efficiency. More efficient parts result in higher performance for a given thermal envelope.
If you wanted the most efficiency, you would buy a 4090 and turn the power limit down to 50%. This is very easy to do. Reducing voltage and power limits is extremely common in small case builds.