Unfortunately the gaming hardware prices are also shooting for the top
https://www.pcgamer.com/ultra-enthusiast-hardware-is-strangl...
The ideal for a gamer is that every year the GPU maker pulls another rabbit out of their hat, and doubles the performance from the previous year, with the same manufacturing and 50% less power consumption. Anyone who can remotely afford it and prioritizes gaming could upgrade.
The ideal for the GPU maker is that every year, gamers double in number, and double in hunger for higher performance, and any incremental gain in performance can be sold for double the price of the previous year, all while R&D and manufacturing costs are dwarfed by sales volume.
There were years where it felt like the gamers were winning. You could buy a mid-range card for a couple hundred US dollars that outperformed the previous flagship. Amazing! The past couple of years shifted the other way. Between cryptocurrency mining and machine learning, the appetite for GPU hardware created a seemingly insatiable seller's market.
Fortunately, the mining craze has all but vanished, and GPU hardware is much more powerful than it was a year earlier. However, the GPU makers are still on a high from the wave they were riding.
If you're a gamer (and not a machine learning hardware consumer), you're not in a terrible place right now. There's lots of hardware that can run modern games in 1440p and 4K resolutions with lots of eye candy and plenty of performance for $500 available to buy right now. If you can't afford $500, you can buy a $200 card and easily get excellent 1080p performance. But things aren't advancing as fast as they had at some point in the past, and there is no new midrange ($250-300) card yet available from the current generation that can embarrass old flagships (that used to cost $500-600). So we can be a bit unhappy about that, but I don't think things are terrible. And I think with a little patience, we'll still see performance increase, and affordable options continue to improve.
Except that's the price of a full gaming console that can also get me 1440p/4k. It's ridiculous that I have to pay the same ammount of money for a GPU to get similar experience to my PS5.
Consoles were always cheaper than PC gaming, but midrange GPUs costing more than a console is a new phenomena and I doubt it's sustainable for gaming industry.
There's a mix - some PC gamers just prefer the experience enough to justify the cost. But most people have some other reason to own a PC anyway, so paying for the GPU makes sense for them instead of having a separate console. And then there's the ability to build, customize and upgrade your system, compared to going out and buying a new console entirely. There's also, in general, much more backwards compatibility in PC gaming, though it's not without flaws.
If you isolate it to "playing games on a screen" then the console will almost always seem like the better option. But there are more factors that go into those decisions.
As for the cost comparisons, consoles have always been $200-500, and most GPUs were often in that same range. In other words, you'd almost always have to spend more on a GPU than a console for the higher end PC gaming experience, but that has yet to kill the PC gaming market.
In 2006, a Playstation 3 was $600, but by 2010 it was a $300 system. Video cards were $500 at the top end (GTX 480). So you could argue that the console was a vastly better buy, if you didn't mind it having 4 year old hardware.
Consoles have held their pricing much longer lately, so the $500 PlayStation 5 is still largely $500 over 3 years after launch. Meanwhile, you can get some great graphic cards in the $500-$800 range.
Overall, that's not a vastly different ratio. It's not until you look at the $1000+ video cards that it seems quite skewed, but no one needs those cards to have a good PC gaming experience. Those are extravagances.
I think this is an oversimplification, or a particular perspective. Why is your hardware becoming obsolete every 2-3 years forcing an upgrade if you want to keep playing newer titles an ideal for gamers?
in a backhanded way, the stagnation of the jaguar consoles/quad-core desktop worked out great for consumers, as long as you didn't buy at the very tail end of it. You could spend $200-250 on a CPU and 5 years later the best consumer processor available at the same price point would only be like 20-30% better, meaning you weren't locked out of anything you wanted to play.
Contrast that to the early-mid 2000s in the GPU market where the GPU API specs (DX versions, OpenGL shader model, etc) were moving so fast that 2-3 years later your crappy old GPU might not even be able to launch the game, or would require some hack that allowed you to launch but completely ruined the graphics etc. I remember when hardware T&L became a hard requirement and a bunch of stuff pretty much just stopped working, and there were similar cliffs a couple times with various shader models.
In a technology sense yeah, stagnation is bad and we want it to move forward, but at a consumer level? It's actually kind of good if I don't have to buy a whole new PC every couple years just to keep where I started, the PC market of the 2000s was a massive Red Queen's Race.
That is also why I fundamentally don't buy into the "NVIDIA and AMD don't want to sell you an upgrade" thing. They do, it would be great for them if they could put out a card that was so great that you couldn't even keep your 3080 anymore. That's not the world we live in anymore, moore's law died a long time ago and GPUs fundamentally relied on having more transistors at less cost every product cycle, and we've gotten to the place where all the easy gains have been squeezed. A lot of the more difficult gains were completely shat-upon at first by the public (DLSS and other non-raster tricks were extremely unpopular when the idea was first raised, especially the idea of throwing 5% or so die space at it!) and now they've been largely squeezed too.
Consumers simply just don't want to accept this reality. The entire Turing and Ampere generations were focused on keeping costs down by using cheaper nodes instead of expensive TSMC N7 - that's why AMD (who used TSMC N7) wasn't enthusiastic about competing on price, it's not that "AMD is joining NVIDIA in gouging", it's that those are the actual costs of operating in this market and consumers are still living in 1992 (when they're not making up outright fantasy prices). The x70 segment has always bounced between $349 and $399 even 10+ years ago, with the sole exception of the GTX 970. The "true" 680 was the GTX titan, which launched at $999 (in 2012 dollars!) for a cutdown GK110, a cutdown. Ampere was below the cost curve and a 3070 still cost $499 MSRP, a TSMC GPU would have been even more expensive, and now that NVIDIA is back on TSMC we are seeing prices snap back to the actual cost curve rather than Uncle Jensen's Discount Ampere/Turing Emporium pricing.
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_proces...
The idea that "NVIDIA just wants to sell enterprise GPUs" is completely insane, both these families share architectures and general architectural work and the economics doesn't work without both of them. Look at the revenue over the last few quarters for what happens without consumer sales. NVIDIA would love if consumers had to buy a new GPU every 2 years to stay in the game. And margins have stayed pretty constant, apart from a secular trend towards more datacenter gpus (and the resulting higher margins). When you consider that NVIDIA is selling 3-4x more datacenter cards than even 5 years ago, there is no evidence of some ballooning margin in consumer GPUs, this is just what they cost now.
(In contrast though AMD is in a very real decision-point between enterprise CPUs and everything else... right now they are trying to cement their gains in the datacenter market before sapphire rapids and other competition finally re-enter the market. So they actually are pushing for datacenter CPU in favor of, say, consumer GPU, along with everything else. Their shipments are massively down and that's why their sales share is under 10% now. NVIDIA though? No, their revenue is in the tank because of consumer GPUs and they very much wish they could sell you something twice as good at the same price as last year.)
But like, people just dive into these bizarre conspiracy theories. It's a margin pump that doesn't show up in corporate margins. It's slow-walking the technology at inflated prices because NVIDIA doesn't like selling you a new card every year. These are consumer goods. NVIDIA makes money by selling them to you as often as possible and by delivering gains that make you want to buy as often as possible. This is just the reality of post-moore's-law technology and spiraling costs from R&D, from TSMC, and on everything else during the pandemic. Yes, volume is way down but they can't sell them to you at a loss either, they'll cut shipments instead.
The good news is prices may finally be breaking. Flash and DRAM are way down and those are the canaries in the coal mine for overall consumer demand. TSMC is reportedly looking to cut prices to try and keep customers from cutting orders. VRAM and power ICs are coming down and stabilizing. All of this will take a long time to trickle through (NVIDIA has been producing Ada for over a year now, a lot of stuff was produced with high-cost components/inputs), and the long-term trend is still upwards (until MCM GPUs really get solved - and by MCM everyone means multi-GCD and not just pulling out memory controllers like AMD did). But there will be some near-term relief over the next 2-3 years until consumer demand starts picking back up again.
Long term though? The first 28nm generation was the GTX 670, launched for $399 in 2012 dollars, for a cutdown 300mm2 die. A 1070 was the next leading-node launch at basically $450 street price until prices finally settled, again for a ~300mm cutdown. A 3070 at $500 was under the average cost-trend because it was a trailing node. Fair price according to cost-trends for a 4070/4070 Ti/whatever you want to call it is probably $649 right now - $799 is still a little inflated but not all that much. Consumers really don't like that fact, they want to go back to imaginary $300 prices (which no x70 card ever launched at). You can't make a ferrari for $20k and you can't even make a yugo for $5k, there are simply some consumers that can't be satisfied and companies have to focus on the ones who can be. "sell at a loss and make it up in volume" is a meme not a business strategy.
In some senses the GTX 970 and the Turing/Ampere (particularly Ampere) generations were a colossal mis-step in strategy from NVIDIA, because they mis-calibrated people's pricing expectations. People aren't gonna remember the generation when you saved them from a 15% price increase, they're not even gonna know that, but they're definitely gonna remember it when prices snap upwards 30% in a single generation and eat up all the perf/$ gains. From a market-psychology perspective a steady 10-15% increase every gen is better than 0% and 30% in subsequent generations.
And for as reviled as Turing was - the 2080 Ti at $999 (EVGA sold them as cheap as $949 remember!) is less than the first GK110 cutdowns (GTX titan, the true 680) which launched at $999 in 2012 dollars. Consumers just have no overall perspective on the price trends in the industry, they latch onto the one time you launched at a cheap price and you'll never live it down. Y'all aren't giving NVIDIA any reasons to try and give you a good low-cost generation in the future if it means more generations of mis-calibrated expectations afterwards.
And that's what it comes down to. NVIDIA is obsessed with cloud gaming and other things that don't make sense to ordinary people (like kids cloud gaming in the back seat of a minivan instead of just playing a switch.) Maybe a gamer keeps a GPU busy for a few hours a day, but a cloud GPU could be shared between multiple gamers and train neural networks when gamers are asleep. From a "value-based pricing" perspective a GPU that sells for $1600 and is utilized 10% of the time could be sold for $16,000 to someone who could keep it busy 100% of the time.
By that perspective NVIDIA can't really afford to sell cards to individual gamers.
They can't afford to continue profitable line of business that took them from nothing to multi-billion dollar corporation?
This is not a new problem - why does Intel sell CPUs to gamers for $100 when theu can sell same CPU for servers for $1000?
Narket segmentation, and you try to make sure these markets buy diffetent products that are priced differently. You artifically create differences in priducts. For example Intel withholds ECC memory from consumer products. Whether its rihht or wrong is debatable.
But please let's not use language like 'we can't afford', It implies they are losing money and subsidising gamers.
Whereas in reality they are very profitable and you are saying they could nickel and dime their customers even harder.
For instance on a bad day it seems that Microsoft would be glad to pour lighter fluid all over the Windows ecosystem if it could get just have 10% of the success Android has had with phones. Or look at Meta neglecting it's current products (even market leading VR games) for the sake of one man's dystopian vision.
A common factor in the above is a lack of competition. If Microsoft was really threatened by the Mac or Linux or if there was some real competition for Office, Microsoft wouldn't be able to neglect its customers in favor of daydreams... And the same is true for Facebook. If advertisers voted with their feet you'd see Zuckerberg burn all those VR headsets tomorrow.
Consider the success of the Nintendo switch. The dock for the switch was a bit of a Jedi mind trick, just like the Wiimote was for the Wii. In particular, it changed the competitive psychology of game producers to think that any "big game" that came out for the Playstation or Xbox would also be released for the Switch. Selling the switch as a hybrid device stops gamers and games company from realizing it is what is... the beginning of the end of the TV-attached console.
The Steam Deck is a much more powerful machine for about the same price and it also has a very strong catalog. The Steam Deck can play Control natively and the Switch can't. Yet for the most part the industry is going to keep remaking the same old games and releasing them for the Switch, even most the innovative parts are going to be making games like Genshin Impact for midrange phones.
Sales of the 4090 are driven by prestige, the pleasure you can get out of saying that you own the very best. They are not driven at all by the reality that you can have anywhere near 10x the pleasure from a $4000 gaming PC than from a $400 portable. However a business that revolves around neural networks very well might be able to generate 10x the revenue from a 10x faster graphics card.
It's not a question of what "we" can afford but it is a function of what "they" can justify to their shareholders. I mean, the people in my town could afford 100% fiber optic internet, but the phone company can't afford to stop using the copper network when it can still charge people premium prices for subprime service. If we decided that $100 a month was a fair price for gigabit fiber and that price caps should be scaled accordingly, my telecom would quickly decide that it "can't afford" to offer DSL at $0.10 a month and would start tearing up that copper network pretty quickly.
I have never met these people, and my sample includes four grandmas that are 70 years old. Quality of videocalls so important that 1 Mbps internet is worth $0 to them. Gaming is leisure, that's different - some people buy $1000 golf clubs, and $5,000 bicycles, some don't know what to do with them.
> I mean, the people in my town could afford 100% fiber optic internet, but the phone company can't afford to stop using the copper network when it can still charge people premium prices for subprime service.
Sounds like a monopoly has your town by the balls, that's not a healthy state of capitalism / free market.
These "artificial scarcity" techniques are cancerous tumours, if we allow them to fester and grow they will bring down the whole system.
Only if you (among other things) ignore the other costs and profit of the cloud gaming service.
How much does continuously streaming 4k bandwidth alone cost over the life of a GPU?
Doesn't matter - the bandwidth used by an average user costs a small portion of a montly subscription fee. High prices for GPUs may be a good way to get more subsribers, who will pay a lot more money over the years.
Citation needed.
> High prices for GPUs may be a good way to get more subsribers, who will pay a lot more money over the years.
And they set price of both markets. Sounds like market manipulation to me.
And that's significantly understating the number of subscribers per GPU, if we're trying to pay off $16000.
Having significantly more subscribers wouldn't give each player much time to play.
Whatever you think the real numbers are, you probably need about half a subscription to pay the bandwidth. I'd be more concerned about the electricity bill, since you'd be in the ballpark of 200kWh a month.
Average driver uses car <5% of the time, but you can't replace 20 cars with 1 uber if everyone needs to get to work at 9 am