NVidia may have focused on the high end too much. They have a card for the machine learning crowd that costs $16,000.[1] The product they're now pushing to gamers, the 2080, is about $1000. It has some ray tracing hardware used by nothing. On existing games it's about the same speed as their 2-year old 1080, which NVidia just discontinued.
[1] https://www.dell.com/en-us/shop/accessories/apd/490-bens
I certainly don't intend to upgrade, but I admire nVidia for the gutsy move, and I'm glad they're pushing the industry in this direction.
Hard to fault this. It's reminiscent of when hardware transform & lighting (T&L) was introduced. Nothing used it until the cards existed, but the cards had to come along before anyone would bother coding it into the games.
For those who weren't around during the first wave of hardware 3D acceleration (circa 90s), there was a huge amount of "X brand only accelerates this and that game."
Nowadays, with DirectX and OpenGL (somewhat) winning the API war, a GPU is a GPU for all games.
It seems entirely reasonable raytracing hardware and game support follow the same path. Supported in some hardware, supported in some games. If the market likes it, over time we see fewer and fewer games that don't use it.
* Caveat: Things are a little different this time since DirectX 12 supports Ray tracing today, as far as I can tell. And Microsoft doesn't really care who wins between Nvida and AMD (aside from keeping suppliers in competitive bidding for XBox GPUs).
I wonder if Intel are about to have deep pockets?
You can use accelerated lighting even in a 2D game, but I'm unclear on what casual games would be using ray-tracing for. Have a particular vision of the future you'd like to share?
Indies and less realistic or more "casual" games may not bother with it, just as many of them don't bother with things like parallax mapping and some don't even bother to use textures at all. The features are there and easy to use but developers opt in and out of the different rendering options based on their artistic styling.
There was no pressing need to replace your 2D card, and in fact you might actively dread it (drivers were finicky back then—replacing your 2D card was signing up for a process not dissimilar to that of building a Hackintosh today.) And computers back then had a lot of room and a lot of spare PCI slots. So why not buy a separate card just for 3D, leaving your existing 2D card intact? Smaller change to worry about.
Hmm, certainly in the 80s and maybe very early 90s but by the time Voodoo cards came along things were much better. Still not up to the "plug and play" level of today, however.
I'm predicting this will be the same kind of story playing out again - One vendor kicks it off, some 'demo' games that build heavily on it drop but are quite short-lived, and the tech slowly turns into a commodity feature with in-game settings menu 'on/off' selectors, followed eventually by ubiquitous support to the point that nobody really talks about it any more, it's just part of 'modern graphics'.
Shaders is another good example that followed that kind of pattern, it was all the shit in the late 90s that Q3 had 'shaders', now it's not even a bullet point for most releases, but it's certainly there.
I've rambled, but my point is that I'm not sure a lynchpin 'RTX required' game is needed at all to drive industry-wide adoption here.
On one hand, there is the appreciation for the technical achievement of something that seemed unlikely for a while (real-time ray tracing), plus as you mention, then increased die size. They look to be pretty good for ML applications, as well.
On the other hand, there is the increased price and, frankly, underwhelming results. The only game that uses the RTX feature suffers substantial performance penalties for turning on the RTX feature and most people (including me) just try it out for a few minutes and then turn it off. Is the quality better? Yeah, the reflections are shaper and more consistent when changing views (ie, aiming down a sight), but they aren't game-changing and are definitely not worth the performance penalty.
This generation of cards is worth skipping if you aren't building a new computer.
That said, this comes with a pretty big cost, both in $$$ and performance. If this could be accomplished without the substantial performance hit that it seems to come with, then it would be fairly impressive and worth the $$$.
Or, come to think of it, a "Half-Life 3: Portal 3" where you use gravity+portal+various paint (and other liquid) guns, on an alien ship and in space, would make a lot of sense...
I personally do think the raytracing footage we've seen is a significant step up, however.
However, nothing will be the step up that bumpmapping brought to the table. Doom 3's textures look like garbage with that disabled, due to the bump/normalmapped greebles all over them. Everything looks like clay or cardboard.
I may be (read am) an old git, and ray tracing just makes me think Wolfenstein 3D. I assume this is a different raytracing? Is it the origin of the rays rather than coming from the 'eye', coming from the light source? Isn't that inefficient? Or something else???
According to the wikipedia ray casting article, the two were used interchangeably "in early computer graphics literature", just to make me feel even older.
I think you have it backwards. Ray tracing is O(#of pixels*number of light bounces). More common 3D rendering is based on O(# of objects in the scene). It's conceivable that ray tracing can be more efficient for scenes with many objects.
[1] https://en.wikipedia.org/wiki/Bidirectional_scattering_distr...
Actually, nVidia's raytracing stuff is—somewhat-uncharacteristically for the company—all done in open APIs.
Don't buy the first generation of anything unless you've got money to burn. Get the second or third.
No, you'll get something worn out after running flat out for a year in a Monero mining farm.
I seriously doubt that we can take this for granted: the RTX 2080 will almost certainly remain near the state-of-the-art for several years.
Here's the thing about gaming vs content creation vs ml pricing -- they're fundamentally different activities from a financial standpoint.
Gaming (for the VAST majority) is a cost center. How much are you willing to spend on it?
Content creation, and now ml, are profit centers. How much do you make from it?
The latter question is directly related to your business. If you're running financial trading models and the card makes you 5% faster, that 5% may be worth USD$1M. If you're running a 1 acre farm, maybe it's worth $3.
But by hardening and popularizing CUDA and GPGPU, Nvidia upscaled their most wealthy buyer from sfx studio to any company that makes predictions.
It's a heck of a lot easier to deliver 1% improvement to a $100M company, than to deliver a 200% improvement to a $500k company.
seems to me like they're making products available at all sorts of price points. i'm not an mba, but that strikes me as a good idea.
not making quite enough mid-level products is unfortunate, but since that was temporary, it wouldn't have made sense for them to invest in increased production.
(they should've abandoned the MSRP and just reverse auctioned inventory to the retailers, so that they could capture the value of their products, instead of letting folks selling on ebay get it.)
> They have a card for the machine learning crowd that costs $16,000
they have hardware that costs way more than that.
But it's in my interest to want maximum bang for my buck and it's in Nvidia's interest to extract maximum money from it's customers. The only escape would be if AMD comes up with an equivalently powerful card and I haven't seen an indication of this happening.
But yeah, it's hard to see them at any risk unless they really mismanaged their capital investment, which is always a risk for companies like this.
If you want to double the framerate you are getting with a 1080 and be on one physical card, with reasonable power and heat, that is the 2080ti.
If you ignore the ray tracing and machine learning features, this is still what you're asking for, is it not?
Or do you mean, you want to pay the cost of a 1080, and get a card twice as good as a 1080? Because i would also love to trade my leaf for a tesla but pay nothing...
And sure, I want to Moore's Law pricing overall. Even if each generation is twice the power of the last generation, I still want each generation at the same price. But again, only with competition.
or perhaps you're right and it's divide and conquer plain and simple.
SLI/CF never scaled linearly and in modern games (if they support it at all) you're usually looking at 10-30 % performance increase, for twice the cost and introducing microstuttering.
Yhe ML quote is from dell and looks to be 'appropriately' dell priced at ~50% markup vs mass retail..
but yes $8-10k is still alot.
That said, if you need this stuff and your super fast $5k workstation is basically an otherwise 'worthless' I/O pump to the GPU, it's still worth the price tag.
I was initially thinking the same on Ray Tracing. But having seen more videos and more tuning so less performance loss when using RT, I am now fully convinced Ray Tracing is the future, and it is achievable. The next 5 years, 7nm, 5nm and 3nm from TSMC is going to make RT Gaming Everywhere.
It's going to be purchased mainly by the cloud vendors to build more power-efficient (and vertically-scalable) instances.
See also: the high-end of the NVMe market for the last five years (i.e. since its inception.)
So they've had an upwards trend in the past. That's no guarantee for future performance.
> Ignore the crypto bubble/pump and NVDA is growing quickly
Quoting the article: That bust is obvious in Nvidia’s revenues this year: they are essentially flat for three quarters now, hovering between $3.1 and $3.2 billion.
We need to be able to distinguish between the 2 tailwinds. AI and crypto. Crypto turned out to be more of a factor than analysts anticipated, leading to the huge runup and then fall. Consider that done and dusted.
However the AI story is consistent and strong. Now that the stock price has come back down to earth we're still looking at amazing returns, there's nothing to be displeased with here.
The Q/Q performance is not concerning because of the product cycles. Every generational change there was a big drop in sales of the previous generation. Look at the first quarter of Pascal and first quarter of Maxwell's introduction and you'll see the same pattern. A flat quarter is actually an amazing improvement over the straight up declines of the previous generations.
Major growth will return soon as new products get ramped.
Is anything a guarantee of future performance?
They've had an upwards trend year after year very consistently. Except for the past 12 months.
So weight those trends accordingly.
I acknowledge that the recent stagnation might only be short-lived, but it would be premature to call this new trend temporary based on the past performance.
At the very least, we know that the environment has changed significantly, because the Crypto cash cow seems to have stopped producing milk.
Also, their P/E is at an healthy 19.99, reasonable.
The article says that Apple is making their own GPUs, that doesn't matter, because Apple is not a nVidia client for a very very long time, and was never a big client.
The others are even less suitable comparasion as those companies aren't making GPUs and they are not just going to walk in that market.
see also: QQQ, XLK, etc.
> It's up 50% over 2 years and 400% over 3. Idiot article.
It's funny because that applies to all crypto articles too.
Fauxbituary
I should buy that domain
The "cryptocoin" itself functions more like a commodity, a market which was never for the passive retail trader because there is no expectation for commodities to increase in value indefinitely, only in ebbs and flows based on supply and demand. So the prevalance of the retail trader in the cryptocoin market explains the misaligned value ideas, which are mostly analogies to a stock market. Oops. Although stocks are the only capital market prevalent amongst retail, try not to get the asset classes confused.
"Cryptocoins" typically have a more transparent supply than non-digital commodities, and upgrades to their utility are also typically in open source projects which means these provides advantages in determining price targets in the future value and the scale of the addressable market participants. As in, how much any future market participant will want to own to access a provision provided by the network. As open source projects, and freely fungible, you can create utility for a new market until your positions are profitable, with near zero overhead costs.
Instead of the government of the United States or China or Britain or India or some other actual entity that is highly likely to survive nearly any catastrophe or change in leadership, cryptocurrencies are backed by shady MLM hucksters who originally met on a minor and very geeky gaming BBS to come up with their scheme, and whose downline marketers hire C- and D-list celebrities to shill for them. You don't even end up with a garage full of awfully-scented soap when it crashes.
But your disdain is still too broad and blanket.
There are plenty of ways to be productively deploy resources for profit in that market. Its typically just not a directional/long-only passive approach, it is more actively creating an asset and market until you are satisfied. Just like in other markets.