Like every single major advance in graphics in the last 20 years.
If you're going to pull 'economics' on me, produce some numbers to back your position up. Otherwise, I'm going to go with historical trends over vague theorycrafting.
A lot of cost is added to the developers with DX12.
The 380X is ~20% faster than the 970 and matching the 980 @ 1440.
A lot of cost is added to the developers with DX12. You not only need to make paths and optimize down to SKU level. You also have to keep patching for newer cards and IGP in the future. Plus there is the extended requirement of development time and manpower.
But its quite simple, unless you find a way to reduce not only the workload, but also the developerment time with DX12. The rule of economics will make it worse than DX11. because those paying for the development isn't going to think its funny to fork out more money for a DX12 path with very little benefit. Or even regression.
Developers dont gain money by having games run on slower CPUs with faster graphics cards due to DX12.
DX12 however is a money and time sinkfor developers. And its no surprise unless you are naive.
MS demands DX12 with Xbox One, hence DX12 in these ports we see. And then there are sponsorships.
What is more, because there is such wide front end with 64 ROPs they are starved for bandwidth. So now they are bottlenecked. But the difference is plain. AMD hardware in DX11 was bottlenecked by software. Maxwell cards are bottlenecked on HARDWARE level. And that is something you cannot mitigate with software.
Its embarrassing when AMD cant even get the install base right, yet alone TAM
They should ask Steam instead as to what their base is.
What's wrong with the 7.5m number Roy Taylor mentioned for VR?
Not only is a bit silly to say the Maxwell is bottlenecked at the HW level (we don't have that sort of data to make such statement) but the notion that you can't help HW bottlenecks with SW is non-sense. Developers using SW pipelines to work around poor GCN geometry culling performance are saying "hello!"
Its not correct. Steam alone got much more.
We do have data, at least two companies worked with Nvidia on async already, one got negative results and the other saw no gains.
I think Roy Taylor will have the exact numbers on that tbh. Look at Jon Peddie's numbers for last year, what was it 6 (5.9) million "enthusiast" cards sold? That's 970 and upward, with another ~1 million later in 2014 (AMD probably sold a few hundred thousand Hawaii cards in all of 2014), the vast majority (80%-90%) of those will be Nvidia cards but 7.5 million would be very close in total.
Example,
GAME A (DX-12)
TAM = 10M
Cost of development = 10M
GAME B (DX-11)
TAM = 3M
Cost Of development = 7M
I believe everyone would go and develop Game A. Dont you think ??
Could you link me where JPR classifies the cards?
Its not that hard to figure out, since we got the numbers in the steam installed base and we know how big the steam user pool is:
http://store.steampowered.com/hwsurvey/videocard/
So we can conclude the result will be greater than Steam alone.
On Steam alone, you get close to 7.5 million just with the GTX970. And you are above 10 million with GTX970, 980 and 980TI.
"MS demands DX12 with Xbox One, hence DX12 in these ports we see. "
Do you have any proof for that statement, or are you just throwing that out for fun?
Still not explaining the economics, which since you know all about, I'd love to hear about for this.
"And then there are sponsorships."
Same here actually, would LOVE to hear the economics of why people are sponsoring DX12 games over DX11 games.
where does it show numbers of people with GPUs there? It only shows percents?
Does it show how many people take the survey?
This is just circumstantial evidence, not proof, in fact we can use it to prove exactly the opposite of the point you are making.
Async compute helps improving performance when part of the GPU is not fully utilized by scheduling independent work, so a very efficient GPU might see little to no improvement with async compute, simply because it doesn't have lots of unused resources in the first place. I have no idea which case applies to Maxwell (perhaps both in different situations) but it doesn't prove what many on this forum think it proves.
% and known user amounts. You should know better
The amount of people taking the survey is so great that its statistically correct.
% and known user amounts. You should know better
The amount of people taking the survey is so great that its statistically correct.
Again, where are the numbers?
Edit: For anyone wondering, Shintai is going to avoid directly answering my questions after what happened last time.
Yes but not everybody gets tested, and it's run every month.
Regardless of that it should be obvious that the reason why Peddie's "Enthusiast" bracket increased by so much in 2015 is due to Nvidia's success with the 970.
Again, where are the numbers?
Edit: For anyone wondering, Shintai is going to continue to avoid directly answering my questions after what happened last time.
That's what Lebaredian said to Hardware.fr
Regardless of the reason behind it, it's quite obvious that Maxwell gets nothing out of async. It is likely a combination of already being better utilised and just not quite having the hardware to do it right, but the end result stays the same.
So you guess on JPRs categorization. While you refuse Steam numbers?
If there was a big variance you would see it month to month. So again, nothing wrong with the Steam numbers
You already got the numbers, that you refuse them is your own fault
But Steam numbers are there.