Was there any speculation on the cost to make the RTX cards?

Nov 18, 2007
10,598
2
91
#1
I'm wondering if adding the RT and DLSS cores adds much to the cost. Also the die is quite bigger so it's fewer per wafer. Will they release the info during the next earnings call? Or at least mention the margins on these cards? I'm really curious if nVidia really jacked up the margins or if the price increase was needed.
 

aigomorla

Cases and Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
17,049
17
126
#2
no it doesnt.

Honestly i think Jenson saw what idiots will pay during the bitcoin mining crazy and thought, hey, if we make them that price, they will buy it.

If i pimp out Ray Tracing, they will buy it
If we show them Ray Tracing is the future they wont see the price and will buy it.
If we tell them Ray Tracing will save them money, they will ignore the fact its almost the price of the launch Titan XP and just buy it!

And hence we get 1200 dollar cards when the Ti's were floating at around 699 for the past 3 generations.
(780ti ~ 649... to 980ti ~ 699 then the 1080ti ~699)

The only way to bring the price back to regular price tags is to just NOT buy them.
Personally im crossing my fingers for AMD to come spank Nvidia like they did Intel.
I really want a good reason to go back to team Red, and that 1200 dollar RTX is a good reason.
 
Mar 10, 2004
27,986
107
126
#3
If he's really pushing RT and Tensor cores, then the 2060 and 2050 might have them. Otherwise it creates 2 distinct classes of cards.
 

Headfoot

Diamond Member
Feb 28, 2008
4,379
27
126
#4
I wouldn't be surprised if it was a little of column A, a little of column B approach. They undoubtedly are paying more on a per die basis given how ungodly enormous that chip is. At the same time, I very much doubt they would intentionally cut their own margins when there is no credible competitor for that tier of performance. So I'm sure it's both. The other thing is that the Turing tech is the beginning of a longer road so the accumulated R&D will have an extended amortization schedule aka not all of the R&D cost needs to be borne by RTX 2080 and RTX 2080 Ti. It will be spread out amongst all product lines and in future chips as well. With something this complex you can do some (above board) magic with the numbers
 

PeterScott

Platinum Member
Jul 7, 2017
2,481
62
96
#5
If he's really pushing RT and Tensor cores, then the 2060 and 2050 might have them. Otherwise it creates 2 distinct classes of cards.
So RTX 2050 would do ray tracing at 320x200? ;)

IMO 2050 won't be an RTX card. I doubt 2060 will either, though there is some chance that 2060 could be a 2070 with disabled cores.
 
Mar 10, 2004
27,986
107
126
#6
So RTX 2050 would do ray tracing at 320x200? ;)

IMO 2050 won't be an RTX card. I doubt 2060 will either, though there is some chance that 2060 could be a 2070 with disabled cores.
At $499 to start, ray tracing would then be out of reach for most pc users.
 

PeterScott

Platinum Member
Jul 7, 2017
2,481
62
96
#7
At $499 to start, ray tracing would then be out of reach for most pc users.
New technology often is, during it's first year or two. Even if there were some token RT cores in 2050, it wouldn't realistically have the performance to run RT games.
 

Malogeek

Golden Member
Mar 5, 2017
1,387
2
96
yaktribe.org
#10
You don't have to buy an FE card, do you?
Do you actually think any AIB cards are going to be cheaper? Has the existing 2080/ti launch not shown otherwise when even basic blower versions are the same price as FE? Are you still blinded by Nvidia's marketing of this magic MSRP?
 

Headfoot

Diamond Member
Feb 28, 2008
4,379
27
126
#11
The only MSRP I believe is on the Arizona Iced Tea cans
 
Nov 27, 2001
28,704
0
126
#12
The per-die price is something that has been on my mind in regard to pricing. I still see people complain about Nvidia's pricing on the 2000 series, but I think it's ignoring two factors: die size and process maturity. Basically, the 2000 series dies are significantly larger than their predecessors with the 2080 Ti (754mm²) compared to the 1080 Ti (471mm²) coming in at 60.1% bigger. The 2080 (545mm²) doesn't fare much better when compared to the 1080 (314mm²) with the difference being 73.6%, but the 2070 (445mm²) is a bit less worse compared to the 1070 (314mm²) with the difference being 41.8%.

Another important factor is whether these big dies are coming out working. Unfortunately, I can't find any yield numbers on TSMC's "12nm" process. Although, given that it's really just a modification of their 16nm process, which Pascal was already using, it shouldn't be that bad on yields.

All in all, you've got more expensive dies on what probably amounts to slightly worse yields. If Nvidia wants to keep up their bottom line, you're going to end up with a more expensive product. Now, what I think you can rightfully be annoyed at is that while Nvidia may be raising their price in line with BoM increases, the problem is that the BoM costs don't change the fact that ray tracing is pretty much useless right now.
 
Mar 10, 2004
27,986
107
126
#13
Nov 27, 2001
28,704
0
126
#15
The 2080 Ti was supposed to have an MSRP of $1000, but we haven't really seen much near that. I don't know if that's strictly due to retailer mark-up or its the MSRP from the AIB partner. I recall Gamers Nexus mentioning how margins were quite low on RTX, and I wouldn't be surprised if that's part of the issue.
 

alcoholbob

Diamond Member
May 24, 2005
5,800
16
91
#16
The per-die price is something that has been on my mind in regard to pricing. I still see people complain about Nvidia's pricing on the 2000 series, but I think it's ignoring two factors: die size and process maturity. Basically, the 2000 series dies are significantly larger than their predecessors with the 2080 Ti (754mm²) compared to the 1080 Ti (471mm²) coming in at 60.1% bigger. The 2080 (545mm²) doesn't fare much better when compared to the 1080 (314mm²) with the difference being 73.6%, but the 2070 (445mm²) is a bit less worse compared to the 1070 (314mm²) with the difference being 41.8%.

Another important factor is whether these big dies are coming out working. Unfortunately, I can't find any yield numbers on TSMC's "12nm" process. Although, given that it's really just a modification of their 16nm process, which Pascal was already using, it shouldn't be that bad on yields.

All in all, you've got more expensive dies on what probably amounts to slightly worse yields. If Nvidia wants to keep up their bottom line, you're going to end up with a more expensive product. Now, what I think you can rightfully be annoyed at is that while Nvidia may be raising their price in line with BoM increases, the problem is that the BoM costs don't change the fact that ray tracing is pretty much useless right now.
Seems like a bad deal still.

980 -> 980 Ti was a 51% die size increase for $100 price increase.

1080 -> 1080 Ti was a 50% die size increase for $150 price increase.

The process at this point should be pretty mature as it's basically a + + + refined process that started off as 20nm + Finfet all the way back in 2015, but 1080 Ti -> 2080 Ti is basically a linear price increase for die size (60% die size increase for $500 price increase).

In a normal release pattern on the same density process we would expect a $879 2080 Ti. Being that 7nm is on the verge of launching $1200 launch price seems like a real attempt at gouging (especially with vendors complaining about low margins on these cards) to get as much money as they can before the products become EOL'd.
 

ASK THE COMMUNITY