Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
I did a Google search for Micron GDDR6X prices. Didn't find any prices on the first search page but did find two different sources (TweakTown & Tom's Hardware) both stating that Micorn had confirmed the memory size of the RTX 3090. The only poriblem was TweakTown said 12 Gb and Tom's Hardware said 21 Gb so it is still a toss up. If the RTX 3090 is the Titan Replacement, I believe it will be 24 Gb but it will probably be more than $1,400 or $1,500. We will jusat need to wait another 12 dayus or so to find out.
Yeah nobody really knows what the new cards will be like so many conflicting leaks, no one really knows at least we wont until Nvidia do their announcement.. The one concern from all the leaks is the chips being used, as if Samsung are going to be used as some of the leaks have suggested, then there might be issues.. With rumors being that we might end up with another Fermi in terms of the problems those cards were supposed to have had..
The other thing that may be a factor is that these new cards are supposed to be PCI-E 4.0 as well.. The one thing is that Intel do not have PCI-E 4.0 boards yet, now whether this is a factor that affects us who knows.. If you have one of the latest AMD systems with PCI-E 4.0 then you wont have to worry about any of this..
Nvidia has said the cards will be PCIE gen 4 but PCIE is backward compatible. It is very unlikely that even the highest end cards will be fast enough to saturate a x16 PCIE gen 3 connection. Making the cards gen 4 is just checking a box for newest tech (basically).
Micron has not announced any prices for GDDR6X prices yet. Current GDDR6 prices are behind a pay wall. You can probably find older prices with some searching. 1Gb modules have been around $11 for a while.
...I am using a PCIE 3.0 x16 Titan-X in a PCIE 2.0 x16 slot and there is only a minimal delay as scene data initially is transferred to the card. After that it renders nice and fast.
Are you sure that wasn't 21GBps, referring to the speed?
My own opinion is the 3090 (should it prove to be real and not vapor-tech), will have maybe 12GB; the ti version, should it be proved a requirement due to what AMD do will have more - maybe 16GB or more.
Question: is it now possible in IRay (and/or Cycles) to have 2x3070 and have access to the combined VRAM? Looking at the proposed prices, that might even be cheaper than one of these top end GPUs.
It is possible in iRay on the 2070 Super, 2080, 2080 Super, 2080 ti, RTX Titan and all the RTX Quadros. I have yet to see any details on whether the cards need to be precisely matched for manufacturer and brand etc. but I'd stick to that if I was buying them, as that was what SLI required. As to Cycles you need to check the Cycles documentation to see if they have enabled it. What Ampere cards will have NVLink connectors is not known. To share VRAM the cards have to have the NVLink connector and an appropriate NVLink bridge connected, be aware these bridges are not cheap.
Sorry - I missed something: which two did you buy? 2x2070? $80 each?
EDIT: I think I get it now - NVLink is not something that comes with the GPU when you purchase it, right? You have to buy it. I thought it was something like a ribbon cable that connected two cards together. Nevertheless, I don't see 2070 mentioned on the sites I've just searched so my idea of 2x3070 might be a non-starter anyhow.
You have 4 cards?
Guys, Nvidia selling a x80ti for a lot less than the Titan class while being extremely close to that Titan (or even beating it) has been a key marketing strategy for years. If you really follow the industry this should be obvious. They have put out Titans at crazy absurd prices, then either in that generation or the next, Huang will go on stage and say how their new x80ti is just 5% less than this Titan while costing so much less. What a deal!
And if $1500 is too low for a GPU that has 24GB, how on this earth did the 1080ti ever work at $700? 11GB isn't that much cheaper. The Titan Pascal launched in 2016 at the price of $1200. It had 12GB VRAM. The 1080ti launched a few months later at $700. That is a $500 difference, correct? The cost of that Titan was significantly higher, yet they were nearly identical. Not only that, but Titan was exclusive to Nvidia, meaning that its 1417 or so clockspeed was all it offered. Meanwhile the 1080ti was sold through 3rd parties who cranked it all the way to 2Ghz. So many 1080tis can out perform the Titan in gaming. Oh, and the 1080ti had 95% of the VRAM, with just 1 single GB less than the Titan offering.
The 1080ti made a GPU that was released just mere months before completely ridiculous. The Pascal Titan was not the previous generation like Turing is about to be, Pascal was in full swing. But it worked. The 1080ti sold many, many units. The Userbenchmark database shows 5,668 user benchmarks for the Pascal Titan. But the 1080ti has almost 490,000 user benchmarks! This was a stroke of brilliance for Nvidia, and if anybody in this thread believes that Nvidia wouldn't do "x" because they wouldn't want to make something else obsolete...you are fooling yourself. The Titan is a marketing tool as much as a GPU. It is their halo product, their pony car, the Titan is an advertisement for how powerful their GPUs are. The Titans have always had crazy high prices, and Nvidia is not focused on how well the Titan sells.
The 1080ti demonstrates several things. That Nvidia will undercut themselves at will. It also shows that a top end GPU could be built for $700, its only been 3 years, people. You are telling me that they cannot build a GPU with 24GB and sell it for $1500? Come on. And they can still maintain a huge mark up. Pretty much any company would be envious of Nvidia's mark ups! LOL.
But these rumors do not point to a Titan. They point to a 3090. Just a few posts ago I told you that the name is key to understanding. The name 3090 says two things: it is not a Titan, but it is higher than a 3080ti would be. Thus it will be priced accordingly. It will cost more than a 3080ti would (if it exists), but it is STILL a gaming card, and thus will not cost what a Titan does. The last Titan was $2500.
And as I said, we are on the brink of a new generation of gaming. It is time to level up. The idea that 8GB will be enough is nonsense. That is about to change in a big hurry with what is coming. Cyberpunk's specs have not been posted yet. If you found specs, those are purely guess work...funny to talk about that in this thread. I will tell you this, there is no way that Cyberpunk is going to play at its 4K max settings on a 2060. Sorry, not happening. Maybe at 1080p it might be able to, LOL. But Cyberpunk is also a game that has been in development for almost a decade. This game has actually been built around aging hardware, not new hardware. So it is totally logical that the game is designed for less VRAM because that is all they had. A number of things have been holding gaming back, and VRAM is a big one. Developers have been building games and trying to make them work on freakin' potatoes for years. On PC, most publishers focus on the mass market to sell as many copies as they can. That results in games with less variety in their environments than there could be. You want more foliage? Sorry, you aren't getting that because so many people own 1060s. It is a chicken and egg situation.
The consoles are going to shift that quickly. Not overnight, but I believe it will be a quick transition as more games release that take better advantage of the console's capability. PC ports might be downgrades rather than upgrades!
And besides, if Nvidia isn't going to put more VRAM on a GPU, AMD looks to be ready to do so. The rumors point to them using 16, and they have done it before. And oh, the Radeon 7 was $700, with 16GB of HBM memory. If AMD is indeed competing with Nvidia this time around, Nvidia shipping with just 12GB on their top card would look pretty silly. AMD right now is riding a huge hype wave. They have people anticipating their card, so if they are really close to the 3090, offer 16GB and do so at a reasonable price, then Nvidia might be in trouble.
This is unknown, but I believe it is very unlikely. The original 2070 does not support Nvlink. The 2070 Super only supports Nvlink because it is a cut down 2080 and uses the same board that just happens to have Nvlink connectors on it. That is the only reason it has it. The original 2070 is a totally different chip and board. The 3070 so far appears to be a different chip and board from the 3080, so I would be really shocked if it did.
I'm so sick of bangingmy head against this wall but once again. MATH!
GDDR module prices have been reasonably stable for a long time. At launch they are around $12 or $13 per module (at base speed the higher speed modules cost more) (both GDDR5 and GDDR6 were at those points. I don't have the GDDR4 prices but those were on cards back in 2007) and then dropped to $10 to $11.
So the first Pascal Titan had 12 Gb at around $11 each equals $132 so not a significant part of a cards cost. A few months later the 1080 ti is out with 11Gb so assuming they're now paying in that same ballpark that's $121. It would be 17.28% of the total cost of the card.
$400 of a $1400 card is 28.57% of the total cost of the supposed 3090. That would be more than an 11% reduction in the profit margin. Do you really thing the margin on these cards was that large?
With respect - because I know nothing about these things - you are asking everyone to take as gospel your numbers. Are you sure that NVidia are paying your repeatedly stated $400 - if so, what is your source? I would find it hard to believe that this would be public knowledge but, again, I know nothing.
The math is pretty simple. Even at $13 a Gb 24Gb is $312. Micron themselves have said BDDR6X is higher than their usual cost. I simply put it at 16 per. 16 x 24 = 384. I just used $400 as a round number. You can easily find plenty of places that quote GDDR5 and GDDR6 prices. If you want tomorrow mornings quotes you'll have to pay but old prices are available Google has them "GDDR5 price" certainly works as does "GDDR6 price"
I still find it hard to believe that NVidia - a very big concern - are tied to the pricing that anyone can look up on the internet. Surely they have some bargaining clout?
Why else would I need two bridges?
OK, so I did. Perhaps not surprisingly, nothing came up to indicate that the price is published. To the contrary, Digitimes went through the trouble to publish a report to estimate the cost of GDDR5, a strange thing to do if the information is publically available. The report also said that it could not even estimate the cost of GDDR5X, an even stranger thing to say if the vendors are required to publish their prices.
Another 3DCenter.org report on the cost of memory was said to have used list prices "sourced" from DigiKey. "sourced" is not a journalistic term for publicly available information.
Can you provide any evidence to support your claims?
I also find it difficult to believe that NVidia and their partners have to go to micron.com just like everyone else. As far as I know, the DoJ's definition of "price fixing" is when multiple vendors agree, for anti-competitive reasons, to mutually configure their prices, not when a single customer negotiates with a single vendor, i.e. does what every business does every day.
well they actually were not, as the Titan series could be configured to TCC mode (which sidestepped windows WDDM) while the TI couldn't. that is a major game player as one could use the entire 12 GB VRAM for rendering.
So hard to find
https://www.guru3d.com/news-story/gddr6-significantly-more-expensive-than-gddr5.html
Also sourced from Digikey means they got it from Digikey which is a distributor of semiconductors.
You did see in there the rather important cavet:
"The mentioned prices correspond to a purchase quantity of 2,000 pieces. Manufacturers of video cards are likely to buy larger quantities, which means they could get the parts cheaper. 3dcenter.org estimates the possible discount at 20 to 40 percent, the prices in the table below would only be estimates"
Which means that major manufactures of video cards could get GDDR 6 for as low as $6 a GB, making a price for 24GB of $144
That was of course a guess based on no knowledge of antitrust. The makers cannot make any such price fixing deals any more. That's what they keep getting busted for doing.
Price-fixing is collusion to maintain artificially high prices and is aimed at harming customers. It's not a conspiracy to lower prices. That is predatory pricing, the opposite of price-fixing, which is done in order to harm competitors and create barriers to entry, and one of the things anti-dumping laws are meant to prevent. The kind of thing ride-hailing companies have done to the taxi industry.
IANAL, but there's an interesting blog that discusses just how complicated all this antitrust stuff including predatory discounts gets, with discussion of actual cases, which typically drag on for years. https://www.theantitrustattorney.com/
Anyway, discounts are common in all industries. Daz offers loyalty/volume discounts in the form of purple banners, December discounts, etc in addition to all the other sale discounts.
My experience is in fashion apparel wholesale, where in most cases discounts are dictated by the customer, rather than the supplier and take the form of payment discounts when payments aren't early at all, distribution discounts, margin guarantees, abusive non-compliance chargebacks, advertising discounts, new-store (including renovated stores) discounts, loyalty club discounts (you didn't think the retailers ate those, did you?), priority shelving contributions, and e-commerce discounts. One big box retailer in particular extorts vendors into massive discounts for goods going into their me-too e-commerce channel by making purchases for their physical stores contigent upon them.
Again and again.
There are 3 comapnies that make RAM. They have been sued repeatedly for price fixing of every kind (and not just in the US most notably by China).
The claims have ranged from charging more to certain companies, Chinese ones by China, to charging some less, Samsung charging itself less by Taiwan, and pretty much everything in between. You're of course welcome to look all this up. Some of this is widely thought to be attempts to get these companies to give up their ip to a certain large nation where none of them are headquartered but that doesn't change the facts. They cannot charge anyone anything but the flat rate.
I'm not going to explain what price-fixing is and isn't again. If all these things are so easy to look up, it shouldn't be difficult to provide sources. I mean it's easy to look up proof that the earth is flat, too.
https://en.wikipedia.org/wiki/DRAM_price_fixing#:~:text=On%2027%20April%202018%2C%20Hagens,price%20of%20DRAM%20nearly%20tripled.
https://www.pcgamesn.com/samsung-micron-sk-hynix-dram-fine
I wonder if Iray will get NVCache, or Nvidia will neuter it like in the past.
Both of thse are regarding price-fixing, but only to keep prices high. I even linked the first article in my post.
Again, list prices. Out of the three paragraphs, two of them referred to the price for 2000 units, meaning volume matters.