If prices are not dropping, there are only two reasons. Either supply is lower than expected or demand is higher than expected. We know that supply is going to vastly increase over the next few weeks and months with used mining GPUs. That means demand is rising as prices are dropping.The prices have been speculated to fall for years now and they did to some degree but it seems like they won't fall as fast as people thought or down enough to justify the wait.
It's also governed by profits and loss. If it's true that Nvidia GPUs have less profit built-in then it's unlikely the prices will fall significantly until companies start to feel the pinch from lower sales. At some point they may have to dump 30 series if the new AMD cards show better price/performance. For example, if I can get 3080Ti performance at $700 instead of $800, you may see some adjustments. On the other hand if AMD is at parity in price and performance, then prices will remain high, until such time as companies need to reduce inventory.I would imagine these cards will start to fall, probably dramatically but that will be governed by the market. Used cards should start to show up on eBay soon ....
I would argue that after the 40 series announcement they are not getting steadily lower. In fact, I've seen some price increases in the last week. The 2 GPUs I'm tracking haven't move a penny in the past 2 weeks. Mobos have dropped a tiny bit, maybe $10-20. Otherwise, things are either out of stock, or at the same price.Let's look on the bright side. Prices are steadily getting lower and we are not all lamenting the insanity of the crypto boom anymore. I for one am happy. My target price range is around $500. I paid around $400 for a 1070Ti in 2017, so for me something that is equivalent to a GPU just shy of the $500 range. I am really hoping that the new AMD cards can make this a real option for me.
Am I happy with the new nvidia pricing, not at all. More than just the price, it is the size and power needs of the card. The joke now seems to be that you need to plug your MB/CPU into your GPU card.
For some context:
1) 1080 Ti was released in 2017 at $699. Inflation adjusted for today it would cost $850.
2) The Titan Xp (maybe this is a more realistic comparison to the 4090?) was released in 2017 at $1200 and is inflation adjusted to $1500. The MSRP for the 4090 will be $1,600, which is actually pretty close.
I guess it is all about perspective.
They're all sh1tcoins. People always make this mistake: they expect past performance to be indicative of future results. ETH's story of success may never be replicated. There's a big chance ETH's merge could throw all POW under a bridge. But tell that to a guy who spent a ton of money on a mining rig.A lot of former GPU miners are probably holding out hope that another POW will moon, but I think the landscape is too fragmented; Ravencoin, Ergo, Eth Classic (and technically, ETHW) are all jockeying for it and all of them seem to have pretty entrenched communities. At some point they'll have to liquidate something to keep running.
As long as their is a gap in pricing like the price model Nvidia shown off 4000 series on top and 3000 series on the bottom. Nvidia wants them to coexist 1) probably because Ampere yields are so good that they are still making a profit and have room to move. 2) they want to sell as many as 4000 series cards as possible especially before AMD can launch and have competitive products.My guess is that both Nvidia and AMD will create a false shortage of the new products until they can sell off their remaining inventory of the previous gen. By that time, the gullible will be bursting to give them their cash for the new (very expensive) stuff, that gives them 10 more FPS in their games.
If they aren't bottoming out by Cyber Monday they'll probably keep gouging right through the holidays, and this is exactly what I expect. Jan 1st will probably be magic time.
Nobody can give you an actual answer, because such information is propitiatory and kept under close supervision. One can make some educated guesses though. The AD102 has a 608 square millimetre die and is fabricated on TSMC's N4 process - this is based on their N5 node, but is a little cheaper to manufacture. A single 300 mm wafer, on the N4 process, can produce approximately 90 dies in total, although the percentage that will be usable, let alone 4090-usable, is unknown.Does anyone knows to give me an answer !?How much does it cost to manufacture a 4090 Gpu (end product)!?
Thanks for the effort!Nobody can give you an actual answer, because such information is propitiatory and kept under close supervision. One can make some educated guesses though. The AD102 has a 608 square millimetre die and is fabricated on TSMC's N4 process - this is based on their N5 node, but is a little cheaper to manufacture. A single 300 mm wafer, on the N4 process, can produce approximately 90 dies in total, although the percentage that will be usable, let alone 4090-usable, is unknown.
But let's say 70% of them are good enough to be used in the relevant models and out of those, say 40% are for the 4090 (the remainder will get used for the professional line of Lovelace-powered cards, and the new 4080s). A single wafer can cost up to $15k to manufacture, and given that N4 is relatively new, there's not going to be any huge cost savings yet. But let's take a round number of $10k - that means each AD102 die costs around $110 to make. However, since 25 dies from each wafer end up in a 4090, Nvidia will charge a higher price for them to AIB vendors, simply because the supply of them is less than the more damaged ones that get used for the 4080s.
The dies need to be packaged and it's anyone's guess as to how much that costs, in terms of materials and manufacturing, but then one needs to add in the mainboard, all of the components on it, 24GB of 21Gbps GDDR6x, and the cooling system being used. Once again, the price of that RAM isn't made public, but given that all electronic components are still in relatively short supply/priced higher than normal.
Nvidia stores and distributes its packaged GPUs from a center in Hong Kong, as the manufacturer of every other part is based in China or Taiwan. So those costs need to be included, as well. So one could be looking as high as $400 to $500 to manufacture and distribute a 'standard' 4090, perhaps considerably more.
Nvidia charges a small fortune for that model for three reasons: (1) because they can; (2) because it's an LVS (low volume seller); (3) market positioning.