Blackwell based consumer GPUs have come to market, and unlike previous generation where datacenter products had slightly different architectures with Volta/Turing and Ada/Hopper, or used different nodes within Ampere, nvidia brought everything together this time… and guess what, all the developement went in the datacenter direction.
Gaming results are underwhelming: the performance gains over Ada are not even proportional to the massive bandwith and the slight cuda core count increase, meaning the new architecture has actually regressed in that space!
The DIE SIZE have not changed between 5080 and 4080, therefore we have not seen a shift in the product stack as some reported: GB203 was always going to be a 80 class SKU unlike what happened with the 4070ti being intended as 4080 12gb . Being on the same TSMC 4n, cost for the cards has just gone up because of the new GDDR7 modules, and even efficiency has not improved!
Now, don’t pretend the second biggest company in the world can make “mistakes”: launch shortage and inflated prices are planned because nVidia does not actually want to sell the cards at MSRP, and is moving steadily to a luxury brand mindset. No reasonable person would spend time to find a 16gb 5080 and buy it for more than 1200 euros when the performance is within 10% of a 24gb 7900xtx readily avaiable withou any hassle for 900. Another clear indication of this strategy is the evergrowing price gap between the x080 and x090 class cards wich is now close to 100%.
The 32gb framebuffer on the 5090 will make it an ideal target as compute cards anyway, and i predict its price will never reach MSRP in its entire lifespan just like the 4090 before never did.
How does this makes sense for Nvidia?
The answer is – this has been building up at least since Pascal but the company has always had such a tight grip into consumer mindset that reviewers are screwed anyway; they cannot speak the truth out loud as that would lose both access to nvidia cards for preview benchmarking and support from the very same audience they should inform, as i face whenever i make videos showing the issue of nvidia driver overhead. For this generation, the cards were incredibly bad value even at sauggested price wich is frankly insane and never seen before in the tech industry, wich shows that nvidia doesn’t even care that much about good reviews, as their strategy in the consumer market has fully moved to that of a luxury brand. Their target are not looking at reviews; team green has built a cult following during the years, It would be stupid to let other players take their place in the market without cashing on this while keeping the whole supply chain alive with minimal developement time and wafer allocation possible.
You have the adults that were raised with the dream of owning high end GeForce wich in their head is the highest number they can get their hands on, and 5080 sounds good enough for them – and if their life has been good enough they might even get a 5090; the fact that they are so hard to find and expansive becomes a feature of the product. It’s not like they can see the difference between 100 and 150 fps or tell the input lag from frame generation when you are 35 old anyways. HW Unboxed – one of the few media outlet that had the balls to speak out against nvidia in the past, and faced important backlash doing so – clearly said we also have to look down the stack to board partners: if the supply of chips is so low, they have to make up the loss of income from the low volume by raising the price of individual cards. This is fully endorsed by nvidia – if they wanted to punish for not selling cards at MSRP they’d have full power to do so.
Then you have the kids that just care about having a GeForce at all – maybe the previous guy children. These will be feed by super tiny lower end cores that sell like hotcakes also in prebuilts machines and keep the volumes for the AIB partners. That explains how bad the lower end 5070 and below will perform, as i predict that they will be avaiable in higher volume to grab on this demographic.