NVIDIA is obviously not above market segmentation via dubious means (see: driver limitations for consumer GPUs), but I think binning due to silicon defects is a more likely explanation in this case.<p>Some 4090s have this extra fp16 -> fp32 ALU disabled because it's defective on that chip.<p>Other 4090s have it disabled because it failed as an Ada 6000 for some other reason, but NVIDIA didn't want to make a 4095 SKU to sell it under.<p>Or if you generalize this for every fusable part of the chip: NVIDIA didn't want to make a 4094.997 SKU that only one person gets to buy (at what price?)
This is why Nvidia needs competition. I love the performance of their hardware and the quality of their drivers, but I don't love being their customer. They have a long, long history of price discrimination using techniques like this. Back in the day it was "workstation" graphics for CAD programs that they would nerf for consumer cards by downgrading various features of OpenGL.<p>Different markets, same techniques. It's in the company DNA. That and their aversion to open source drivers and various other bad decisions around open source support that make maintaining a working Linux GPU setup way harder than it should be even to this day.
There was a time when Intel seemed unbeatable. In 2000 they had a 500 billion USD valuation. That's almost a trillion dollars in today's (2024) USD. Today they are valued at 90 billion USD and Broadcom was thinking about buying them...<p>My point is these things don't seem to last in tech.
Between EVGA getting out of the Nvidia card business, Nvidia continuing to be problematic under Linux (even if that’s improving), all the nonsense with the new power connector, and the company’s general sliminess, I’m increasingly leaning towards an AMD (or potentially Intel) card for my next tower upgrade.<p>AMD and Intel might only be competing in the entry-to-midrange market sector but my needs aren’t likely to exceed what RX 8000 or next-gen Intel cards are capable of anyway.
I'm ambivalent about this sort of thing (or, as another example, Intel's CPUs many years ago that offered paid firmware upgrades to enable higher performance).<p>On one hand, it's very bad because it reduces economic output from the exact same input resources (materials and labor and r&d).<p>On the other hand, allowing market segmentation, and more profits from the higher segments, allows more progress and scaling for the next generation of parts (smaller process nodes aren't cheap, and neither is chip R&D).
Binning and market segmentation are not mutually exclusive. Of course they're going to put their best-performing chips in the most expensive segment.
There is competition, but the competition isn't winning 1st, 2nd or 3rd.<p>There was a day in which Intel was top dog and for a long time, now AMD are competing in 1st place.<p>AMD and/or other competitors will have their day, but today its Nvidia.
It's interesting how we the people (broadly) accept this practice in software and even some hardware, but not in other areas. Note how frustrated people are when you hear about "unlocking" sensors and services available on cars.<p>If a product is made, and the cost to provide that product is the same one way or the other but you cripple it to create segmentation, then that is greed. Period. Objectively. And if you're okay with that, then fine, no problem. Just don't try to tell me it isn't maximization of profit.<p>There are no heroes in the megacorp space, but it would be nice for AMD and Intel to bring Nvidia to heel.