Shrinkflation is a variant of inflation where manufacturers reduce the quantity of a product while keeping the price the same or slightly higher. This is commonly seen in consumables like soda bottles, bags of chips, soap bars, etc. What you might have missed is that shrinkflation has also been sneakily affecting GPUs, with manufacturers offering less GPU power for the same or even more money.

If you examine the core configuration, memory bandwidth, and VRAM capacity of Nvidia GPUs over the last few generations, it's clear that you've been paying more for less. This trend is also present on AMD GPUs, but to a lesser extent. Hence, I focus only on Nvidia GPUs in this article, paying special attention to the latest RTX 50 series GPUs.

3 Nvidia GPUs have fewer cores than ever

The issue at the core

It used to be that when you bought an 80-class or 70-class Nvidia GPU, you expected a certain level of performance, relative to the flagship SKU. However, since 2022, we have seen the nomenclature become meaningless, as Nvidia actively cuts down its GPUs more aggressively compared to the most powerful card in the lineup.

If you consider the GTX 900, GTX 10, RTX 20, and RTX 30 series, you will see the 80-class card featuring around 70% of the CUDA cores of the flagship card, with the RTX 3080 being the exception, as it received around 83% of the cores found on the RTX 3090 (something that hadn't happened since the GTX 700 series). Nvidia immediately followed up with a drastic core count reduction on the RTX 4080, which featured only 59% of the CUDA cores available on the RTX 4090 — the lowest ever in recent memory.

Things became worse with the latest RTX 5080, as Nvidia deemed fit to reward it with only 49% of the CUDA cores found on the monstrous RTX 5090. The price of the RTX 5080 is also half that of the RTX 5090, but how impressive is that when you aren't getting a decent 80-class product, relatively speaking? The story is darker with the 70-class cards — the RTX 4070 and RTX 5070 featured only 36% and 28% (respectively) of the cores found on the top card in the lineup, a massive cut compared to the 55% average seen till the RTX 30 series.

This shrinkflation has slowly but surely ensured that consumers pay more for less GPU power year after year. AMD has also been guilty of this, considering the RX 6800 XT had around 90% of the cores on the RX 6900 XT, but in the next generation, the RX 7800 XT received only around 70% of the cores on the RX 7900 XT.

2 Memory bandwidth is at an all-time low

You call it sandbagging, I call it shrinkflation

CUDA cores aren't the only thing getting the ax on Nvidia GPUs; memory bandwidth is another area where the 80-class and 70-class GPUs have become less exciting than ever. Since the GTX 900 series, Nvidia's 80-class GPUs have featured around 70% of the memory bandwidth of the flagship GPU, with the RTX 3080 again being the exception with 81%. The RTX 5080 only has 54% of the memory bandwidth of the RTX 5090, falling somewhere between a 60-class and 70-class SKU.

And once again, the RTX 5070 fares worse, as it receives only 38% of the memory bandwidth seen on the RTX 5090, compared to the 61% average seen till the RTX 40 series. This number makes it worse than a 60-class card, since the latter has received, on average, around 45% of the memory bandwidth of the flagship card. The disappointing bandwidth, combined with the historically low core count, makes the RTX 5070 one of the worst Nvidia GPUs to ever come out.

1 We're also getting lower VRAM per dollar

This will surprise no one

Finally, Nvidia GPUs have been offering less VRAM per dollar over the years, robbing consumers of yet another important element of a GPU. Nvidia's VRAM problem has been discussed ad nauseam, but what you might have overlooked is that Team Green has cut down on the amount of VRAM compared to previous generations, especially on the 70-class cards. Even the upcoming RTX 5060 is all but confirmed to offer 8GB of VRAM — a mere 25% of that on the RTX 5090, well below the historical 41% average for the 60-class cards.

Coming back to the RTX 5070, its 12GB framebuffer equates to around 38% of the VRAM on the RTX 5090, whereas the average for 70-class cards has been around 50%. Even the RTX 5080 falls slightly below average, since its 50% number is lower than the 55% average for the 80-class SKUs. Nvidia has been hitting all its SKUs from all sides; whether it's the core count, memory bandwidth, or VRAM, consumers are getting the short end of the stick.

Nvidia's shrinkflation is worse than others

Shrinkflation reduces the amount of product the consumer receives for the same price. However, in Nvidia's case, gamers have been paying more for less GPU in many instances. This makes the story much worse than what regular shrinkflation would lead you to believe. Consumers have little control over such tactics; the most they can do is vote with their wallets against disappointing SKUs, but in the face of little competition, we haven't seen that happening. Maybe AMD's RX 9070 and RX 9070 XT will convince gamers to avoid Nvidia's mid-range options like the terrible RTX 5070.