Nvidia could bring back the 12GB RTX 3060 as supply issues disrupt GPU roadmap

Daniel Sims

Posts: 2,472   +74
Staff
Rumor mill: Rumors that Nvidia might restart production of a 5-year-old graphics card that currently tops the Steam Survey have circulated for months. The latest claims, however, push the relaunch to the summer, while another rumored mainstream GPU refresh could be delayed or canceled.

Prominent leaker MEGAsizeGPU recently claimed that a long-rumored version of Nvidia's RTX 5050 with increased memory capacity has been delayed and might never see release. Meanwhile, the still-popular RTX 3060, originally expected to have returned to the market by now, could instead fill the gap in the release schedule in June.

As AI data centers consume a growing share of memory production capacity, maintaining a steady supply of consumer graphics cards has become more difficult. This is especially true for Nvidia's high end RTX 50 series GPUs, which use GDDR7 memory.

The RTX 3060 and RTX 5050, by contrast, rely on GDDR6, which remains more affordable and widely available. After earlier rumors suggested that Nvidia might relaunch the RTX 3060 in the first quarter of this year, CEO Jensen Huang expressed tentative support for the idea and even hinted that a new version could include AI features from the RTX 40 and 50 series.

Despite the RTX 3060's age, it remains at the top of the Steam Survey, likely due to its 12GB of VRAM compared with the 8GB found on the RTX 4060 and RTX 5060.

Nvidia reportedly stopped producing the RTX 3060 in 2024, and inventory likely ran out by December of the following year. Earlier rumors of a revival suggested that Nvidia might reintroduce an 8GB version, but the latest information indicates that the new 3060 would retain its full 12GB memory.

MEGAsizeGPU also recently claimed that a new version of the RTX 5060 is in development, using binned chips originally intended for the RTX 5070, which could improve performance. Despite moving away from the original GB205 configuration, the updated 5060 is expected to retain its 8 pin power connector. New laptops from Lenovo and Asus are also expected to feature a 12GB version of the mobile RTX 5070.

Graphics cards are not the only legacy chips that the memory crunch might bring back. Another leaker, HXL, recently shared packaging photos of a new edition of the Ryzen 7 5800X3D, which AMD may release to mark the AM4 socket's 10th anniversary.

The CPU, which introduced the company's successful 3D V-Cache technology, remains the best processor for AM4 motherboards, which remain popular after the DDR5 RAM required for AM5 boards became prohibitively expensive.

Permalink to story:

 
My God! That's 5 year old GPU. In the past manufacturers at least would gave recycled chip a new name and some speed bump, coming from node maturation. Now I'm sure We will see a price bump.
 
So a delayed 5050 9GB that probably won't be any faster than the current 5050 8GB, which in return is only about 7-12% faster (depending on resolution) than a 5 year old 3060 12GB. Neither card is well suited for trying to run the half dozen or however many software options to enable RT/PT/extra frames and so on. It's almost a wash between the two cards.

Only thing the 5050 has going for it is the lower power draw when gaming (roughly 50W) and if anyone cares, AV1 encoding capabilities. Otherwise it has less VRAM and is priced practically the same.

If you're into 1080p resolution and just need an entry level GPU for gaming, these will work for you. It just depends on what you feel is the better choice. A current gen with less VRAM and probably a longer support life or a "new" 5 year old architecture with more VRAM and probably shorter support.
 
Bring the Linux drivers up to speed and we might have a winner. It's really only 90 class card owners that hate on affordable GPUs. Not everyone needs or can afford 5090 so for people who can't afford one or simply sees spending that much money on a GPU as stupid, these cards are great and 90% thinks 60 class GPUs are great.

And the fact that it has 12gigs of RAM makes it a decent AI GPU.

The 30 series GPUs are lacking in the Linux driver area, but maybe we'll see some focus from NV on the 30 series if they are going to re-release them.

I wonder if they're using Samsung memory because they recently announced they were stopping production of LPDDR4 and, I'm shooting from the hip here, I think GDDR6 uses the same node as LPDDR4
 
We made Nvidia what it is today - the gamers, developers, renderers, professionals, etc. I understand that a company can choose to pursue other forms of business anytime but at least give us some damn respect where it's due!

This level of dereliction has consequences. Think about the people that came to work for Nvidia, people with a passion to see graphics capabilities expand, to see a better future. Your losing that idea of the 'pursuit of dreams'
 
Using the relatively less expensive VRAM and, probably more importantly, more available process node has cost advantages.

However, modern GPUs aren’t that overpriced. I find it hard to believe that enough people will buy a 5 year old GPU to save $70. Most consumers are tech illiterate but understand 5060 > 3060. The only way this makes sense is if Nvidia believes things are going to get substantially worse. Luckily there are increasing cracks showing in the hype bubble.
 
Last edited:
AMD uses GDDR6 (not GDDR6X), but their video cards are not much "worse", it seems that video memory speed is not the most important thing.
 
Last edited:
And guess what?? Nvidia will charge over $700 for it!!
How about bringing back the 3050 for $500+ ??
Too bad there is no 3040!

People will still buy it, because you know, Nvidia and "AMD drivers".....as they have been repeatedly told by reporters here 100s of times.
 
Here is something I want to know

HBM4 memory is better for AI, and its ok for Gaming, DDR is cheaper to make than HBM for cost per meg
here is official Costing
💡 Why DDR5 is cheaper to manufacture
Simple 2D design: DDR5 chips are made using traditional planar DRAM processes and then mounted on PCBs (DIMMs).
High volume production: It’s used everywhere (PCs, servers), so fabs are optimized for mass production.
No advanced packaging required: No stacking or exotic interconnects.

👉 Typical cost:

~$2–$5 per GB

With this price, lets go for $20 a Gig to be safe
RTX5070 - $499
+12GB x 20 = $240
= $739
This should be the price and I am happy to pay 800 for a 24GB Card that will last 3-5 years with that RAM and also do AI

Why do they charge $900 for a 16GB.... Because people keep paying for it and never challenge being ripped off
 
It's not just that the older GPU uses older RAM where the supply isn't as tight. It's also built on an older process node, and their fab partner may have more available capacity for that. The newer GPU has a smaller die that would normally be less expensive to make, but not if all the fab capacity is being bid up to high prices because of demand for AI chips.
 
And guess what?? Nvidia will charge over $700 for it!!
How about bringing back the 3050 for $500+ ??
Too bad there is no 3040!

People will still buy it, because you know, Nvidia and "AMD drivers".....as they have been repeatedly told by reporters here 100s of times.
Yes, until the price is known it's difficult to make a decent rounded out opinion about this.

It's old tech, should be cheapish. But it's NV, likely to cost orders of magnitude above what would be reasonable. Possibly as much as a really high quality leather jacket?
 
Back