NVIDIA's product strategy: one step forward, 4GB of VRAM back
Description
This is a four-panel meme using the 'Car Reversing' format to criticize NVIDIA's product decisions. In the top-left panel, a man wearing a cap with the NVIDIA logo looks forward while driving, with the text 'RTX 3060 12 GB' below him. The top-right panel provides the context: 'people not upgrading because it has enough VRAM'. The bottom-left panel shows a hand shifting an automatic car's gear into reverse. In the final panel, the man, again representing NVIDIA, looks back over his shoulder with a sly smirk, and the text now reads 'RTX 4060 8 GB'. The meme humorously illustrates the community's perception that NVIDIA deliberately released its newer-generation RTX 4060 with less video memory (VRAM) than its predecessor. This is seen as a cynical business move to create an artificial bottleneck, especially for VRAM-heavy tasks like AI/ML model training and high-resolution gaming, thereby encouraging users to buy more expensive cards or upgrade sooner
Comments
35Comment deleted
Upgrading from a 3060 to a 4060 is like refactoring your code and ending up with fewer features and more memory swapping
Convincing a 3060-12 GB owner to “upgrade” to a 4060-8 GB is like mandating a Rust rewrite for performance but capping the heap at 512 MB - looks great in the micro-benchmark, face-plants the moment a transformer wanders in
NVIDIA's product team discovering that developers actually use VRAM for more than just storing their collection of unhandled exceptions and memory leaks - turns out those ML models and 4K textures need actual memory, who knew?
When your 'upgrade' path involves explaining to stakeholders why the newer GPU generation has less memory than the old one, and they ask if you're sure you read the spec sheet correctly. RTX 4060: proof that in GPU land, progress isn't always linear - especially when product segmentation meets the reality of LLM fine-tuning requirements
4060 8GB is the PM’s idea of an upgrade: bigger number, smaller heap - great in the deck, same OOMs in production; I’m keeping the 3060 12GB
NVIDIA's genius: Shrink VRAM to 8GB on 4060 so your fine-tuned Llama finally learns humility via OOM
Calling 8GB an upgrade over a 12GB predecessor is the GPU equivalent of moving to microservices and halving every service’s heap - enjoy torch.cuda.OutOfMemoryError
Actually, the 4060 (8GB) seems to be better than the 3060 (12GB) https://gpu.userbenchmark.com/Compare/Nvidia-RTX-4060-vs-Nvidia-RTX-3060/4150vs4105 Comment deleted
Don't use userbenchmark Comment deleted
Why do you point out to benchmark when the meme is about the VRAM size? A bit better performance not gonna help if you are getting out of fast memory located next to the chip Comment deleted
Because the “extra” memory is valuable, but it’s useless if the performance is not good. In my experience, having more is not always a synonym for better, but having the proper balance is what outstands overall. Comment deleted
With the current gen of consoles the minimum requirement for vram on PC is 12 GB, which comes from the way consoles memory subsystem works. This may partially change with dataset streaming but it's yet to be seen. That's why any new GPU having less than that and not being lowest-end is a waste of sand. It's Nvidia's/AMD/Intel job to design GPU in a way that it can efficiently use such memory buffer Comment deleted
true, as far as you want to play PS5 games on PC through Play Station PC distributor - but if you want to buy a graphics card that is as expensive as a next-gen console, then maybe consider if you can’t just buy the console itself, as it has been created for games and it doesn’t have the performance issues reported on PC Comment deleted
ps5 has no games Comment deleted
Switch has all the games Comment deleted
There’s just too many tasks when 8Gb is just not enough anymore, especially if we are talking about gaming Best reference made in the chat is that video memory size on consoles is what should be used as reference for middle-level renderers and nvidia specing 8Gb model for -60 model is nothing but a dirty game Comment deleted
https://www.youtube.com/watch?v=RQSBj2LKkWg Comment deleted
It's perhaps the worst benchmark out there Comment deleted
You're probably better off reading marketing materials from corporations than this crap Comment deleted
I honoured owner of 3060 12G, and only my next upgrade going to be 4080/4090 in September after (I hope 😭) price drop when "new" ti cards showing Comment deleted
Why not Radeon? Comment deleted
some people are just Nvidia fan-boys (in the same way that you can find a “war” between Intel and AMD followers) Comment deleted
And some have use cases like professional workload or streaming and they require CUDA or part of the drivers suite that Radeon doesn't have Comment deleted
Fuck 4060 Comment deleted
consoles usually are better for gaming; PC wins if you want to mod games - unless you need that GPU for work Comment deleted
Me crying with my 3070 and 8gb ram:/ Comment deleted
you see, it runs a whole GHz faster Comment deleted
Yes especially if clown influencer on YouTube try to compare arm benchmark scores with amd64/x86_64/i64 Comment deleted
muh m1 is fastor intel iz ded letz boy obble silicon for 5k$ Comment deleted
M is still pretty for now it will have the same overblowting broblem and if actually decides to remove hardware features because they become legacy then say welcome to a new locked down shit that wont even support apps that are older than 6 years lmao Comment deleted
apple user spotted the apple symbol is not unicode standardized and only appears on (most) apple products. everyone else sees the "unknown symbol" box Comment deleted
I know don’t worry I set up my keyboard to replace every time I write apple to the logo lmao. Btw 🍎🍏🍏 Comment deleted
🍏 Comment deleted
Is there a ascii apple art? Comment deleted
M1 is not technically ARM (they'd have to pay licensing fees if it was ;P) Comment deleted