RTX 5060s Announced.

Almost any GPU made in the past 10 years can run "4K", but running it well on recent games not so much.

I have this card downstairs in my HTPC connected to a 4K HDTV

https://www.gigabyte.com/Graphics-Card/GV-N3050OC-6GL#kf

It can run Trails games, emulators and some platformers fine in medium settings. Anything heavier and I need to run it as 1080p medium and use 2x upscaling for playable frame rates.

Running a modern game at 3840x2160 on Ultra/Legendary/Fusion Reactor settings for 60+ FPS requires a very beefy, into the $800~1000+ range.

Does it include the portable nuclear reactor it needs to power it?

I think 4k/60 at moderate settings is respectable for an entry level GPU. The fact that you can run 1080 and most 1440 in eye candy mode seems pretty solid to me for sub 500, but my GPU is about 4 years old.

Should be fine on normal wall outlet. It is pull start like a lawn mower, so the engine carborator handles most of the power. The only real issue you will have is when it needs cleaning or repairs

They're RRP, Nvidia hasn't sold for that for years. The values are meaningless. 5090 RRP is £1800-2000, sells for £2400++

They're RRP, Nvidia hasn't sold for that for years. The values are meaningless. 5090 RRP is £1800-2000, sells for £2400++If they were going to switch they'd likely have done it across the board. The $600 it'll sell for isn't "dirt cheap" either.

OEMs only just stopped shipping laptops and mini PCs with 8gb RAM as the base model just this year when they should have stopped 3-4 years ago really.

I'm not really up on the business outside of the US, but I know that most everyone else is subject to various additional fees. I thought it was only slightly worse than tariffs. +40% is brutal if that's a retailer selling that.

That would depend on the game. When it comes to raw rasterization resolution matters a lot. 1920x1080 is 2,073,600 pixels per frame that need rendered sixty times per second for a total of 124,416,000 pixels/s. 3840x2160 is 8,294,400 pixels per frame, a 4x increase in raw power requirement. Now there are lots of other things involved but that gives you a good idea of the kind of power jump that happens from 1080p to 2160p. 1440p is 3,686,400, less then half the requirement of "4K". Expecting 3840x2160x32 @60fps on medium from an entry level GPU is absurd. 1440p is more palatable and 1080p upscaled is doable. 1920x1080 is integer compatible with 3840x2160 making post process upscaling computationally cheap. This is how consoles can get away with "doing 4K".

It's also super game dependent, doing Trails of Cold Steel II, Grandia II HD Remake, Astlibra, etc on my 3050 is fine at 4k. Visions of Mana on the other hand is one of those games I'm best at doing 1080p and x2 upscaling. Stepping up to Tales of Arise and it's pretty much unplayable. xx60 series GPUs are similar only with a higher bar one set of games with lower demands might play fine, but then Cyber Punk 2077 starts up and the GPU emits a death cry.

:Note:
60 models are main stream, 30,40 and 50 are entry level. 70 are upper mainstream enthusiast cards with 80 being the high end. 90's are just renamed Titans and for extreme setups.

It's just supply and demand, that's all. First you have those AI companies willing to throw big money away on those chips, occupying the production capacity. then you have rich gamers willing to pay anything for a 5090.

So there are room to jack up the price because the market demand is strong.

The scarcity came from the demand from the consumer.

If nobody is buying any of those the price will drop. You can't have cheap stuff if someone else are willing to pay more money than you.

The problem is the prices will not come down no matter what happens.

They'll just keep pushing up and that's it. You either pay $600 or you buy a console/laptop.

30 series are still selling for $300+ and they're 2 generations old. You could've literally bought a 3090/4090, used it, and resold it for profit.

There is no real shortage of 5090s. It is entirely fake like when Apple purposely only made limited numbers of the gold iPhone years ago, or complete *** morons paying £12 a bottle for Prime which tastes like piss+sugar.

Actually no, the shortage is quite real but for a very specific reason:

https://www.visualcapitalist.com/nvidia-revenue-by-product-line/

As you can see there, Data Cards closed out 78% of Nvidia's revenue for 2024 and the market for AI/Data is likely only going to increase from there. I myself am a Steamfitter and we're building Data Centers non-stop in Virginia. Can't seem to build the damn things fast enough lol so almost every ounce of good Silicon Nvidia can buy from TSMC is mostly going to the A100 and H100 Cards.

Gamers and Creators simply got shafted this time around. Just a simple business move was all it was and the numbers make perfect sense :( They are almost certainly only making the bare minimum of 50-series cards just so they don't lose too much market share to AMD and the rest are getting sent to the Data pile full-send lol.

The market is finally starting to break "A Little bit" and more 5090s are showing up at local MicroCenters more often but it will be a while before it's stabilized. I'm just glad I was able to grab mine haha.

Aren't you the one who said manufacturing industry is just bunch of cheap labour and 5 year old working in a factory? How come such problem still exist?

Maybe one day when China can finally mass produce 3050 or equivalent with their cheap labour and armies of 5 yr old workers, then you can get cheap 3050 with everyone undercutting in the market!

These just don't make sense between pricing and bad drivers, unless your GPU literally dies and you have no choice. My local Microcenter has the 16GB 5060 Ti broadly in stock, for the 5070's MSRP. I built a PC a couple of weeks ago. There was a ton of 5070's in stock when I picked up my order, for almost $800. You can get a 5070 Ti for about the 5080's MSRP. The 5080 is in stock today, for approaching the 5090's MSRP. And the 5090's are commanding an enormous premium, if you can find them at all. All of this is exacerbated by the 5k-series features like MFG being less valuable as you go down the product stack due to latency.

I'm feeling pretty good about my $600 9070XT. Of course, pricing today on those is way out of wack now as well. The gamers who are really sitting pretty are those who got a 4090 or even 4080 Super at MSRP.

I mostly agree with this, but the way these GPUs are named is inconsistent. The 8GB card should be named "5060" and the 16GB should be the "5060 TI". Typical gamer is going to believe they can plug in any 5060 TI and crank it up to 4k out of the box. If nVidia was cooperative with their marketing then we wouldn't have so many people thinking that is, or should be a thing.

The xx70 tier is another example. The 5070 is mid, but I would qualify the 5070 TI as a high tier GPU. 5090 is overkill for gaming so I push that off into it's own space. Can be used for gaming, but probably better deserves the 'productivity' flag.




VAT fees? Isn't this a fee people in EU pay on top of the suggested retail? Because that would explain why a 1800-2000 MSRP is actually costing 2500. This is all after the 15-20% AIB upcharge for the bells and whistles.

I'm definitely with you on the supply shortage though. All the big manufacturers have their own fabs, but they're still relying on TSMC for a good part of their supply.

Another thing to note regarding MSRP:

The only cards that are subject to these prices are the base models of the GPU. If it's labeled with any marketing at all (ie. Overclocked/Superclocked, RGB, white shroud/PCB, replaceable/magnetic/triple/windofrce fans, ect.) you should expect a price for that premium.

I'm seeing a lot of complaints about 'fake MSRP' but they're referencing and even picturing models of the GPU that have obviously been modified and are no longer 'stock' models.

Yes, it sucks because we see less and less of the real MSRP models, but I think it's important to draw the distinction. It's misleading when AIBs label their products as MSRP when they're just signaling that this is their 'normal' price for the GPU after they've added all the ***.

If there's an over supply of 5070s and 5080s but a shortage of 5090s then yes, it is a manufactured problem. 5070 and 5080 are barely better than the 4070 and 5080 and they knew people wouldn't be buying them en masse. They didn't make enough 5090 to keep the price up.

At the time of launch the 2GB chips were £0.5 and the 3GB chips were £2.50 (an extra £2 per chip). There are 4 2GB chips in the 5060 and 8 in the 5060TI so using the 3GB chips would only have added £2 to the 5060 manufacturing cost and using 6 3GB chips for 18GB in the 5060TI would have been +£11. Of course this is an oversimplification and there's probably more to it like just reusing the same PCBs.

I'd suggest everyone skips this generation.

The founders edition cards are selling over MSRP

Did you read any of the articles people provided? 3nm and 5nm chip productions are all full because of AI industry demand. And AI industry just pays more money for those chips than gamers. Of course the production will prioritize AI industry over gamer's need since it's better money.

They didn't produce less 5090 to "keep the price up", they only choose to produce AI hardware over graphic cards for better money instead.

5070, 5080 and 4070 all uses 5nm. It doesn't matter which card is better by how much percentage because the price is determined by supply and demand. The issues is that there aren't enough 3nm and 5nm chips being produced for both gamers and AI industry.

I honestly didn't even know founder's edition was a thing for 5060s, but I doubt this would have any impact on the taxes, tariffs and other fees associated with your region. Again, I'm assuming this is tacked on the the MSRP so that's about as 'real' as you're going to get.

edit*
No founder's card for 5060 ti, guess you were still talking about 5090.

When I went to Microcenter in the Flushing NYC location on launch day they only had 4 5090s and 64 5080s

was nutty - I didnt get a 5090 till march but I was only going on the weekend after launch.

Exactly. Just by last year's numbers alone, Nvidia made 3.54 times as much money on Data Center Cards than all of their other Graphical and/or Gaming Cards COMBINED. They could go 100% Data Center if they wanted to and made us all wait another 2 or 3 years for 50-series release but they probably don't want to leave the door too open or else AMD and now Intel might gain an opportunity to seize market share.


Nice, stopped by Microcenter VA for something unrelated and saw an MSI 5090 Vanguard edition in the window and grabbed a rep immediately lol. 4090 resale was surprisingly fast too. Ebay seems to have quite a few 5090s for sale at 2500-2800 is why I said it appears that the market is starting to break a little bit. So anyone after a specific card, you can probably wait another month or two and find it just fine.

1. Noone is making AI chips on 5nm now, that's absurd.
2. Google use their own TPUs for inference.
3. Groq runs on LPUs for inference.
4. Deepseek and Qwen will be running on third party servers, probably Mistral too. Deepseek and Qwen were trained on old hardware.

They very clearly made too many 5070+5080 and not enough 5090s. There is no legitimate reason for this unless they really did *** up their own understanding of the GPU market (unlikely).

5070+5080+5090 all have FE, all selling over MRSP

It all takes the same Silcon... So if your market for Data Center Cards is ENORMOUSLY LARGER THAN ANY VERSION OF CONSUMER CARD it doesn't make any business sense NOT to prioritize that. TSMC has a limited number of wafers they can produce in a cycle; a wafer cycle is generally around 14-20 weeks, they're not just gonna yeet that ***carelessly lol. I've been in these places they don't mess around (Micron VA and Intel Ronler Acres Oregon), Nvidia most certainly took like 90% of their stock and said "A100 and H100 Pile" and left the rest for Consumer Cards.


This is certainly possible but also not likely because of the Limited Availability of Wafers vs Die Sizes:

The NVIDIA RTX 5090 GPU die size is significantly larger than the RTX 5070 GPU die size. The RTX 5090's GB202 GPU has a die size of 750mm² (or 761mm² according to another source), while the RTX 5070's GB205 GPU has a die size of 263mm². The RTX 5090 also features a higher number of transistors (92.2 billion) compared to the RTX 5070 (31.0 billion)

So Nvidia very likely banking on selling Quantity instead of Quality to make their money. I wholeheartedly believe they severely underestimated the demand for the 5090 and the consumer's ability to put money aside to purchase them, but it doesn't look like any sort of scheme - at least not to me. Just plain cash grab, man. I don't know how else to say it.

I think you misunderstand these cards, the 5060 TI 8/16GB models are identical GPUs.

GDDR DRAM chips take two 16-bit data paths each, making it one chip per 32-bit memory bus. The 5060 has a 128-bit memory bus meaning four chips per card. Current densities are 16Gb (@GB) per chip, meaning four chips at 2GB for a total of 8GB of VRAM. Now GDDR has the ability to operate in what's called clamshell mode. In clamshell mode they will operate in two 8-bit channels per chip allowing double the total number of DRAM chips at half the bandwidth per chip for the same total bandwidth. This is what the 4060 and 5060 16GB models are doing, its the exact same chip but with the DRAM operating in different modes. Clamshell is rarely used because it reduces the chips effective bandwidth in half and bandwidth is usually far more important to performance then raw capacity.

The 4070/5070 have 192-bit memory bus's for six total chips (twelve in clamshell mode), thus 6 * 2GB = 12GB VRAM. The 4080/5080 use 256-bit memory bus's for eight total chips, 8 * 2GB = 16GB VRAM. You can always tell memory capacity by the memory bus width.

https://en.wikipedia.org/wiki/GeForce_RTX_50_series

5060
GPU Die: GB206-250
Core: 3,840/etc..
Clock: 2.28/2.50

5060 TI
GPU Die: GB206-300
Core: 4,608
Clock: 2.41/2.57

The 5060's are basically defective 5060 TI's that have had part of their die's either fused off or disabled. The 5070 is using a GB205-300 while the 5070 TI is a GB203-300 die (defective 5080).

At least you can say you saw them. None of those GPUs even exist on this side of the pond unless they're from resellers. What does the retailer say when you ask them if it comes with a reach around?

I think people just do not get how silicon allocation works.

All the vendors bid for silicon allocation from TSMC and from those bids win various allotments on various process nodes. Older nodes are obviously cheaper but it's still a fixed amount of total capacity. From that wafer allotment you can then cut the disks into various chips with smaller chips being way easier to fit onto a die.

A GH200 is 814 mm2 worth of silicon and makes a product that sells for 30~40,000 USD depending. The GB200 is the newer version and it's larger and makes a product that sells for upwards of $50K or more (they don't sell individual units so per-GPU price is hard to get).

The 5090 use's the GB202-300 die at 750 mm2 and makes a product that sells for 2~3,000 USD. 5080 is half it's size at 378 mm2 and so forth. With datacenter GPU's selling faster then nVidia can make them, they are losing money by making any consumer GPU's.

https://wccftech.com/nvidia-ship-150k-200k-blackwell-gb200-ai-servers-q4-2024-500-550k-units-q1-2025/

Seriously, Jensen Hwang can close nVidia's entire consumer division and transfer that allocation to Hopper/Blackwell AI GPU's and make more money. The only reason nVidia is still selling any consumer GPU's is that Jensen likes to wear leather jackets and stand on stage acting like he's Steve Jobs.

This is part of the reason why I wanted to see the head to head, specifically in instances where they're not spilling. I'm curious why 28G/s and not 32 tho? It makes my brain cell itch.

Here you go:

https://www.trendforce.com/news/2025/03/12/news-meta-reportedly-tests-in-house-ai-chip-on-tsmcs-5nm-with-mass-production-set-for-2026/