![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Interesting how the 4060 will have only 8GB VRAM, but to be fair the 3060Ti has only 8 yet it runs considerably better than the base 3060. So I wouldn't pan it yet.
970 gtx had 4 GB the 1060 GTX had 6 gig of ram.
That's true. It makes more sense that it has 8, seems like a downgrade to the 'masses' who saw the 3060 has 12 but don't know the other factors.
I think this is what they're getting at. Lower RAM with a similar memory configuration. But the CUDA core count is higher by over 25% so that would be a boost, as well as the GDDR6X over non-X with the 3060 which should definitely give it a considerable gain. Also 36 SMs compared to 28 on the 3060.
The 1060 had 3 and 6GB variants with the 6GB ones being significantly more popular since they could also double for efficient mining cards
Not necessarily. It is low for a higher end card but if the memory clock is high enough, it will balance out some.
How so? Doesn’t 192 have to be in multiples of 3 to work properly? The leak says 8GB at 192 bits which would be a very odd configuration. It would either be 8 GB at 256 bits or something like 6 or 12 at 192.
6GB was also more popular because it wasn't castrated by Nvidia's deceptive marketing department.
It does not. Generally, this would be the optimal case as it allows each 32 bits of memory bus to connect to one package of GDDR (being either one or two GB VRAM). A card that does not follow this will access some of the memory at a partial bandwidth. In the case of the 4060, 8 GB would be split as such: 4 GB would run on their own 32-bit lanes, the remaining 4 would run with 2 GB per 32-bit lane. GDDR has something called clamshell mode, which allow two packages to connect to one channel, doubling memory. It is less efficient than if you had 8 GB on 256 bits, so it certainly isn't optimal, but it is a viable option to raise the performance of the card somewhat than being forced to limit it to 6 GB VRAM, which would be a significant step down.
60 series have never been considered a higher end card.
I wasn't referring specifically to the 60s with this in mind. I was thinking more about the RTX line as a whole, compared to the 1600s or other less powerful card lines. Maybe I'm looking at it a little too black and white.
Isn’t that what they did with the 970? It was hardly a good idea
No, the 970 was 256-bit and 4 GB (512 MB per 32-bit lane), ostensibly. Nvidia settled a class-action lawsuit over the 970 because 512 MB was separated from the main line so the card effectively ran as with 3.5 GB. I remember getting a $30 'consolation prize' due to that.
I fully agree. The 970 was a fantastic card for my build several years ago. Of course the 3060 I have now is a big step up, but for its time the 970 was an excellent mid-range card. I could play almost anything maxed out on 1080p with it.
How is this a worst case scenario?
Only for people who rely on resale value to bankroll their hobby.
But if prices remain consistent then new tech is just costing more.
Some people get really obsessed with GPU resale value and try to time it just correctly. Sometimes to get burned like selling 2080tis for $400 only to find rtx 3000 cards noexistant lol.
To be fair, that happened once. Time will tell if that becomes the norm moving forward
To tell you again, I’d had to have told you at least once before.
Well that's because the 3060 Ti is using the same chip as the 3070 while the 3060 is a tier below chip (GA104 vs GA106)
My system is to sell a few months prior to the next launch (and use a back up card and/or my gaming laptop in the interim), thereby maximizing resale value. People who wait within a month or two to sell normally get burnt because, at that point, rumors are starting to solidify on performance and price of the next gen and those with the current gen cards panic. The situation with the 2080 ti selling for $500 a few weeks before the 3xxx series launch (which were extremely hard to get) is the prime example. It's just like any investment market: you need to not be greedy holding (and be willing to go a few months without assets).
True. If you want to see some absurd return-of-Jesus-and-Buddha optimism (some members still here I won't call out), go look up the old Vega thread.
If everything is made at TSMC then 30% could very well be on the low side. The samsung process seems to be quite inferior to what is available at TSMC. Also keep in mind that ray-tracing is still immature so we might see massive gains there as well.
Idk. Never viewed a GPU as an investment. I've never gimped myself to a backup card because, well, I still want to be able to play the same stuff.
And the absurdity of 30% being a "good" outcome for a revised architecture on a new node isn't lost on us old timers. As late as 5 years ago, a 30% uplift after two years would have been considered a middling disappointment -- but here we are.
So true. Back in the days we used to be disappointed when below 50% and happy when you saw 70% or higher increase over previous architectures initial release. Anything in-between was as expected. It used to be new architecture->refresh->new architecture with 2 years between architectures and refresh in-between. Now the refreshes barely offer up any increase and the architecture to architecture gap has been fairly low despite being 2 years or so between each new architecture. E.g. 3080->4080 would be an architecture jump and way back in the days we would have usually gotten somewhere between 50 and 70% gains, if not more 15+ years ago.
Are you using some actual pricing tier system or looking at 1% low, ms for a frame instead of average fps to say that ? Really depend on resolution
