THANK YOU NVIDIA

Especially if you want it silent. Either 4 slot designs with tons of copper so say goodbye to your pcie slot or an fucking aio.

Not that i have anythign against the idea of an aio on an gpu. But normies can fuck up anything

>160bit
That's the same bus width as a 1060, 2060 or 3060
nvidia's so cocky that they're selling the xx60 class lovelace die as a 4070

videocardz.com/newz/geforce-rtx-40-ada-gpus-to-feature-very-large-l2-caches-nvidias-own-infinity-cache

still 1ghz lower then AMD rdna3

I'm more interested in the cuda core count and vram since I'm going to have to undervolt the hell out of this because of the ridiculous power requirements

Until amd makes ROCm not a complete fucking nightmare to work with I'm going to move along.

POODNA3 is not hitting 3GHz, fuck off faggot

Will this make up for the low memory bandwidth?

I swear to god, I’m fucking sick of this old ass architecture. When will this shit stop spiralling out of control?

You'd have to either introduce some sort of regulations through laws, or you'd have to convince everyone to stop buying. Neither will happen.

Are you retarded? Ampere GPUs used 14Gbps memory, Ada GPus are using 18Gbps memory, there's no lower memory bandwidth

Can’t they start using ARM to solve this shit? I pay more in electricity than internet atm.

to some extent, yes

Attached: E2rEszJXEAA2mDS_format=jpg&name=large.jpg (1818x1024, 221.54K)

check the bus widths of the xx80, and xx70 gpus

How do you imagine ARM solving anything for GPUs?

I guess that makes sense, I was more concerned with the lower 160 bit bus, I wasn't thinking about the faster ram.

You mean the "old-ass" Lovelace architecture that hasn't even been released yet?

Still younger than x86

>old ass architecture
that has nothing to do with it. it's a conscious decision by nvidia to fatten their chips to the breaking point of pcie (and soon, socket) power limits

And what does x86 have to do with this thread?