Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by justinclift | Hacker News Reader

Full thread

justinclift·

From the data sheet:

https://resources.nvidia.com/en-us-design-viz-stories-ep/pro...

* NVIDIA Ada Lovelace architecture-based CUDA Cores: 14,080

* NVIDIA fourth-generation Tensor Cores: 440

* NVIDIA third-generation RT Cores: 110

* Single-precision performance: 69.3 TFLOPS

* RT Core performance: 160.2 TFLOPS

* Tensor performance: 1108.4 TFLOPS

Seems like it'd be in between the 4080 and 4090 raw power wise, but with double the memory of the 4090?

View on HN

What really surprised me was the 285W TGP.

That's mighty impressive considering that the previous generation equivalent (the RTX A6000) used 300W at less than 60% the (theoretical) performance.

Ada Lovelace is really power-efficient. It's a shame it's so expensive because otherwise it's a great GPU architecture.

I heard the power bill for this AI stuff is pretty ridiculous too though.

This is a cut-down RTX A6000 (i.e. same generation) made to fit within the limits imposed by US sanctions against China. I think you're misinterpreting the performance numbers.

The current gen is the RTX 6000 Ada - the RTX A6000 is Ampere, i.e. previous gen. Confusing naming scheme, I know.

Yeah, I meant that one. So confusing and easy to typo...

Sadly, those are FP8 FLOPS in tiled matrix multiplications with structured sparsity, according to footnotes.

Very hard to estimate the value of this GPU without third party benchmarks.

And comes with a price 5 times that of the 4090 for no apparent reason, I assume?

Doesn't seem to have pricing info anywhere obvious online, but yeah it probably is because "professional market" plus "its the only high end part they're allowed to get in China now". ;)

Probably not 5x, more like 5-6k. The a6000 is $7k and this is a cut down version.

Reply on news.ycombinator.com