←back to thread

65 points fidotron | 2 comments | | HN request time: 0s | source
Show context
littlestymaar ◴[] No.43575210[source]
Re-using a comment a wrote some time ago:

Tenstorrent really needs to put more VRAM on their cards.

If chinese companies can hack Nvidia GPUs with 48 or 96GB vram at a competitive price, surely Tensorrent can too.

Variants of n300d at $2500 for 48GB and $3900 for 96GB would be instant hits.

~~24GB for $1500 simply isn't gonna do it.~~ (old part of the comment related to the old n300 which can be update with: 32B for $1400 still isn't enough for success. There's some progress, but that's still too low considering it's exotic hardware that will lead to tons of compatibility issues).

replies(4): >>43575246 #>>43575326 #>>43575727 #>>43579793 #
1. aseipp ◴[] No.43575326[source]
The new p150 cards linked in the OP have 32GB GDDR6 @ 512GB/s for $1,300. Which isn't bad on paper, I guess. They're meant to be networked (quad 800GB QSFP-DD) like Nvidia GPUs, so two of them would get you 64GB of VRAM at $2600 for ~600W which is basically what you're asking for? The power usage isn't good enough yet at scale I think, but for a workstation it's quite manageable.

Real workloads remain to be seen, but if they can actually get a working build of vLLM and their cards remain actually buyable, well, they're doing better than some of the competition...

replies(1): >>43575428 #
2. littlestymaar ◴[] No.43575428[source]
> so two of them would get you 64GB of VRAM at $2600 for ~600W which is basically what you're asking for?

Almost, except with respect to space in the box and power usage, which are critical IMHO.

> but if they can actually get a working build of vLLM and their cards remain actually buyable, well, they're doing better than some of the competition...

That's a big if though, poor software support is to be expected and you'll need to factor that in IMHO, and that's why they need to beef up the memory. Of course if software support is stellar then it may be good enough of a deal.