So i was sitting here wondering, if there was a market for a completely insane price-does-not-matter machine from hell running a suped up version of flame , what would that look like?
My Guess?
Heavy focus on realtime very-large-model-ML Tools as that seems to be the realtime “frontier” right now and that stuff is expeeensive.
Pretty much would be the spiritual successor to SGI … NVIDIA DGX supercomputers.
maybe something like a DGX H100 Server
built with 8 H100 accelerators, for a total of 32 PFLOPs of FP8 AI compute and 640 GB of HBM3 Memory .
would be siiick duuude. price still cheaper than high end inferno systems of the time
the DGX H100 was priced at £379,000 or ~US$482,000 at release
As far as I recall, the last iteration of Inferno was a special for Japan, which comprised Flame with on-board burn.
Of course, I may be totally wrong…
As to the initial question it’s interesting because all the things Inferno did are doable in a Mac Mini now, and a full spec flame is leagues past one of the old SGI refrigerators and their accompanying drive arrays.
I don’t spend very much time rendering, so my limited imagination is really struggling with building out something. Now most of my time is setting up camera projections and tracking, so if the supercomputer could background process many tracks both 2d and 3d while I’m doing other stuff so I could just attach objects without managing tracks, that would feel powerful and inferno-like.
It would do what every non-Flame artist thinks Flame does: realtime rendering. A system that could render in realtime to super high res systems like The Sphere would probably command $1M+
Its more about what a spiritual successor would look like, what would a “flame ultimate 3000” even be, what would you want it to do.
There honestly wasnt much fancy schmanzy expensive hardware out there but with these super expensive AI cards from nvidia …
Its all a joke anyhow nobody will pay 2million for a machine to render a commercial.
HOWEVER ML has maybe opened a door back into the high end world that has been shunned for so long from our part of the business - as you can actually not run large models on current “consumer” or even “prosumer” hardware , at all.
i think a flame ultimate 3000 machine wouldnt just do what we have always done but faster, id like to see it do multiple aspects of a job behind the scenes. i did a panasonic ad once where i bounced to houdini and after quite a few youtube clips i learnt how to do a body of water to replace a swimming pool surface. imagine if this machine spun up an instance of houdini or unreal and using chatgpt/ai created what you needed and rendered it.. ie assets, environments but specific and dialed in, etc. that would be cool and left us to do the creative more. automatic tracking in the bg, ordering flat whites when my mouth was dry.. this would also lead to our inevitable demise and destroy more peoples jobs so im not sure if any of this is a GOOD idea.
The new NVIDIA DGX Station is a first-class AI computer mainly used for AI training and inference, but it is not suitable for graphics rendering and video encoding and decoding. NVIDIA AI GPU cards are mainly Tensor Core， RT Core is rare, so at least for now, it is not suitable for applications like Flame that heavily rely on graphics rendering and video encoding and decoding. So I think Flame should add multi GPU support features, such as an 8 GPU Server, and the latest RTX PRO GPU Cards are very suitable.
Yes however , if you add a simple GPU for legacy stuff, you would have a realtime AI workstation that can do fast and large genAI stuff .
I dont know about you but if I had a AI workstation for a sort of client-attending AI sessions in confyUI or whatever (there isnt s tool yet hence “inferno”) I think i could have that thing booked in no time.
It catches the spirit of inferno, the best machine humanly possible to help bring a creative vision to life and that live!
I wouldn’t be surprised to see some kind of NVIDIA Grace or SOC solution that is somewhere between the DGX behemoths mentioned on this thread and the DGX Spark. I kind of feel like Intel is in trouble whilst AMD is trying to pivot.
It is an AI system but the DGX Spark is a pretty cool bit of kit, but not for what we are doing.
