128GB is all you need.
A few more generations of hardware and open models will find people pretty happy doing whatever they need to on their laptop locally with big SOTA models left for special purposes. There will be a pretty big bubble burst when there aren't enough customers for $1000/month per seat needed to sustain the enormous datacenter models.
Apple will win this battle and nvidia will be second when their goals shift to workstations instead of servers.
My guy, look around.
They are coming for personal compute.
Where are you going to get these 128GBs? Aquaman? [0]
The ones who make RAM are inexplicably attaching their fate to the future being all LLMs only everywhere.
End users will still get access to RAM. The cloud terminal they purchase from Apple, Google, Samsung, or HP will have all the RAM it will ever need directly soldered onto it.
For inference, there is already a 10x improvement possible over a setup based on NVIDIA server GPUs, but volume production, etc... will take a while to catch up.
During inference the model weights are static, so they can be stored in High Bandwidth Flash (HBF) instead of High Bandwidth Memory (HBM). Flash chips are being made with over 300 layers and they use a fraction of the power compared to DRAM.
NVIDIA GPUs are general purpose. Sure, they have "tensor cores", but that's a fraction of the die area. Google's TPUs are much more efficient for inference because they're mostly tensor cores by area, which is why Gemini's pricing is undercutting everybody else despite being a frontier model.
New silicon process nodes are coming from TSMC, Intel, and Samsung that should roughly double the transistor density.
There's also algorithmic improvements like the recently announced Google TurboQuant.
Not to mention that pure inference doesn't need the crazy fast networking that training does, or the storage, or pretty much anything other than the tensor units and a relatively small host server that can send a bit of text back and forth.
Isn't reading from flash significantly more power intensive than reading DRAM? Anyway, the overhead of keeping weights in memory becomes negligible at scale because you're running large batches and sharding a single model over large amounts of GPU's. (And that needs the crazy fast networking to make it work, you get too much latency otherwise.)
> becomes negligible at scale
Nothing is negligible at scale! Both the cost and power draw of the HBMs is a limiting factor for the hyperscalers, to the point that Sam Altman (famously!) cornered the market and locked in something like 40% of global RAM production, driving up prices for everyone.
> sharding a single model over large amounts of GPUs
A single host server typically has 4-16 GPUs directly connected to the motherboard.
A part of the reason for sharding models between multiple GPUs is because their weights don't fit into the memory of any one card! HBF could be used to give each GPU/TPU well over a terabyte of capacity for weights.
Last but not least, the context cache needs to be stored somewhere "close" to the GPUs. Across millions of users, that's a lot of unique data with a high churn rate. HBF would allow the GPUs to keep that "warm" and ready to go for the next prompt at a much lower cost than keeping it around in DRAM and having to constantly refresh it.
Flash has no idle power being non-volatile (whereas DRAM has refresh) but active power for reading a constantly-sized block is significantly larger for Flash. You can still use Flash profitably, but only for rather sparse and/or low-intensity reads. That probably fits things like MoE layers if the MoE is sparse enough.
Also, you can't really use flash memory (especially soldered-in HBF) for ephemeral data like the KV context for a single inference, it wears out way too quickly.
However, for old-style 1-bit per cell flash memory I do not see any reason for differences in power consumption for reading.
Different array designs and sense amplifier designs and CMOS fabrication processes can result in different power consumptions, but similar techniques can be applied to both kinds of memories for reducing the power consumption.
Of course, storing only 1 bit per cell instead of 3 or 4 reduces a lot the density and cost advantages of flash memory, but what remains may still be enough for what inference needs.
Difference is that the current prices have a lot of subsidies from OPM
Once the narrative changes to something more realistic, I can see prices increase across the board, I mean forget $200/month for codex pro, expect $1000/month or something similar.
So its a race between new supply of hardware with new paradigm shifts that can hit market vs tide going out in the financial markets.