- *for certain definition of self hosted.
Developed by Chinese open weights model dev Z.ai, GLM 5.1 isn’t exactly a small model. At roughly 754 billion total parameters (40 billion activated for each token generated), the model requires roughly 380 GB of GPU memory just to hold the weights and another 6-26 GB per user to maintain context depending on precision.
To serve GLM-5.3 at scale, Revolut deployed 56 Nvidia B200 GPUs, equivalent in GPU count to seven eight-GPU DGX B200 systems. But as Abitbol noted, the greatest performance gains didn’t come from adding GPUs, but rather tuning the model’s hyperparameters along with switching to NVFP4 quantization, implementing disaggregated prefill and decode, and tuning context lengths.
The quoted numbers are the interesting part. 380 GB just for weights plus several GB of context per user means this only pays off at Revolut’s scale. What stood out to me is that the article credits the biggest gains to NVFP4 quantization, splitting prefill and decode, and context tuning rather than adding GPUs. Those tricks carry over to much smaller setups too.
Agreed. Economies of scale are about the only way this makes sense.
I was just working thru the use case for purchasing a 64gb Strix (vs the typical coding I might do using OR or Open Code Go). The ROI is somewhere in the order of 7-10 years for me. So, on a purely dollar cost basis, local inference is $$$. OTOH, from a sovereignty and “own your own code” basis…there’s something to be said for it.
Probably I will run a few tests using OR with a few outstanding tickets of mine, to kick the tyres of Qwen3.8-27B. I’ve tried Qwen3.6-35B and frankly it’s…not very good at code gen or patching.
To boot, code gen isn’t the only thing I use LLM for, so I suppose it depends on how bloody minded I want to be about future use case (I do actually have one).
The investment of just under £500,000
And so, the slow but inexorable clawing back what was taken begins.
Oh, wait. Fintech? Shit.
Cute that they called it $0.3 / 1 M tokens.
GPU WORLD
US$559,950 per unit: https://gpuworld.ai/gpu/nvidia-blackwell-b-series-gpusSo, IRL…about $3.5 million dollary-doos capex. Given they probably have thousands of engineers on staff, yeah, I can see it being worth it.
That username’s great, so no shade here: I don’t often engage with fresh accounts until they’re no longer wet from the hatching. Nothing personal, just basic hygiene. Thank you for the reply, though!🤘🏼
No dramas. This is probably my 3rd or 4th Lemmy account in last 2yrs; trying dbzer0.com to see what it’s like.
We’re a pretty decent bunch, IMHO. This is my fourth or so, I think? I hear ya. Welcome!
In the spirit of things, I asked GLM5.3 what that set up actually looks like. Ouch.
Fetched that page for real. Specs (from gpuworld’s DGX B200 datasheet page):
Dimensions: 17.5" W × 87.5" D × 42.9" H (cm: 44.5 × 222.3 × 109.0) Weight: 275 lb (124.7 kg) Rack height: 10U equivalent Power: up to 14.3 kW max per systemIn real-world terms:
Height: 10U = 10 rack units = 10 × 1.75" = 17.5" tall — about the height of a bar fridge standing on its side. But that's the chassis; you'll also want ~2U of clearance for airflow/management, so plan for ~1.5U of breathing room above it in a rack. Depth: 87.5cm = just under a metre — so it'll stick out of a standard 1m-deep server rack unless you've got deep racks (or you let the back panel hang open). Standard 42U racks are typically 1.1–1.2m deep to handle this. Weight: 275 lb / 125kg. That's a hefty man's weight — two people minimum to lift, ideally four or a rack lift. Standard datacentre floor loading is usually fine at this density but worth checking if you're stacking multiple in a rack.Rack math for 7 units: 70U of just these systems — so a single 42U rack can’t hold them all (you’d need 2 racks), and at ~14kW each you’re looking at ~100kW total draw — that’s serious power/cooling, way beyond a standard office closet.
Short version: think “fridge-sized box, half a tonne with rails, eats 14kW” — not a pizza box.
Ha! I’ll look forward to your onramping, fellow human. Good luck!
BRB, just getting some scratch lotto tickets.
14.3 kw. That’s fucking obscene.
And by all rights, the Fintech datacentre is what…small? medium?
GLM again:
One DGX B200 flat-out: 14.3 kW × 24h = ~343 kWh/day ≈ what a typical UK house uses in a month
Datacentre size: a full rack of AI gear like this is ~50–100kW. Classification-wise:
Your average office server closet: 5–20 kW Small datacentre: <1 MW Mid-size / enterprise: 1–10 MW Hyperscale (AWS/Google/Microsoft): 50–500+ MW per campusRevolut’s 56 GPUs ≈ one dense rack or two — well under 1 MW, so firmly “small,” and that’s exactly why they colocate/rent space (“a US-based service”) rather than owning a facility. Self-hosting the GPUs doesn’t mean self-hosting the building. You don’t put 100kW of fridges in your London office — you rent rack space in a proper colo where the power and cooling already exist.
Enjoy the lotto. If you win, you can buy an eighth of a DGX.
Fuuuuuuuuuck
What I don’t unterstand about stories like this: Why aren’t the US companies making a fuckton of money then? If you can self-host something at 10% of the costs, shouldn’t that mean that a datacenter can make around 90% margin with interference?
Ah…that’s just the one-off costs. Remember…they still have to pay monstrous electric bills. Maybe in the order of $100K or more a month.
Doesn’t Revolut pay for electricity either?
Of course they do…but it’s not stated in the costs they cutely described as $0.3/1M tokens.
The ones GLM5.3 worked thru ($3.5M capex, based on GPU costs) didn’t include ongoing colo costs. So, figure on $3.5K + whatever site rental and electricity cost (I read the article to suggest they were using a colo not on site, but the same idea stands either way: no free lunch just because you have the hardware)
