Limited-time dealSupermicro HGX B200, 8 GPUs8-GPU NVIDIA HGX B200 server in stock
View the specification Call +91 92058 93902innovations@udhyamlabs.tech
AI hardware advisory
This practice answers one question: what hardware you should buy and house yourself, buy and colocate, or rent, for what you intend to run. If you already hold a quotation, we check it. We take no commission, referral fee or rebate from any supplier, and an assessment never recommends equipment that we have listed for sale.
Sizing a GPU purchase, or checking a quotation before you sign it
We size from the work you intend to run, and not from a catalogue of parts. A GPU server, the kind of machine that runs AI models, has several limits: how fast it calculates, how fast it moves data through its memory, and how much memory it holds. We test which limit your work reaches first and specify the servers for that limit. No supplier pays us, so the recommendation can be to buy what was quoted, to buy less or more, to buy a different part, or to rent first.
What the work is planned to deliver
- A written recommendation, with the arithmetic shown in your currency
- The cost of owning the servers, in your own facility or colocated in another firm's data centre, and of renting, compared on one basis in your currency
- A specification for the servers and their network, where you ask for one, with the power and cooling the data centre must supply
- NVIDIA
- AMD
- Bottleneck profiling
- Network design
- Power and cooling envelope
- Cost per million tokens
The chart shows what renting, colocating and buying cost at different levels of use. The three bars show which limit of the hardware the work reaches first.
What the assessment examines, what you provide, and where the risk is
The assessment has a fixed fee and a fixed length, and it ends with written findings that are yours. You can stop there.
The assessment
We take the quotation, if there is one, and a description of what you intend to run, and we test a representative workload. You receive a written recommendation to buy as quoted, buy less or more, buy a different part or rent first, with the arithmetic shown in your currency. For a quotation check, the assessment is the whole engagement.
What you provide
You provide the quotation and its validity date, if there is a quotation, the models and the number of users you expect, and the response time the use case needs.
Where the risk is
Demand that is not yet known carries the risk. Where there is no workload to measure, we run a representative one on capacity rented in an account you control, and we state how uncertain the result is before any capital is committed.
How the work is done
We start from the workload: which model, at what context length, at what concurrency, and with what response time target. Profiling that workload shows the binding constraint, meaning the resource that is exhausted first. At low batch width, decode is limited by memory bandwidth. As batch width rises it moves towards compute, and for long contexts the memory taken by the KV cache usually caps the batch width first. Multi-node training is frequently limited by the interconnect. The measurement shows which of these binds at your concurrency.
We cover accelerator selection across NVIDIA and AMD, the specification of the host and of the network around the accelerators, and the power and cooling envelope for the facility that will hold them. A measurement on capacity rented in an account you control characterises the accelerator. The quoted host and network are checked by calculation, and the findings say which figures are measured and which are calculated.
Before committing capital it can be cheaper to rent the same capacity for a period, observe how heavily it is used, and buy only if use stays high. We work out the level of sustained use at which buying becomes cheaper than renting for your workload, and the same for colocation.
The comparison is worked through as capital cost depreciated over a stated period, the cost of financing it, power, space, support contracts and the staff to run the hardware. Dividing the total by the tokens served at the utilisation you expect is how idle capacity enters the figure. It is delivered as a spreadsheet with every assumption visible, so that your finance team can check it line by line.
Questions about AI hardware advisory
The Working with us page answers the rest. If yours is not there, ask it on the contact page and an engineer will reply.
Can you check a GPU quotation before it expires?
Tell us the validity date and what you intend to run. An engineer replies within one working day and tells you whether the check can be completed in the time you have. The quotation itself can follow once the NDA is signed.
Nothing is running yet. What do you measure?
For an AI use case we run the pilot with a sample of your real requests, on capacity rented in an account you control. Where there is no pilot yet, we work from stated inputs and say so. For a hardware purchase we run a representative workload, built from the model, the context lengths and the number of users you expect, on capacity rented in the same way. In both cases we state how uncertain the result is and what would reduce the uncertainty.
How is the work priced?
The assessment has a fixed fee and a fixed length, and both are quoted before it starts. The price of the build is set out in the proposal that follows the assessment. We do not publish rates, because the fee follows the scope.
Articles on AI hardware advisory
Written by the engineers who do this work.
One team runs all five
Describe the problem, and an engineer will reply within one working day
Tell us in general terms what you are trying to do and what has stopped it. If it is not ours to take on, we will say so.