Your cart
Your cart is empty
Welcome to VDR's latest news & guides
Expert guides on selling used IT and telecoms equipment, buying certified refurbished hardware, WEEE recycling and secure data centre decommissioning - insights from the UK's IT asset recovery specialists.
Which GPU Do You Actually Need? A Guide For Business And Creative Workloads

VRAM matters more than the model name. Here is how to choose without overspending.
Most GPU Buying Advice Is Written For Gamers
Search for graphics card guidance and you will find frame rates, benchmarks and enthusiast opinion. Almost none of it addresses what a business or creative studio actually needs to decide.
The questions that matter in a working environment are different: how much video memory does this workload need, will it fit in the chassis, what does it draw, will the power supply carry it, does the software require certified drivers, and how many of these can be run in one machine.
This guide answers those.
Start With VRAM. Everything Else Is Secondary.
If you take one thing from this article, take this: video memory capacity is the constraint that stops work, and raw speed is not.
A card that is slightly slower will finish the job. A card without enough VRAM will not run it at all - it will fail outright, or fall back to a slower path that wipes out any advantage the GPU offered.
Roughly how requirements scale:
| Workload | Practical VRAM Guidance | Notes |
|---|---|---|
| Office, CAD viewing, multi-display | Entry level is sufficient | Display outputs matter more than compute |
| Photo editing, 1080p video | 8GB and up | Comfortable headroom at this tier |
| 4K video editing and colour work | 12-16GB and up | Timeline complexity drives the requirement |
| 3D rendering and simulation | 16-24GB and up | Whole scene must fit in memory |
| AI inference on smaller models | 16-24GB and up | Model size sets the floor absolutely |
| AI training and larger models | 48GB and up, often multi-GPU | Data centre accelerator territory |
| Virtual desktop hosting | Depends on user density | Licensing and certification are decisive |
Buy for the workload you actually run. Buying for a hypothetical future workload is how budgets get spent on capability that never gets used.
Consumer, Workstation Or Data Centre?
Consumer cards - the GeForce RTX line and equivalents - offer the strongest raw performance per pound. For rendering, video, design and a great deal of smaller-scale AI work, they are frequently the correct answer. Blower-style or compact variants matter if the card is going into anything other than a roomy tower.
Workstation cards - the RTX A-series and Quadro lineage - carry certified drivers for professional applications, larger memory options and better multi-card behaviour. If your software vendor certifies specific cards, or your support agreement depends on it, this is not really optional.
Data centre accelerators - A100, H100, L40S, A30, A40, V100, T4 and the AMD Instinct line - are built for sustained parallel compute in rack servers. They are typically passively cooled, relying on chassis airflow, which means they will overheat in a normal desktop. They also carry the highest prices, new and used.
The most common and most expensive mistake we see is buying a data centre accelerator for a workload a workstation card would have handled comfortably.
The Practical Constraints People Forget
Physical fit. Length, height and slot width. High-end cards are frequently three slots wide and long enough to foul drive cages. Measure before you order.
Power supply capacity and connectors. Check both total wattage headroom and whether you have the right connectors. Adapters are a common cause of instability.
Cooling. Passive server cards need directed chassis airflow. Multiple active cards in one case need a plan for the heat they generate.
Drivers and certification. Check what your applications support before committing, especially in regulated or vendor-supported environments.
Displays. Confirm output types and counts. It is a mundane detail that derails desk deployments regularly.
New Or Refurbished?
Both are legitimate, for different reasons.
New makes sense for current-generation consumer and workstation cards where you want full warranty, the latest architecture and known provenance. It is the straightforward choice for a small number of desk-side machines.
Refurbished makes sense when you need capacity rather than the newest capability, when matching cards already in service, when a previous-generation card comfortably meets the requirement at a fraction of the cost, or when lead times on new stock do not fit the project.
A previous-generation professional card with ample VRAM will often serve a real workload better than a newer card with less memory at similar cost. Memory is the constraint. Generation is a preference.
You can see current stock in our graphics cards range, which includes new retail cards as well as enterprise hardware.
Buying More Than One
If you are considering multiple cards in a single system, check three things early.
First, whether your application actually scales across GPUs - many do not, and a second card sits idle. Second, spacing and cooling, because cards packed together throttle. Third, power draw under simultaneous full load, which is where undersized supplies fail.
For anything beyond two cards, a purpose-built chassis is usually the right answer rather than a workstation with more slots.
And When You Replace Them
Worth planning at the point of purchase rather than three years later.
The GPU secondary market is unusually strong, and enterprise and workstation cards hold value well. The cards you are replacing are very likely worth a meaningful contribution towards the ones you are buying - which changes the net cost of the upgrade considerably.
Our guide to selling enterprise GPUs covers what drives that value, including why faulty cards are still worth listing.
How We Can Help
We supply graphics cards to UK businesses, studios and research groups, and we buy them back when they are replaced.
Tell us the workload rather than the model you had in mind, and we will tell you what actually fits it - including when the cheaper option is the right one. Browse our graphics cards, or send us the specification you are considering.
Maximise Value. Minimise Waste.
Frequently Asked Questions
Is a gaming card acceptable for professional work?
Frequently, yes - particularly for rendering, video and smaller AI workloads. The exceptions are applications requiring certified drivers and environments where vendor support depends on approved hardware.
How much VRAM do we need for AI work?
The model you intend to run sets the floor, and it is a hard floor rather than a preference. Tell us the models and we will advise.
Can we put a data centre GPU in a desktop?
Generally not. Most are passively cooled and depend on server chassis airflow. In a standard tower they overheat and throttle.
Do older professional cards still make sense?
Often, yes. A previous-generation card with more VRAM regularly outperforms a newer one with less on memory-bound work.
Will you take our old cards in part exchange?
Tell us what you are replacing and what you are buying, and we will look at both sides together.
Do you supply single cards or only volume?
Both. Tell us what you need.
Got equipment to sell?
Turn the hardware you just read about into cash - free valuation, free collection, same-day payment.
Get a Free ValuationSend us a message
We reply by email, usually within a few working hours.
Message sent
We have it, and we will reply to your email address shortly.
- Choosing a selection results in a full page refresh.
- Opens in a new window.











