Having fallen for the Mac hype I am completely out of touch with PC specs. etc…

I need to build a linux box to process some hardcore GIS based queries. It will run linux and postgres and would benefit from shed loads of memory + multi-processors + fast disks. I need to start smallish and then upgrade as we get more business. Initial budget is £5-800’ish.

What should I be looking at? Is it still basically a choice between Dell/HP?

The only reason to look at Dell/HP would be if you want a server rather than desktop/workstation, it does sound like a server would be a good fit but you aren’t going to get one for your budget (although your requirements are a bit woolly…).
If I were you I’d look at a cloud service (IaaS) if it’s for a business requirement but you want to minimise your capex until it’s successful. That way you won’t have to worry (much) about resilience and being able to scale up quickly if needed.

I can’t go for cloud service as I may need to install a graphics card to use the GPU for processing (but not display)

Let me re-phrase then. What PC for budget of £800 best capable of handling data processing rather than graphics/widgets etc..

I’d just get a system from somewhere like overclockers.co.uk, with a decent CPU & memory and a low-end GPU. With your budget you’re better off just going with a single CPU (6 core or more). RAM depends on how much you need obviously, the drives will be tricky depending on how the application uses them. If most of the data is static with only a small part requiring decent I/O speed then go for an SSD + SATA drive. If the app needs a lot of high-speed disk then you’d really need a SAS array (with the RAID type dependant on the read/write bias) but again I can’t see you getting that within your budget.

Will you go second hand, or new only?

You could get a stonking server for £800.

Agree with FuzzyWuzzy about buying from a non-major h/w vendor. You’ll be able to fine tune the h/w to suit your workload.

You say you need fast disk. How large is your data ?. If it’s relatively small i’d look at a couple of the cheaper SSD’s, split them into 2 PV’s. A small one mirrored for the O/S VG & a larger one raid0 for the data.

Check whatever GPU you buy has CUDA/OpenCL support.

I assume the workload is parallel. Have you thought about using a grid engine, get a free EC2 account & split the workload across your local compute & also Amazon ?.

Some good stuff there, thanks..I’d forgotten overclockers
I can’t control the workload as parallel jobs, the main grunt is computing a load of Thiessien polygons from within postgris and the geometries are interdepent.