Configure SPM Thor

GPUs, blade memory and systems. What you choose becomes your enquiry.

1

GPUs

Design intent

Two, four, six or eight liquid-cooled RTX PRO-class GPU blades on the same platform. Empty slots are blanked, ready for more.

Cut-away of SPM Thor with 8 GPU blades populatedConcept render
2

GPU memory per blade

Concept

Each GPU has its own memory. The blade is modelled with 48 GB; a 96 GB clamshell variant fits the same envelope.

3

Systems

More than one Thor behind a gateway adds capacity and failover. Each model is best kept within one system.

1

8 GPUs in all

4

What Sarja does around it

Everything beyond a configured machine is scoped in your quote.

Also scope

5

Your site

Estimate

At 8 GPUs Thor is estimated to draw about 3.0 kW at full load. A comms or server room: server-class fan noise at full load. What do you already have?

SPM Thor

THOR-8x48

8 GPU blades · every slot populated
GPUs per system
8 RTX PRO-class GPU bladesEvery slot populated
GPU memory
48 GB per GPUAs modelled: 16 × 3 GB GDDR7
GPU memory per system
384 GB across 8 GPUs8 separate 48 GB pools, not one shared memory
Estimated load
About 3.0 kW DC per system at full loadEstimate
Power
A 200–240 V / 16 A circuit per systemEstimate
Deployment
Configured machine (included with every SPM Thor)
Services to scope
None selected
Site has
Not stated

Flagship conceptIn development · taking enquiries

Thor is quoted per configuration. There is no list price while it is in development. Estimates are confirmed against your site before anything is quoted.

THOR-8x488 GPUs · flagship conceptEnquire →