4× MI300X768 GB HBM3 · CDNA 3
768 GB of HBM3 for 405B-class models at FP8.
$6,996 /mo41 available
Configure
Solutions · Inference
Serve models at a cost you can predict.
A flat monthly price turns inference into a fixed cost: no surprise invoice when traffic spikes. Size the server to the model, from an L40S for small models to B300 for FP4 serving.
- Servers from $1,199/mo
- At least 25% below market
- Root access in < 60 s
The problem
What slows teams down today
Spiky bills
Per-hour and per-token pricing make monthly inference costs hard to forecast.
Over-provisioned GPUs
Choosing a flagship GPU for a 13B model burns budget.
Cold starts and quotas
Autoscaling on shared clouds adds latency and capacity risk.
How CoreGPU fits
Dedicated GPUs, a monthly price, no surprises
Every server is racked, tested and imaged before it goes on sale, so you get root access in under a minute and a bill you know in advance.
- Dedicated GPUs that stay warm all month
- Right-sized options from 48 GB to 288 GB per GPU
- vLLM, SGLang and TensorRT-LLM ready images
- Scale out by adding servers at each renewal
Recommended servers
Configurations our engineers recommend
Live stock and the monthly price of the whole server. Open a GPU page to configure it.
FAQ
Inference at scale: common questions
Another question? Our engineers answer within one business day.
Contact usWhich GPU gives the lowest cost per token?
For models that fit, memory bandwidth per dollar drives decoding cost. MI300X and H200 are strong for 70B–405B models; RTX PRO 6000 and L40S win for smaller models.
Can I run several models on one server?
Yes. With root access you can split GPUs between models, or use MIG on RTX PRO 6000 for isolated slices of a card you rent.
Do you charge for bandwidth?
Network use is included in the monthly price under our acceptable use policy.
Other solutions
Tell us about your workload
Model size, context length, deadline and budget: we reply with the right GPU and server count within one business day.