Skip to content
State of GPU Compute 2026
Cloud ServersNVIDIA B300GIGABYTE

GIGABYTE G894-SD3-AAX7

No. 19·16x 128GB RDIMM 6400MHz (Support up to 32 DIMM DDR5 up to 8192 Gb)

On demand · 8× B300
€48.00/server·hr
Per GPU
€6.00/GPU·hr

GPU as listed: B300 HGX

The newest silicon we rent, with the most memory per GPU. Built for models that should not be cut into pieces. You get the whole machine, billed per second in euros.

Rent this serverCompare

Spec sheet updated 27 Oct 2026

Exploded view of a Nexpod node: an 8U server with eight NVIDIA B300 GPUs on an HGX baseboard, an NVLink switch tray above it, and data paths to the EU-AMS-1, EU-FRA-1 and US-NYC-1 regions.
GPUs
8× B300
One NVLink fabric
Memory per GPU
288GB
HBM3e
GPU memory, server
2,304GB
8 × 288 GB
System memory
16x 128GB
16x 128GB RDIMM 6400MHz (Support up to 32 DIMM DDR5 up to 8192 Gb)
Front elevation · GIGABYTE8U · 19-inch rack
U8U7U6U5U4U3U2U1
GPU tray8× B300
CPU2 sockets
CPU0
CPU1
Memory16 × 128GB
Drives2 + 2 bays

This chassis

The whole host, not a slice

The CPUs, system memory and local NVMe come with the GPUs. Nothing is carved up for other tenants.

nexctl · describe
$ nexctl servers describe gigabyte-g894-sd3-aax7-hgx
model
GIGABYTE G894-SD3-AAX7
gpu
B300 HGX
cpu
2x Intel Xeon 6767P
memory
16x 128GB RDIMM 6400MHz (Support up to 32 DIMM DDR5 up to 8192 Gb)
storage
2 x 1.92TB M.2 / 2x 3.84TB U.2 (up to 8 HDD 2.5" NVMe)
form
8U
● 8/8 GPUs healthy · NVLink all-to-all

Blackwell Ultra

Eight B300 GPUs. One machine.

What the B300 brings to this chassis, from the package to the fabric between all eight.
Blackwell Ultra
HBM3e
HBM3e
HBM3e
HBM3e
NVIDIAB300288 GB
HBM3e
HBM3e
HBM3e
HBM3e
Silicon

NVIDIA Blackwell Ultra

Each of the eight packages is an NVIDIA B300 with 288 GB of HBM3e next to the die.

B300288 GB
B200180 GB
H200141 GB
H10080 GB
Memory

288 GB on every GPU

2,304 GB across the server. Here is how B300 sits in the fleet.

NVLINKGPU 0GPU 1GPU 2GPU 3GPU 4GPU 5GPU 6GPU 7
Interconnect

All-to-all NVLink

Every GPU talks to every other one directly, so tensor-parallel runs stay inside the box.

Use it for

Where the B300 earns its rate.

4 workloads / scroll

Technical specs

Every field on the spec sheet.

Server values appear exactly as the builder lists them, spelling included. Anything they leave open, we configure to order.

Talk to an engineer
Server
ModelGIGABYTE G894-SD3-AAX7The exact chassis you get. No substitutes after you deploy.
BrandGIGABYTERacked, cabled and burn-in tested by us before it goes live.
GPUB300 HGXAll eight GPUs are yours. Nothing is split or shared.
GPU familyB300NVIDIA Blackwell Ultra generation.
CPU2x Intel Xeon 6767PHost cores for data loading, tokenising and preprocessing.
System memory16x 128GB RDIMM 6400MHz (Support up to 32 DIMM DDR5 up to 8192 Gb)Room for datasets, page cache and CPU offload.
Local storage2 x 1.92TB M.2 / 2x 3.84TB U.2 (up to 8 HDD 2.5" NVMe)Local NVMe for checkpoints and scratch. Network volumes mount on top.
Form factor8URack height in the data hall. It does not change how you use it.
Model IDgigabyte-g894-sd3-aax7-hgxPin this chassis with nexctl --model or the API.
Catalogue no.19Its position in the Nexpod server list.
GPU · NVIDIA B300
ArchitectureNVIDIA Blackwell UltraThe generation behind every B300 we rent.
Memory per GPU288 GB HBM3eHolds weights, activations and KV cache on the GPU itself.
GPU memory per server2,304 GB8 GPUs × 288 GB, reachable over NVLink.
GPU interconnectNVLink, all-to-allEvery GPU reaches every other GPU without going through the host.

Pricing

Two ways to pay. Same machine.

Run it on demand by the second, or reserve it for a year at a lower rate. Both are invoiced in euros, excluding VAT.
On-demand

Start now, stop any time.

€48.00/server·hr
Per GPU·hr
€6.00
Per second, full server
€0.01333
Month at 730 h
€35,040
Rent on demand
Reserved · 12 months

The same server, held for you all year.

−17%
€40.00/server·hr
Per GPU·hr
€5.00
Per second, full server
€0.01111
Month at 730 h
€29,200
Reserve this server
billing · gigabyte-g894-sd3-aax7-hgx
metering
€0.01333/s
8 × B300 · on demand · per second
0s€0.0015s€0.2030s€0.4045s€0.6060s€0.80
1 minute
€0.80
20-min test
€16.00
1 hour
€48.00

Related servers

Also worth a look.

All 20 servers
  • NVIDIA DGX B300
    NVIDIA · B300
    €48.00/server·hr€6.00/GPU·hr
    2000GB (DDR5 ECC REG)
  • GIGABYTE G894-ZD3-AAX7
    GIGABYTE · B300 · 8U
    €48.00/server·hr€6.00/GPU·hr
    6144GB (DDR5 ECC REG)
  • GIGABYTE G894-AD3-AAX7
    GIGABYTE · B300 · 8U
    €48.00/server·hr€6.00/GPU·hr
    6144GB (DDR5 ECC REG)
Compare withNVIDIA DGX B300GIGABYTE G894-ZD3-AAX7GIGABYTE G894-AD3-AAX7

FAQ

Before you rent it.

Short answers. The docs have the long ones.
Do I share the GIGABYTE G894-SD3-AAX7 with anyone?
No. You rent the whole machine: all eight B300 GPUs, the host and its system memory. Nobody else runs work on it while it is yours.
How is it billed?
Per second while it runs. On demand it is €48.00 per hour for the full server, or €40.00 per hour on a 12-month reservation. Invoices are in euros.
What does “Configurable on request” mean?
The builder's listing leaves that part open. We fit it to your order, for example CPU or local storage, before handover and confirm the exact configuration in writing.
Can I join it with other servers?
Yes. Instant Clusters link several 8-GPU servers over a high-speed fabric. Start on one node and add more when the job needs them.
Which regions can I deploy in?
Nexpod runs in EU-AMS-1, EU-FRA-1 and US-NYC-1. You pick the region at deploy time and the console shows current capacity for this model.
Placeholder

Placeholder quote. A customer describes, in one or two plain sentences, what changed after moving a real workload to Nexpod.

Customer nameRole, Company

Rent the GIGABYTE G894-SD3-AAX7

Eight B300 GPUs, ready in minutes.

Create an account, pick a region and connect over SSH. Stop the server when the job is done and the meter stops with it.

Get startedTalk to sales
nexctl · eu-ams-1
$ nexctl servers rent gigabyte-g894-sd3-aax7-hgx --region eu-ams-1
reserved 1 × GIGABYTE G894-SD3-AAX7
status provisioning → ready
ssh root@node-07.ams.nexpod.ai
RelatedAll serversGPU comparePricingCloud Servers