Together HGX H100 SXM / 8 GPU node

Name: Together HGX H100 SXM / 8 GPU node
Brand: Together

Together HGX H100 SXM / 8 GPU node allocates 8 NVIDIA H100 SXM accelerators.

Together

Verified 11 Nov 2026

GPUs / instance: 8
VRAM / GPU: 80 GB
Total nominal VRAM: 640 GB

View provider specifications Compare alternatives Request a quote

$31.92 per instance-hour ÷ 8 GPUs = $3.99 per GPU-hour, calculated from the published rate and GPU count. Scope: Public USD GPU Clusters list.

GPU Clusters on-demand per-GPU rate multiplied by eight. Dedicated Inference and prepaid reservation tariffs are separate; shared storage is extra.

Instance specification

Source per value

Cloud Cloud operator publishing this instance or preset.: Together 1
Instance / preset Official SKU or provider pricing-table preset; descriptive configuration names are used when no API SKU is published.: HGX H100 SXM / 8 GPU node 2
GPU GPU model and form factor where stated by the cloud. Models and memory variants are not interchangeable.: NVIDIA H100 SXM 1
GPUs Physical GPUs allocated to one instance, not an entire cluster.: 8 2
VRAM / GPU Numeric capacity per GPU in the provider’s stated unit. Read VRAM unit or GPU memory before comparing GB and GiB. Aggregate-only specifications are divided by GPU count.: 80 1
vCPUs Provider-advertised vCPU count. Oracle publishes OCPUs, so this field is left empty for Oracle.: Not published
Host RAM System memory separate from VRAM. Source units are preserved; Google JEDEC GB are GiB.: Not published
Interconnect Published GPU or node fabric. EFA / RDMA describe scale-out networking; NVLink / NVSwitch describe GPU communication within a node. Empty means not captured for this configuration.: 3.2 Tb/s InfiniBand 1
Network Published VM NIC limit, including baseline / burst or aggregate labels where provided. This is separate from GPU-to-GPU bandwidth.: Not published
Instance / hour $/hr Whole-instance current public USD on-demand rate in the stated region. GPU-hour quotes are multiplied by GPU count. Future, Spot, Flex-start, reservation and inference-only rates are excluded.: $31.92 3
Rate scope Region or published pricing geography attached to the rate; not a list of every supported region.: Public USD GPU Clusters list 3
Price terms Purchase model, inclusions, exclusions, effective dates and any reason the current rate is empty.: GPU Clusters on-demand per-GPU rate multiplied by eight. Dedicated Inference and prepaid reservation tariffs are separate; shared storage is extra. 3
VRAM unit Provider-stated nominal GPU capacity unit. GB and GiB are preserved, not silently converted.: GB 1
Memory / GPU Capacity per GPU with its source unit; calculated from aggregate memory and count where necessary.: 80 GB 1
GPU / hour $/hr Calculated whole-instance USD hourly rate divided by physical GPU count, with the same purchase terms and price scope.: $3.99 2
Published regions Published family or preset deployment regions/zones. Listing does not confirm stock, account quota or a regional tariff. Empty means not independently captured.: Not published
Native currency / hour Whole-instance hourly rate in Native currency. No currency conversion is made; read setup fees and monthly caps in Price terms.: Not published
Native currency ISO currency code of the native hourly rate.: Not published
CPU source terms CPU cores or other provider CPU measures retained without converting them to vCPUs.: Not published

Read the memory correctly

80 GB per GPU × 8 GPUs = 640 GB nominal total, calculated from the two sourced inputs. That capacity is split across devices. Check the interconnect and your parallelism strategy before selecting a configuration.

Memory units, pricing rules and verification method

Source documents

Together instance specifications accessed 2026-11-11
Together whole nodes of eight GPUs accessed 2026-11-11
Together GPU Clusters hourly tariff accessed 2026-11-11

Matched by GPU model, count and memory per device. Cross-cloud matches break ties.

Provider coverage and missing offers · Neocloud companies · HPC providers and cluster services

Request a quote

Send your GPU requirement by email: instance, region, duration and start date. We reply to discuss the request before sharing contact details with a provider.