GPUBank vs TensorWave
TensorWave is a US cloud founded in 2023 that rents only AMD Instinct accelerators, as 8-GPU bare-metal MI300X, MI325X and MI355X nodes sold through a sales call. It has no NVIDIA GPU; its MI300X starts at $1.71 per GPU-hour on an 8-GPU node, with the term not stated, against $1.65 for a single MI300X at GPUBank. TensorWave offers newer AMD parts, RoCEv2 clusters and SOC 2, ISO 27001 and HIPAA; GPUBank offers self-serve access, NVIDIA GPUs and published billing rules.
GPUBank
Tallinn, Estonia
MI300X
$1.65 / h
- vCPU, RAM, NVMe, IPv4 and traffic included
- Per-second billing after the health check
- Top up in Bitcoin, USDT and 12+ other coins
vCPU, RAM, NVMe, IPv4 and traffic included
TensorWave
Las Vegas, United States · since 2023
MI300X · reserved
$1.71 / h
- MI325X from $2.25 and MI355X from $2.95, up to 288 GB per GPU
- Bare-metal 8-GPU nodes with RoCEv2 at 3.2 Tb/s
- Managed Kubernetes and managed Slurm on multi-node clusters
TensorWave
AMD cloud
United States16 points compared15 of their pages readprices last checked on 29 September 2026page updated 2 November 2026
The verdict
Newer AMD Instinct parts and clusters at TensorWave. Self-serve single GPUs and published billing at GPUBank.
TensorWave suits ROCm teams that need multi-node MI300X, MI325X or MI355X capacity under contract, with managed Slurm or Kubernetes and SOC 2 Type II, ISO 27001 and HIPAA paperwork. On the one GPU both of us rent, the MI300X, prices are close: $1.71 'starting at' per GPU-hour on an 8-GPU node there, $1.65 for one GPU here. TensorWave publishes no billing granularity, payment methods, regions or SLA, and access goes through a form asking for company details and a phone number. GPUBank opens with an email and password, bills by the second and adds NVIDIA GPUs from the RTX 4090 to the B200.
TensorWaveGPUBank 01The real cost
The same GPUs, side by side, per hour.
Every GPU model both of us rent, at each provider's own published price for one card, in US dollars, excluding VAT. Prices last checked on 29 September 2026.
| GPU | TensorWave per hour | GPUBank per hour | Difference | 100 h at TensorWave | 100 h at GPUBank |
|---|---|---|---|---|---|
 MI300X192 GB · their reserved price |
$1.71 | $1.65 | −4 % | $171.00 | $165.00 |
TensorWave publishes 'Starting at' per-GPU prices for 8-GPU bare-metal nodes. The pages do not say whether they are on demand or reserved, or for which term.
Where TensorWave publishes no on-demand price for a model, its lowest published reserved price is used, as labelled on the row.
A large-model run, 100 h on an MI300X
TensorWave
$171.00
GPUBank
$165.00
Saves $6.00 at GPUBank (−4 %)
02Their catalogue
Every GPU TensorWave lists.
TensorWave lists 4 GPU models. The 3 below are the ones recorded in our check, each at its lowest public price per GPU-hour. 1 of them is also rented at GPUBank (in blue).
- MI355X288 GB · reserved$2.95
- MI325X256 GB · reserved$2.25
- MI300X192 GB · reserved$1.71
03What the price includes
What appears on the bill besides the GPU.
The same MI300X session of 2 h 14 m 03 s, line by line, as each provider describes its billing.
TensorWave
2 h 14 m 03 s on MI300X
- GPU · $1.71 / h
- $3.82
- vCPU and RAM
- Included
- Storage
- Not published
- Egress
- Not published
- Public IPv4
- Not published
- Billing granularity
- Not published
- Minimum charge
- Not published
GPUBank
2 h 14 m 03 s on MI300X
- GPU · $1.65 / h
- $3.69
- vCPU and RAM
- Included
- Storage
- Local NVMe included; persistent disk $0.03 per GB-month
- Egress
- Included
- Public IPv4
- Included
- Billing granularity
- Per second, from the health check
- Minimum charge
- None
While stopped
What keeps running on the bill.
TensorWave
Not published.
GPUBank
Only the persistent disk, $0.03 per GB-month. Compute, IPv4 and scratch cost nothing.
04Where the GPUs run
Where TensorWave runs GPUs is not published.
Countries where TensorWave says its GPUs run, from its own region list or documentation, against GPUBank's four regions. A pin marks a country, not the exact site.
Paris
Frankfurt
Ashburn
Singapore
TensorWave0
No country list on its public pages.
GPUBank4
Paris
Frankfurt- Ashburn
Singapore
TensorWave · 0 countriesGPUBank · 4 regions
05Paying and signing up
How each one takes your money.
TensorWave
Payments and sign-up
Payment methods
No cryptoMinimum top-up Not published
To start
- 1Sales contact form
- 2Sales contract
Billing modelNot published
GPUBank
Payments and sign-up
Payment methods
A prepaid USD balance, topped up in crypto. Every GPU-second is drawn from it.
Minimum top-up $25
To start
- 1Email and password
- 2Top up in crypto
- 3Launch a GPU
Billing modelPrepaid balance
06Trust and support
Certificates, uptime and who answers.
| TensorWave | GPUBank | |
|---|---|---|
| SOC 2 | Type II | Not published |
| ISO 27001 | Yes | Not published |
| HIPAA | Yes | No |
| GDPR DPA | DPA available | Published DPA, applies automatically |
| Availability target | No SLA found | 99.9 % monthly, per region |
| Status page | None found | Public status page |
| Support | Email, chat, phone, tickets; solution engineers | Tickets from a $25 balance |
07Their site, our site
What you see before you pay.
Both sites as a visitor sees them, without an account: the TensorWave home page captured on 29 September 2026, and GPUBank's live board. Cookie banners and chat bubbles are closed; nothing else is edited.
- 1tensorwave.com/pricing returns 'Page not found'; prices appear only on each accelerator page as 'Starting at' per GPU-hour.
- 2Every call to action leads to 'Book A Call' or 'Request POC'.
- 3The MI455X is listed with 'Pricing TBD'.
- 1Live stock per region and every price, without an account.
- 2Per-hour and per-second prices, with vCPU, RAM, NVMe, IPv4 and traffic in them.
- 3A machine-readable price feed at /prices.json.
08The spec sheet
16 points, read off both sites.
GPUBank aheadTensorWave aheadEven
| Criterion | GPUBank | TensorWave |
|---|---|---|
| Price and catalogue | ||
| H100 SXM, per GPU-hourCompared with our MI300X at $1.65; GPUBank's H100 SXM is $1.85. | $1.85, all-in | No NVIDIA; MI300X from $1.71 (8-GPU node) |
| Cheapest GPU-hour | $0.23, RTX 4090 | $1.71, MI300X, term not stated |
| GPU models listed | 10 | 4, AMD Instinct only |
| What the GPU price includes | vCPU, RAM, NVMe, IPv4, traffic | Dual EPYC CPUs, DDR5, local NVMe per node |
| Billing | ||
| Billing granularity | Per second | Not published |
| Minimum chargeEarly cancellation of a reserved contract may incur a fee. | None | Not published |
| Stopped instance | Disk only | Not published |
| Persistent storage | $0.03 per GB-month | Not published |
| Egress traffic | Included | Not published |
| Public IPv4 | Included | Not published |
| Billing model | Prepaid balance | Not published |
| Payment and sign-up | ||
| Crypto payments | Bitcoin, USDT, Ethereum and 12+ more | Not published |
| Card payments | No | Not published |
| Minimum deposit | $25 | Not published |
| Platform | ||
| API and CLI | REST API, CLI, Python SDK | No public API or CLI documentation |
| Terraform | Official provider | None documented |
| Serverless GPUs | No | No |
| Managed Kubernetes | No | Managed Kubernetes and managed SlurmAhead |
| Ready-made templates | 25 | PyTorch, TensorFlow and Jupyter environments |
| Interruptible or spot | Yes, 40 % off | No |
| GPUs per instance | Up to 8 | 8-GPU nodes, multi-node clustersAhead |
| Multi-node interconnect | Not offered | RoCEv2, 3.2 Tb/s per nodeAhead |
| Network and regions | ||
| Regions | 4 | Not published |
| Countries | France, Germany, US, Singapore | Not published |
| EU regions | Paris, Frankfurt | Not published |
| Trust and support | ||
| Availability target | 99.9 % monthly | No SLA found |
| Compliance reports | None published | SOC 2 Type II, ISO 27001:2022, HIPAAAhead |
| Public status page | Yes | None found |
| Support | Tickets from a $25 balance | Email, chat, phone, tickets; solution engineersAhead |
09The right fit
Different needs. A clear choice.
What TensorWave does better
- MI325X from $2.25 and MI355X from $2.95, up to 288 GB per GPU
- Bare-metal 8-GPU nodes with RoCEv2 at 3.2 Tb/s
- Managed Kubernetes and managed Slurm on multi-node clusters
- SOC 2 Type II, ISO 27001:2022 and HIPAA on the Trust Center
- Support by email, chat and phone, with solution engineers
Choose GPUBank if
- One MI300X at $1.65 per GPU-hour, no 8-GPU node or contract
- NVIDIA GPUs too: B200, H200, H100, L40S, RTX 5090 and RTX 4090
- Account in minutes with email and password, no sales call
- Per-second billing, IPv4 and traffic included, published rules
- Pay in Bitcoin, USDT and 12+ other coins, four named regions
Moving from TensorWave takes an afternoon.
TensorWave nodes run Ubuntu 22.04 with ROCm, Docker, Kubernetes or Slurm, so workloads are containers that move to another MI300X host as they are. Moving to NVIDIA GPUs means porting ROCm code to CUDA, and the contract terms govern exit.
- 1Launch an MI300X instance on GPUBank, up to 8 GPUs, with a persistent disk.
- 2Copy model weights and datasets from the node with rsync over SSH.
- 3Run the same ROCm container images, then end the TensorWave contract under its terms.
# 1. a GPU and a disk on GPUBank gpubank launch --gpu mi300x --region eu-paris --disk data # 2. copy your data from the old machine rsync -azP old-host:/workspace/ root@<new-ip>:/workspace/ # 3. run the same image or template ssh root@<new-ip> 'cd /workspace && ./start.sh'
A few things you might be wondering.
Slightly. TensorWave lists the MI300X at 'Starting at' $1.71 per GPU-hour on an 8-GPU bare-metal node, with the term not stated. GPUBank charges $1.65 per GPU-hour for a single MI300X, billed per second, with vCPU, RAM, NVMe, IPv4 and traffic included.
No. TensorWave rents only AMD Instinct accelerators: MI300X, MI325X and MI355X, with MI455X Helios racks announced. GPUBank rents the H100 SXM at $1.85 per GPU-hour, along with the MI300X and eight other NVIDIA models.
No self-serve sign-up is published. The contact form asks for name, company, company email, country and phone, and access is governed by a sales contract. GPUBank opens an account with an email and password.
TensorWave. It sells bare-metal 8-GPU nodes in multi-node clusters with RoCEv2 at 3.2 Tb/s, managed Kubernetes and managed Slurm, plus the MI325X and MI355X. GPUBank rents up to 8 GPUs per instance and lists only the MI300X on the AMD side.
TensorWave does not publish its data center locations; its Terms give a notice address in Las Vegas. GPUBank runs in Paris, Frankfurt, Ashburn and Singapore.
Without. Every price on this page excludes VAT and sales tax, for TensorWave and for GPUBank, so the two columns compare like for like. Tax is added at checkout according to where the buyer is based.
Sources
Every figure about TensorWave was read on its own pages or official documentation. Prices last checked on 29 September 2026.
- 01tensorwave.com/products/accelerators/amd-mi300xMI300X starting price and node spec
- 02tensorwave.com/products/accelerators/amd-mi325xMI325X starting price and node spec
- 03tensorwave.com/products/accelerators/amd-mi355xMI355X starting price and node spec, Series B news headline
- 04tensorwave.com/products/accelerators/amd-mi455xMI455X Helios, pricing TBD
- 05tensorwave.com/products/bare-metalBare-metal offer, SOC 2 Type II
- 06tensorwave.com/aboutFunding over USD 500M, leadership
- 07security.tensorwave.com/Founded 2023, SOC 2, ISO 27001:2022, HIPAA, pentests
- 08docs.tensorwave.com/welcome-to-tensorwave/faq.mdAMD only, SSH bare metal, support channels, reserved contracts, on-demand planned
- 09docs.tensorwave.com/welcome-to-tensorwave/bare-metal-quickstart.mdUbuntu 22.04, ROCm, root SSH
- 10docs.tensorwave.com/policies/terms-of-serviceLas Vegas notice address, no payment terms
- 11tensorwave.com/connectSales form fields
- 12tensorwave.com/products/managed-services/managed-kubernetesManaged Kubernetes
- 13tensorwave.com/products/storage/high-speed-network-storageStorage offer
- 14tensorwave.com/pricingConfirms no public pricing page
- 15tensorwave.com/GPU line-up, platform stack, compliance claims
What we could not verify (6)
- Mode and commitment term for the 'Starting at' prices not stated; recorded as reserved with term unknown.
- Whether on-demand or hourly self-serve rental exists today not published (FAQ describing it as planned is about 2 years old).
- No pricing page (tensorwave.com/pricing is 404); no storage, egress or IPv4 prices.
- Data center locations and countries not published.
- API, CLI, SDK and Terraform availability not documented publicly.
- SLA and status page not found.
In short
GPUBank vs TensorWave, in facts.
The essentials of this comparison in one table, for quick readers and AI assistants. Prices last checked on 29 September 2026.
Read as Markdown- H100 SXM, per GPU-hour
- TensorWave: No NVIDIA; MI300X from $1.71 (8-GPU node)
GPUBank: $1.85, all-in - Cheapest GPU-hour
- TensorWave: $1.71, MI300X, term not stated
GPUBank: $0.23, RTX 4090 - Billing granularity
- TensorWave: Not published
GPUBank: Per second - Stopped instance
- TensorWave: Not published
GPUBank: Disk only - Persistent storage
- TensorWave: Not published
GPUBank: $0.03 per GB-month - Egress traffic
- TensorWave: Not published
GPUBank: Included - Crypto payments
- TensorWave: Not published
GPUBank: Bitcoin, USDT, Ethereum and 12+ more - Card payments
- TensorWave: Not published
GPUBank: No - Minimum deposit
- TensorWave: Not published
GPUBank: $25 - GPU models listed
- TensorWave: 4, AMD Instinct only
GPUBank: 10 - Regions
- TensorWave: Not published
GPUBank: 4 - Serverless GPUs
- TensorWave: No
GPUBank: No - Availability target
- TensorWave: No SLA found
GPUBank: 99.9 % monthly - Compliance reports
- TensorWave: SOC 2 Type II, ISO 27001:2022, HIPAA
GPUBank: None published - Support
- TensorWave: Email, chat, phone, tickets; solution engineers
GPUBank: Tickets from a $25 balance - Headquarters
- TensorWave: Las Vegas, United States
GPUBank: Tallinn, Estonia
How this page was made
Every price and policy of TensorWave was read on its own website, documentation or public API on 29 September 2026, with the address of each page listed above. A value we could not read on a primary source is marked "Not published" rather than estimated. GPUBank figures come from this site and its terms. Prices change: check both providers before you buy. TensorWave, its logo and its website belong to their owner and are shown only to identify the service compared. GPUBank is not affiliated with TensorWave.
