Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
NVIDIA announced on September 20, 2022, that its H100 Tensor Core GPU had entered full production. That was not the same as saying finished DGX H100 systems were already shipping. DGX H100 systems were orderable at the time, but NVIDIA announced their full production in March 2023 and said they were shipping worldwide on May 1, 2023. The original September announcement did not promise that DGX H100 would ship in Q1 2023.
What NVIDIA announced in September 2022
NVIDIA’s September 20, 2022 announcement said that the H100 Tensor Core GPU, based on the Hopper architecture, was in full production. It described a staged rollout: partner products were expected to start appearing in October, H100-based systems were expected in the coming weeks, and more than 50 server models were expected by the end of 2022, with additional models in the first half of 2023. NVIDIA also said DGX H100 systems could be ordered.
Those statements refer to different stages and products. GPU production does not establish that every server design is built, delivered, or available in every region. “Orderable” is not the same as “shipping,” either. The September release described the H100’s production status and the broader ecosystem’s expected rollout; it was not a specific Q1 2023 shipping promise for DGX H100.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesH100 GPU, HGX platform and DGX system: what’s the difference?
| Term | What it is | What availability means |
|---|---|---|
| H100 | An individual NVIDIA accelerator, offered in SXM and PCIe forms. | GPU production can be underway before a particular finished server is ready or available to a buyer. |
| HGX H100 | A multi-GPU server platform used by system manufacturers to build their own systems. | Availability depends on the manufacturer’s completed server, configuration, region and supply. |
| DGX H100 | NVIDIA’s integrated, supported eight-GPU enterprise AI system. | System production and delivery are separate milestones from H100 GPU production. |
| DGX SuperPOD | A larger infrastructure design that connects multiple DGX systems. | It adds facility, networking and deployment requirements beyond one DGX node. |
| DGX Cloud | A hosted NVIDIA AI infrastructure service, rather than a DGX server installed at the customer’s site. | Access depends on the commercial offer, provider, region and capacity. |
SXM and PCIe H100s are not interchangeable parts in every machine. SXM modules are intended for tightly integrated platforms such as DGX and HGX designs; PCIe cards suit a wider range of compatible servers. Form factor affects power, cooling, memory and interconnect. A server must be designed and validated for the GPU type it uses.
#1 Best Overall
- Standard Memory: 40 GB
- Host Interface: PCI Express 4.0
- Cooler Type: Passive Cooler
- Product Type: Graphics Card
Why H100 mattered
H100 introduced NVIDIA’s Hopper architecture for data-center computing. Its Transformer Engine is designed to accelerate transformer workloads using FP8 and mixed-precision computation. That can reduce the amount of time spent on supported AI operations, but realized speed depends on the model, software, precision choices and system configuration.
Memory and communication matter alongside arithmetic throughput. H100 uses HBM3 memory, while NVLink and NVSwitch provide high-bandwidth communication among GPUs in supported systems. NVIDIA described up to 900 GB/s of GPU-to-GPU connectivity for the DGX H100 architecture. In multi-GPU training, data movement between accelerators and across nodes can constrain performance, so a fast GPU alone does not guarantee fast end-to-end training. Hopper also includes confidential-computing support for workloads that need additional protection.
NVIDIA advertised up to 9× faster AI training and up to 30× faster large-language-model inference than A100 in selected comparisons. These are vendor claims, not universal application results. The outcome depends on the model and sequence length, batch size, precision, software and kernels, sparsity, GPU count and networking. Likewise, the DGX H100’s 32-petaflops figure is a system-level FP8 performance specification, not a promise that every application will sustain that rate.
Rank #2
- 3rd generation Tensor core and 2nd generation RT core provide 1.5 times more graphic CAD performance and 3 times more rendering and generating AI performance than previous model T1000
- It delivers up to twice the real-time lay-tracing performance of previous generations, allowing you to perform complex 3D model processing and more realistic image processing in half the time
- Up to 3.6 times higher generation AI performance than previous generations, creating high-quality images/videos, and generating 3D assets quickly
- Equipped with 4 Mini DisplayPort connectors for increased productivity for multi-application workflow. 8K output can also output 2 screens simultaneously
DGX H100 specifications and operating implications
The DGX H100 is an integrated enterprise system, not simply a workstation with a powerful GPU. NVIDIA’s datasheet lists eight H100 GPUs with 640 GB of aggregate GPU memory, 2 TB of system memory, two x86 CPUs, four NVSwitch devices and eight 3.84 TB NVMe U.2 drives. It lists 32 petaflops of FP8 performance, ConnectX-7 networking, NVIDIA Base Command and NVIDIA AI Enterprise, and three-year business-standard hardware and software support. Configurations support high-speed InfiniBand or Ethernet networking; check the exact system specification for the adapters and network setup being quoted.
The listed maximum system power is approximately 10.2 kW. That makes facilities planning part of the purchase decision: validate rack power and redundancy, cooling, rack space and depth, network switches and cabling, maintenance access, and backup power. Confirm requirements for the exact configuration and installation rather than treating a DGX as ordinary office equipment.
The actual production and shipping timeline
| Date | Milestone | What it establishes |
|---|---|---|
| March 22, 2022 | NVIDIA introduced the Hopper architecture and H100. | Announcement of the GPU generation and its planned products—not a claim that customer systems were shipping. |
| September 20, 2022 | NVIDIA said H100 was in full production. | GPU production status; NVIDIA also outlined partner rollouts and said DGX H100 systems were orderable. |
| October 2022 and after | Partner products and services were expected to roll out in stages. | A forecast for ecosystem availability, not proof of immediate supply in every market. |
| March 21, 2023 | NVIDIA said DGX H100 AI supercomputers were in full production and becoming available to enterprise customers. | A separate production milestone for the complete DGX system. |
| May 1, 2023 | NVIDIA said DGX H100 systems were shipping worldwide. | The clearest public confirmation in NVIDIA’s announcements that completed DGX systems were shipping globally. |
Sources: NVIDIA’s Hopper announcement, September 2022 production announcement, March 2023 announcement and May 2023 shipping update.
Rank #3
- Support for upto four 5K displays
- Four Mini DisplayPort 1.4 connectors with latching mechanism1
- DisplayPort with audio
- HDCP 2.2 support
- 3 Years warranty
“Shipping worldwide” does not mean every configuration was in stock for every buyer, or that delivery times, quantities and local availability were uniform. For a procurement decision, confirm current availability with NVIDIA or the relevant system supplier.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow organizations could obtain H100 capacity
There are three broad routes: purchase an integrated DGX system, buy an OEM server built around H100/HGX, or rent GPU capacity from a cloud provider. They solve different problems. DGX bundles hardware, networking, software and enterprise support; OEM systems may offer different configurations; cloud instances avoid an on-premises installation but bring region, capacity, billing and service constraints.
- AWS EC2 P5: AWS lists configurations from a one-H100 P5.4xlarge to an eight-H100 P5.48xlarge. Check the current instance page and the region-specific pricing and capacity terms. Capacity Blocks prices are a distinct purchasing mode and should not be mistaken for a universal on-demand rate.
- Google Cloud A3: A3 High and A3 Mega configurations include eight H100 GPUs per listed machine type. See Google’s accelerator-optimized pricing page for current rates and regional terms.
- Azure, Oracle Cloud and specialist GPU clouds: H100 offerings and availability vary by region and time. Verify the provider’s live product and capacity details rather than relying on an old launch announcement.
- DGX Cloud: A hosted NVIDIA environment, which avoids installing a DGX system locally. Any historical launch pricing should not be assumed to describe the current commercial offer; request current terms.
Cloud prices and capacity change. Compare the full cost for the required GPU count, runtime, storage, data transfer, networking, support and minimum commitment—not just a per-GPU headline figure. A one-GPU development job and distributed multi-node training job have very different infrastructure needs.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.When does DGX H100 make sense?
A DGX H100 is most defensible when an organization has sustained demand for large-model training or other workloads that benefit from an eight-GPU node, needs an integrated and supported NVIDIA stack, and has the facilities and staff to operate it. Standardized on-premises infrastructure can also suit organizations with data-control requirements or a plan to scale into DGX SuperPOD-class deployments.
It can be excessive for occasional experiments, small-scale inference, or fine-tuning that fits on one GPU. It is also a poor fit if the organization lacks rack power, cooling, networking or a credible utilization plan. An expensive accelerator that sits idle can cost more than renting capacity when needed.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Before committing, validate CUDA and driver compatibility; framework and Transformer Engine support; FP8 behavior for the actual workload; NCCL performance on the intended topology; containers and scheduler integration; checkpoint and storage throughput; and software licensing and support terms. Peak specifications do not replace testing with the intended model and software stack.
Best Value
- 900-5G172-2280-000
Should a new deployment still use H100?
As of September 2026, H100 remains listed in cloud and enterprise offerings, but it is no longer NVIDIA’s newest data-center accelerator. For a new deployment, compare it with H200 and Blackwell-generation systems using the actual workload, required memory capacity, software compatibility, delivery timing and total cost. H200 may suit memory-capacity or bandwidth-constrained workloads; newer systems may be preferable for other requirements, but the right answer needs model-specific benchmarks rather than a generation label alone. A100 can also be a lower-cost option when its capabilities meet the job.
For purchase versus rental, estimate utilization over the equipment’s useful life and include acquisition, support, power, cooling, space, networking and operations. Rent when demand is bursty, benchmarking is still underway, or capital and deployment lead time are concerns. Buy when use is predictably high, on-premises control matters, and the organization can operate the system. In either case, verify current regional availability and commercial terms before deciding.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
