I'd love to hear what you think! Please drop me a line and let me know what you like and what could be better. 🙏
NVIDIA H100 PCIe 80GB Specs, Benchmarks & Pricing
The NVIDIA H100 PCIe 80GB is a professional graphics card designed primarily for machine learning and high-performance computing applications. Announced in March 2022 and shipping from fall 2022, it is built on the innovative Hopper GH100 architecture and utilizes the 4 nm process by TSMC. The H100 PCIe 80GB stands out for its substantial memory capacity and high-speed memory bandwidth, making it exceptionally suitable for handling large datasets and complex machine learning models. Its tensor cores significantly accelerate machine learning operations. The Hopper architecture, as seen in the H100 and H200 GPUs, introduces several technological innovations aimed at enhancing performance in AI training and inference. These GPUs are distinguished by their massive transistor count and advanced memory technologies, like HBM3e and HBM2e, supporting up to 80 GB of memory. The H100 supports HBM2e memory, while the H200 supports the faster HBM3e memory system, which can deliver up to 3 TB/s, a significant increase over the previous generation's capabilities. A common nvidia part number for this card is 900-21010-0000-000.
- Release Date: October 1, 2022
- MSRP: $99,696 USD†
- GPU Architecture: hopper
- Hardware-Accelerated GEMM Operations:FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
- CUDA Compute Capability : 9
† Estimated MSRP of $99,696 USD, based on an OEM list price. NVIDIA does not publish an MSRP for data center GPUs. This is the Cisco list price (GPL) for UCSC-GPU-H100-80=, and the only OEM list price found. Lenovo and HPE figures seen (for example Lenovo 4X67A82257, HPE R9S41C) are reseller selling prices, not list prices, so they are not used. This is a current list price rather than a 2022 launch price; the only launch-time price found is the GDEP Advance pre-order in April 2022 at ¥4,745,950 (about $36,550). Pricing varies by form factor (PCIe vs SXM) and memory size.
Strengths
- Excellent FP32 compute performance (top 6% of GPUs)
- Excellent FP16 compute performance (top 9% of GPUs)
Specifications for NVIDIA H100 PCIe
| Specification | Performance Ranking |
|---|---|
| FP32 TFLOPs | |
| FP16 TFLOPs | |
| Tensor Core Count | |
| Memory Capacity (GB) | |
| Memory Bandwidth (GB/s) | |
| Int8 TOPs | |
| FP8 TFLOPs | |
| FP4 TFLOPs | Data not available |
Real-time NVIDIA H100 PCIe GPU Prices
We're tracking 6 of the NVIDIA H100 PCIe GPUs currently available for sale. The lowest average price is $39,211 (what's this?)
Buy NowCompare Price/Performance to other GPUs
We track real-time prices of other GPUs too so that you can compare the price/performance of the NVIDIA H100 PCIe GPU to other GPUs.
Compare GPU Price/PerformanceCompare NVIDIA H100 PCIe to Another GPU
Compare the NVIDIA H100 PCIe directly to another GPU to see specs, benchmarks, and prices side-by-side.
Compare GPUs Side-by-SidePrice History
NVIDIA H100 PCIe Price History
This chart tracks NVIDIA H100 PCIe prices over the past 6 months based on eBay listings. The yellow line shows the average of the three lowest-priced listings each day—a useful indicator for finding deals.
Trend: Over this period, the lowest average price has been down 12.7%. The current lowest average is $33,236, compared to a typical $34,280 over the period. The lowest average price is calculated as the average of the three lowest-priced listings each day.
gpupoet.com
Product Identifiers
Manufacturer Part Numbers (3)
- NVIDIA Part Number
- 900-21010-0000-000
- Product SKU
- P1010 SKU 200
- NVIDIA Part Number
- 699-21010-0200-xxx
Available from 7 Partners (15 products)
Dell
- NVIDIA H100 PCIe Tensor Core GPU 80GB
- DGP4C(part number)
- NVIDIA H100 PCIe Tensor Core GPU 80GB with Passive Heatsink
- DP44G(part number)
- NVIDIA H100 PCIe Tensor Core GPU 80GB
- 900-21010-0100-030-DELL(part number)
HPE
- NVIDIA H100 80GB PCIe Accelerator for HPE
- R9S41C(part number)
HP
- NVIDIA H100 PCIe Tensor Core GPU 80GB
- P54751-001(part number)
PNY
- NVIDIA H100 Tensor Core GPU 80GB PCIe
- NVH100TCGPU-KIT(sku)
- NVIDIA H100 Tensor Core GPU 80GB PCIe Retail
- XNVH100TCGPU-KIT-RETAIL(sku)
Supermicro
- NVIDIA H100 80GB PCIe 5.0 x16 Passive Cooling
- GPU-NVH100-80(part number)
- NVIDIA H100 80GB PCIe 5.0 x16 Passive Cooling
- GPU-NVH800-80(part number)
- NVIDIA H100 80GB GPU PCIe 5.0
- 699-21010-0200-600(part number)
Lenovo
- ThinkSystem NVIDIA H100 80GB PCIe Gen5 Passive GPU
- 4X67A82257(part number)
Cisco
- NVIDIA H100 PCIe 350W 80GB 2-Slot FHFL GPU
- UCSC-GPU-H100(part number)
- NVIDIA H100 350W 80GB 2-Slot FHFL GPU
- UCSC-GPU-H100-80(part number)
- NVIDIA H100 350W 80GB 2-Slot FHFL GPU (Spare)
- UCSC-GPU-H100-80=(part number)
- Cisco UCS NVIDIA H100 80GB Passive 350W DW FH/HL GPU
- UCSX-GPU-H100-80(part number)
References
- https://resources.nvidia.com/en-us-tensor-core/nvidia-tensor-core-gpu-datasheet
- https://www.nvidia.com/en-us/data-center/h100/
- https://www.techpowerup.com/gpu-specs/h100-pcie-80-gb.c3899
- https://nvidianews.nvidia.com/news/nvidia-announces-hopper-architecture-the-next-generation-of-accelerated-computing
- https://nvidianews.nvidia.com/news/nvidia-hopper-gpus-expand-reach-as-demand-for-ai-grows
- https://nvidianews.nvidia.com/news/nvidia-hopper-in-full-production
- https://en.wikipedia.org/wiki/Hopper_(microarchitecture)
- https://www.nvidia.com/en-us/data-center/technologies/hopper-architecture/
- https://www.nvidia.com/en-us/data-center/tensor-cores/
- https://www.tech-america.com/item/cisco-nvidia-h100-graphic-card/ucsc-gpu-h100-80
- https://itprice.com/cisco-gpl/ucsc-gpu-h100-80=
- https://news.mynavi.jp/article/20220502-2336156/
- https://www.tomshardware.com/news/nvidia-hopper-h100-80gb-price-revealed
- https://www.serversupply.com/GPU/HBM2E/80GB/NVIDIA/900-21010-0000-000_387928.htm
- https://www.itcreations.com/product/140934
- https://www.itcreations.com/product/149409
- https://www.hpe.com/us/en/product-catalog/options/pip.specifications.nvidia-accelerators.1014716083.html
- https://buy.hpe.com/us/en/Options/Server-Accelerators/Computational-and-Graphics-Accelerators-for-Servers/Computational-and-Graphics-Accelerators-with-HPE/NVIDIA-H100-80GB-PCIe-Accelerator-for-HPE/p/R9S41C
- https://www.itcreations.com/product/146020
- https://www.centralcomputer.com/pny-nvidia-h100-tensor-core-gpu-accelerator-80gb-hbm2e-nvh100tcgpu-kit.html
- https://www.itcreations.com/product/146438
- https://www.renewtech.com/supermicro-699-21010-0200-600.html
- https://www.ahead-it.eu/en/shop/hardware/supermicro/supermicro-gpu-nvh100-80-nr-nvidia-h100-80gb-pcie-5-0-x16-passive-cooling
- https://lenovopress.lenovo.com/lp1732-thinksystem-nvidia-h100-pcie-gen5-gpu
- https://serverproven.lenovo.com/option/4X67A82257/
- https://www.itcreations.com/product/145735
- https://www.provantage.com/cisco-systems-ucsc-gpu-h100-80~4CSC651P.htm
- https://www.connection.com/product/cisco-nvidia-h100-pcie-graphics-card-80gb/ucsc-gpu-h100-80/41789563
- https://expresscomputersystems.com/products/cisco-ucs-nvidia-h100-80gb-passive-350w-dw-fh-hl-gpu-ucsx-gpu-h100-80
- https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf
- https://server2u.com/shop/900-21010-0000-000-900-21010-0300-030-nvidia-h100-pcie-80gb-60144
- https://e4company.com/wp-content/uploads/nvidia-h100-datasheet-nvidia-us-2287922-web-sep22-partner-r6.pdf
Notes
- fp32TFLOPS of 378 is TF32 Tensor Core throughput without sparsity: half the 756 TFLOPS sparse TF32 figure NVIDIA published for the H100 PCIe in its H100 datasheet (sep22 r6, e4company.com mirror in references). The current nvidia.com/en-us/data-center/h100/ page lists only SXM and NVL, and 378 has not been re-read from a vendor source since. FP32 vector (CUDA core) throughput is 51.2 TFLOPS.
- fp16TFLOPS: NOTE: FP16 Tensor Core
- int8TOPS of 1513 represents non-sparse (dense) INT8 Tensor Core performance, derived from sparse INT8 of 3026 TOPS ÷ 2 per NVIDIA Hopper architecture datasheet. With 2:4 structured sparsity, INT8 performance is 3026 TOPS. Dense value used for consistent cross-vendor comparison.
- fp8TFLOPS of 1513 represents non-sparse (dense) FP8 Tensor Core performance, derived from sparse FP8 Tensor Core of 3,026 TFLOPS ÷ 2, per the NVIDIA H100 Tensor Core GPU Datasheet (mirrored at e4company.com/wp-content/uploads/nvidia-h100-datasheet-nvidia-us-2287922-web-sep22-partner-r6.pdf, which lists "FP8 Tensor Core 3,026 TFLOPS*" for H100 PCIe with the footnote indicating 2:4 structured sparsity). This matches int8TOPS exactly (1513), consistent with the H100/H200 datasheets showing identical peak throughput for FP8 Tensor Core and INT8 Tensor Core on Hopper. Dense value used for consistent cross-vendor comparison.
- fp4TFLOPS is null: Hopper predates FP4 support, which was introduced in Blackwell. Confirmed hardware negative per the NVIDIA Hopper architecture datasheet, which lists no FP4 Tensor Core row (resources.nvidia.com/en-us-tensor-core/nvidia-tensor-core-gpu-datasheet). Not added to unverifiedSpecs since this is a confirmed negative, not a data gap.
- supportedHardwareOperations re-verified 2026-09-02 against the canonical Hopper baseline (FP64, TF32, BF16, FP16, FP8, INT8) documented in .claude/agents/gpu-researcher.md, sourced from the NVIDIA H100 Tensor Core GPU Datasheet. No FP32 or INT4 tensor-core row exists per the GEMM-only definition of this field (plain FP32 CUDA-core throughput is a separate, non-tensor spec). No changes were required for this file; it already matched the canonical set.
- Estimated MSRP of $99,696 USD, based on an OEM list price. NVIDIA does not publish an MSRP for data center GPUs. This is the Cisco list price (GPL) for UCSC-GPU-H100-80=, and the only OEM list price found. Lenovo and HPE figures seen (for example Lenovo 4X67A82257, HPE R9S41C) are reseller selling prices, not list prices, so they are not used. This is a current list price rather than a 2022 launch price; the only launch-time price found is the GDEP Advance pre-order in April 2022 at ¥4,745,950 (about $36,550). Pricing varies by form factor (PCIe vs SXM) and memory size.
- releaseDate of 2022-10-01 is month-level: NVIDIA announced the H100 on 2022-03-22 for availability from the third quarter, its March 2023 release says H100 began shipping in fall 2022, and the 2022-09-20 Hopper in Full Production release (nvidianews.nvidia.com/news/nvidia-hopper-in-full-production) names October as when partners roll out the first systems. The exact shipping day is unconfirmed. 2023-03-21 is the H100 NVL announcement date, not the PCIe card.
- Manufacturer identifiers sourced from NVIDIA official documentation and OEM product listings. Primary NVPN 900-21010-0000-000 confirmed across multiple resellers. Product SKU P1010 SKU 200 and alternate NVPN 699-21010-0200-xxx found in NVIDIA product briefs and OEM documentation.
- Third-party product identifiers sourced from official OEM websites and authorized reseller listings. Dell part numbers (DGP4C, DP44G, 900-21010-0100-030-DELL) from ITCreations and Express Computer Systems. HPE part number (R9S41C) from HPE official product catalog. HP part number (P54751-001) from ITCreations. PNY SKUs (NVH100TCGPU-KIT, XNVH100TCGPU-KIT-RETAIL) from Central Computer and ITCreations. Supermicro OEM part numbers (GPU-NVH100-80, GPU-NVH800-80, 699-21010-0200-600) from Ahead-IT and Renewtech. Lenovo part number (4X67A82257) from Lenovo Press and ServerProven. Cisco part numbers (UCSC-GPU-H100, UCSC-GPU-H100-80, UCSC-GPU-H100-80=, UCSX-GPU-H100-80) from Provantage, Connection, and Express Computer Systems.
- Note: Part number 900-21010-0300-030 is a variant SKU found in some reseller listings but specific differences from 900-21010-0000-000 are not documented in available sources.