SECTIONS
EDITIONS
NETWORK
AIXX Pedia The encyclopedia of compute hardware

AIXX Pedia › Model Entry › H100

DGX H100 eight-GPU platform official image (vendor image)
Vendor image · AIXX Pedia (NVIDIA official site)

H100

NVIDIA Hopper generation · Volume spot-market workhorse · 80GB HBM3 · Model Entry #003

Overview OVERVIEW

H100 (Full name NVIDIA H100, part of the "Hopper" series) is a data center GPU announced by NVIDIA at GTC on March 22, 2022 with mass production officially confirmed in September 2022 (shipping from October): the opening act of the Hopper generation, the first GPU with a Transformer Engine (FP8 debut), the first with HBM3 memory and PCIe Gen5[1][2]。

This is the card that "pried open the LLM era": after ChatGPT ignited the global compute arms race, the H100 briefly became the world's most sought-after unit of AI compute, and due to export controls it also became a symbol of the US-China chip standoff (see "History" and "Limitations & Controversies").

In the 2026 spot market, the H100's role is a volume workhorse : its compute is still full-fat Hopper, its ecosystem the most mature, and rental demand actually rebounded through 2025-2026 (public index basis[7][8]), but the hard 80GB memory ceiling is being undercut generation by generation by the H200 (141GB) and B300 (288GB) — before buying, first confirm that memory isn't your bottleneck.

Use Cases FIT

AIXX Index (0–100): estimated by this Pedia via weighted public specifications — memory 40% + interconnect 30% + ecosystem 30%, with weights adjusted by use case. 90+ flagship tier · 75+ Excellent · 60+ usable · below 60 not recommended. The methodology is transparent and verifiable, and does not constitute a performance guarantee.

LLM Inference 85AIXX Index
A single 80GB card can host FP8 deployment of 70B-class models; beyond that memory gets tight — for 100B and up, look at the H200.
Model Training / Fine-tuning 88AIXX Index
The most mature Hopper ecosystem and the largest global installed base; the mainstay platform for 1B-to-100B fine-tuning and LoRA pipelines.
HPC / FP64 Computing 92AIXX Index
FP64 34 TFLOPS / Tensor 67 TFLOPS — scientific computing and simulation with no equivalent substitute among the new consumer flagships.
Cost-sensitive Inference 90AIXX Index
2026 rental rates of $1.8-3.5/GPU·hr (public aggregation basis); its cost per token remains the pricing anchor of the spot market.
Large-memory scenarios (long context / full R1) 58AIXX Index
80GB×8=640GB can't fit full-precision R1 FP8 (~690GB) — for this class of need, go straight to the H200 entry.

History HISTORY

GTC on March 22, 2022 , GTC unveiled the Hopper architecture and the H100: 80 billion transistors on a TSMC 4N custom process, a Transformer Engine that brought FP8 into the GPU's primary precision lineup for the first time, and the first GPU with HBM3 + PCIe Gen5[2][3]。

September 20, 2022 GTC fall keynote confirmed mass production, shipping from October[4]; March 21, 2023 DGX H100 entered volume production and the first unit was delivered to OpenAI[5]; over the following year the H100 became the absolute mainstay of global AI compute expansion.

October 7, 2022 : US export controls placed the H100 under China licensing; NVIDIA responded with the A800/H800 China-market variants (late 2022) and the H20 China-market line (shipped 2024) to keep supplying China, with repeated interruptions from the October 2023 control escalation and the April 2025 H20 licensing rules[6]. The H100 itself never returned to the approved China list — what was approved for China in December 2025 was the H200[9]。

2025-2026 : the H100 entered its "installed-base" phase. The H200/B300 siphoned off new purchase orders generation by generation, but inference demand pushed rental prices back up — the rental index rose roughly 10% from December 2025 to January 2026[7], and in April 2026 one-year contract rates broke $2/GPU·hr (public analysis basis[8]). It remains the highest-volume Hopper model in the spot market.

Technical Specifications SPECIFICATIONS

ItemSpecifications (H100 SXM · official figures)
GPU CoreNVIDIA Hopper GH100(SXM5)
Board Memory80GB HBM3
Memory Bandwidth3.35TB/s per card
FP8 Tensor Core3,958 TFLOPS (with sparsity)
FP16 / BF16 Tensor Core1,979 TFLOPS (with sparsity)
TF32 Tensor Core989 TFLOPS (with sparsity)
FP64 / FP64 Tensor34 / 67 TFLOPS
GPU-to-GPU InterconnectNVLink 4th gen · 900GB/s + NVSwitch full interconnect
Board Power (TBP/TGP)Up to 700W (configurable; PCIe version 350W / NVL version 350-400W)
Mainstream System Form FactorsHGX H100 8-GPU baseboard (Air-cooled / Liquid-cooled) · DGX H100
Previous Generation / Competing ProductA100 80GB(Ampere)/ H200、B300

Specifications are subject to NVIDIA's official public documentation[1]; all Tensor compute figures are official "with sparsity" numbers — halve them for dense; full-system configs vary by OEM and batch.

80GBH100 · HBM3141GBH200 · HBM3e288GBB300 · HBM3eMemory per Card · Generation Comparison (NVIDIA official specs)

Generational Comparison · Four Generations Side by Side GENERATIONS

ItemA100 80GBH100 80GBH200B300
Board Memory80GB HBM2e80GB HBM3141GB HBM3e288GB HBM3e
Memory Bandwidth~2TB/s3.35TB/s4.8TB/s~8TB/s
Flagship precisionFP16FP8(Transformer Engine)FP8FP4
GPU-to-GPU InterconnectNVLink 3 · 600GB/sNVLink 4 · 900GB/sNVLink 4 · 900GB/sNVLink 5 · 1.8TB/s
One-line buying adviceInstalled-base continuationBudget ceiling · largest ecosystemLarge memory · volume mainstayFlagship new purchase · next generation

Specifications per NVIDIA official public materials; see the A100/H200/B300 entries for their respective columns.

Family Members FAMILY

The H100 family splits into four branches by package and form factor; in mainland China and globally, the spot market is dominated by HGX 8-GPU systems (this catalog's stock consists of HGX H100 8 systems from Supermicro / Dell / H3C / Gigabyte):

H100 SXM 80GB (on-board package)HGX H100 8-GPU baseboard (NVSwitch full interconnect)H100 PCIe 80GB HBM2e (350W add-in card)H100 NVL 94GB (dual PCIe cards bridged at 600GB/s)DGX H100 (NVIDIA official 8-GPU system)

Cooling & Deployment COOLING

The mainstream H100 is the air-cooled HGX 8-GPU form factor : the 700W-class SXM5 cooling tower solution is mature and deployable in existing server rooms — one reason it remains "compute with the lowest retrofit barrier" to this day. Liquid-cooled variants rolled out from 2023 with high-density projects (e.g., the Supermicro SYS-821GE-TNHR liquid-cooled system[12]), while the H100 PCIe version is a 350W dual-slot air-cooled card with no baseboard interconnect.

Liquid-cooled 8-GPU/NVSwitch tray, real-world photo
Liquid-cooled 8-GPU/NVSwitch tray · Actual photo
SXM module power delivery side, real-world photo
SXM module power delivery (VRM) side · Actual photo

Model Compatibility MODEL FIT

Model / ScenarioMemory requirement (FP8 estimate)H100 deployment (8 cards, 640GB)AIXX Index
DeepSeek R1 full (671B)~690GBDoesn't fit on eight cards — requires INT4 quantization (~350GB, fits) or multi-node deployment; for a full-precision single node, see H200/B30041
R1-Distill 70B-class~70GBDeployable on a single 80GB card, though headroom is tight (KV cache needs careful budgeting)82
72B-class open-source models~72GBSingle-card borderline; in practice usually run with two-card tensor parallelism75
7B-14B class (incl. small video-generation models)~8-30GBEasy on a single card; batch pipelines scale concurrency with multiple cards90

Memory estimated as FP8 weights ≈ parameter count × 1 byte (rough estimate, excluding KV cache and activation headroom); the AIXX Index is this catalog's own estimation basis, updated with measurements. Single-card 70B deployment is a borderline case — go by measured results.

H100 SXM module front, real-world photo
H100 SXM module front (GH100 die) · Actual photo
H100 SXM module back, real-world photo
SXM module rear power delivery · Actual photo
Liquid-cooled 8-GPU tray, real-world photo
Liquid-cooled 8-GPU/NVSwitch tray · Actual photo
H100 SXM module official image (vendor image)
H100 SXM module · Vendor image (NVIDIA official site)
DGX H100 eight-GPU platform official image (vendor image)
DGX H100 eight-GPU platform · Vendor image (NVIDIA official site)
Supermicro H100 server cluster data center deployment, real-world photo
Supermicro H100 server cluster, data center deployment · Actual photo (F5 BNK Lab)

Actual photos are added continuously as inspection batches are stocked; AI-generated images are never passed off as real photos.

Video Overview VIDEO

This video is the general stress-test reference footage (the six-test protocol is model-agnostic); the H100-specific version is in the production queue and will replace this one with an H100-anchored cover once live.

Community real hardware · Source: YouTube COMMUNITY HANDS-ON

Playback may not be available in some regions

Software & Ecosystem ECOSYSTEM

ItemRequirements / Status
CUDAHopper supported since 11.8 (launch version), fully covered by the current 12.x line
DriverFrom the 520 series (launch), current 550 series and above
Training frameworksPyTorch / TensorFlow officially support Hopper — the most mature ecosystem
Inference enginesvLLM / SGLang / TensorRT-LLM all use the H100 as their baseline supported model
NetworkingConnectX-7 400G or an equivalent interconnect solution
Operating systemMainstream Linux distributions (Ubuntu 22.04+ etc.)

With over three years of maturation, the Hopper ecosystem is the most developed; most public inference-optimization benchmarks use the H100 as the baseline model. Software stack versions evolve over time — defer to official NVIDIA releases.

Purchasing & Inspection PROCUREMENT

Three inspection points when buying an H100:

1
First, distinguish SXM / PCIe / NVL
"Low-price H100" quotes are often for the PCIe version (80GB HBM2e, 350W, 2TB/s, no NVSwitch full interconnect) — a full performance tier below the SXM version. The package form factor must be stated in quotes and contracts.
2
Verify serial numbers independently
The H100 and H200 look highly similar, and "passing off an H100 as an H200" has appeared in the spot market; check silkscreen markings against official firmware reports and cross-verify serial numbers through independent channels.
3
Measured memory bandwidth
Check the 80GB HBM3 and 3.35TB/s bandwidth against spec with a real measurement — memory is the core value of a generation and a hotspot for refurbished-unit tampering.

The full six-test stress-test protocol (figures anchored to the H100) is in this entry's "How to Stress-Test" section.

How to Stress Test BENCH TEST

Six-test protocol, all numeric examples anchored to the H100: unboxing identification → single-card Burn-in Test (700W air-cooled tower, ≤85°C, zero ECC errors to Pass) → Full Load on all cards → memory bandwidth (3.35TB/s spec, ≥85% to Pass) → Interconnect Bandwidth (nccl-tests ≥750GB/s vs. 900GB/s spec) → inference throughput (≥80% of public baselines).

① Unboxing identification nvidia-smi
Watch: card count · per-card VRAM · whether driver/CUDA versions match what was promised
Missing cards8/8 cards · 80GB+ firmware report
Missing cards or reduced VRAM → reject outrightAll 8/8 cards · 80GB/card+ official firmware report
Fail : missing cards, less VRAM than stated, or version mismatch; Pass : all 8 cards present, full VRAM, versions consistent; Excellent : official firmware report available for cross-reference.
Plain talk: count the cards first, then check VRAM — this step alone filters out half the tricks.
② Single-card Burn-in Test gpu-burn 3600 or dcgmi diag -r 3 · 30–60 minutes
Watch: temperature · power draw · ECC errors (700W at Full Load, air-cooled tower)
75℃80859095℃
Excellent ≤80°C, no throttlingPass ≤85°C, zero errorsFail ≥90°C, hard throttle
Fail : any error, Card Drop, or artifacts; Pass : sustains Full Load, zero ECC errors for 1 hour; Excellent : temperature ≤80°C with no throttling throughout.
Plain talk: a Burn-in Test makes the card sprint at full effort for an hour — weak silicon shows its true colors on the spot.
③ Full-system Full Load 8-card parallel Burn-in Test · ≥2 hours
Watch: ECC counters across all 8 cards · whether frequency curves stay aligned
0.5h1h2h
Any card lagging behind → rejectZero fatal ECCTemperature curves flat as a line
Fail : any card errors or falls behind; Pass : zero fatal ECC, no abnormal throttling; Excellent : all eight curves smooth and consistent.
Plain talk: examine all eight cards together — whoever falls behind is the problem. A single good card isn't good enough; the whole family has to keep up.
④ Measured Bandwidth of VRAM bandwidthTest(CUDA bundled sample)
Watch: measured bandwidth ÷ spec bandwidth (H100 HBM3 spec: 3.35TB/s)
70%808595%
Fail <70%Pass ≥85%Excellent ≥95%
Fail : below 70% of rating — strong evidence the memory was swapped or degraded; Pass :≥85%; Excellent :≥95%。
In plain terms: if the memory has been tampered with, the bandwidth gives it away immediately — this is the hurdle refurbished cards fear most.
⑤ Interconnect Bandwidth Measured Bandwidth nccl-tests all_reduce benchmark
Watch: eight-GPU fabric bandwidth (NVSwitch; per-GPU NVLink rated at 900GB/s)
0.650.750.810.9TB/s
Fail <650GB/sPass ≥750GB/sExcellent ≥810GB/s
Fail : Below 650GB/s — some GPUs have downgraded interconnect, or there's a cable/substrate issue; Pass :≥750GB/s; Excellent :≥810GB/s。
In plain terms: the speed at which eight cards work hand in hand — one poor connection and it shows instantly.
⑥ Inference Throughput Measured Bandwidth vLLM Run a real model under load
Watch: tokens/s throughput ÷ the model's public benchmark
50%80%100%Benchmark +10%
Fail: won't run / throughput cut in halfPass: ≥80% of benchmarkExcellent: within ±10% of benchmark
Fail : Service won't start or throughput is cut in half; Pass : Reaches 80% of the public benchmark; Excellent : Within ±10% of the benchmark.
In plain terms: passing everything above isn't the end — actually running a large model once is the report card buyers want.
⑦ Stability Wrap-up (Required for Large Orders) Full Load burn-in for 24–72 hours
Watch: hourly temperature/power/error logs
24h48h72h
Any card drop or reboot → rejectZero fatal ECCZero abnormal downclocking throughout
Fail : Card drops, reboots, fatal ECC errors; Pass : Triple zero (zero fatal ECC, zero abnormal downclocking, zero card-drop reboots); Excellent : Monitoring curves attached for every hour.
In plain terms: winning a sprint isn't winning — running three days straight without incident is true stability.

Methodology note: The Pass/Excellent thresholds were compiled by this database based on common engineering practice (bandwidth as a share of rated values, temperatures per standard datacenter limits), and will be calibrated gradually as Measured Bandwidth data from Inspection Reports accumulates; this table constitutes no acceptance commitment — specifics are subject to contract terms. The measured items in this database's Inspection Reports are recorded per this table — NOT_EXECUTED means the item was not run; "not tested" is itself important information.

Shipping & Delivery LOGISTICS

1
Customs & Compliance Documents
The H100 is an export-controlled model; the customs pathway and compliance status of cross-border units directly affect warranty and resale value. All accompanying documents must be verified and archived; compliance status should follow the relevant authorities' policies and professional advice.
2
Shipping Insurance
Insuring high-value equipment at full invoice value is standard industry practice; shipping insurance terms and claims procedures should be written into the contract.
3
Joint Inspection on Arrival
Check seals, exterior, and serial numbers one by one against the Inspection Report; photograph and document any discrepancy on the spot.

OEM & Variants VENDORS

On the same HGX H100 board, OEM systems differ in form factor, cooling, and service — start by comparing the differences (four of them are currently listed in our catalog):

System VendorRepresentative Form FactorDifferentiatorsBest For
SupermicroHGX H100 8 Systems (Air-cooled / Liquid-cooled)Volume workhorse; the liquid-cooled model has been fully reviewed by professional media[12]Spot Purchase · Best Value
DellPowerEdge Series (HGX H100 8)Global tier-one service network and enterprise-grade supportMultinational Deployment · Service First
H3CHGX H100 8 SystemLocal service and delivery network, common in APAC projectsAPAC Channels · Local Service
GigabyteHGX H100 8 SystemFlexible configurations, strong APAC channel presenceAPAC Channel Purchase
NVIDIA officialDGX H100 (8-GPU System)Official integrated delivery, full-stack software with vendor backing; first unit delivered to OpenAI[5]Turnkey Training · Vendor Support

AIXX Index · Four-Dimension Deployment Comparison AIXX INDEX

System VendorEase of InstallationMaintenance ConveniencePerformance DensityExpected Lifespan
Supermicro88848480
Dell90928485
H3C86848382
Gigabyte85808580
NVIDIA official92888285

The AIXX Index is our catalog's estimation methodology (0–100, based on public form factor, cooling solution, and service network differences), updated with hands-on verification, and does not constitute a purchase commitment. Refer to official vendor releases for specific models.

Glossary Cards GLOSSARY

HBM3 Memory

High-bandwidth memory stacked next to the GPU. Each H100 carries 80GB of HBM3 at 3.35TB/s — HBM3e only arrived with the H200 generation. Memory capacity is the H100's biggest selection drawback today.

NVLink 4 Interconnect

Fourth-generation NVLink, 900GB/s per-GPU direct interconnect; paired with NVSwitch, eight GPUs act as one. The H100 and H200 share the same generation and speed.

SXM5 Mezzanine Package

GPUs are soldered directly onto the board (not plug-in cards); the HGX H100 8-GPU board is the SXM5 form factor — inspect the board together with the GPUs during verification. The plug-in counterpart is the PCIe version (350W, HBM2e).

Transformer Engine / FP8

The H100's launch Mixed Precision engine made 8-bit floating point a mainstream precision for training and inference. Estimate memory footprint: FP8 weights ≈ parameter count × 1 byte.

Export Controls & Compliant Batches

The H100 has been subject to U.S. export controls on China since October 2022; the China-market Variant line (A800/H800/H20) are all adjusted models derived from these controls. Batch and customs pathway affect warranty and resale value — always check during inspection. All statements in this catalog reflect public reporting and do not constitute legal advice.

News & Related Events TIMELINE

Limitations & Controversies LIMITS

  • Hard Memory Ceiling : 80GB HBM3 / 3.35TB/s, substantially surpassed by the H200 (141GB / 4.8TB/s). Full DeepSeek R1 in FP8 is roughly 690GB, which the H100's 640GB across eight GPUs cannot fit (estimated; see "Model Fit") — the root cause of the selection divide between it and its successor.
  • The Face of China Export Controls : First added to the control list in October 2022, after which the China-market Variant line (A800/H800/H20) was adjusted several times under policy changes; the H100 itself has yet to return to the approved list for China (the H200 was approved in December 2025) — no public sales through legitimate mainland channels, spot availability is gray-market[6][9]. This catalog records public reporting only and does not constitute legal advice.
  • Blackwell Generation Pressure : The B300, with 288GB of HBM3e, FP4 precision, and NVLink 5 (1.8TB/s), opens up a two-precision-tier generational gap; the H100's share of new compute purchases keeps eroding.
  • Version Traps : The PCIe version (HBM2e / 2TB/s / 350W / no full NVSwitch interconnect) differs greatly in performance from the SXM version — most "cheap H100" quotes are for PCIe versions. Quotes and inspection must first pin down the package form factor.
  • Refurb & Impersonation Risk : The H100 and H200 look highly similar, and "passing off an H100 as an H200" has been reported in the spot market (per inspection agency bulletins); independent serial-number verification plus measured memory checks are essential steps.
  • Generational Debate : "The H100 is outdated" and "the H100 holds its value" camps coexist — rental prices rebounding in 2025–2026 (public index data[7]) supports the latter, while Blackwell's replacement curve supports the former; we take no stance here. See "Market Pulse" for the data.

Market Band MARKET DATA

H100 · Reference Price Band
DateBasisRegionReference Price BandNotes
2026-10-20Hall Real-TimeUS (USD)$1.87M – 1.87M1 quote › (batch asking prices)
2026-10-20Hall Real-TimeMainland China¥50.24M – 50.24M1 quote › (batch asking prices)
2026-10-20Hall Real-TimeSingapore (SGD)S$3.6M – 4.14M3 quotes (batch asking prices)
2026-10-20Hall Real-TimeMiddle East (AED)AED 3.6M – 3.6M1 quote (batch asking prices)
2026-10-20Hall Real-TimeEurope (EUR)€2.91M – 2.91M1 quote (batch asking prices)
2026-10-20Hall Real-TimeJapan (JPY)JPY 3.73M – 3.73M1 quote (batch asking prices)
2026-10Channel ReferenceGlobalNew single GPU around $31K eachAggregated public quotes[10] · Systems on Sale ›
2026-04Rental MarketNorth America CloudOne-year contracts >$2 / GPU·hr (+40%)Public analysis[8]
2026-01Rental MarketGlobalRental price index +10% in one monthSilicon Data Rental Index[7]
2026-10Policy StatusMainland ChinaNo approved public salesThe H100 is not on the approved list for China; spot availability is gray-market (public reporting; not legal advice)
Methodology: The market band aggregates publicly available channel information and listed price ranges in our catalog, for market reference only, and does not constitute any offer to trade. Prices fluctuate with batch, condition, quantity, and channel terms; actual price subject to the seller's quote. Listed systems are HGX H100 8-GPU systems (batch asking prices). Cross-currency bands are recorded in each region's native currency and are not converted between one another. Price data source for this catalog: AIXX Exchange (Zero-commission Matching; does not participate in transactions). Market snapshots have been accumulated daily since 2026-10-20.
Time Reference price band (estimated)
Region

Methodology: North America is an estimated band — anchored to publicly reported system prices of $280K–350K/unit at initial 2023 shipment, and to the on-sale observed price of $1.874M/unit (1 listing) on the October 2026 anchor date, with slow-rise interpolation in between, replaced point-by-point as daily snapshots accumulate. Mainland China/Singapore/Europe/Japan/Middle East are on-sale observed snapshots (CNY 5.024M/unit · 1 listing · SGD 3.604–4.142M · 3 listings · EUR 2.915M / JPY 3.734M / AED 3.601M · 1 each), all in native currency, not converted between one another. Price data source for this catalog: AIXX Exchange (Zero-commission Matching; does not participate in transactions, does not possess or speculate on transaction prices — hence the data's neutrality). The estimated band is for trend reference only; actual price subject to the seller's quote.

Price milestone timeline MILESTONES

FAQ FAQ

Is the H100 still worth buying?

Look at the bottleneck: for budget-sensitive builds, sub-70B-scale inference and fine-tuning, and HPC/FP64 workloads, the H100 remains the spot market's largest ecosystem with the deepest installed base — and rental prices even rebounded in 2025–2026 (public index data). But 80GB of memory is a hard ceiling for tens-of-billions-parameter models — if memory is the bottleneck, go straight to the H200.

How big is the gap between the H100 and H200?

Compute is on par and the interconnect is the same generation (NVLink 4 · 900GB/s); the difference is all in memory: 80GB HBM3 / 3.35TB/s vs 141GB HBM3e / 4.8TB/s; 640GB vs 1.128TB across eight GPUs — full R1 in FP8 (about 690GB) won't fit on an H100 but deploys on a single H200 node. In one line: the H200 sells memory, the H100 sells budget.

Can the H100 be bought for China?

The H100 has been under U.S. export controls on China since October 2022; the H200 was approved for China in December 2025, but the H100 is not on the approved list. There are no public sales through legitimate mainland channels; spot availability is gray-market. This reflects public reporting and is not legal advice — defer to the relevant authorities' policies and professional compliance advice before purchasing.

How to tell an H100 from an H200 (anti-impersonation)?

The two generations look highly similar, and photos alone aren't reliable: check silkscreen markings against official firmware reports, verify serial numbers through independent channels, then settle it with a measured memory check — 80GB vs 141GB shows up immediately (see "Purchasing & Inspection" and "How to Stress Test").

H100 SXM vs PCIe: what's the difference?

SXM version: 700W, HBM3 at 3.35TB/s, full NVSwitch interconnect — the mainstream HGX 8-GPU form factor. PCIe version: 350W, HBM2e at 2TB/s, and no high-speed interconnect without a dual-GPU bridge. Most "cheap H100" listings are PCIe versions; quotes and contracts must specify the package form factor.

H200B300HBM3eNVLinkFP8HGX H100 80GBA100 (Previous Generation)

Model comparison and selection COMPARE

References REFERENCES

  1. NVIDIA Official Site — H100 Product Page (primary spec source, verified 2026-10) https://www.nvidia.com/en-us/data-center/h100/
  2. Tom's Hardware — Nvidia Reveals Hopper H100 GPU With 80 Billion Transistors (2022-03 GTC launch coverage) https://www.tomshardware.com
  3. Nvidia Launches Hopper H100 GPU, New DGXs and More(2022-03-22) https://www.hpcwire.com
  4. CRN — Nvidia H100 GPUs In Full Production, Shipping In October(2022-09-20 keynote) https://www.crn.com/news/components-peripherals/nvidia-h100-…
  5. NVIDIA Investor Relations — NVIDIA Hopper GPUs Expand Reach as Demand for AI Grows (2023-03-21, DGX H100 mass production / first unit to OpenAI) https://investor.nvidia.com/news/press-release-details/2023/…
  6. CRS R48642: U.S. Export Controls and China: Advanced Semiconductors (Export Control Timeline) https://www.congress.gov/crs-product/R48642
  7. Silicon Data — H100 Rental Price Index +10% (2026-01-09) https://www.silicondata.com
  8. H100 rental prices +40%, one-year contracts break $2/GPU·hr (2026-04, secondary reporting) https://semianalysis.com
  9. H200 approved for China; channel module prices around RMB 1.4M per module (2025-12-24, same as H200 entry [2]) 
  10. Channel Quote Aggregation — New H100 around $31,000/GPU (2026 data) https://benchgecko.ai
  11. Checking out the NVIDIA H100 in Our First Look at Hopper (module photos) https://www.servethehome.com/checking-out-the-nvidia-h100-in…
  12. A Look at the Liquid Cooled Supermicro SYS-821GE-TNHR 8x NVIDIA H100 AI Server (liquid-cooled system photos) https://www.servethehome.com/a-look-at-the-liquid-cooled-sup…
  13. Market band data — AIXX Exchange (listings on this platform and channel-informed ranges, updated daily) /hall.html
Versions and errata | This entry is maintained by the AIXX Pedia editorial team, AI-assisted drafting with human fact-checking. Updated 2026-10-20 · Version v1.0 · Full change history auditable. Spotted an error? Reach us via Corrections Channel ; corrections are published after verification and recorded in version history.

This Library
Editorial PolicyAbout This Library (Disclosure of Interests)Corrections Channel
Editorial Constitution Summary
Every claim sourcedDrawbacks always stated · corrections publicly loggedNo paid entries acceptedZero traffic-funneling or sales elements within entriesMarket data attached only as data modules
AIXX Family
AIXX PediaAIXX ExchangeAIXX News
Market data source: AIXX Exchange · © 2026 AIXX Pedia · An AIXX Product