GPU MODELS / H800
H800 cloud GPU pricing
2 listings across 2 providers. Cheapest published rate: $1.560/GPU/hr at AutoDL. Sorted by per-GPU hourly price. Every row links to the provider's own published price.
WHAT IS THE H800
The H800 was the first export-compliant Hopper for the China market: near-H100 tensor throughput but NVLink cut from 900 to 400 GB/s and FP64 essentially removed. It trains well inside a server but scales out worse than an H100 because the intra-node fabric is halved. Where the panel lists it, it sits between the H100 and the later, more restricted H20.
MEMORY BANDWIDTH
3.35 TB/s
TENSOR COMPUTE
1,513 TFLOPS FP16/BF16 tensor (3,026 TFLOPS FP8, TF32 756 TFLOPS)
POWER DRAW
700 W (SXM5; 350-400 W PCIe/NVL SKUs)
INTERCONNECT
NVLink 400 GB/s (SXM5), PCIe 5.0
WHAT IT IS GOOD AT
LLM training
Capable at single-node scale: roughly three quarters of the H100's FP16 throughput. Multi-node training works but leans harder on external InfiniBand/Ethernet because NVLink is capped.
Large-batch inference
Strong. H100-class compute with an 80 GB HBM3 memory system serves large models at high batch sizes.
FP64 HPC
Do not. FP64 is cut to the 1 TFLOPS class for compliance.
WHO SHOULD NOT RENT THE H800
FP64 scientific computing - the double-precision units are fused off to roughly 1 TFLOPS.
Very large training clusters where NVLink domain size drives the design; the 400 GB/s cap changes cluster economics versus H100.
| Provider | Config | VRAM | Location | $/hr | $/GPU/hr | Source |
|---|
 AutoDL | H800 × 1 | 80 GB | China | $1.560 | $1.560 | receipt |
 CloudGPU | H800 × 1 | 80 GB | Not published | $1.790 | $1.790 | receipt |
THE H800 MARKET, BY THE NUMBERS
2 providers publish 2 H800 configurations on the panel today. Published per-GPU hourly rates run from $1.560 (AutoDL) to $1.79 (CloudGPU), a 1x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $1.79/GPU/hr. 1 rows were reported in stock by the provider at last observation; the rest are published catalog rates. All rates are on-demand - committed-use and reserved pricing is excluded.
H800 PRICE TREND, DAILY PANEL SNAPSHOTS
Across every H800 row on the panel, the daily floor moved from $1.790 to $1.560/GPU/hr over 10 days of tracking (-12.8%). Snapshots run daily; the full history is public in the repo.
H800 PROVIDER NOTES
AutoDL · 1 config · 1 regionSets the panel floor for this model.
from $1.560/hrCloudGPU · 1 config · 1 in stockWithin 15% of the panel floor.
from $1.790/hrBUYING H800: WHAT THE PANEL SAYS
At the current floor, one H800 running around the clock costs about $1,139 a month (730 hours of on-demand arithmetic). Compare the alternatives below before committing - the same budget often buys more than one chip class.
H800 VS THE ALTERNATIVES
The H800 keeps about 76% of the H100's FP16 tensor throughput (1,513 vs 1,979 TFLOPS) but halves NVLink to 400 GB/s and drops FP64. Single-node training is close; scale-out training and HPC are not.
The H20 is the later, tighter compliance cut: 96 GB of memory but only 148 TFLOPS FP16. The H800 has roughly 10x the compute; the H20 has the newer memory system. Training says H800, memory-per-dollar inference says H20.
Compare live rates: H800 vs H20 · H800 vs H100 SXM
H800 PRICING FAQ
What is the cheapest H800 cloud GPU right now?
The cheapest published H800 rate on the panel is $1.560 per GPU-hour at AutoDL (H800 x 1, China). Every price links to the provider's own published rate.
How much does a H800 cost per hour?
Across 2 providers tracked today, published per-GPU hourly rates for H800 run from $1.56 to $1.79, with a median of $1.79. Committed-use and reserved rates are excluded; everything shown is on-demand.
How many providers rent H800 GPUs?
2 providers publish 2 distinct H800 configurations on Compute Cafe. 1 of those rows were reported in stock by the provider's live endpoint at last observation.
Are H800 prices going up or down?
Over the 10 days Compute Cafe has tracked the panel, the H800 floor is falling: $1.790 to $1.560/GPU/hr (-12.8%). Daily snapshots are public in the repo, so the trend is auditable.
What does a H800 cost per month?
At the current floor of $1.560/GPU/hr, a H800 running 24/7 costs about $1,139 per month (730 hours of on-demand arithmetic on the cheapest published rate). Committed-use contracts can price lower; the panel tracks on-demand rates only.
How much slower is the H800 than the H100?
On raw tensor compute, about 25-35% depending on precision (FP16: 1,513 vs 1,979 TFLOPS). The bigger difference is NVLink at 400 GB/s versus 900 GB/s, which shows up in multi-GPU training efficiency, and FP64, which is essentially absent on the H800.
Can the H800 train large models?
Yes, and it does across the China market. Single 8-GPU nodes perform near H100 levels; at cluster scale the interconnect cap costs efficiency, so the same model needs more careful parallelism or more GPUs than an equivalent H100 deployment.
See also alternatives to the H800 · cheapest GPU-hour per model · inference chips · training chips · by location