Compare · prices checked 2026-09-03
H100 NVL vs H100 SXM: specs, price per hour, which to rent
NVIDIA H100 NVL (80 GB, $1.877/hr) against NVIDIA H100 SXM (80 GB, $1.587/hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.
H100 NVL vs H100 SXM specifications
Public NVIDIA figures (dense, non-sparsity). The last column is H100 NVL relative to H100 SXM.
| Spec | H100 NVL | H100 SXM | Difference |
|---|---|---|---|
| Architecture | Hopper (2023) | Hopper (2022) | — |
| VRAM | 80 GB HBM3 | 80 GB HBM3 | same |
| Memory bandwidth | 3,900 GB/s | 3,350 GB/s | +16% |
| FP16 tensor (dense) | 835 TFLOPS | 990 TFLOPS | −16% |
| FP32 | 60.0 TFLOPS | 67.0 TFLOPS | −10% |
| CUDA cores | 14,592 | 16,896 | −14% |
| TDP | 400 W | 700 W | −43% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 588 | 697 | −16% |
| Max GPUs per machine | 4× | 8× | — |
H100 NVL vs H100 SXM price per hour
Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.
| Rate | H100 NVL | H100 SXM | Cheaper |
|---|---|---|---|
| On-demand, per GPU-hour | $1.877 | $1.587 | H100 SXM (−15%) |
| Interruptible, per GPU-hour | $0.938 | $0.793 | H100 SXM |
| Reserved (3 mo), per GPU-hour | $1.220 | $1.031 | H100 SXM |
| On-demand, per month | $1,370 | $1,159 | H100 SXM |
| Market median (reference) | $2.68 | $2.27 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.25 | $1.60 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0235 | $0.0198 | H100 SXM (better value) |
Try a full month with storage and bandwidth in the GPU cost calculator.
What fits in VRAM: 80 GB vs 80 GB
| Workload | H100 NVL | H100 SXM |
|---|---|---|
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |
Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the VRAM guide.
Verdict: which should you rent?
The H100 NVL is a PCIe card bridged in pairs with higher clocks than the H100 PCIe; the H100 SXM has full 900 GB/s NVLink across 8 GPUs and 3.35 TB/s. NVL for inference on PCIe hosts, SXM for multi-GPU training.
- Cheaper per hour: H100 SXM ($1.587 vs $1.877, −15%).
- More VRAM: H100 NVL (80 GB vs 80 GB).
- More FP16 throughput: H100 SXM (about 1.2×).
- Best value per TFLOP-hour: H100 SXM.
- Best value per GB of VRAM: H100 SXM.
- Multi-GPU: H100 NVL with NVLink · H100 SXM with NVLink.
H100 NVL vs H100 SXM: FAQ
Deeper reading: H100 vs H200 vs B200, RTX 4090 vs RTX 5090, how cloud GPU pricing works.
Is the H100 SXM faster than the H100 NVL?
On dense FP16 tensor throughput the H100 SXM leads by about 1.2× (990 vs 835 TFLOPS). Memory bandwidth matters as much for inference: H100 NVL 3,900 GB/s vs H100 SXM 3,350 GB/s.
Which is cheaper to rent, the H100 NVL or the H100 SXM?
The H100 SXM: $1.587/hr on-demand versus $1.877/hr — 15% less. Interruptible rates are $0.938 (H100 NVL) and $0.793 (H100 SXM). Per TFLOP-hour the better value is the H100 SXM.
Which has more VRAM and what does that change?
The H100 NVL has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.
Can I rent both on PowerGPU right now?
Yes — 15 × H100 NVL and 69 × H100 SXM are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.