Compare · prices checked 2026-09-03
RTX 5090 vs L40S: specs, price per hour, which to rent
NVIDIA RTX 5090 (32 GB, $0.318/hr) against NVIDIA L40S (48 GB, $0.466/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.
RTX 5090 vs L40S specifications
Public NVIDIA figures (dense, non-sparsity). The last column is RTX 5090 relative to L40S.
| Spec | RTX 5090 | L40S | Difference |
|---|---|---|---|
| Architecture | Blackwell (2025) | Ada Lovelace (2023) | — |
| VRAM | 32 GB GDDR7 | 48 GB GDDR6 | −33% |
| Memory bandwidth | 1,792 GB/s | 864 GB/s | +107% |
| FP16 tensor (dense) | 419 TFLOPS | 362 TFLOPS | +16% |
| FP32 | 104.8 TFLOPS | 91.6 TFLOPS | +14% |
| CUDA cores | 21,760 | 18,176 | +20% |
| TDP | 575 W | 350 W | +64% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 295 | 255 | +16% |
| Max GPUs per machine | 8× | 8× | — |
RTX 5090 vs L40S price per hour
Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.
| Rate | RTX 5090 | L40S | Cheaper |
|---|---|---|---|
| On-demand, per GPU-hour | $0.318 | $0.466 | RTX 5090 (−32%) |
| Interruptible, per GPU-hour | $0.159 | $0.233 | RTX 5090 |
| Reserved (3 mo), per GPU-hour | $0.206 | $0.302 | RTX 5090 |
| On-demand, per month | $232 | $340 | RTX 5090 |
| Market median (reference) | $0.45 | $0.67 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.76 | $1.29 | RTX 5090 (better value) |
| $ per GB of VRAM per hour | $0.0099 | $0.0097 | L40S (better value) |
Try a full month with storage and bandwidth in the GPU cost calculator.
What fits in VRAM: 32 GB vs 48 GB
| Workload | RTX 5090 | L40S |
|---|---|---|
| Largest LLM in FP16, one card | ~13B | ~14B |
| Largest LLM at 4-bit, one card | ~49B | ~72B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | 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 RTX 5090 (32 GB GDDR7, 1.79 TB/s) is faster and cheaper; the L40S (48 GB, passive, ECC) fits bigger models and is built for continuous datacenter duty. Choose by VRAM and uptime needs rather than raw speed.
- Cheaper per hour: RTX 5090 ($0.318 vs $0.466, −32%).
- More VRAM: L40S (48 GB vs 32 GB).
- More FP16 throughput: RTX 5090 (about 1.2×).
- Best value per TFLOP-hour: RTX 5090.
- Best value per GB of VRAM: L40S.
- Multi-GPU: RTX 5090 over PCIe · L40S over PCIe.
Related comparisons
- RTX 5090 vs RTX 4090 $0.318 vs $0.262 per hour Compare
- RTX 4090 vs L40S $0.262 vs $0.466 per hour Compare
- L40S vs A100 PCIE $0.466 vs $0.662 per hour Compare
- L40S vs H100 PCIE $0.466 vs $2.147 per hour Compare
- RTX 5090 vs A100 SXM4 $0.318 vs $0.583 per hour Compare
- RTX PRO 6000 WS vs RTX 5090 $1.097 vs $0.318 per hour Compare
RTX 5090 vs L40S: FAQ
Deeper reading: H100 vs H200 vs B200, RTX 4090 vs RTX 5090, how cloud GPU pricing works.
Is the RTX 5090 faster than the L40S?
On dense FP16 tensor throughput the RTX 5090 leads by about 1.2× (419 vs 362 TFLOPS). Memory bandwidth matters as much for inference: RTX 5090 1,792 GB/s vs L40S 864 GB/s.
Which is cheaper to rent, the RTX 5090 or the L40S?
The RTX 5090: $0.318/hr on-demand versus $0.466/hr — 32% less. Interruptible rates are $0.159 (RTX 5090) and $0.233 (L40S). Per TFLOP-hour the better value is the RTX 5090.
Which has more VRAM and what does that change?
The L40S has 48 GB versus 32 GB. In LLM terms that is roughly a 14B FP16 model (or ~72B in 4-bit) on one card against 13B FP16 (~49B 4-bit). If the model does not fit, speed is irrelevant.
Can I rent both on PowerGPU right now?
Yes — 516 × RTX 5090 and 59 × L40S are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.