A100 SXM vs L40S: specs, price and which to rent
GPU Rent Hub rents both cards as dedicated bare metal in three United States datacenters, paid in Bitcoin, Ethereum, USDT, USDC, Monero, Solana or Litecoin with no identity verification. The A100 SXM is $990 a month and the L40S is $520 a month. Everything below is measured, not argued.
What is the difference between the A100 SXM and the L40S?
Every figure is the one we rent against, taken from the same catalogue the dashboard prices from. Per-GPU-hour is the flat monthly price divided by 720 hours; you are never billed by the hour.
| A100 SXM | L40S | |
|---|---|---|
| Memory | 80 GB HBM2e | 48 GB GDDR6 ECC |
| Memory bandwidth | 2.04 TB/s | 864 GB/s |
| Architecture | Ampere (2020) | Ada Lovelace (2023) |
| Cores | 6,912 CUDA · 432 Tensor | 18,176 CUDA · 568 Tensor · 142 RT |
| FP8 | not supported | 1,466 TFLOPS (sparse) |
| FP16 | 624 TFLOPS (sparse) | 733 TFLOPS (sparse) |
| Interconnect | NVLink · 600 GB/s | PCIe Gen4 peer-to-peer |
| Board power | 400 W | 350 W |
| Host vCPU | 16 | 16 |
| Host RAM | 125 GB | 94 GB |
| Local NVMe | 1 TB | 500 GB |
| Weekly price | $280 | $148 |
| Monthly price | $990 | $520 |
| Per GPU-hour | $1.38 | $0.72 |
| Price per GB of VRAM | $12.38 | $10.83 |
| In stock now | 78 | 74 |
Which should you rent?
The A100 SXM carries 32 GB more memory, which is the only thing that decides whether a model loads at all. The L40S supports FP8 and the A100 SXM does not, which is unusual for the cheaper card and makes it the better inference buy. Bandwidth is 2.04 TB/s against 864 GB/s, and for single-stream token generation it is bandwidth that sets the speed, not the FLOPS figure on the spec sheet. Only the A100 SXM has NVLink, so multi-card training scales on it and struggles on the L40S. The A100 SXM costs 90% more per month, $990 against $520, or $1.38 against $0.72 per GPU-hour.
Rent the A100 SXM when
The cheapest 80 GB card with NVLink. 80 GB of HBM2e under $1,000 a month, the value pick for multi-week runs.
$990/mo · $280/wk · specifications and stock
Rent the L40S when
Serving 7B–34B models in production. 48 GB with FP8 support and RT cores, the best inference card per dollar in our fleet.
$520/mo · $148/wk · specifications and stock
Where each card is in stock
A dash means the card is on the waitlist in that region. The price is identical in all three, so choose on latency and stock, nothing else.
| Region | A100 SXM | L40S |
|---|---|---|
| DFW-1Dallas, TX | 40 | 26 |
| IAD-1Ashburn, VA | 22 | 18 |
| PDX-1Hillsboro, OR | 16 | 30 |
A100 SXM against L40S
Is the A100 SXM better than the L40S?
What is the price difference between the A100 SXM and the L40S?
How much VRAM does the A100 SXM have compared to the L40S?
Can I rent either card without KYC?
Which one is available today?
Can I put several A100 SXM cards in one machine?
Rent either one by the week or the month
No identity check, no email address, no card. Fund a balance in any of seven coins and deploy in about ninety seconds.