RTX A6000 vs RTX A5000
The RTX A5000 rents for 52% less per hour today — $0.27 against $0.56 — with 24 GB against 48 GB of memory.
Prices collected 2026-09-18 · 8 providers price both
Index median per GPU-hour, on-demand, per snapshot.
Side by side
| RTX A6000 | RTX A5000 | |
|---|---|---|
| Median rental, per GPU-hour | $0.56 | $0.27 |
| Cheapest listing | $0.33 | $0.16 |
| Per month at 720 h | $403 | $194 |
| Providers pricing it | 16 | 8 |
| Memory | 48 GB GDDR6 | 24 GB GDDR6 |
| Memory bandwidth | 768 GB/s | 768 GB/s |
| GB of memory per $/hr | 85.7 | 88.9 |
| Architecture | Ampere | Ampere |
| Class | Workstation | Workstation |
| Released | 2020-10 | 2021-04 |
Same provider, both GPUs
The cleanest comparison: 8 providers price both cards, so the difference is the GPU and nothing else. The RTX A6000 is cheaper at 1 of them.
| Provider | RTX A6000 | RTX A5000 | RTX A5000 premium |
|---|---|---|---|
 RunPod Community Cloud | $0.33 | $0.16 | -52% |
 GPU.ai | $0.44 | $0.27 | -38% |
 io.net | $0.49 | $0.19 | -62% |
 RunPod Secure Cloud | $0.53 | $0.27 | -49% |
 Runcrate | $0.55 | $1.56 | +184% |
| $0.57 | $0.44 | -23% | |
 Vast.ai | $0.60 | $0.22 | -64% |
 Paperspace (DigitalOcean) | $1.89 | $1.24 | -34% |
What each one runs, and what a million tokens costs
Weights + KV cache at 8K context + overhead against 90% of each card's memory. Where a model needs more than one card the cell says how many and prices the set at today's median. $/M tokens is the hourly median divided by estimated single-stream output, from memory bandwidth — an estimate for ranking, not a benchmark.
| Model | Bits | RTX A6000 | RTX A5000 |
|---|---|---|---|
| Kimi K3 | 4-bit | 37× GPU$20.72/hr | 74× GPU$19.98/hr |
| Kimi K3 | 16-bit | 130× GPU$72.80/hr | 260× GPU$70.20/hr |
| DeepSeek V4.1-Flash | 4-bit | 8× GPU$4.48/hr | 15× GPU$4.05/hr |
| DeepSeek V4.1-Flash | 16-bit | 26× GPU$14.56/hr | 52× GPU$14.04/hr |
| Llama 4 Maverick | 4-bit | 6× GPU$3.36/hr | 11× GPU$2.97/hr |
| Llama 4 Maverick | 16-bit | 19× GPU$10.64/hr | 38× GPU$10.26/hr |
| gpt-oss-120b | 4-bit | 2× GPU$1.12/hr | 4× GPU$1.08/hr |
| gpt-oss-120b | 16-bit | 6× GPU$3.36/hr | 11× GPU$2.97/hr |
| Llama 3.1 70B | 4-bit | 2× GPU$1.12/hr | 3× GPU$0.81/hr |
| Llama 3.1 70B | 16-bit | 4× GPU$2.24/hr | 7× GPU$1.89/hr |
| Gemma 4 31B | 4-bit | ✓ 21.2 GB~35 tok/s · $4.44/M | ✓ 21.2 GB~35 tok/s · $2.14/M |
| Gemma 4 31B | 16-bit | 2× GPU$1.12/hr | 4× GPU$1.08/hr |
| Qwen3 Coder 30B-A3B | 4-bit | ✓ 19 GB~331 tok/s · $0.47/M | ✓ 19 GB~331 tok/s · $0.23/M |
| Qwen3 Coder 30B-A3B | 16-bit | 2× GPU$1.12/hr | 3× GPU$0.81/hr |
| Qwen3.8 27B | 4-bit | ✓ 18.7 GB~40 tok/s · $3.89/M | ✓ 18.7 GB~40 tok/s · $1.88/M |
| Qwen3.8 27B | 16-bit | 2× GPU$1.12/hr | 3× GPU$0.81/hr |
| gpt-oss-20b | 4-bit | ✓ 13.7 GB~303 tok/s · $0.51/M | ✓ 13.7 GB~303 tok/s · $0.25/M |
| gpt-oss-20b | 16-bit | 2× GPU$1.12/hr | 3× GPU$0.81/hr |
| Llama 3.1 8B | 4-bit | ✓ 6.9 GB~137 tok/s · $1.14/M | ✓ 6.9 GB~137 tok/s · $0.55/M |
| Llama 3.1 8B | 16-bit | ✓ 18.4 GB~38 tok/s · $4.09/M | ✓ 18.4 GB~38 tok/s · $1.97/M |
Buy or rent?
Hours of renting at today's median that add up to the card's approximate list price. List prices are typical quoted figures, not tracked, and do not include the server, power or the person to run it — so the break-even is a floor, not a verdict.
Questions
Is the RTX A6000 or the RTX A5000 cheaper to rent?
On 2026-09-18 the RTX A6000 has a market median of $0.56 per GPU-hour across 16 providers and the RTX A5000 $0.27 across 8, so the RTX A5000 is 52% cheaper per hour. Among the 8 providers that price both, the RTX A6000 is cheaper at 1 and the RTX A5000 at 7. Per hour is only half the answer — the table below prices a whole model on each.
Which should I rent for a 70B model?
At 4-bit, Llama 3.1 70B needs about 44.5 GB. On the RTX A6000 that is 2 cards at $1.12/hr; on the RTX A5000 3 cards at $0.81/hr.
What is the real difference between them?
48 GB GDDR6 at 768 GB/s (Ampere) against 24 GB GDDR6 at 768 GB/s (Ampere). Memory decides what fits; bandwidth decides how fast it generates, because each token streams the active weights once. Compute matters for prompt processing and training, not for how fast text appears.