RTX A6000 Price, 48GB VRAM & Local LLMs
Price alert
We'll email you when the eBay (US) price reaches your target.
NVIDIA's RTX A6000 (2020) offers 48GB of ECC VRAM and runs models up to 35B at 8-bit, but its modest bandwidth favors MoE models over dense ones. It takes a CPU 8-pin power cable, not the usual GPU 8-pin.
eBay
RTX A6000 price history
eBay
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
RTX A6000 VRAM & full specs
- Memory48 GB · GDDR6 ECC
- Bandwidth768 GB/s
- FP16 / BF16154.9 TFLOPS
- INT8309.7 TOPS
- Bus width
- 384-bit
- Architecture
- Ampere (GA102)
- Cores
- 10,752
- Tensor cores
- 336
- Ecosystem
- CUDA
- Interface
- PCIe 4.0 x16
- NVLink
- 112.5 GB/s
- Power
- 300 W
- Power connector
- 8-pin EPS
- Slots
- 2
- Length
- 267 mm
- Type
- Workstation
- Launch date
- Oct 5, 2020
Best RTX A6000 deals
Powered by VRAMGlass real-time price tracking — handpicking the best-value hardware deals across the web
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
UsedNVIDIA RTX A6000 48GB GDDR6 Graphics Card - Mounting Bracket BentTop Rated$4,399Buy on eBay
UsedNVIDIA Quadro RTX A6000 48GB RAM - Graphics Card - Good - Bent FrameTop Rated$4,400Buy on eBay
UsedLenovo RTX A6000 48GB GDDR6 Graphics Card PG133CTop Rated$4,400Buy on eBay
UsedNVIDIA Quadro RTX A6000 48GB - Graphics Card - Very GoodTop Rated$4,440Buy on eBay
UsedNVIDIA RTX A6000 48GB GDDR6 Graphics Card - Good ConditionTop Rated$4,499Buy on eBay
UsedNVIDIA RTX A6000 48GB GDDR6 Quadro Graphics CardTop Rated$4,600Buy on eBay
As an Amazon Associate I earn from qualifying purchases. The price you pay stays the same.
What LLMs can a single RTX A6000 run?
One A6000 is best with 30B-class MoE at 4-bit, about 125 tokens/s. The dense Qwen3.8 27B manages about 20 at 8-bit and low 30s at 4-bit.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 12B12B | 4.7 GB | Runs | 102.3 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 129.8 | 256K |
 Qwen3.6 27B27.8B | 11.8 GB | Runs | 45.7 | 256K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 54.1 | 256K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 141 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 44.1 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 151.9 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 12B12B | 7.1 GB | Runs | 72.2 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 107 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 32.9 | 256K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 33.5 | 256K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 122.3 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 29.6 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 125.4 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 12B12B | 12.7 GB | Runs | 42.8 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Runs | 84 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Runs | 19.8 | 256K |
| Qwen3.8 27B27.8B | 29 GB | Runs | 19.6 | 256K |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Runs | 95.5 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Runs | 17.2 | 202K |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Runs | 99.3 | 256K |
RunsFits fully in VRAM; speed is estimated from VRAM bandwidth.
Needs RAM offloadMoE only: expert weights sit in system RAM, so speed is estimated from 70 GB/s RAM bandwidth.
Too bigWon’t fit on one card at this quantization, even with RAM offload.
Full calculation rules are on the Methodology page.
What LLMs can multiple RTX A6000 cards run?
Two A6000s (96GB) mostly run MoE, gpt-oss 120B at about 97 tokens/s; four run DeepSeek V4 Flash at 4-bit. NVLink works, but only in pairs.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 2 cards | $9,950 | 96 GB | 100.6 | 600 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 4 cards | $19,900 | 192 GB | 80.8 | 1,200 W |
 Hy3295B (21B active) | 107.2 GB | 4 cards | $19,900 | 192 GB | 37.6 | 1,200 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 4 cards | $19,900 | 192 GB | 65 | 1,200 W |
| GLM 5.3744B (40B active) | 253.9 GB | 8 cards | $39,800 | 384 GB | 33 | 2,400 W |
| Hy4 preview770B (49B active) | 307.2 GBest. | 8 cards | $39,800 | 384 GB | 24.8 | 2,400 W |
 Muse Spark Open816B (42B active) | 311.2 GB | 8 cards | $39,800 | 384 GB | 29.4 | 2,400 W |
 Kimi K2.7 Code1,000B (32B active) | 339.5 GB | 8 cards | $39,800 | 384 GB | 39.9 | 2,400 W |
| Kimi K2.61,000B (32B active) | 340.5 GB | 8 cards | $39,800 | 384 GB | 39.8 | 2,400 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 4 cards | $19,900 | 192 GB | 84.6 | 1,200 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 4 cards | $19,900 | 192 GB | 59.1 | 1,200 W |
| Hy3295B (21B active) | 182.2 GB | 8 cards | $39,800 | 384 GB | 28 | 2,400 W |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | 8 cards | $39,800 | 384 GB | 41.5 | 2,400 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 4 cards | $19,900 | 192 GB | 57.3 | 1,200 W |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | 8 cards | $39,800 | 384 GB | 61.5 | 2,400 W |
| Hy3295B (21B active) | 317.7 GB | 8 cards | $39,800 | 384 GB | 19.2 | 2,400 W |
| GLM 5.3 Flash320B (18B active) | 341 GB | 8 cards | $39,800 | 384 GB | 26.6 | 2,400 W |
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- You want 48GB plus the full CUDA ecosystem
Qwen3.8 27B at 8-bit (29 GB) fits on one card, at about 20 tokens/s
- Mainly running MoE models
The bandwidth shortfall matters less for them, and Qwen3.6 35B A3B at 4-bit runs at about 125 tokens/s, nearly four times the dense Qwen3.8 27B
- Planning to pair two cards
It supports an NVLink bridge, for 96GB across the two
- System power and size matter to you
At 300W and two slots, it is easier to fit than two 3090s
Not a good fit
- Generation speed is your priority
Qwen3.8 27B at 4-bit runs at about 34 tokens/s, slower than both the 3090 and the RTX 6000 Ada
- You need FP8 or FP4
This generation has no FP8 compute, so FP8 models only save VRAM and don't run any faster
Before you buy
The power connector is CPU 8-pin
The A6000's power connector is a CPU 8-pin (EPS-12V), not the PCIe 8-pin common on graphics cards, and the two plugs are not interchangeable. NVIDIA's power guide allows an adapter that combines “two PCIe 8-pin into one CPU 8-pin”. Check your cables before changing cases or power supplies.
Buy the NVLink bridge to match your slot spacing
Official bridges come in 2-slot and 3-slot spacings, and each connects only two cards. Before buying one, measure the spacing between the two x16 slots on your motherboard; with the wrong spacing, the bridge won't fit.
Which quantization format to use
This generation has no FP8 compute, so with FP8 models vLLM only stores the weights in FP8 and still computes in 16-bit, saving VRAM without speeding anything up. On this card, prefer GGUF at 4-bit to 8-bit, or 4-bit formats such as AWQ and GPTQ.
Same chip as the 3090
The A6000 uses the full GA102, and the 3090 a cut-down version of the same chip. But the 3090 pairs it with faster GDDR6X memory, so it actually runs models faster. The A6000's 48GB has ECC, which automatically corrects single-bit errors in memory.
Compared with similar GPUs
RTX A6000 vs RTX 6000 Ada, GeForce RTX 3090, and GeForce RTX 4090 48GB: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | RTX A6000 | RTX 6000 Ada | GeForce RTX 3090 | GeForce RTX 4090 48GB |
|---|---|---|---|---|
| Memory | 48 GB | 48 GBSame | 24 GB50% less | 48 GBSame |
| Bandwidth | 768 GB/s | 960 GB/s25% higher | 936 GB/s22% higher | 1,008 GB/s31% higher |
| BF16 / FP16 compute | 155 TFLOPS | 364 TFLOPS135% higher | 142 TFLOPS8% lower | 330 TFLOPS113% higher |
| INT8 compute | 310 TOPS | 729 TOPS135% higher | 285 TOPS8% lower | 661 TOPS113% higher |
| Qwen3.8 27B speed (2-bit) | 54.1 t/s | 66.3 t/s23% faster | 64.8 t/s20% faster | 69.3 t/s28% faster |
| Largest dense model (4-bit) | 79B | 79BSame | 37B53% less | 79BSame |
| Power | 300 W | 300 WSame | 350 W17% higher | 450 W50% higher |
| GPU interconnect | NVLink 112.5 GB/s | Not supported | NVLink 112.5 GB/s | Not supported |
| Interface | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Released | 2020-10 | 2022-09 | 2020-09 | 2022-10 |
| eBay median |
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
RTX A6000 rental price per hour
On-demand rental of the RTX A6000 starts at $0.33 per hour (RunPod), about $80/month at 8 hours a day. Buying one (eBay median asking price $4,975) pays off in about 12.3 years, 6.2 years or 2.1 years at 4, 8 or 24 hours a day.
| Option | Price | 4 h/day | 8 h/day | 24 h/day | View |
|---|---|---|---|---|---|
 RunPodRentUpdated 29 min. ago | $0.33/hr | $40/mo | $80/mo | $241/mo | Rent |
 Vast.aiRentUpdated 29 min. ago | $0.43/hr | $52/mo | $105/mo | $314/mo | Rent |
 eBayUpdated 3 hr. ago | $4,975 | Payback ~12.3 yrsPower $7/mo | Payback ~6.2 yrsPower $13/mo | Payback ~2.1 yrsPower $39/mo | Buy |
Break-even uses the US residential average ($0.18/kWh) and 300 W rated power, excluding resale value and the rest of the build.
Rental links are referral links: we may receive a commission or credit if you sign up through them. As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
Amazon.jp
Xianyu