GeForce RTX 4090 Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
The RTX 4090 is NVIDIA's 2022 24GB flagship gaming card, running Qwen3.8 27B at 4-bit at about 43 tokens/s. Some used ones lack a chip; test under load.
eBay
GeForce RTX 4090 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.
Full GeForce RTX 4090 specifications
- Memory24 GB · GDDR6X
- Bandwidth1,008 GB/s
- FP16 / BF16330.3 TFLOPS
- FP8 / INT8660.6 / 660.6 TOPS
- Bus width
- 384-bit
- Architecture
- Ada Lovelace (AD102)
- Cores
- 16,384
- Tensor cores
- 512
- Ecosystem
- CUDA
- Interface
- PCIe 4.0 x16
- NVLink
- Not supported
- Power
- 450 W
- Power connector
- 16-pin 12VHPWR
- Slots
- 3
- Length
- 304 mm
- Type
- Consumer
- Launch date
- Oct 12, 2022
- MSRP
- $1,599
Best GeForce RTX 4090 deals on eBay
Hand-picked deals from trusted eBay sellers. Shop with confidence.
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.
UsedZOTAC Gaming GEFORCE RTX 4090 AMP EXTREME AIRO 24GB GDDR6X Graphics Card *READ*$3,200+$20 shippingBuy on eBay
UsedGALAX GeForce RTX 4090 SG GDDR6X 24GB RGB 384 Bit Graphics Card GPU Complete$3,200+$39 shippingBuy on eBay
UsedGIGABYTE Gaming OC GeForce RTX 4090 24GB - Graphics Card - Display Port IssueTop Rated$3,300Buy on eBay
UsedASUS ROG Strix RTX 4090 BTF OC 24GB GDDR6X Video Graphics Card GPUTop Rated$3,300Buy on eBay
UsedNVIDIA GeForce RTX 4090 24GB GDDR6X Founders EditionTop Rated$3,399Buy on eBay
UsedMSI GeForce RTX 4090 Gaming X Trio 24GB GDDR6X GPU *Used - Excellent*$3,399+ shippingBuy on eBay
What LLMs can a single GeForce RTX 4090 run?
One RTX 4090 fits up to 35B-class at 4-bit, 27B at about 43 tokens/s, with little room left for context. 8-bit fits only 12B-class.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 155.8 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 145.4 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 128.5 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 143.1 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 58.8 | 192K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 69.3 | 221K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 153.4 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 56.7 | 143K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 163 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 125.5 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 111.8 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 91.9 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 121.4 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 42.6 | 117K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 43.3 | 122K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 136.1 | 115K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 38.4 | 66K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 139 | 123K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 82.6 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 72.1 | 256K |
| Gemma 4 12B12B | 12.7 GB | Runs | 55.1 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 26.4 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 29 GB | Too big | — | — |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Needs RAM offload | 22.9 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 35 | 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 GeForce RTX 4090 cards run?
Two 4090s suit 48B-class MoE; four (96GB) are needed for Qwen3.8 Flash Next at 2-bit, and DeepSeek V4 Flash takes eight. The 4090 has no NVLink.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 4 cards | $15,032 | 96 GB | 115.1 | 1,800 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 8 cards | $30,064 | 192 GB | 94.8 | 3,600 W |
 Hy3295B (21B active) | 107.2 GB | 8 cards | $30,064 | 192 GB | 46.8 | 3,600 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 8 cards | $30,064 | 192 GB | 77.8 | 3,600 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 8 cards | $30,064 | 192 GB | 98.8 | 3,600 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 8 cards | $30,064 | 192 GB | 71.4 | 3,600 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 8 cards | $30,064 | 192 GB | 69.4 | 3,600 W |
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Running 27B to 31B dense models on one card
Qwen3.8 27B at 4-bit runs at about 43 tokens/s with 122K of context, and Gemma 4 31B at about 38
- Fitting a 30B-class MoE model entirely on the card
Qwen3.6 35B A3B at 4-bit runs at about 139 tokens/s, and GLM 4.7 Flash at about 136
- Regularly feeding it long documents or batched requests
2.3 times the compute of a 3090, plus FP8 support
Not a good fit
- Single-user chat only
Generation speed depends on bandwidth, which is only 8% higher than the 3090's, so the same dense model runs only about 7% faster
- Running 27B models at 8-bit
Qwen3.8 27B needs 29 GB at 8-bit, so you would need the 32GB 5090
- Using NVLink across two cards
The 40 series dropped NVLink, so cards can only talk over PCIe
Before you buy
Watch out for cards stripped of their GPU
In China, large numbers of 4090s have had their GPU and memory removed to build dual-slot blower cards or 48GB cards. As a result, the used market has seen 4090s that are just empty boards, or even carry fake chips, and repair shops outside China have reported them many times. When inspecting a card, always power it on and run it at full load, and use nvidia-smi to confirm the model and the 24GB of VRAM. Don't go by photos alone.
Seat the 12VHPWR plug all the way
In late 2022 NVIDIA received about 50 reports of melted connectors, and its investigation concluded that the main cause was plugs not fully inserted. Gamers Nexus's testing found the same thing: if the plug isn't fully seated, or the cable is pulled at an angle after it is plugged in, the plug can back out a few millimeters and heat up. So push it in until there is no gap on any side, avoid sharp bends right at the plug, and don't keep unplugging it to check, which actually makes it more likely to work loose. On a used card, first check the connector for yellowing or scorch marks.
Power supply and size
For the 450W 4090, NVIDIA recommends an 850W power supply. The Founders Edition is three slots thick and 304 mm long, and many partner cards are thicker, taking up as many as four slots. Measure your case and your motherboard's slot spacing before you build.
China also has a 4090 D
The 4090 D, launched in China in late 2023, is a cut-down version: about 10% fewer cores and 425W of power, with the same 24GB of VRAM and the same bandwidth. Generation speed is basically the same as the 4090, while long prompts are processed slightly slower. Check which version a listing is for.
Compared with similar GPUs
GeForce RTX 4090 vs GeForce RTX 3090, GeForce RTX 5090, and GeForce RTX 4090 48GB: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | GeForce RTX 4090 | GeForce RTX 3090 | GeForce RTX 5090 | GeForce RTX 4090 48GB |
|---|---|---|---|---|
| Memory | 24 GB | 24 GBSame | 32 GB33% more | 48 GB100% more |
| Bandwidth | 1,008 GB/s | 936 GB/s7% lower | 1,792 GB/s78% higher | 1,008 GB/sSame |
| BF16 / FP16 compute | 330 TFLOPS | 142 TFLOPS57% lower | 419 TFLOPS27% higher | 330 TFLOPSSame |
| INT8 compute | 661 TOPS | 285 TOPS57% lower | 838 TOPS27% higher | 661 TOPSSame |
| Qwen3.8 27B speed (2-bit) | 69.3 t/s | 64.8 t/s6% slower | 114.6 t/s65% faster | 69.3 t/sSame |
| Largest dense model (4-bit) | 37B | 37BSame | 51B38% more | 79B114% more |
| Power | 450 W | 350 W22% lower | 575 W28% higher | 450 WSame |
| GPU interconnect | Not supported | NVLink 112.5 GB/s | Not supported | Not supported |
| Interface | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 5.0 x16 | PCIe 4.0 x16 |
| Released | 2022-10 | 2020-09 | 2025-01 | 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.
GeForce RTX 4090 rental price per hour
On-demand rental of the GeForce RTX 4090 starts at $0.34 per hour (RunPod), about $83/month at 8 hours a day. Buying one (eBay median $3,758) pays off in about 9.9 years, 5 years or 1.7 years at 4, 8 or 24 hours a day.
| Option | Price | 4 h/day | 8 h/day | 24 h/day | View |
|---|---|---|---|---|---|
 RunPodRentUpdated 47 min. ago | $0.34/hr | $41/mo | $83/mo | $248/mo | Rent |
 Vast.aiRentUpdated 47 min. ago | $0.62/hr | $75/mo | $151/mo | $452/mo | Rent |
 eBayUpdated 6 hr. ago | $3,758 | Payback ~9.9 yrsPower $10/mo | Payback ~5 yrsPower $20/mo | Payback ~1.7 yrsPower $59/mo | Buy |
Break-even uses the US residential average ($0.18/kWh) and 450 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