GeForce RTX 5090 Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
NVIDIA's RTX 5090 (2025) is a 32GB flagship gaming card that fits every model up to 35B at 4-bit, the fastest consumer card for local LLMs. It draws 575W.
eBay
GeForce RTX 5090 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 5090 specifications
- Memory32 GB · GDDR7
- Bandwidth1,792 GB/s
- FP16 / BF16419 TFLOPS
- FP8 / INT8838 / 838 TOPS
- Bus width
- 512-bit
- Architecture
- Blackwell (GB202)
- Cores
- 21,760
- Tensor cores
- 680
- Ecosystem
- CUDA
- Interface
- PCIe 5.0 x16
- NVLink
- Not supported
- Power
- 575 W
- Power connector
- 16-pin 12V-2x6
- Slots
- 2
- Length
- 304 mm
- Type
- Consumer
- Launch date
- Jan 30, 2025
- MSRP
- $1,999
Best GeForce RTX 5090 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.
UsedMSI GeForce RTX 5090 32GB GDDR7 VENTUS 3X OC Graphics CardTop Rated$7,000+$70 shippingBuy on eBay
UsedNVIDIA GeForce RTX 5090 32GB Founders Edition Graphics Card$7,196Buy on eBay
UsedNEW ASUS ProArt GeForce RTX 5090 32GB GDDR7 OC Edition Graphics Card$7,199Buy on eBay
NewGigabyte GeForce RTX 5090 OC 32GB GPU Graphics Card - New No Box (From prebuilt)$7,229Buy on eBay
New🔥ASUS RTX 5090 TUF 32GB NVIDIA GeForce Graphics Card - New Sealed - Free Ship!$7,300Buy on eBay
New🔥MSI RTX 5090 Gaming Trio OC 32GB NVIDIA GeForce Graphics Card - New Sealed!$7,400Buy on eBay
What LLMs can a single GeForce RTX 5090 run?
One RTX 5090 fits every model up to 35B at 4-bit, Qwen3.8 27B at about 74 tokens/s. At 8-bit it tops out at 27B-class.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 236.9 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 223.2 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 200.4 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 167 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 98.2 | 256K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 114.6 | 256K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 174.7 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 95 | 246K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 181.6 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 196.3 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 177.1 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 148.5 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 149.5 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 72.3 | 245K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 73.5 | 250K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 161.5 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 65.5 | 168K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 163.9 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 134.7 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 118.9 | 256K |
| Gemma 4 12B12B | 12.7 GB | Runs | 92.5 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Runs | 128.4 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Runs | 44.6 | 69K |
| Qwen3.8 27B27.8B | 29 GB | Runs | 44 | 63K |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Runs | 139.4 | 26K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 36.5 | 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 5090 cards run?
Two RTX 5090s (64GB) run 70B-class dense models at about 28 tokens/s; 2026's large MoE models need four. The 5090 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 | $33,188 | 128 GB | 144 | 2,300 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 4 cards | $33,188 | 128 GB | 125.1 | 2,300 W |
 Hy3295B (21B active) | 107.2 GB | 4 cards | $33,188 | 128 GB | 71 | 2,300 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 4 cards | $33,188 | 128 GB | 107.7 | 2,300 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 4 cards | $33,188 | 128 GB | 129 | 2,300 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 8 cards | $66,376 | 256 GB | 100.7 | 4,600 W |
| Hy3295B (21B active) | 182.2 GB | 8 cards | $66,376 | 256 GB | 55.7 | 4,600 W |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | 8 cards | $66,376 | 256 GB | 76.9 | 4,600 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 8 cards | $66,376 | 256 GB | 98.4 | 4,600 W |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | 8 cards | $66,376 | 256 GB | 103.5 | 4,600 W |
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Running 27B-class models at 8-bit on one card
Qwen3.8 27B (29 GB) runs at about 44 tokens/s with 63K of context
- Long context at 4-bit
Qwen3.8 27B can run up to 250K, and Gemma 4 31B still gets 168K
- The fastest speeds on 30B-class MoE models
Qwen3.6 35B A3B at 4-bit runs at about 164 tokens/s, and GLM 4.7 Flash at about 162
- You need a dual-slot card
The Founders Edition is only two slots thick, one slot thinner than the 4090 Founders Edition
Not a good fit
- Running 70B-class dense models on one card
32GB can't hold a 70B model even at 4-bit, so you need two cards
- Tight power or cooling
The card draws 575W, putting out 125W more heat than a 4090
Before you buy
The 12V-2x6 connector can still melt
der8auer tested the Founders Edition and found current unevenly distributed across the cable: a single wire carried more than 22 A, and the plug at the power supply end hit a 150°C hot spot. The cause is that the reference design doesn't monitor current on each pin separately, so when a few pins make poor contact, all the current crowds onto the others. So use the power supply maker's native 12V-2x6 cable, seat it fully, and keep an eye on the plug temperature during long full-load model runs.
Power supply and cooling
The card draws 575W, and NVIDIA recommends a 1000W system power supply. The Founders Edition is dual-slot and 304 mm long. The GPU uses liquid metal as its thermal interface, so taking the card apart to replace the thermal paste yourself is not recommended.
Compared with similar GPUs
GeForce RTX 5090 vs GeForce RTX 4090, GeForce RTX 5090 D, and RTX PRO 6000: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | GeForce RTX 5090 | GeForce RTX 4090 | GeForce RTX 5090 D | RTX PRO 6000 |
|---|---|---|---|---|
| Memory | 32 GB | 24 GB25% less | 32 GBSame | 96 GB200% more |
| Bandwidth | 1,792 GB/s | 1,008 GB/s44% lower | 1,792 GB/sSame | 1,792 GB/sSame |
| BF16 / FP16 compute | 419 TFLOPS | 330 TFLOPS21% lower | 297 TFLOPS29% lower | 500 TFLOPS19% higher |
| INT8 compute | 838 TOPS | 661 TOPS21% lower | 594 TOPS29% lower | 1,000 TOPS19% higher |
| Qwen3.8 27B speed (2-bit) | 114.6 t/s | 69.3 t/s40% slower | 114.6 t/sSame | 114.6 t/sSame |
| Largest dense model (4-bit) | 51B | 37B27% less | 51BSame | 165B224% more |
| Power | 575 W | 450 W22% lower | 575 WSame | 600 W4% higher |
| Interface | PCIe 5.0 x16 | PCIe 4.0 x16 | PCIe 5.0 x16 | PCIe 5.0 x16 |
| Released | 2025-01 | 2022-10 | 2025-01 | 2025-03 |
| 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 5090 rental price per hour
On-demand rental of the GeForce RTX 5090 starts at $0.69 per hour (RunPod), about $168/month at 8 hours a day. Buying one (eBay median $8,297) pays off in about 9.7 years, 4.8 years or 1.6 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 | $0.69/hr | $84/mo | $168/mo | $504/mo | Rent |
 Vast.aiRentUpdated 47 min. ago | $0.97/hr | $118/mo | $235/mo | $705/mo | Rent |
 eBayUpdated 6 hr. ago | $8,297 | Payback ~9.7 yrsPower $13/mo | Payback ~4.8 yrsPower $25/mo | Payback ~1.6 yrsPower $76/mo | Buy |
Break-even uses the US residential average ($0.18/kWh) and 575 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