RTX PRO 6000 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 PRO 6000 Blackwell (2025) is a 96GB workstation flagship running a 176B MoE at 2-bit at about 149 tokens/s. It comes in three versions.
eBay
RTX PRO 6000 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 RTX PRO 6000 specifications
- Memory96 GB · GDDR7 ECC
- Bandwidth1,792 GB/s
- FP16 / BF16500 TFLOPS
- FP8 / INT81,000 / 1,000 TOPS
- Bus width
- 512-bit
- Architecture
- Blackwell (GB202)
- Cores
- 24,064
- Tensor cores
- 752
- Ecosystem
- CUDA
- Interface
- PCIe 5.0 x16
- NVLink
- Not supported
- Power
- 600 W
- Power connector
- 16-pin 12V-2x6
- Slots
- 2
- Length
- 304 mm
- Type
- Workstation
- Launch date
- Mar 18, 2025
Best RTX PRO 6000 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.
NewNVIDIA RTX PRO 6000 Blackwell 96GB GDDR7 Workstation Edition BULK OEM$17,500Buy on eBay
NewNVIDIA RTX PRO 6000 Blackwell 96GB GDDR7 ECC AI Workstation GPU NEWTop Rated$17,500Buy on eBay
New1 x NVIDIA RTX PRO 6000 Blackwell 96GB GDDR7 Workstation Edition (600W) WARRANTY$22,199Buy on eBay
UsedPNY NVIDIA RTX PRO 6000 Blackwell Workstation 96GB GDDR7 VCNRTXPRO6000B-PB$22,900Buy on eBay
NewPNY NVIDIA RTX PRO 6000 Blackwell Workstation 96GB GDDR7 VCNRTXPRO6000B-PB$23,900Buy on eBay
UsedNVIDIA RTX PRO 6000 Blackwell Workstation Edition Graphics Card - 600 W - 96 GB$24,995Buy on eBay
What LLMs can a single RTX PRO 6000 run?
One PRO 6000 runs Qwen3.8 Flash Next and DeepSeek V4 Flash at 2-bit; at 4-bit they spill into system RAM, near 20 tokens/s. Up to 35B, 8-bit is easy.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 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 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 181.6 | 256K |
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | Runs | 148.6 | 256K |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | Runs | 125.9 | 918K |
 Hy3295B (21B active) | 107.2 GB | Needs RAM offload | 10.8 | 256K |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | Needs RAM offload | 15.6 | 1M |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| 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 | 256K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 73.5 | 256K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 161.5 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 65.5 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 163.9 | 256K |
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | Needs RAM offload | 21 | 256K |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | Needs RAM offload | 13.1 | 1M |
| Hy3295B (21B active) | 182.2 GB | Needs RAM offload | 6.5 | 256K |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | Needs RAM offload | 8.8 | 1M |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| 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 | 256K |
| Qwen3.8 27B27.8B | 29 GB | Runs | 44 | 256K |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Runs | 139.4 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Runs | 39 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Runs | 142.8 | 256K |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | Needs RAM offload | 13 | 256K |
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | Needs RAM offload | 12.6 | 1M |
| Hy3295B (21B active) | 317.7 GB | Needs RAM offload | 3.8 | 256K |
| GLM 5.3 Flash320B (18B active) | 341 GB | Needs RAM offload | 5.2 | 1M |
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 PRO 6000 cards run?
Two PRO 6000s (192GB) run DeepSeek V4 Flash at 4-bit at about 100 tokens/s; from four, trillion-parameter models fit at 2-bit. No NVLink.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 1 card | $22,350 | 96 GB | 144 | 600 W |
| DeepSeek V4 Flash284B (13B active) | 96.8 GB | 2 cards | $44,700 | 192 GB | 125.1 | 1,200 W |
| Hy3295B (21B active) | 107.2 GB | 2 cards | $44,700 | 192 GB | 71 | 1,200 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 2 cards | $44,700 | 192 GB | 107.7 | 1,200 W |
| GLM 5.3744B (40B active) | 253.9 GB | 4 cards | $89,400 | 384 GB | 63.8 | 2,400 W |
| Hy4 preview770B (49B active) | 307.2 GBest. | 4 cards | $89,400 | 384 GB | 50 | 2,400 W |
 Muse Spark Open816B (42B active) | 311.2 GB | 4 cards | $89,400 | 384 GB | 57.9 | 2,400 W |
 Kimi K2.7 Code1,000B (32B active) | 339.5 GB | 4 cards | $89,400 | 384 GB | 74.5 | 2,400 W |
| Kimi K2.61,000B (32B active) | 340.5 GB | 4 cards | $89,400 | 384 GB | 74.3 | 2,400 W |
| DeepSeek V4 Pro1,600B (49B active) | 638.4 GBest. | 8 cards | $178,800 | 768 GB | 52.6 | 4,800 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 2 cards | $44,700 | 192 GB | 129 | 1,200 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 2 cards | $44,700 | 192 GB | 100.7 | 1,200 W |
| Hy3295B (21B active) | 182.2 GB | 4 cards | $89,400 | 384 GB | 55.7 | 2,400 W |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | 4 cards | $89,400 | 384 GB | 76.9 | 2,400 W |
| GLM 5.3744B (40B active) | 467.3 GB | 8 cards | $178,800 | 768 GB | 41.6 | 4,800 W |
| Hy4 preview770B (49B active) | 467.3 GB | 8 cards | $178,800 | 768 GB | 36.5 | 4,800 W |
| Muse Spark Open816B (42B active) | 505.5 GB | 8 cards | $178,800 | 768 GB | 40.7 | 4,800 W |
| DeepSeek V4.1 Flash763B (8 / 16B active) | 510.3 GB | 4 cardsEngram in RAM | $89,400 | 384 GB | 68.1 | 2,400 W |
| Kimi K2.61,000B (32B active) | 583.7 GB | 8 cards | $178,800 | 768 GB | 52.3 | 4,800 W |
| Kimi K2.7 Code1,000B (32B active) | 583.7 GB | 8 cards | $178,800 | 768 GB | 52.3 | 4,800 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 2 cards | $44,700 | 192 GB | 98.4 | 1,200 W |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | 4 cards | $89,400 | 384 GB | 103.5 | 2,400 W |
| Hy3295B (21B active) | 317.7 GB | 4 cards | $89,400 | 384 GB | 40 | 2,400 W |
| GLM 5.3 Flash320B (18B active) | 341 GB | 4 cards | $89,400 | 384 GB | 53.2 | 2,400 W |
| Kimi K2.61,000B (32B active) | 594.5 GB | 8 cards | $178,800 | 768 GB | 51.6 | 4,800 W |
| Kimi K2.7 Code1,000B (32B active) | 594.5 GB | 8 cards | $178,800 | 768 GB | 51.6 | 4,800 W |
Rows marked “Engram in RAM” need about 203 GB of extra system RAM. Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Running 2026's large MoE models on one card
Qwen3.8 Flash Next at 2-bit runs at about 149 tokens/s, and DeepSeek V4 Flash at about 126 with up to about 918K of context
- Top speed on 27B to 35B models
Qwen3.8 27B runs at about 74 tokens/s at 4-bit and about 44 at 8-bit
- Sharing one card among several people or services
MIG splits it into up to four 24GB instances
- Putting 384GB of VRAM in one tower workstation
Four 300W Max-Q cards
Not a good fit
- Running large MoE models at 4-bit
Qwen3.8 Flash Next at 4-bit has to put its experts in system RAM and manages only about 21 tokens/s, while the 128GB DGX Spark holds it entirely and actually gets about 43
- Using NVLink across cards
No version of this generation has it, so cards can only talk over PCIe 5.0
- Packing several Workstation Editions side by side
At 600W, its fans blow hot air back into the case, so for multiple cards switch to the Max-Q version
Before you buy
Know the three versions first
All three have 96GB, and listings often just say “PRO 6000 96G”, but they are used very differently. The Workstation Edition draws 600W, has dual axial fans, and is noticeably taller than a standard full-height card. The Max-Q version draws 300W, uses a blower and is standard size, with the same VRAM and bandwidth as the Workstation Edition. The Server Edition is passively cooled, with power adjustable between 400W and 600W, about 10% less bandwidth than the other two, and video outputs disabled by default.
Feeding 600W
The Workstation Edition pulls its full 600W through a single 16-pin (12V-2x6) connector, so a native 12V-2x6 cable from the power supply is best. With the adapter, all 4 PCIe 8-pin cables must be connected: if any are missing, the adapter tells the card there isn't enough power available (with only one plugged in, it reports just 150W), and the card won't power on.
sm_120 is not a B200
This card's architecture code is sm_120, unlike the data center B200 (sm_100). When a new model launches, vLLM often has to wait for support or rely on community patches before it runs on this card. For example, GitHub user Infatoshi put together 13 patches to run DeepSeek V4 Flash on two PRO 6000s, because DeepGEMM had no sm_120 kernels yet.
Compared with similar GPUs
RTX PRO 6000 vs GeForce RTX 5090, H100 PCIe, and RTX PRO 6000D: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | RTX PRO 6000 | GeForce RTX 5090 | H100 PCIe | RTX PRO 6000D |
|---|---|---|---|---|
| Memory | 96 GB | 32 GB67% less | 80 GB17% less | 84 GB13% less |
| Bandwidth | 1,792 GB/s | 1,792 GB/sSame | 2,000 GB/s12% higher | 1,398 GB/s22% lower |
| BF16 / FP16 compute | 500 TFLOPS | 419 TFLOPS16% lower | 756 TFLOPS51% higher | 385 TFLOPS23% lower |
| INT8 compute | 1,000 TOPS | 838 TOPS16% lower | 1,513 TOPS51% higher | 771 TOPS23% lower |
| Qwen3.8 27B speed (2-bit) | 114.6 t/s | 114.6 t/sSame | 125.6 t/s10% faster | 92.7 t/s19% slower |
| Largest dense model (4-bit) | 165B | 51B69% less | 136B18% less | 143B13% less |
| Power | 600 W | 575 W4% lower | 350 W42% lower | 600 WSame |
| GPU interconnect | Not supported | Not supported | NVLink 600 GB/s | Not supported |
| Interface | PCIe 5.0 x16 | PCIe 5.0 x16 | PCIe 5.0 x16 | PCIe 5.0 x16 |
| Released | 2025-03 | 2025-01 | 2022-03 | 2025-09 |
| 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 PRO 6000 rental price per hour
On-demand rental of the RTX PRO 6000 starts at $1.6 per hour (Vast.ai), about $390/month at 8 hours a day. Buying one (eBay median $22,350) pays off in about 10.3 years, 5.1 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 |
|---|---|---|---|---|---|
 Vast.aiRentUpdated 48 min. ago | $1.6/hr | $195/mo | $390/mo | $1,169/mo | Rent |
 RunPodRentUpdated 48 min. ago | $1.69/hr | $206/mo | $411/mo | $1,234/mo | Rent |
 eBayUpdated 6 hr. ago | $22,350 | Payback ~10.3 yrsPower $13/mo | Payback ~5.1 yrsPower $26/mo | Payback ~1.7 yrsPower $79/mo | Buy |
Break-even uses the US residential average ($0.18/kWh) and 600 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