← All GPUs
NVIDIA · Consumer · Blackwell (GB206)
NVIDIA GeForce RTX 5060 Ti 16GB
128-bit bus; NVIDIA's page does not state the PCIe lane count.
VRAM
16 GB
GDDR7
Memory bandwidth
448 GB/s
FP16 compute
not published
Price
~$429
new, maker or suggested price
What fits at 4-bit
The 4-bit file plus 8,192 tokens of context and 1.5 GB of overhead, against the VRAM of 1 and 2 cards. Speeds are measured medians where someone has published a run at that card count; a blank speed means no run, not a slow one. Machine pages carry estimates for the gaps.
| Model | Needs | 1 card (16 GB) | 2 cards (32 GB) |
|---|---|---|---|
| Qwen3.5 4B4B · Q4_K_M | 4.3 GB | fits | fits |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | 55 tok/smedian of 2 runs | fits |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | 69 tok/smedian of 1 run | fits |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GB | fits | fits |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | fits | fits |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GB | fits | fits |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GB | fits | fits |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | fits | fits |
| Phi-414.7B · Q4_K_M | 11.3 GB | fits | fits |
| gpt-oss-20b21B, 3.6B active · MXFP4 (native) | 12.5 GB | 102 tok/smedian of 2 runs | fits |
| Devstral Small 2 24B (2512)24B · Q4_K_M | 16.1 GB | too large | fits |
| Qwen3.6 27B27B · Q4_K_M | 17.7 GB | too large | fits |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M | 17.4 GB | too large | fits |
| Qwen3.8 27B27B · Q4_K_M | 18.2 GB | too large | fits |
| Granite 4.2 30B29.3B · Q4_K_M | 20.3 GB | too large | fits |
| Gemma 4 31B IT30.7B · Q4_K_M | 18.9 GB | too large | fits |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M | 19.5 GB | too large | fits |
| Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M | 21.9 GB | too large | fits |
| Qwen3.6 35B-A3B35B, 3B active · Q4_K_M | 22.4 GB | too large | fits |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M | 25.3 GB | too large | fits |
Measured speeds
One card
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| gpt-oss-20b | MXFP4 MoE | llama.cpp b6208 CUDA | 112 tok/s | 3,839 tok/s2,048-token prompt | 0 | llama.cpp benchmark discussion2025-08 |
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 92 tok/s | 3,585 tok/s4,096-token prompt | 4,096 | review2026-03 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 59 tok/s | 2,365 tok/s | 0 | LocalScore2026-10 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 51 tok/s | 2,314 tok/s | 0 | LocalScore2026-10 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 69 tok/s | 2,965 tok/s4,096-token prompt | 4,096 | review2026-03 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp (commit 23b0202a, CUDA) | 28 tok/s | – | 0 | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp b11707 (CUDA) | 25 tok/s | – | 0 | llama.cpp benchmark discussion2026-10 |
Speculative decoding, several users, or long context
Kept apart because each measures a different workload from one person chatting near the start of a conversation.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 82 tok/s | 2,753 tok/s16,384-token prompt | 16,384 | RTX 5060 Ti 16GB | review2026-03 |
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 73 tok/s | 1,738 tok/s32,768-token prompt | 32,768 | RTX 5060 Ti 16GB | review2026-03 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 51 tok/s | 1,448 tok/s16,384-token prompt | 16,384 | RTX 5060 Ti 16GB | review2026-03 |
| Qwen3 8B | Q4_K | llama.cpp (build not stated) | 39 tok/s | 915 tok/s32,768-token prompt | 32,768 | RTX 5060 Ti 16GB | review2026-03 |
| Qwen3.8 27B | NVFP4 | vLLM (version not stated) | 59 tok/s | – | 0 | 2× RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp (commit 23b0202a, CUDA) | 50 tok/s | – | 0 | RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp (commit 23b0202a, CUDA) | 48 tok/s | – | 6,600 | RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp ('RVN build', version not stated, CUDA) | 47 tok/s | 1,220 tok/s30,000-token prompt | 0 | 2× RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp (commit 23b0202a, CUDA) | 46 tok/s | – | 0 | RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp b11707 (CUDA) | 41 tok/s | – | 0 | RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
| Qwen3.8 27B | UD-IQ4_XS | llama.cpp (commit 23b0202a, CUDA) | 27 tok/s | – | 6,600 | RTX 5060 Ti 16GB | llama.cpp benchmark discussion2026-10 |
Specifications
- Memory
- 16 GB GDDR7
- FP16 basis
- NVIDIA publishes only 759 AI TOPS (FP4, sparse) for this card; no dense FP16 tensor figure.
- Board power
- 180 W
- PCIe
- Gen 5
- NVLink
- No
- Cooling
- Open-air
- Software
- CUDA, Vulkan
- Released
- 2025
- Availability
- sold new
- Price basis
- NVIDIA launch price for the 16 GB model; partner cards only. Jan 2026 reports say ASUS stopped making new 16 GB batches.
- Maker page
- nvidia.com
Sources
- nvidia.com: 16 GB/8 GB GDDR7 128-bit, 180 W, 759 AI TOPS, PCIe Gen 5
- nvidia.com: NVIDIA GeForce compare page: RTX 50 series memory bandwidth row; no RTX 50 SUPER listed
- nvidia.com: 8 GB $379 / 16 GB $429 launch prices
- hostzealot.com: Jan 2026 report (ASUS / Hardware Unboxed) that 16GB RTX 5060 Ti and 5070 Ti production stopped