NVIDIA · workstation

RTX 5880 Ada Generation

RTX 5880 Ada Generation has 48 GB of VRAM at 960 GB/s — about 44.64 GiB usable after driver and compositor overhead. 2037 of 2118 indexed models fit at 16K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
48 GB
GDDR6
Bandwidth
960 GB/s
384-bit bus
Tensor FP16
277 TF
dense
TDP
285 W
$6999 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1749vision language 184image 2video 16audio tts 21embedding 26audio asr 39

What fits at 16K context

largest quantization that fits, per model · 2037 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Devstral-2-123B-Instruct-2512UD-IQ2_M125B40.55 GiB2.92 GiB44.63 GiB0.01 GiB13±22%
Qwen3-72B-SynthesisQ4_K_S72.7B40.80 GiB2.66 GiB44.58 GiB0.06 GiB13±22%
GLM-4.5-Air-DerestrictedMoEIQ2_M110B42.02 GiB1.53 GiB44.58 GiB0.06 GiB47±37%
GLM-4.5-AirMoEIQ2_M110B42.02 GiB1.53 GiB44.57 GiB0.07 GiB47±37%
Apertus-70B-Instruct-2509Q4_K_M70.6B40.72 GiB2.66 GiB44.56 GiB0.08 GiB13±22%
Mistral-Medium-3.5-128BIQ2_XS128B40.41 GiB2.92 GiB44.49 GiB0.15 GiB13±22%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedIQ3_XXS109B41.87 GiB1.59 GiB44.49 GiB0.15 GiB47±37%
Huihui-GLM-4.5-Air-abliterated-lossytensorsMoEI1-Q2_K110B41.88 GiB1.53 GiB44.44 GiB0.20 GiB47±37%
Qwen3.5-122B-A10BMoEQ2_K125B43.21 GiB0.20 GiB44.43 GiB0.21 GiB74±37%
Trinity-2-Codestral-22B-v0.2F1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
Mistral-Small-Drummer-22BF1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
Cydonia-v1.3-Magnum-v4-22BF1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
Mistral-Small-Instruct-2409F1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
Mistral-Small-22B-ArliAI-RPMax-v1.1F1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
magnum-v4-22bF1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
Codestral-22B-v0.1BF1622.2B41.44 GiB1.86 GiB44.36 GiB0.28 GiB13±22%
dolphin-2.9.1-mixtral-1x22bMoEBF1622.2B41.42 GiB1.86 GiB44.34 GiB0.30 GiB7±37%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEBF1623.0B42.85 GiB0.44 GiB44.30 GiB0.34 GiB48±37%
GLM-4.7-Flash-REAP-23B-A3BMoEBF1623.0B42.85 GiB0.44 GiB44.30 GiB0.34 GiB48±37%
Llama-3_1-Nemotron-51B-InstructIQ3_M51.5B21.88 GiB21.25 GiB44.27 GiB0.37 GiB13±22%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEBF1623.6B43.09 GiB0.17 GiB44.26 GiB0.38 GiB64±37%
NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16MoEQ4_K_S75.4B43.15 GiB0.00 GiB44.22 GiB0.42 GiB111±37%
GLM-4.6VMoEQ2_K108B41.64 GiB1.53 GiB44.20 GiB0.44 GiB47±37%
XORTRON-NXTXPRTXXLI1-Q2_K_S128B40.05 GiB2.92 GiB44.13 GiB0.51 GiB13±22%
L3.3-Electra-R1-70bQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Llama-3.3-70B-Instruct-abliteratedQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Llama-3.3-70B-InstructQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Llama-3.1-Nemotron-70B-Instruct-HFQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
L3.3-70B-Euryale-v2.3Q4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Rombos-LLM-70b-Llama-3.3Q4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Anubis-70B-v1.2Q4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Tess-R1-Limerick-Llama-3.1-70BQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
functionary-medium-v3.2KV unresolvedQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Athene-70BQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Infinity-Instruct-7M-Gen-Llama3_1-70BQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Hermes-3-Llama-3.1-70BQ4_K_L70.6B40.33 GiB2.66 GiB44.11 GiB0.53 GiB13±22%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEQ8_042.4B41.98 GiB1.11 GiB44.09 GiB0.55 GiB50±37%
Qwen3-Coder-NextMoEIQ4_NL79.7B42.20 GiB0.80 GiB43.99 GiB0.65 GiB74±37%
Qwen3-Next-80B-A3B-ThinkingMoEIQ4_NL81.3B42.20 GiB0.80 GiB43.99 GiB0.65 GiB74±37%
Qwen3-Next-80B-A3B-InstructMoEIQ4_NL81.3B42.20 GiB0.80 GiB43.99 GiB0.65 GiB74±37%
Midnight-Miqu-70B-v1.5I1-Q4_169.0B40.20 GiB2.66 GiB43.98 GiB0.66 GiB13±22%
Huihui-Qwen3-Coder-Next-abliteratedMoEQ4_079.7B42.78 GiB0.20 GiB43.97 GiB0.67 GiB84±37%
c4ai-command-r-plus-08-2024IQ3_XS104B40.61 GiB2.13 GiB43.91 GiB0.73 GiB13±22%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q2_K125B42.67 GiB0.20 GiB43.89 GiB0.75 GiB75±37%
Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliteratedMoEQ2_K123B42.66 GiB0.20 GiB43.89 GiB0.75 GiB75±37%
Llama-4-Scout-17B-16E-Instruct-abliterated-v2MoEKV unresolvedI1-IQ3_XS109B41.25 GiB1.59 GiB43.87 GiB0.77 GiB47±37%
L3-DARKEST-PLANET-16.5BQ6_K16.5B40.47 GiB2.36 GiB43.87 GiB0.77 GiB13±22%
HarmonicHarlequin_v5-20BI1-Q6_K33.3B25.46 GiB17.27 GiB43.77 GiB0.87 GiB13±22%
Qwen3.5-88BMoEI1-Q3_K_L87.7B42.43 GiB0.20 GiB43.66 GiB0.98 GiB67±37%
CalmeRys-78B-Orpo-v0.1I1-IQ4_XS78.0B39.63 GiB2.86 GiB43.62 GiB1.02 GiB13±22%
calme-2.3-rys-78bIQ4_XS78.0B39.63 GiB2.86 GiB43.62 GiB1.02 GiB13±22%
Qwen3-Coder-Next-Opus-4.6-Reasoning-DistilledMoEQ4_K_S—42.37 GiB0.20 GiB43.56 GiB1.08 GiB85±37%
Phi-3.5-MoE-instructMoEKV unresolvedQ8_041.9B41.44 GiB1.06 GiB43.51 GiB1.13 GiB37±37%
Llama-3_3-Nemotron-Super-49B-v1_5IQ3_M49.9B21.10 GiB21.25 GiB43.49 GiB1.15 GiB13±22%
Valkyrie-49B-v2.1I1-IQ3_M49.9B21.10 GiB21.25 GiB43.49 GiB1.15 GiB13±22%
Llama-3_3-Nemotron-Super-49B-v1IQ3_M49.9B21.10 GiB21.25 GiB43.49 GiB1.15 GiB13±22%
Meta-Llama-3-70B-InstructQ4_K_M70.6B39.61 GiB2.66 GiB43.39 GiB1.25 GiB13±22%
Maenad-70BI1-Q4_K_M70.6B39.60 GiB2.66 GiB43.38 GiB1.26 GiB13±22%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-Q4_K_M70.6B39.60 GiB2.66 GiB43.38 GiB1.26 GiB13±22%
calme-2.4-llama3-70bQ4_K_M70.6B39.60 GiB2.66 GiB43.38 GiB1.26 GiB13±22%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a RTX 5880 Ada Generation run?
2037 of 2118 indexed open-weight models fit a RTX 5880 Ada Generation at 16,384 context with q8_0 KV cache, the largest being Devstral-2-123B-Instruct-2512 at UD-IQ2_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX 5880 Ada Generation actually have?
Its nameplate is 48 GB, but about 44.64 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX 5880 Ada Generation fast for local AI?
Its memory bandwidth is 960 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.