NVIDIA · workstation

RTX 5000 Ada Generation

RTX 5000 Ada Generation has 32 GB of VRAM at 576 GB/s — about 29.76 GiB usable after driver and compositor overhead. 2011 of 2118 indexed models fit at 32K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
GDDR6
Bandwidth
576 GB/s
256-bit bus
Tensor FP16
261 TF
dense
TDP
250 W
$4000 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
video 16text 1726vision language 181image 2embedding 26audio tts 21audio asr 39

What fits at 32K context

largest quantization that fits, per model · 2011 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.76 GiB0.00 GiB12±22%
Assistant_Pepe_70BQ2_K70.6B25.79 GiB2.81 GiB29.73 GiB0.03 GiB12±22%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB2.18 GiB29.71 GiB0.05 GiB37±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB2.18 GiB29.71 GiB0.05 GiB37±37%
Hermes-4-70BUD-IQ3_XXS70.6B25.76 GiB2.81 GiB29.69 GiB0.07 GiB12±22%
Llama-3.3-70B-InstructUD-IQ3_XXS70.6B25.76 GiB2.81 GiB29.69 GiB0.07 GiB12±22%
DeepSeek-R1-Distill-Llama-70BUD-IQ3_XXS70.6B25.76 GiB2.81 GiB29.69 GiB0.07 GiB12±22%
Apriel-1.6-15b-ThinkerBF1614.9B26.88 GiB1.69 GiB29.62 GiB0.14 GiB12±22%
Salience-1.5-ProMoEQ6_K36.0B28.43 GiB0.18 GiB29.61 GiB0.15 GiB67±37%
Qwable-v1MoEQ6_K36.0B28.43 GiB0.18 GiB29.61 GiB0.15 GiB67±37%
T-SearchMoEQ6_K36.0B28.43 GiB0.18 GiB29.61 GiB0.15 GiB67±37%
CalmeRys-78B-Orpo-v0.1I1-IQ2_XXS78.0B25.43 GiB3.02 GiB29.58 GiB0.18 GiB12±22%
calme-2.3-rys-78bIQ2_XXS78.0B25.43 GiB3.02 GiB29.58 GiB0.18 GiB12±22%
ALIA-40b-fc-2606I1-Q5_K_M40.4B26.78 GiB1.69 GiB29.58 GiB0.18 GiB12±22%
ALIA-40b-instruct-2606I1-Q5_K_M40.4B26.78 GiB1.69 GiB29.58 GiB0.18 GiB12±22%
Devstral-2-123B-Instruct-2512IQ1_S125B25.33 GiB3.09 GiB29.58 GiB0.18 GiB12±22%
Mistral-Medium-3.5-128BI1-IQ1_S128B25.33 GiB3.09 GiB29.58 GiB0.18 GiB12±22%
XORTRON-NXTXPRTXXLI1-IQ1_S128B25.33 GiB3.09 GiB29.58 GiB0.18 GiB12±22%
Delphi-25B-SimpleRL-MathI1-Q6_K25.0B19.08 GiB9.41 GiB29.57 GiB0.19 GiB12±22%
Qwen3.5-88BMoEI1-Q2_K_S87.7B28.30 GiB0.21 GiB29.54 GiB0.22 GiB59±37%
Kimi-Linear-48B-A3B-InstructMoEQ4_K_L49.1B28.26 GiB0.27 GiB29.53 GiB0.23 GiB12±22%
Apertus-70B-Instruct-2509IQ3_XXS70.6B25.53 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Maenad-70BI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Rombos-LLM-70b-Llama-3.3I1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
L3.3-Electra-R1-70bI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
L3.3-70B-Magnum-v4-SEIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Llama-3.3_70_b_uncensored_continuedI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Llama-3.3-70B-Instruct-abliteratedI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
grok-oss-Revenant-70BI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
L3.3-70B-Euryale-v2.3I1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Hermes-3-Llama-3.1-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Hermes-4-70B-hereticI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Llama-3.1-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Anubis-70B-v1.2IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Golem-70B-v1bI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Legion-V2.1-LLaMa-70BI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Tess-R1-Limerick-Llama-3.1-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
SEMIKONG-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
functionary-medium-v3.2KV unresolvedIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Llama-3.1-WhiteRabbitNeo-2-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Infinity-Instruct-7M-Gen-Llama3_1-70BI1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
New-Dawn-Llama-3-70B-32K-v1.0I1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Meta-Llama-3-70B-Instruct-abliterated-v3.5I1-IQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Athene-70BIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
L3.3-70B-Magnum-DiamondIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Meta-Llama-3-70B-InstructIQ3_XXS70.6B25.58 GiB2.81 GiB29.52 GiB0.24 GiB12±22%
Hunyuan-A13B-InstructMoEQ2_K80.4B27.40 GiB1.13 GiB29.52 GiB0.24 GiB12±22%
Qwen3-Coder-NextMoEUD-IQ3_S79.7B27.65 GiB0.84 GiB29.49 GiB0.27 GiB61±37%
Wizard-Vicuna-30B-UncensoredI1-Q3_K_M32.5B14.69 GiB13.71 GiB29.47 GiB0.29 GiB12±22%
archangel_sft-kto_llama30bI1-Q3_K_M32.5B14.69 GiB13.71 GiB29.47 GiB0.29 GiB12±22%
Yi-34B-200K-DARE-megamerge-v8I1-Q6_K34.4B26.28 GiB2.11 GiB29.47 GiB0.29 GiB12±22%
dolphin-2.9.1-yi-1.5-34b-hereticQ6_K34.4B26.28 GiB2.11 GiB29.47 GiB0.29 GiB12±22%
dolphin-2.9.1-yi-1.5-34bI1-Q6_K34.4B26.28 GiB2.11 GiB29.47 GiB0.29 GiB12±22%
OrionStar-Yi-34B-Chat-LlamaI1-Q6_K34.4B26.28 GiB2.11 GiB29.47 GiB0.29 GiB12±22%
Yi-34B-200K-LlamafiedI1-Q6_K34.4B26.28 GiB2.11 GiB29.47 GiB0.29 GiB12±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 5000 Ada Generation run?
2011 of 2118 indexed open-weight models fit a RTX 5000 Ada Generation at 32,768 context with q4_0 KV cache, the largest being Bernini-R at Q8_0. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX 5000 Ada Generation actually have?
Its nameplate is 32 GB, but about 29.76 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX 5000 Ada Generation fast for local AI?
Its memory bandwidth is 576 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.