Skip to content
Run LLM Locally

Mixtral 8x7B Instruct on the H100 PCIe

With 80 GB of VRAM, the H100 PCIe runs Mixtral 8x7B Instruct at Q4_K_M entirely on the GPU with an 8K context (about 27.5 GB needed). At Q8_0 it fits entirely.

ShareFacebookXWhatsAppLinkedInEmail

Mixtral 8x7B Instruct

Q4_K_M · 8K context

Excellent

The model needs about 27.5 GB and fits in the 77.6 GB available, with 50.1 GB to spare.

Your hardware
H100 PCIe 80 GB · 64 GB RAM
Estimated memory
~27.5 GB estimate
VRAM (usable memory)
77.6 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
112–168 tok/s estimate
Largest context in fast memory
32K tokens
Confidence: low

Based on: memory bandwidth (2,000 GB/s) and model size

Memory breakdown

~27.5 GB (26.8 GB–28.3 GB) needed · estimate

Fits entirely in VRAM · 50.1 GB left

  • Model weights24.6 GB
  • KV cache1 GB
  • Runtime overhead0.6 GB
  • Safety margin1.3 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size

EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • no benchmark for this hardware and model yet
  • mixture-of-experts: speed depends on expert routing

Other quantizations on the H100 PCIe

Mixtral 8x7B Instruct on H100 PCIe + 64 GB RAM · 8K context
Mixtral 8x7B Instruct Q8_08K~50.2 GB · 72–102 tok/s estimateExcellent
Mixtral 8x7B Instruct Q6_K8K~39.2 GB · 87–126 tok/s estimateExcellent
Mixtral 8x7B Instruct Q5_K_M8K~33.2 GB · 98–145 tok/s estimateExcellent
Mixtral 8x7B Instruct Q4_K_M8K~27.5 GB · 112–168 tok/s estimateExcellent
Mixtral 8x7B Instruct IQ4_XS8K~27.1 GB · 113–170 tok/s estimateExcellent
Mixtral 8x7B Instruct Q3_K_M8K~21.6 GB · 131–202 tok/s estimateExcellent

Alternatives

GPUs that run Mixtral 8x7B Instruct Q4_K_M entirely in VRAM

Other models that fit the H100 PCIe

Hardware that runs Mixtral 8x7B Instruct Q4_K_M entirely in memory

Prices and links follow the market you choose.

GeForce RTX 5090

NVIDIA

CompatibleRuns Mixtral 8x7B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

Radeon PRO W7900

AMD

ExcellentRuns Mixtral 8x7B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

MacBook Pro 14 M3 Max

Apple Silicon

ExcellentRuns Mixtral 8x7B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

AMD Ryzen AI Max+ 395

Unified memory PC

ExcellentRuns Mixtral 8x7B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the H100 PCIe run Mixtral 8x7B Instruct?
The model needs about 27.5 GB and fits in the 77.6 GB available, with 50.1 GB to spare.
Which quantization should I use on the H100 PCIe?
Q8_0 is the highest-precision quantization that fits entirely in VRAM at 8K context. ~50.2 GB / 77.6 GB.
What is the largest context on the H100 PCIe?
At Q4_K_M, about 32K tokens fit entirely in VRAM with an f16 KV cache. A q8_0 KV cache roughly doubles that.
How fast is Mixtral 8x7B Instruct (H100 PCIe)?
We estimate 112–168 tok/s (estimate, confidence low). Based on: memory bandwidth (2,000 GB/s) and model size.