Local LLM configurator

Model

Quantization

Q4_K_M is the right trade-off for most uses. What is this?

Target speed

Around 10 tok/s reads comfortably; 30+ feels instant.

Platform

Advanced options8K tokens · KV cache f16

The defaults suit most uses. You can skip this section.

Context length

How much text the model keeps in memory. 8K is enough for chat; long documents need 32K or more. What is this?

KV cache type

Leave f16 if unsure. q8_0 roughly halves the conversation's memory, q4_0 nearly quarters it; the option must also be enabled in your app. What is this?

Compatible machines

What the model needs

Llama 3.3 70B Instruct Q4_K_M with a 8K context needs about 45.8 GB (up to 48.7 GB) of GPU or unified memory.

Compatible configurations for Llama 3.3 70B Instruct Q4_K_M · 8K context
	NVIDIARTX 6000 AdaCompatible RTX 6000 Ada Generation · 48 GB GDDR6 ECC 960 GB/s System RAM 64 GB Smallest configuration of this platform that runs the model entirely in fast memory.	AMDRadeon PRO W7900Compatible Radeon PRO W7900 · 48 GB GDDR6 ECC 864 GB/s System RAM 64 GB Smallest configuration of this platform that runs the model entirely in fast memory.	Apple SiliconM1 Max · 64 GBCompatible MacBook Pro 14, MacBook Pro 16, Mac Studio 32 GPU · 400 GB/s Smallest configuration of this platform that runs the model entirely in fast memory.	Unified memory PCAMD Ryzen AI Max+ 395 · 64 GBCompatible Radeon 8060S 256 GB/s Smallest configuration of this platform that runs the model entirely in fast memory.	Highest estimated speedH100 PCIeExcellent H100 PCIe 80GB · 80 GB HBM2e 2,000 GB/s System RAM 64 GB Highest estimated generation speed among configurations that fit.
Memory for the model	47 GB	47 GB	48 GB	48 GB	78 GBMost memory
Estimated speed	16–21 tok/sestimate	13–18 tok/sestimate	6–8.5 tok/sestimate	3.5–5.5 tok/sestimate	30–42 tok/sestimateFastest
Power draw (GPU/board)	~300 W	~295 WLowest power	–	–	~350 W
Price	No current offer	No current offer	No current offer	No current offer	No current offer
Confidence	low	low	low	low	low

Leaders are marked per criterion; there is no overall winner. Speeds are estimates unless marked measured. Power is the GPU's rated board power (or the platform's, when known).

Suggested PC build

GPU: RTX 6000 Ada Generation
Processor: Ryzen 7 9700X
RAM: 48 GB DDR5
Power supply: 600 W

The processor matters little when the model runs on the GPU; any current 8-core CPU is enough. PSU sized with 30% headroom for power spikes.

Compatibility checks: No compatibility problem detected between these parts.

Where to buy

Your market

Prices and links follow the market you choose.

RTX 6000 Ada Generation

RTX 6000 Ada Generation · 48 GB GDDR6 ECC · 960 GB/s

CompatibleRuns Llama 3.3 70B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

Radeon PRO W7900

Radeon PRO W7900 · 48 GB GDDR6 ECC · 864 GB/s

CompatibleRuns Llama 3.3 70B Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

AMD Ryzen AI Max+ 395

Radeon 8060S · 256 GB/s

CompatibleRuns Llama 3.3 70B 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.