DeepSeek · 70.6B total / 70.6B active · dense · reasoning
What hardware do you need to run DeepSeek R1 Distill Llama 70B locally?
You need about 46 GB of GPU or unified memory to run DeepSeek R1 Distill Llama 70B at Q4_K_M (4-bit), or 80 GB at Q8_0 (8-bit). 13 of the 18 rigs we track can hold it. The cheapest is the GMKtec EVO-X2 (64 GB) at $2,200, at an estimated ~5 tok/s. The fastest is the PNY NVIDIA RTX PRO 6000 Blackwell at an estimated ~34 tok/s. Prices & stock checked 2026-10-24. Prices change daily and some retailers block our checker, so always confirm on the store page.
Memory needed
| Q4_K_M (4-bit) | Q8_0 (8-bit) | |
|---|---|---|
| Memory needed (est.) | ~46 GB | ~80 GB |
| Weights only | ~43 GB | ~75 GB |
| Read per token | ~43 GB | ~75 GB |
| Rigs that fit | 13 of 18 | 11 of 18 |
It's a dense model, so every one of its 70.6B parameters is read for each token. Memory bandwidth, not compute, sets the speed.
DeepSeek-R1's reasoning distilled into Llama-3.3-70B-Instruct, so it has the same 70.6B dense architecture (and the same memory and speed) as Llama 3.3 70B. It's not the full 671B DeepSeek-R1. Model card
13 rigs that run DeepSeek R1 Distill Llama 70B
Sorted by today's lowest price. Quant is what we'd use on that machine: Q8_0 when it fits comfortably and still runs 10+ tok/s, otherwise Q4_K_M.
| Rig | Price today | Quant | Memory used | Est. tokens/sec | Buy |
|---|---|---|---|---|---|
| GMKtec EVO-X264 GB · 256 GB/s | $2,200GMKtec store · 2026-10-24 | Q4_K_M | ~46 of 48 GB | ~5 est. | Check price |
| GMKtec EVO-X2128 GB · 256 GB/s | $3,500GMKtec store · 2026-10-24 | Q4_K_M | ~46 of 96 GB | ~5 est. | Check price |
| Minisforum MS-S1 Max128 GB · 256 GB/s | $3,919Minisforum store · 2026-10-24 | Q4_K_M | ~46 of 96 GB | ~5 est. | Check price |
| Beelink GTR9 Pro128 GB · 256 GB/s | $4,349Beelink store · 2026-10-24 | Q4_K_M | ~46 of 96 GB | ~5 est. | Check price |
| Framework Desktop DIY128 GB · 256 GB/s | $4,839Framework store · 2026-10-24 | Q4_K_M | ~46 of 96 GB | ~5 est. | Check price |
| Mac Studio M5 Ultra96 GB · 1,200 GB/s | $5,499Apple Store · 2026-10-24 | Q4_K_M | ~46 of 72 GB | ~23 est. | Check price |
| Lenovo ThinkStation PGX128 GB · 273 GB/s | $5,859B&H Photo · 2026-10-24 | Q4_K_M | ~46 of 109 GB | ~5 est. | Check price |
| ASUS Ascent GX10128 GB · 273 GB/s | $5,999B&H Photo · 2026-10-24 | Q4_K_M | ~46 of 109 GB | ~5 est. | Check price |
| NVIDIA DGX Spark128 GB · 273 GB/s | $6,950NVIDIA Marketplace · 2026-10-24 | Q4_K_M | ~46 of 109 GB | ~5 est. | Check price |
| HP ZGX Nano G1n128 GB · 273 GB/s | $7,699B&H Photo · 2026-10-24 | Q4_K_M | ~46 of 109 GB | ~5 est. | Check price |
| Gigabyte AI TOP ATOM128 GB · 273 GB/s | $8,499B&H Photo · 2026-10-24 | Q4_K_M | ~46 of 109 GB | ~5 est. | Check price |
| PNY NVIDIA RTX PRO 6000 Blackwell96 GB · 1,792 GB/s (card only, needs a host PC) | $17,899B&H Photo · 2026-10-24 | Q8_0 | ~80 of 89 GB | ~19 est. | Check price |
| Mac Studio M5 Ultra256 GB · 1,200 GB/s | No price seen | Q8_0 | ~80 of 192 GB | ~13 est. | Check price |
Tokens/sec are estimates from our bandwidth formula (one user, ~8K context, llama.cpp-class runtime), not benchmarks. GPU prices are for the card only. Prices change daily, so confirm on the store page. How we estimate
Measured speeds from public benchmarks
No public runs of the distill itself in our set. These are Llama 3.x 70B runs: the distill is built on Llama-3.3-70B-Instruct, so it has identical size and speed. These are measured, not our estimates.
| Hardware | Model / run | Measured tok/s | Source |
|---|---|---|---|
| Strix Halo256 GB/s | Llama 3.3 70B Q4_K_Mllama.cpp Vulkan tg128 | 4.8 | source |
| M3 Ultra 80c819 GB/s | Llama 3.3 70B 4-bitMLX (oMLX) 1K ctx | 17.0 | source |
| RTX PRO 60001,792 GB/s | Llama 3.3 70B Q4_K_Mllama.cpp tg128 | 34.0 | source |
| Framework Desktop (Strix Halo)256 GB/s | Llama 3.1 70B Q4_K_MOllama (Geerling) | 5.0 | source |
| Dell Pro Max GB10273 GB/s | Llama 3.1 70B Q4_K_MOllama (Geerling) | 4.7 | source |
| M3 Ultra 512GB819 GB/s | Llama 3.1 70B Q4_K_MOllama (Geerling) | 14.1 | source |
From the benchmark set we use to calibrate our speed formula. All 52 runs
Every rig we track, 4-bit and 8-bit
| Machine | Price | Q4_K_M | Q8_0 |
|---|---|---|---|
| Mac mini M5 Pro15-core CPU / 16-core GPU, 24GB | $1,699 | Won't fitneeds ~46 GB | Won't fitneeds ~80 GB |
| GMKtec EVO-X2Ryzen AI Max+ 395, 64GB / 1TB | $2,200 | Tight fit~5 tok/s46 of 48 GB | Won't fitneeds ~80 GB |
| Mac Studio M5 Max18-core CPU / 32-core GPU, 36GB | $2,499 | Won't fitneeds ~46 GB | Won't fitneeds ~80 GB |
| Mac Studio M5 Max18-core CPU / 40-core GPU, 48GB | $3,099 | Won't fitneeds ~46 GB | Won't fitneeds ~80 GB |
| GMKtec EVO-X2Ryzen AI Max+ 395, 128GB / 1TB | $3,500 | Fits~5 tok/s46 of 96 GB | Fits~3 tok/s80 of 96 GB |
| Minisforum MS-S1 MaxRyzen AI Max+ 395, 128GB / 2TB | $3,919 | Fits~5 tok/s46 of 96 GB | Fits~3 tok/s80 of 96 GB |
| Beelink GTR9 ProRyzen AI Max+ 395, 128GB / 2TB | $4,349 | Fits~5 tok/s46 of 96 GB | Fits~3 tok/s80 of 96 GB |
| Framework Desktop DIYRyzen AI Max+ 395, 128GB | $4,839 | Fits~5 tok/s46 of 96 GB | Fits~3 tok/s80 of 96 GB |
| Gigabyte GeForce RTX 4090 Windforce V224GB card | $4,899 | Won't fitneeds ~46 GB | Won't fitneeds ~80 GB |
| Mac Studio M5 Ultra30-core CPU / 64-core GPU, 96GB | $5,499 | Fits~23 tok/s46 of 72 GB | Won't fitneeds ~80 GB |
| Lenovo ThinkStation PGX1TB | $5,859 | Fits~5 tok/s46 of 109 GB | Fits~3 tok/s80 of 109 GB |
| ASUS Ascent GX101TB | $5,999 | Fits~5 tok/s46 of 109 GB | Fits~3 tok/s80 of 109 GB |
| NVIDIA DGX SparkFounders Edition, 4TB | $6,950 | Fits~5 tok/s46 of 109 GB | Fits~3 tok/s80 of 109 GB |
| HP ZGX Nano G1n4TB | $7,699 | Fits~5 tok/s46 of 109 GB | Fits~3 tok/s80 of 109 GB |
| Gigabyte AI TOP ATOM4TB | $8,499 | Fits~5 tok/s46 of 109 GB | Fits~3 tok/s80 of 109 GB |
| NVIDIA GeForce RTX 5090 Founders Edition32GB card | $8,500 | Won't fitneeds ~46 GB | Won't fitneeds ~80 GB |
| PNY NVIDIA RTX PRO 6000 Blackwell96GB card | $17,899 | Fits~34 tok/s46 of 89 GB | Fits~19 tok/s80 of 89 GB |
| Mac Studio M5 Ultra256GB | — | Fits~23 tok/s46 of 192 GB | Fits~13 tok/s80 of 192 GB |
Fit and tokens/sec are estimates. "Tight fit" means over 90% of usable memory: it loads, but leaves little room for long context.
FAQ
How much memory do I need to run DeepSeek R1 Distill Llama 70B locally?
About 46 GB at Q4_K_M (4-bit) and 80 GB at Q8_0 (8-bit). That's the weights plus runtime overhead and roughly 8K tokens of context for one user. Longer context needs more. These are estimates.
What's the cheapest hardware that runs DeepSeek R1 Distill Llama 70B?
Of the rigs we track, the GMKtec EVO-X2 64 GB at $2,200 (GMKtec store, checked 2026-10-24). We estimate about 5 tokens/sec at 4-bit.
How fast does DeepSeek R1 Distill Llama 70B run locally?
We estimate about 5 to 34 tokens/sec at 4-bit across the rigs that fit, fastest on the PNY NVIDIA RTX PRO 6000 Blackwell. Published runs we've collected range from 4.7 to 34.0 tok/s (on related models; see the measured table).
Can I run DeepSeek R1 Distill Llama 70B on a single gaming GPU like the RTX 4090 or 5090?
Not fully in VRAM. It needs about 46 GB at 4-bit, more than a 24 or 32 GB card. You can offload part of it to system RAM, but it gets much slower. The cheapest rig we track that holds it fully is the GMKtec EVO-X2 64 GB at $2,200.
Can a Mac run DeepSeek R1 Distill Llama 70B?
Yes. The cheapest Mac we track that fits it is the Mac Studio M5 Ultra 96 GB at $5,499, at an estimated ~23 tok/s. macOS lets the GPU use about 75% of memory by default.
Are these speeds measured or estimated?
The table shows estimates from memory bandwidth, calibrated against 43 public llama.cpp/MLX benchmarks (median error about 5%). The measured-speeds table shows published runs with links.