readygpu

Intel Arc Pro B70 vs NVIDIA RTX PRO 4000 Blackwell

For running models locally, priced in the US. At 32k context the first runs 61 models fully in memory, the second 57.

In short

The Intel Arc Pro B70 has more memory for models: 32 GB against 24 GB. At 32k context the Intel Arc Pro B70 runs 61 of the models we track fully in memory and the NVIDIA RTX PRO 4000 Blackwell 57; only the Intel Arc Pro B70 holds Kimi Linear 48B-A3B. On the models both run fully in memory, the NVIDIA RTX PRO 4000 Blackwell is about 1.4× as fast as the Intel Arc Pro B70 (estimated, median across models).

Updated . Speeds are estimates; how they are worked out.

Intel Arc Pro B70

Not priced here

Add to build
NVIDIA RTX PRO 4000 Blackwell

Not priced here

Add to build

Specifications

SpecIntel Arc Pro B70NVIDIA RTX PRO 4000 Blackwell
GPUArc Pro B70RTX PRO 4000 Blackwell
Memory32 GB GDDR624 GB GDDR7
Memory bus256-bit192-bit
Memory bandwidth608 GB/s672 GB/s
FP16 compute45.8 TFLOPS46.9 TFLOPS
SoftwareSYCL / oneAPICUDA
Board power230 W145 W
PCIe generationGen 5Gen 5
PCIe lanesx16x16

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelIntel Arc Pro B70NVIDIA RTX PRO 4000 Blackwell
Qwen 3.8 27B10tok/sQ8_018tok/sQ5_K_M
Qwen 3.6 27B10tok/sQ8_019tok/sQ5_K_M
Gemma 4 31B11tok/sQ6_K19tok/sQ4_K_M
Qwen 3.6 35B-A3B50tok/sQ6_K96tok/sQ4_K_M
Qwen 3.5 27B10tok/sQ8_019tok/sQ5_K_M
Gemma 4 26B-A4B33tok/sQ8_056tok/sUD-Q6_K
Qwen 3.5 35B-A3B53tok/sQ6_K97tok/sQ4_K_M
GLM 4.7 Flash29tok/sQ6_K47tok/sQ5_K_M
Devstral Small 2 24B10tok/sQ8_019tok/sQ5_K_M
Nemotron 3.5 Lightning 30B-A3B54tok/sUD-Q5_K_M102tok/sMXFP4
Kimi Linear 48B-A3B78tok/sQ4_K_Moffload
gpt-oss 20B52tok/sMXFP477tok/sMXFP4
Qwen3 Coder 30B-A3B29tok/sQ6_K50tok/sQ4_K_M
Qwen3 32B10tok/sQ5_K_Moffload
Gemma 4 12B22tok/sQ8_031tok/sQ8_0
Qwen 3.5 9B29tok/sQ8_039tok/sQ8_0
Mistral Small 3.2 24B10tok/sQ8_019tok/sQ5_K_M
Gemma 3 27B10tok/sQ8_019tok/sQ5_K_M
DeepSeek R1 Distill Qwen 32B10tok/sQ5_K_Moffload
Ministral 3 14B15tok/sQ8_021tok/sQ8_0
Qwen3 14B14tok/sQ8_020tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B22tok/sQ8_031tok/sQ8_0
Qwen3 8B22tok/sQ8_031tok/sQ8_0
Phi-4 14B14tok/sQ8_019tok/sQ8_0
Qwen 3.5 4B54tok/sQ8_074tok/sQ8_0
Ministral 3 8B22tok/sQ8_031tok/sQ8_0
Granite 4.2 8B21tok/sQ8_028tok/sQ8_0
Gemma 4 E4B33tok/sQ8_045tok/sQ8_0
LFM2.5 8B-A1B63tok/sQ8_094tok/sQ8_0
Llama 3.1 8B28tok/sQ6_K38tok/sQ6_K
Qwen2.5 0.5B326tok/sQ8_0446tok/sQ8_0
Qwen3.5 0.8B247tok/sQ8_0338tok/sQ8_0
Qwen3.5 2B164tok/sQ5_K_M224tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B107tok/sQ8_0147tok/sQ8_0
MiniCPM5 2B74tok/sQ8_0102tok/sQ8_0
Spark X2.5 4B71tok/smlx-4bit97tok/smlx-4bit
LFM2.5 2.6B62tok/sQ8_085tok/sQ8_0
LFM2.5 VL 3B62tok/sQ8_085tok/sQ8_0
FrogNano 4B 260953tok/sQ8_073tok/sQ8_0
Mellum2.1 12B A2.5B Thinking53tok/sQ8_079tok/sQ8_0
granite vision 4.1 4b48tok/sQ8_066tok/sQ8_0
granite 4.2 3b46tok/sQ8_063tok/sQ8_0
Ministral 3 3B Instruct 251243tok/sQ8_058tok/sQ8_0
Kimi VL A3B Thinking 250641tok/sQ8_061tok/sQ8_0
Kimi VL A3B39tok/smlx-8bit58tok/smlx-8bit
LFM2.5 24B A2B35tok/sQ8_056tok/sQ6_K
GLM 4 9B 041432tok/smlx-5bit43tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B31tok/sQ8_042tok/sQ8_0
GLM 4.6V Flash27tok/sQ8_037tok/sQ8_0
DeepSeek R1 Distill Llama 8B24tok/sQ8_033tok/sQ8_0
GLM 4.1V 9B Thinking23tok/smlx-8bit32tok/smlx-8bit
Ministral 3 8B Reasoning 251222tok/sQ8_031tok/sQ8_0
Ministral 3 14B Reasoning 251215tok/sQ8_021tok/sQ8_0
DeepSeek R1 Distill Qwen 14B14tok/sQ8_019tok/sQ8_0
Qwen3.8 27B OBLITERATED13tok/smlx-4bit18tok/smlx-4bit
Muse Glimmer 30B13tok/sQ6_K17tok/sQ6_K
granite 4.2 30b10tok/sQ5_K_Moffload
Devstral Small 250710tok/sQ8_019tok/sQ5_K_M
Magistral Small 250610tok/sQ8_019tok/sQ5_K_M
Magistral Small 250710tok/sQ8_019tok/sQ5_K_M
Magistral Small 250910tok/sQ8_019tok/sQ5_K_M
Qwen 3.5 122B-A10Boffloadoffload
Qwen3 Coder Nextoffloadoffload
Nemotron 3 Super 120B-A12Boffloadwon't run
Mistral Small 4 119Boffloadoffload
GLM 4.5 Airoffloadoffload
Qwen3 Next 80B-A3Boffloadoffload
Llama 3.3 70Boffloadwon't run
DeepSeek R1 Distill Llama 70Boffloadwon't run
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
gpt-oss 120Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5won't runwon't run
GLM 4.5Vwon't runwon't run
GLM 4.6won't runwon't run
GLM 4.6Vwon't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi Dev 72Bwon't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the Intel Arc Pro B70 or the NVIDIA RTX PRO 4000 Blackwell?
Choose the Intel Arc Pro B70 to run bigger models. Choose the NVIDIA RTX PRO 4000 Blackwell for speed.
Which can run bigger models, the Intel Arc Pro B70 or the NVIDIA RTX PRO 4000 Blackwell?
At 32k context the Intel Arc Pro B70 runs 61 of the models we track fully in memory and the NVIDIA RTX PRO 4000 Blackwell 57; only the Intel Arc Pro B70 holds Kimi Linear 48B-A3B.
Which is faster for local AI, the Intel Arc Pro B70 or the NVIDIA RTX PRO 4000 Blackwell?
On the models both run fully in memory, the NVIDIA RTX PRO 4000 Blackwell is about 1.4× as fast as the Intel Arc Pro B70 (estimated, median across models).

More comparisons

More hardware for local AI