Builds / 4x Intel Arc Pro B70 32 GB
4x Intel Arc Pro B70 (Xeon workstation): Ripple-65B Preview Q4_K_M on llama.cpp SYCL at 13.0 tok/s, layer split, 200K context
Last updated:
Source: intelinside.ai ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
- GPU(s)
- 4x Intel Arc Pro B70 32 GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 128
- CPU
- Intel Xeon 676X
Amazon · eBay (used) · Newegg - Motherboard
- not stated
- RAM
- not stated
- PSU
- not stated
- Other
- Each B70 on its own PCIe Gen5 x16 link; stock llama.cpp master b12314 (e1f7593c) built with the upstream intel.Dockerfile (oneAPI 2026.2.0); llama-server -ngl 999 --split-mode layer -c 200000 -np 4 --kv-unified, f16 KV, flash attention auto-disabled
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Ripple 65B Preview | Q4_K_M (AtomicChat GGUF, 38.8 GB) | llama.cpp b12314 (SYCL), layer split across 4 GPUs | 13.0 | not stated | context: 200,000 ctx, one request at a time📏 Server-side decode on streamed chat tasks (code and prose prompts, temperature 0.6, top_p 0.95): 13.0 tok/s on both of 2 single-request runs (446 and 350 tokens); smoke-test scope, no power logging |
Owner-submitted result on intelinside.ai (run 12 Nov 2026, submitter bhannath69). Warm first token 271 ms; the cold first request took 18.8 s. Flash attention was disabled automatically because K/V-row FA is unsupported with this split.
Source: https://www.intelinside.ai/results/335
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)
Quick answers: Can 4x Intel Arc Pro B70 run Ripple 65B Preview?