cd /news/ai-infrastructure/sparkle-arc-pro-b60-only-150-gb-s-vr… · home › topics › ai-infrastructure › article
[ARTICLE · art-156041] src=forum.level1techs.com ↗ pub= topic=ai-infrastructure verified=true sentiment=↓ negative

Sparkle Arc Pro B60: only 150 GB/s VRAM bandwidth (spec 456)

A Sparkle Arc Pro B60 owner reports the card delivering only 150 GB/s of VRAM bandwidth in clpeak's --global-memory-bandwidth test, against a 456 GB/s specification, and is asking other B60 owners to run the same test to confirm. The owner says dense LLMs run at roughly half the speed of other reports, with Qwen3-27B Q4_K_M reaching 8.6 t/s versus 16 t/s on other B60s using the same llama.cpp build and flags. The discrepancy matters because if the 150 GB/s figure is reproducible, the card's memory bandwidth is about a third of its rated spec and would explain the slower token generation.

read1 min views1 publishedOct 23, 2026
Sparkle Arc Pro B60: only 150 GB/s VRAM bandwidth (spec 456)
Image: Forum (auto-discovered)

I’ve been testing dense LLMs on my B60 and they seem to run at half the speed of some other reports - Qwen3.8-27B Q4_K_M at 8.6 t/s, against 16 t/s on other B60s with the same llama.cpp build and flags.

Troubleshooting with claude led me to a potential issue with vram bandwidth.

Testing with clpeak –global-memory-bandwidth I got the result - 150 GB/s.

If another B60 owner can run the same test and let me know I would really appreciate it.

── more in #ai-infrastructure 4 stories · sorted by recency
── more on @sparkle 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/sparkle-arc-pro-b60-…] indexed:0 read:1min 2026-10-23 · —