f41n
15 submissions
  • gpt-oss 20B — 1xRX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 131 t/s generation, 2762 t/s prompt
  • Nemotron3 Nano 4B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 123 t/s generation, 2412 t/s prompt
  • Gemma 4 26B-A4B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 73 t/s generation, 2278 t/s prompt
  • Qwen3.8-27B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · \~51 t/s generation
  • Qwen3.8-27B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 27 t/s generation, 441 t/s prompt
  • Qwen3.8-Flash-Next — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · \~12 t/s generation (warm), \~12 t/s prompt, \~5 t/s first cold prompt
  • gpt-oss 20B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · 89 t/s generation, 858 t/s prompt
  • Nemotron3 Nano 4B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · 77 t/s generation, 783 t/s prompt
  • Qwen3.6 35B-A3B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · 68 t/s generation, 586 t/s prompt
  • Gemma 4 26B-A4B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · 57 t/s generation, 678 t/s prompt
  • Qwen3.8-27B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · \~23 t/s generation
  • Qwen3.8-27B — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · 17 t/s generation, 134 t/s prompt
  • Qwen3.8-Flash-Next — 1x Radeon Instinct MI25 16GB HBM2 @ 170W (WX 9100 BIOS) + Ryzen 5 3600 + 32GB DDR4-3200 · \~17 t/s generation (warm, 15.5-18), \~14 t/s prompt, \~10 t/s first cold prompt
  • LFM2.5 8B-A1B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 221 t/s generation, 4718 t/s prompt
  • LFM2.5 8B-A1B — 1x RX 6800 XT 16GB + Ryzen 7 7700X + 32GB DDR5-5200 · 240 t/s generation
manoreken2
11 submissions
  • DeepSeek-R1:671b-0528_q8_0 — RTX 5090 + IceLake Xeon · 1 tokens/sec
  • DeepSeek-V3.1 Terminus:671b_q8_0 — RTX 5090 + IceLake Xeon · 1 tokens/sec
  • DeepSeek-V3.2:671b_UD-Q8_K_XL — RTX 5090 + IceLake Xeon · 1 tokens/sec
  • GLM-5.2-UD-Q6_K_XL — RTX 5090 + IceLake Xeon · 1 tokens/sec
  • Kimi-K2.6-UD-Q8_K_XL — RTX 5090 + IceLake Xeon · 1 tokens/sec
  • Qwen3:235b-a22b-thinking-2507-q4 — RTX 5090 + IceLake Xeon
  • nemotron-3-super:120b — RTX 5090 + IceLake Xeon
  • GPT-OSS:120b — RTX 5090 + IceLake Xeon
  • DeepSeek-V4-Pro-0813-UD_K_XL.gguf — 1x RTX5090 (32 GB) + IceLake Xeon CPU (1024 GB)
  • DeepSeek-V4.1-Flash
  • gemma-4-31B-it-BF16.gguf — Intel Arc B60 24GB x4 · 2.5
mattlach
5 submissions
  • Gemma4:26B — 1x AMD Instinct MI210 passed through to VM w. 32GB of RAM · Pre-fill: ~2,200 tok/s. Eval ~75-80 tok/s
  • Gemma4:12B — 1x AMD Instinct MI210 passed through to VM w. 32GB of RAM · Pre-fill: ~1,300 tok/s. Eval ~40-45 tok/s
  • Gemma4:E4B — 1x AMD Instinct MI210 passed through to VM w. 32GB of RAM · Pre-fill: ~2,500 - 3,000 tok/s. Eval ~85-90 tok/s
  • Jina Embeddings V3 — 1x AMD Instinct MI210 passed through to VM w. 32GB of RAM
  • Poolside Laguna S2.1 — AMD Epyc 7543 (Milan, 32C/64T), LXC container with 128GB of RAM (no GPU) · Pre-fill 60-62 tok/s. Evaluation 12-14 tok/s
s1ugh34d
5 submissions
  • Carnice-Qwen3.6-MoE-35B-A3B — Radeon VII 16GB HBM2 (gfx906), 32-core EPYC host, 67GB RAM · 54.4
  • gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-GGUF — Radeon VII 16GB HBM2 (gfx906) · 86.7
  • Qwen3.8-27B-GGUF — Radeon VII 16GB HBM2 (gfx906) · 13.9
  • DeepSeek-R1-0528-Qwen3-8B-GGUF — Radeon VII 16GB HBM2 (gfx906) · 84.4
  • Qwen_Qwen3.5-4B-GGUF — Radeon VII 16GB HBM2 (gfx906) · 96.4
Burhan
4 submissions
  • Qwen3.6-27b-q4_KM — 1x RTX-4000 Ada SFF (20 GB) · ~20-30 t/s
  • LFM2.5-thinking:1.2b-q8_0 — 1x RTX-4000 Ada SFF (20 GB) · ~35 t/s
  • Gemma4:26b-a4b-it-mtp-q4_K_M — 1x RTX-4000 Ada SFF (20 GB) · ~25-30 t/s
  • Qwen3.5:9b-q8_0 — 1x RTX-4000 Ada SFF (20 GB) · ~30 t/s
Kif
3 submissions
  • Qwen3.8-Flash-Next — 1x RTX 4090 24GB, Ryzen 9 5900X, 128GB DDR4-3600 CL18 (2x CMK64GX4M2D3600C18, 4x32GB double-rank, holding 3600 MT/s XMP), Gigabyte Aorus X570 Master, Samsung 990 Pro NVMe (PCIe 4.0) · 128
  • Qwen3.8-Flash-Next — 1x RTX 4090 24GB, Ryzen 9 5900X, 128GB DDR4-3600 CL18 (4x32GB double-rank), Gigabyte Aorus X570 Master, Samsung 990 Pro NVMe · 107
  • Qwen3.8-Flash-Next — 1x RTX 4090 24GB, Ryzen 9 5900X, 128GB DDR4-3600 CL18 (4x32GB double-rank), Gigabyte Aorus X570 Master, Samsung 990 Pro NVMe · 39.9
digitalscream
2 submissions
  • Swift-Qwen3.8-27B-Uncensored-FP8 — 2 x R9700 · 120t/s code, 70t/s chat
  • Swift 1.5 Flash Next — 2 x R9700 32GB, Threadripper 3960X, 128GB DDR4 · 200
PawingUnplug
2 submissions
  • Kwaipilot_KAT-Coder-V2.5-Dev-Q5_K_M.gguf — 2 x R9700 · ~50t/s chat
  • Qwen3.8-Flash-Next-Q4_K_M — 2 x w7900 · ~35t/s chat
Novasty
1 submission
  • Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive — RTX 3060 12GB · 14.8
SgtAwesomesauce
1 submission
  • qwen3.8-flash-next — 4x AMD v620 · 68-92
Yehya_Karout
1 submission
  • Qwen3.8-Flash-Next — 1x RTX 5080 16GB + 64GB DDR5-6400, Core Ultra 9 285K · ~170 tok/s tg / ~3,000 pp at 4K, ~150 tg / ~3,950 pp at 180K, 152 tg / 3,800 pp at 250K. Real opencode agent session: ~165-170 tok/s avg

Want to be here?

Submit your setup in the forum thread using the Quick Submit template in Post #2. It takes 30 seconds, and your data appears in the matrix on the next update. Every row in the matrix started exactly this way.

Submit your setup →