Verified 3w ago Mainly llama.cpp 200 approved runs
Submissions
2,318
200 approved public runs
Models
16
unique models
Avg output
24.0tok/s
Best 103 tok/s
Avg prefill
72.7tok/s
Best 281 tok/s
Avg total
43.4tok/s
Best 215 tok/s
Avg TTFT
10,220ms
Best 1,110 ms
Hardware
2
distinct setups
Matrix runs
36
speed test matrices
Matrix runs36+
Matrix runs 36
A matrix run measures prompt reading (prefill) and generation (decode) speed in one go, across several context lengths and numbers of simultaneous requests. Showing the newest 10.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 14.6 tok/s is the aggregate decode rate across all 4 streams, about 3.6 tok/s per stream, and the 2880 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 93 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47348 MiB of VRAM in use after load; a model download was running on the VM during this run, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF unsloth/Qwen3.8-Flash-Next-GGUF @ 766911a6b7369840a91dbcd95f9f997acaab6cd6, file Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 14.6, 12, 12.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 113 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q8_0-00001-of-00003.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 9.1, 8.7, 8.8 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 69 tok/s is the aggregate decode rate across all 4 streams, about 17.2 tok/s per stream, and the 1537 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 78 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47388 MiB of VRAM in use after load, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF unsloth/Qwen3.8-Flash-Next-GGUF @ 766911a6b7369840a91dbcd95f9f997acaab6cd6, file Qwen3.8-Flash-Next-UD-Q2_K_XL-00001-of-00003.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 88.8, 69, 67.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 25.7 tok/s is the aggregate decode rate across all 8 streams, about 3.2 tok/s per stream, and the 5856 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 118 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47748 MiB of VRAM in use after load, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF ggml-org/Step-3.5-Flash-GGUF @ c439ad64d667632f3e1507757cbe225b6a46e6b8, file Step-3.5-Flash-Q4_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 25.7, 26.2, 29 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 26.5 tok/s is the aggregate decode rate across all 16 streams, about 1.7 tok/s per stream, and the 26805 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 29.1, 22.8, 26.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 23.7 tok/s is the aggregate decode rate across all 12 streams, about 2.0 tok/s per stream, and the 19290 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 23.7, 28.6, 24.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 16.1 tok/s is the aggregate decode rate across all 8 streams, about 2.0 tok/s per stream, and the 15535 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 16.1, 23.3, 17.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 15.2 tok/s is the aggregate decode rate across all 4 streams, about 3.8 tok/s per stream, and the 7283 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 15.2, 18.3, 15.4 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 40.6 tok/s is the aggregate decode rate across all 16 streams, about 2.5 tok/s per stream, and the 15203 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 41.6, 40.6, 41.3 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 14.5, 12.9, 13.8 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 30.1 tok/s is the aggregate decode rate across all 8 streams, about 3.8 tok/s per stream, and the 9115 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 30, 36.8, 30.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 34.3 tok/s is the aggregate decode rate across all 12 streams, about 2.9 tok/s per stream, and the 11370 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 34.3, 39.1, 34.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 16.5, 16.6, 16 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 27.4 tok/s is the aggregate decode rate across all 12 streams, about 2.3 tok/s per stream, and the 12268 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 27.4, 33.8, 27.4 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 20.8 tok/s is the aggregate decode rate across all 8 streams, about 2.6 tok/s per stream, and the 7881 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 20.2, 20.8, 20.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 39.7 tok/s is the aggregate decode rate across all 16 streams, about 2.5 tok/s per stream, and the 25423 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 40, 39.5, 39.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 19.7 tok/s is the aggregate decode rate across all 4 streams, about 4.9 tok/s per stream, and the 5202 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 19, 18.7, 19.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 14, 13.5, 13.3 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 28.6 tok/s is the aggregate decode rate across all 12 streams, about 2.4 tok/s per stream, and the 19654 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 29.5, 36.1, 28.6 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 22.1 tok/s is the aggregate decode rate across all 8 streams, about 2.8 tok/s per stream, and the 13792 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 20.1, 22.1, 20 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 54.6 tok/s is the aggregate decode rate across all 16 streams, about 3.4 tok/s per stream, and the 14546 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 55.2, 54.6, 54.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 18.9 tok/s is the aggregate decode rate across all 4 streams, about 4.7 tok/s per stream, and the 7253 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 18.9, 17.9, 19.2 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 13.5, 13.6, 13.4 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 53.6 tok/s is the aggregate decode rate across all 8 streams, about 6.7 tok/s per stream, and the 6975 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 53.9, 53.5, 53.6 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 52.5 tok/s is the aggregate decode rate across all 12 streams, about 4.4 tok/s per stream, and the 9446 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 53.4, 52.8, 52.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 14.6 tok/s is the aggregate decode rate across all 4 streams, about 3.6 tok/s per stream, and the 2880 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 93 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47348 MiB of VRAM in use after load; a model download was running on the VM during this run, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF unsloth/Qwen3.8-Flash-Next-GGUF @ 766911a6b7369840a91dbcd95f9f997acaab6cd6, file Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 14.6, 12, 12.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 113 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q8_0-00001-of-00003.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 9.1, 8.7, 8.8 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 69 tok/s is the aggregate decode rate across all 4 streams, about 17.2 tok/s per stream, and the 1537 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 78 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47388 MiB of VRAM in use after load, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF unsloth/Qwen3.8-Flash-Next-GGUF @ 766911a6b7369840a91dbcd95f9f997acaab6cd6, file Qwen3.8-Flash-Next-UD-Q2_K_XL-00001-of-00003.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 88.8, 69, 67.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 25.7 tok/s is the aggregate decode rate across all 8 streams, about 3.2 tok/s per stream, and the 5856 ms time to first token is under that load. KASA cuda-vm: 1 x NVIDIA RTX A6000 48 GB (PCIe passthrough, driver 615.71.10) in a VM with 16 vCPU pinned to 8 of the 16 cores (16 threads) of an AMD EPYC 7343 (another VM holds the other 8 cores) and 128 GiB. llama.cpp llama-server b11889 (CUDA image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 118 GB model does not fit the 48 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU (loaded with -lm none, so the CPU part is in ordinary RAM, not memory-mapped); the exact split is not logged at the default verbosity; 47748 MiB of VRAM in use after load, 8 generation threads, 16 batch threads, flash attention on, 8 server slots of 4096 tokens each, context 32768. GGUF ggml-org/Step-3.5-Flash-GGUF @ c439ad64d667632f3e1507757cbe225b6a46e6b8, file Step-3.5-Flash-Q4_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 25.7, 26.2, 29 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 26.5 tok/s is the aggregate decode rate across all 16 streams, about 1.7 tok/s per stream, and the 26805 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 29.1, 22.8, 26.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 23.7 tok/s is the aggregate decode rate across all 12 streams, about 2.0 tok/s per stream, and the 19290 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 23.7, 28.6, 24.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 16.1 tok/s is the aggregate decode rate across all 8 streams, about 2.0 tok/s per stream, and the 15535 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 16.1, 23.3, 17.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 15.2 tok/s is the aggregate decode rate across all 4 streams, about 3.8 tok/s per stream, and the 7283 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 15.2, 18.3, 15.4 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 40.6 tok/s is the aggregate decode rate across all 16 streams, about 2.5 tok/s per stream, and the 15203 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 41.6, 40.6, 41.3 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 70 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/GLM-4.6V-GGUF @ 7de678a269f576ae4090efbf59b558a993386fff, file GLM-4.6V-Q4_K_M.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 14.5, 12.9, 13.8 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 30.1 tok/s is the aggregate decode rate across all 8 streams, about 3.8 tok/s per stream, and the 9115 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 30, 36.8, 30.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 34.3 tok/s is the aggregate decode rate across all 12 streams, about 2.9 tok/s per stream, and the 11370 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 34.3, 39.1, 34.1 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 42 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q2_K.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 16.5, 16.6, 16 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 27.4 tok/s is the aggregate decode rate across all 12 streams, about 2.3 tok/s per stream, and the 12268 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 27.4, 33.8, 27.4 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 20.8 tok/s is the aggregate decode rate across all 8 streams, about 2.6 tok/s per stream, and the 7881 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 20.2, 20.8, 20.8 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 39.7 tok/s is the aggregate decode rate across all 16 streams, about 2.5 tok/s per stream, and the 25423 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 40, 39.5, 39.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 19.7 tok/s is the aggregate decode rate across all 4 streams, about 4.9 tok/s per stream, and the 5202 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 19, 18.7, 19.7 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 59 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-IQ4_XS-00001-of-00002.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 14, 13.5, 13.3 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 28.6 tok/s is the aggregate decode rate across all 12 streams, about 2.4 tok/s per stream, and the 19654 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 29.5, 36.1, 28.6 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 22.1 tok/s is the aggregate decode rate across all 8 streams, about 2.8 tok/s per stream, and the 13792 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 20.1, 22.1, 20 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 1 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 16 requests sent at once (lmx --concurrency 16); 54.6 tok/s is the aggregate decode rate across all 16 streams, about 3.4 tok/s per stream, and the 14546 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 55.2, 54.6, 54.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 4 requests sent at once (lmx --concurrency 4); 18.9 tok/s is the aggregate decode rate across all 4 streams, about 4.7 tok/s per stream, and the 7253 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 18.9, 17.9, 19.2 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 67 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF bartowski/meta-llama_Llama-4-Scout-17B-16E-Instruct-old-GGUF @ 91d32823a1855f7ca12919dffa86701254271eb9, file meta-llama_Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf. Canonical reasoning-v1, lmx single-shot requests after one discarded warm-up. Measured requests: 13.5, 13.6, 13.4 tok/s; this is the median (tie to the earlier). Prompt and output samples were added from the lmx run file, engineTimingsRaw from exact llama-server /metrics deltas around the request.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 8 requests sent at once (lmx --concurrency 8); 53.6 tok/s is the aggregate decode rate across all 8 streams, about 6.7 tok/s per stream, and the 6975 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 53.9, 53.5, 53.6 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1.
CONCURRENCY TEST, NOT A SINGLE-USER RESULT. 12 requests sent at once (lmx --concurrency 12); 52.5 tok/s is the aggregate decode rate across all 12 streams, about 4.4 tok/s per stream, and the 9446 ms time to first token is under that load. KASA b65-vm: 1 x Intel Arc Pro B65 32 GB (PCIe passthrough, xe driver, Ubuntu 26.04 kernel 7.0) in a VM with 16 vCPU pinned to 8 cores (16 threads) of an AMD EPYC 7343 and 128 GiB. llama.cpp llama-server b11913 (image ghcr.io/ggml-org/llama.cpp), placement by llama.cpp's default --fit (1 GiB margin): the 63 GB model does not fit the 32 GB GPU, so llama.cpp's --fit kept the layers on the GPU and placed part of the MoE expert weights in system RAM on the CPU; the exact split is not logged at the default verbosity, 8 generation threads, 16 batch threads, flash attention on, 16 server slots of 4096 tokens each, context 65536. GGUF ggml-org/gpt-oss-120b-GGUF @ 7200b74bea78b0a2964ef9f3897eefa401e890ff, file gpt-oss-120b-MXFP4.gguf. Each request: LocalMaxxing's canonical reasoning-v1 prompt with a unique nonce, 256 output tokens. One discarded warm-up batch, then 3 measured batches: 53.4, 52.8, 52.5 tok/s; this is the median. Not eligible for the Verified badge, which requires batch 1. Selection rule applied to every run in this round: the median of the 2 measured requests whose lmx promptTokens is a whole number; the others report a fractional median, which LocalMaxxing refuses.
