Qwen3.8 27B
Qwen/Qwen3.8-27B
Qwen · 22 recorded configurations
Measured setup
Hardware
Runtime and tool
Quantisation
MTP draft depth
Every option opens a recorded setup. Where the corpus recorded nothing at this page's other settings, the option names the fields it switches with it. Allocation and runtime detail stay in the technical record below.
1× RTX 3090 · ExLlamaV3
EXL3 5.00bpw · MTP 2
PP · left axisTG · right axis
View exact measurements
| Context depth (tokens) | Fresh prefill PP (tok/s) | TG (tok/s) |
|---|---|---|
| 0 | 890.55 | 55.28 |
| 16,384 | 736.88 | 45.36 |
| 105,728 | 367.89 | 35.13 |
Concurrent-stream measurements
| Streams | Context depth (tokens) | Aggregate PP (tok/s) | Aggregate TG (tok/s) | TG / stream (tok/s) | Aggregate all (tok/s) |
|---|---|---|---|---|---|
| 2 | 8,192 | 856.67 | 98.08 | 49.04 | 239.42 |
| 3 | 8,192 | 886.55 | 131.64 | 43.88 | 304.49 |
| 4 | 8,192 | 911.69 | 171.90 | 42.98 | 374.49 |
| 5 | 8,192 | 916.76 | 196.39 | 39.28 | 412.46 |
Technical record and provenance
Source 1 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-context-fit-1-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- context
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · context-fit 106496-token allocation
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 106,496 tokens
- Actual context per slot
- 106,496 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Throughput spreads
- KV 0, 1 stream: PP mean 890.55 t/s (min 883.13, max 897.49, SD 7.19, n=3); TG mean 55.28 t/s (min 54.91, max 55.48, SD 0.32, n=3); All mean 147.52 t/s (min 146.52, max 148.05, SD 0.87, n=3) · KV 16,384, 1 stream: PP mean 736.88 t/s (min 736.80, max 737.02, SD 0.12, n=3); TG mean 45.36 t/s (min 45.28, max 45.40, SD 0.07, n=3); All mean 121.16 t/s (min 120.98, max 121.27, SD 0.16, n=3) · KV 105,728, 1 stream: PP mean 367.89 t/s (min 367.79, max 368.07, SD 0.16, n=3); TG mean 35.13 t/s (min 35.10, max 35.14, SD 0.02, n=3); All mean 88.48 t/s (min 88.44, max 88.51, SD 0.04, n=3)
- Measurement outcomes
- 0 tokens · 1 stream: measured; 16,384 tokens · 1 stream: measured; 105,728 tokens · 1 stream: measured
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · context-fit 106496-token allocation. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-context-mtp-2-c1; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 640298efcc4dab1a26a659a7be939d02d2b771ea1f4fb5aec59b936f92d949f3. Separately allocated 106496 tokens per stream for this MTP setting. The allocation search advances in 8192-token steps; its recorded fitted capacity is not a token-exact VRAM maximum. The selected boundary and the 0/16K samples are retained from their recorded source attempts.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-exl-context-fit/5bpw-gpu1/runs/mtp-2-106496-boundary/5.00bpw/context/mtp-2-c1
Source 2 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-2-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 2, 8192-token prefix, 8960-token allocation per stream
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 8,960 tokens
- Actual context per slot
- 8,960 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Throughput spreads
- KV 8,192, 2 streams: PP mean 856.67 t/s (min 855.87, max 857.26, SD 0.72, n=3); TG mean 98.08 t/s (min 97.96, max 98.23, SD 0.13, n=3); All mean 239.42 t/s (min 239.14, max 239.72, SD 0.29, n=3)
- Measurement outcomes
- 8,192 tokens · 2 streams: measured
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 2, 8192-token prefix, 8960-token allocation per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-concurrency-mtp-2-c2; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation; this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c2
Source 3 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-3-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 3, 8192-token prefix, 8960-token allocation per stream
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 8,960 tokens
- Actual context per slot
- 8,960 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Throughput spreads
- KV 8,192, 3 streams: PP mean 886.55 t/s (min 885.08, max 887.73, SD 1.35, n=3); TG mean 131.64 t/s (min 131.25, max 131.88, SD 0.34, n=3); All mean 304.49 t/s (min 303.83, max 305.01, SD 0.61, n=3)
- Measurement outcomes
- 8,192 tokens · 3 streams: measured
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 3, 8192-token prefix, 8960-token allocation per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-concurrency-mtp-2-c3; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation; this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c3
Source 4 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-4-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 4, 8192-token prefix, 8960-token allocation per stream
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 8,960 tokens
- Actual context per slot
- 8,960 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Throughput spreads
- KV 8,192, 4 streams: PP mean 911.69 t/s (min 907.61, max 914.87, SD 3.71, n=3); TG mean 171.90 t/s (min 171.42, max 172.22, SD 0.43, n=3); All mean 374.49 t/s (min 373.26, max 375.12, SD 1.07, n=3)
- Measurement outcomes
- 8,192 tokens · 4 streams: measured
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 4, 8192-token prefix, 8960-token allocation per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-concurrency-mtp-2-c4; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation; this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c4
Source 5 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-5-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 5, 8192-token prefix, 8960-token allocation per stream
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 8,960 tokens
- Actual context per slot
- 8,960 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Throughput spreads
- KV 8,192, 5 streams: PP mean 916.76 t/s (min 915.59, max 917.51, SD 1.03, n=3); TG mean 196.39 t/s (min 195.79, max 196.74, SD 0.53, n=3); All mean 412.46 t/s (min 411.41, max 413.07, SD 0.91, n=3)
- Measurement outcomes
- 8,192 tokens · 5 streams: measured
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 5, 8192-token prefix, 8960-token allocation per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-concurrency-mtp-2-c5; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation; this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c5
Source 6 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-6-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 6, 8192-token prefix, 8960-token allocation per stream
- Measured date
- 2026-10-15
- Hardware
- 1× RTX 3090
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- One physical RTX 3090; target model and native MTP compute tensors were observed on CUDA0, with no CPU expert offload or model split.
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Requested context per slot
- 8,960 tokens
- Actual context per slot
- 8,960 tokens
- Native model context
- 262,144 tokens
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Measurement outcomes
- 8,192 tokens · 6 streams: oom — OutOfMemoryError: CUDA out of memory. Tried to allocate 170.00 MiB. GPU 0 has a total capacity of 23.56 GiB of which 177.06 MiB is free. Including non-PyTorch memory, this process has 23.38 GiB memory in use. 23.56 GiB allowed; Of the allocated memory 23.00 GiB is allocated by PyTorch, and 54.35 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
- Notes
- 1× RTX3090 — no splitting · EXL3 5.00bpw · MTP-2 · concurrency 6, 8192-token prefix, 8960-token allocation per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. Runtime evidence maps that physical GPU to one logical CUDA0. Source configuration 500bpw-concurrency-mtp-2-c6; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation; this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c6
Source 7 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-7-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- Single RTX3090 requested — placement unverified · EXL3 5.00bpw · MTP-2 · concurrency 7, 8192-token prefix, requested 8960-token allocation (actual unknown) per stream
- Measured date
- 2026-10-15
- Hardware
- Not observed (outcome-only startup failure)
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- Unverified (startup failed)
- Observed devices
- None observed
- Requested devices
- CUDA0
- Split mode
- None
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Measurement outcomes
- 8,192 tokens · 7 streams: oom — RuntimeError: Insufficient VRAM in split for model and cache
- Notes
- Single RTX3090 requested — placement unverified · EXL3 5.00bpw · MTP-2 · concurrency 7, 8192-token prefix, requested 8960-token allocation (actual unknown) per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. No successful model placement was observed for this outcome-only source. Source configuration 500bpw-concurrency-mtp-2-c7; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation (actual allocation unknown); this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c7
Source 8 · 2026-10-15
- Source run ID
- trpro-exl-qwen38-27b-5bpw-gpu1-5bpw-mtp-2-concurrency-8-20261015
- Page entry ID
- trpro-exl-qwen38-27b-5bpw-gpu1-mtp-2
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.8 27B EXL3 5.00bpw · MTP-2
- Configuration label
- Single RTX3090 requested — placement unverified · EXL3 5.00bpw · MTP-2 · concurrency 8, 8192-token prefix, requested 8960-token allocation (actual unknown) per stream
- Measured date
- 2026-10-15
- Hardware
- Not observed (outcome-only startup failure)
- Runtime
- exllamav3 · ExLlamaV3 1.5.0+cu128.torch2.9.0; native extension SHA-256 c1d02b410ef729618077ba287f01f05b229f31d118a7aae04a1b7e56142e3b2e
- Weight placement
- Unverified (startup failed)
- Observed devices
- None observed
- Requested devices
- CUDA0
- Split mode
- None
- KV-cache precision
- EXL native K8/V8 with QSA side planes and recurrent state
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- ExLlamaV3 persistent worker time.perf_counter_ns with torch.cuda.synchronize at phase boundaries; fresh PP excludes prefix preparation; aggregate TG is across streams
- Measurement outcomes
- 8,192 tokens · 8 streams: oom — RuntimeError: Insufficient VRAM in split for model and cache
- Notes
- Single RTX3090 requested — placement unverified · EXL3 5.00bpw · MTP-2 · concurrency 8, 8192-token prefix, requested 8960-token allocation (actual unknown) per stream. Persistent EXL native worker: 512 fresh prompt tokens and 256 generated tokens; one warmup then three measured repetitions. PP is fresh-prompt processing after the populated prefix; TG is aggregate decode across concurrent streams. Throughput means and sample standard deviations are recomputed from the retained raw repetitions. EXL native K8/V8 cache storage includes QSA side planes and recurrent state; this is not GGUF Q8_0. Physical GPU 1 (GPU-0a2bed40-39e4-e67f-5234-868704f9704b) was selected as the single device. No successful model placement was observed for this outcome-only source. Source configuration 500bpw-concurrency-mtp-2-c8; model revision a35e75a73baee51da709329d19294245cbeeb5d8; source manifest SHA-256 f889b720ea21d731842579e93d41eb5e4ed995a63475a7e69874a2330b529192. The concurrency source used an 8192-token prefix and a fixed 8960-token per-stream cache allocation (actual allocation unknown); this allocation is independent of context-fit capacities.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/remaining-campaign/trpro-exl-qwen38-27b-5bpw-gpu1-independent-gpus-validation-20261015/5.00bpw/concurrency/mtp-2-c8