Gemma 4 E2B
google/gemma-4-E2B-it
Google · 4 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 · llama.cpp
UD-Q8_K_XL · MTP 2
PP · left axisTG · right axis
View exact measurements
| Context depth (tokens) | Fresh prefill PP (tok/s) | TG (tok/s) |
|---|---|---|
| 0 | 8,737.11 | 303.45 |
| 2,048 | 7,823.18 | 296.97 |
| 4,096 | 7,370.16 | 285.59 |
| 8,192 | 6,687.51 | 275.05 |
| 16,384 | 5,634.61 | 257.86 |
| 32,768 | 4,293.49 | 229.06 |
| 65,536 | 2,905.51 | 186.85 |
| 130,240 | 1,772.35 | 139.38 |
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 | 6,132.83 | 435.00 | 217.50 | 1,130.45 |
| 3 | 8,192 | 5,785.21 | 484.88 | 161.63 | 1,280.78 |
| 4 | 8,192 | 5,223.09 | 519.38 | 129.85 | 1,424.84 |
| 5 | 8,192 | 5,141.09 | 566.45 | 113.29 | 1,553.32 |
| 6 | 8,192 | 4,877.40 | 579.75 | 96.62 | 1,619.20 |
| 7 | 8,192 | 4,861.23 | 590.99 | 84.43 | 1,665.00 |
| 8 | 8,192 | 4,746.11 | 607.50 | 75.94 | 1,719.45 |
Technical record and provenance
Source 1 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-context-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c1
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- context
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 131,072 tokens
- Actual context per slot
- 131,072 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 0, 1 stream: PP mean 8,737.11 t/s (min 8,727.37, max 8,743.77, SD 7.04, n=3); TG mean 303.45 t/s (min 301.37, max 305.09, SD 1.55, n=3); All mean 821.01 t/s (min 816.31, max 825.19, SD 3.64, n=3) · KV 2,048, 1 stream: PP mean 7,823.18 t/s (min 7,799.77, max 7,850.35, SD 20.82, n=3); TG mean 296.97 t/s (min 296.75, max 297.30, SD 0.24, n=3); All mean 798.55 t/s (min 797.43, max 799.69, SD 0.92, n=3) · KV 4,096, 1 stream: PP mean 7,370.16 t/s (min 7,349.25, max 7,387.31, SD 15.77, n=3); TG mean 285.59 t/s (min 283.89, max 286.83, SD 1.24, n=3); All mean 767.00 t/s (min 764.14, max 770.00, SD 2.40, n=3) · KV 8,192, 1 stream: PP mean 6,687.51 t/s (min 6,634.27, max 6,730.55, SD 39.96, n=3); TG mean 275.05 t/s (min 271.87, max 277.43, SD 2.34, n=3); All mean 735.66 t/s (min 727.67, max 740.84, SD 5.73, n=3) · KV 16,384, 1 stream: PP mean 5,634.61 t/s (min 5,588.48, max 5,690.53, SD 42.23, n=3); TG mean 257.86 t/s (min 256.74, max 259.96, SD 1.49, n=3); All mean 684.01 t/s (min 680.73, max 688.63, SD 3.36, n=3) · KV 32,768, 1 stream: PP mean 4,293.49 t/s (min 4,269.66, max 4,312.31, SD 17.77, n=3); TG mean 229.06 t/s (min 227.72, max 230.60, SD 1.18, n=3); All mean 598.89 t/s (min 596.13, max 603.54, SD 3.31, n=3) · KV 65,536, 1 stream: PP mean 2,905.51 t/s (min 2,883.00, max 2,935.83, SD 22.26, n=3); TG mean 186.85 t/s (min 183.18, max 189.15, SD 2.62, n=3); All mean 479.29 t/s (min 471.02, max 485.20, SD 6.02, n=3) · KV 130,240, 1 stream: PP mean 1,772.35 t/s (min 1,756.50, max 1,781.90, SD 11.29, n=3); TG mean 139.38 t/s (min 139.20, max 139.57, SD 0.15, n=3); All mean 349.08 t/s (min 347.57, max 349.92, SD 1.07, n=3)
- Measurement outcomes
- 0 tokens · 1 stream: measured; 2,048 tokens · 1 stream: measured; 4,096 tokens · 1 stream: measured; 8,192 tokens · 1 stream: measured; 16,384 tokens · 1 stream: measured; 32,768 tokens · 1 stream: measured; 65,536 tokens · 1 stream: measured; 130,240 tokens · 1 stream: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=1). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/e2b-context-gpu1
Source 2 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c2
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 2 streams: PP mean 6,132.83 t/s (min 5,969.75, max 6,218.99, SD 115.37, n=3); TG mean 435.00 t/s (min 433.65, max 436.41, SD 1.13, n=3); All mean 1,130.45 t/s (min 1,124.21, max 1,137.09, SD 5.27, n=3)
- Measurement outcomes
- 8,192 tokens · 2 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=2). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 3 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c3
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 3 streams: PP mean 5,785.21 t/s (min 5,635.95, max 6,037.54, SD 179.42, n=3); TG mean 484.88 t/s (min 471.70, max 504.80, SD 14.33, n=3); All mean 1,280.78 t/s (min 1,266.61, max 1,293.23, SD 10.93, n=3)
- Measurement outcomes
- 8,192 tokens · 3 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=3). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 4 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c4
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 4 streams: PP mean 5,223.09 t/s (min 5,075.42, max 5,358.36, SD 115.84, n=3); TG mean 519.38 t/s (min 506.08, max 527.46, SD 9.48, n=3); All mean 1,424.84 t/s (min 1,391.79, max 1,448.55, SD 24.10, n=3)
- Measurement outcomes
- 8,192 tokens · 4 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=4). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 5 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c5
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 5 streams: PP mean 5,141.09 t/s (min 5,017.50, max 5,244.54, SD 93.78, n=3); TG mean 566.45 t/s (min 561.74, max 575.70, SD 6.54, n=3); All mean 1,553.32 t/s (min 1,544.79, max 1,560.68, SD 6.54, n=3)
- Measurement outcomes
- 8,192 tokens · 5 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=5). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 6 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c6
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 6 streams: PP mean 4,877.40 t/s (min 4,813.90, max 4,934.42, SD 49.42, n=3); TG mean 579.75 t/s (min 570.49, max 584.57, SD 6.55, n=3); All mean 1,619.20 t/s (min 1,595.08, max 1,632.92, SD 17.11, n=3)
- Measurement outcomes
- 8,192 tokens · 6 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=6). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 7 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c7
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 7 streams: PP mean 4,861.23 t/s (min 4,832.35, max 4,908.64, SD 33.79, n=3); TG mean 590.99 t/s (min 570.65, max 605.39, SD 14.79, n=3); All mean 1,665.00 t/s (min 1,609.52, max 1,707.27, SD 40.99, n=3)
- Measurement outcomes
- 8,192 tokens · 7 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=7). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1
Source 8 · 2026-10-15
- Source run ID
- pub-20261016-trpro-gemma-e2b-q8-k-xl-concurrency-2026-10-14-single-gpu-isolated-gpu1-co-scheduled-mtp-2-c8
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-2-af408889ad
- Entry sweep
- concurrency
- Raw entry label
- Gemma 4 E2B UD-Q8_K_XL — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 131,072 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- MTP · draft max 2
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 8 streams: PP mean 4,746.11 t/s (min 4,712.90, max 4,765.01, SD 23.56, n=3); TG mean 607.50 t/s (min 594.73, max 615.29, SD 9.11, n=3); All mean 1,719.45 t/s (min 1,686.24, max 1,737.80, SD 23.53, n=3)
- Measurement outcomes
- 8,192 tokens · 8 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-2, concurrency=8). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-15-independent-gpus/gemma-e2b-q8-k-xl-concurrency-physical1