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 1
PP · left axisTG · right axis
View exact measurements
| Context depth (tokens) | Fresh prefill PP (tok/s) | TG (tok/s) |
|---|---|---|
| 0 | 8,728.73 | 243.68 |
| 2,048 | 7,651.99 | 231.02 |
| 4,096 | 7,361.60 | 232.17 |
| 8,192 | 6,685.24 | 219.66 |
| 16,384 | 5,600.70 | 206.06 |
| 32,768 | 4,298.10 | 187.32 |
| 65,536 | 2,904.62 | 155.97 |
| 130,240 | 1,779.99 | 116.11 |
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 | 6,195.60 | 343.42 | 171.71 | 919.60 |
| 3 | 8,192 | 5,569.91 | 407.16 | 135.72 | 1,115.23 |
| 4 | 8,192 | 5,546.22 | 448.73 | 112.18 | 1,244.56 |
| 5 | 8,192 | 5,386.44 | 473.04 | 94.61 | 1,325.29 |
| 6 | 8,192 | 5,299.56 | 501.02 | 83.50 | 1,415.03 |
| 7 | 8,192 | 5,154.86 | 521.91 | 74.56 | 1,482.32 |
| 8 | 8,192 | 5,100.78 | 536.94 | 67.12 | 1,529.58 |
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-1-c1
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 8,728.73 t/s (min 8,654.64, max 8,771.18, SD 52.57, n=3); TG mean 243.68 t/s (min 243.38, max 244.06, SD 0.28, n=3); All mean 672.43 t/s (min 671.81, max 673.62, SD 0.84, n=3) · KV 2,048, 1 stream: PP mean 7,651.99 t/s (min 7,617.57, max 7,689.30, SD 29.36, n=3); TG mean 231.02 t/s (min 228.22, max 235.17, SD 2.99, n=3); All mean 636.14 t/s (min 629.24, max 646.55, SD 7.49, n=3) · KV 4,096, 1 stream: PP mean 7,361.60 t/s (min 7,317.42, max 7,401.84, SD 34.58, n=3); TG mean 232.17 t/s (min 231.58, max 232.71, SD 0.46, n=3); All mean 636.83 t/s (min 634.41, max 638.51, SD 1.75, n=3) · KV 8,192, 1 stream: PP mean 6,685.24 t/s (min 6,572.95, max 6,761.66, SD 81.11, n=3); TG mean 219.66 t/s (min 214.62, max 222.98, SD 3.62, n=3); All mean 601.33 t/s (min 588.27, max 609.79, SD 9.37, n=3) · KV 16,384, 1 stream: PP mean 5,600.70 t/s (min 5,566.06, max 5,666.10, SD 46.27, n=3); TG mean 206.06 t/s (min 201.33, max 210.46, SD 3.73, n=3); All mean 560.04 t/s (min 548.11, max 571.14, SD 9.42, n=3) · KV 32,768, 1 stream: PP mean 4,298.10 t/s (min 4,276.00, max 4,312.05, SD 15.80, n=3); TG mean 187.32 t/s (min 185.21, max 188.76, SD 1.52, n=3); All mean 502.52 t/s (min 497.01, max 506.50, SD 4.02, n=3) · KV 65,536, 1 stream: PP mean 2,904.62 t/s (min 2,851.88, max 2,933.93, SD 37.37, n=3); TG mean 155.97 t/s (min 155.83, max 156.26, SD 0.20, n=3); All mean 410.63 t/s (min 409.39, max 411.86, SD 1.01, n=3) · KV 130,240, 1 stream: PP mean 1,779.99 t/s (min 1,772.64, max 1,786.54, SD 5.70, n=3); TG mean 116.11 t/s (min 115.93, max 116.43, SD 0.23, n=3); All mean 299.36 t/s (min 298.87, max 300.26, SD 0.64, 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-1, 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-1-c2
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 6,195.60 t/s (min 6,177.86, max 6,212.46, SD 14.14, n=3); TG mean 343.42 t/s (min 341.32, max 344.56, SD 1.48, n=3); All mean 919.60 t/s (min 913.44, max 922.88, SD 4.36, n=3)
- Measurement outcomes
- 8,192 tokens · 2 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c3
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,569.91 t/s (min 5,536.67, max 5,599.86, SD 25.90, n=3); TG mean 407.16 t/s (min 405.16, max 409.64, SD 1.86, n=3); All mean 1,115.23 t/s (min 1,109.35, max 1,121.99, SD 5.20, n=3)
- Measurement outcomes
- 8,192 tokens · 3 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c4
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,546.22 t/s (min 5,370.91, max 5,655.71, SD 125.24, n=3); TG mean 448.73 t/s (min 448.35, max 449.00, SD 0.28, n=3); All mean 1,244.56 t/s (min 1,240.20, max 1,248.09, SD 3.27, n=3)
- Measurement outcomes
- 8,192 tokens · 4 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c5
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,386.44 t/s (min 5,287.71, max 5,449.31, SD 70.68, n=3); TG mean 473.04 t/s (min 470.71, max 474.95, SD 1.76, n=3); All mean 1,325.29 t/s (min 1,323.26, max 1,328.52, SD 2.31, n=3)
- Measurement outcomes
- 8,192 tokens · 5 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c6
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,299.56 t/s (min 5,270.77, max 5,320.27, SD 21.00, n=3); TG mean 501.02 t/s (min 498.05, max 503.42, SD 2.23, n=3); All mean 1,415.03 t/s (min 1,406.16, max 1,419.90, SD 6.28, n=3)
- Measurement outcomes
- 8,192 tokens · 6 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c7
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,154.86 t/s (min 5,128.29, max 5,183.03, SD 22.38, n=3); TG mean 521.91 t/s (min 520.93, max 523.32, SD 1.02, n=3); All mean 1,482.32 t/s (min 1,479.40, max 1,485.15, SD 2.35, n=3)
- Measurement outcomes
- 8,192 tokens · 7 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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-1-c8
- Page entry ID
- gemma-4-e2b-ud-q8-k-xl-single-gpu-20261016-mtp-1-3c6cd05243
- 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 1
- 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 5,100.78 t/s (min 5,084.74, max 5,120.19, SD 14.67, n=3); TG mean 536.94 t/s (min 528.64, max 541.71, SD 5.89, n=3); All mean 1,529.58 t/s (min 1,506.22, max 1,543.18, SD 16.59, n=3)
- Measurement outcomes
- 8,192 tokens · 8 streams: measured
- Notes
- Automated llama-server KV-depth sweep (mtp-1, 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