32 GB · 25 reports

V100-PCIE-32GB · 32 GB

Qualified results first; original reports and full tables stay below.

All tasks · Qualified results

Speeds compare only within one exact test. Recipe availability is checked on each report against the published recipe executor.

Qwen3-Embedding-4B

Qwen · 2 tested configurations

fp16 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s

Throughput · Batch 1: 40.9vectors/sec
Best measured · Batch 16: 228vectors/sec
Observed GPU occupancy: 8.2GiB
P95 latency: 25 ms

Use this recipe · Batch 16 ↗Evidence report and run availability ↗Same test on other GPUs

Test details · 3 batch sizes

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

Engine: sentence-transformers · PyTorch 2.12.1+cu126
Protocol: native-forward-v1
Dataset: embedding-mteb-tasks-v1
Workload: embedding

Batch size	Mean throughput (vectors/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	40.94	25 ms	—	8.25	123.4	1	cd3861d6 ↗
B4	100.23	40 ms	—	8.25	146.9	1	cd3861d6 ↗
B16	228.35	70 ms	—	8.25	180.4	1	cd3861d6 ↗

1 other test configurations on this GPU

fp16 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 9 s

Throughput · Batch 1: 41vectors/sec
Best measured · Batch 16: 229vectors/sec
Observed GPU occupancy: 8.2GiB
P95 latency: 25 ms

Use this recipe · Batch 16 ↗Evidence report and run availability ↗Same test on other GPUs

Test details · 3 batch sizes

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

Engine: sentence-transformers · PyTorch 2.12.1+cu126
Protocol: native-forward-v1
Dataset: embedding-mteb-tasks-v1
Workload: embedding

Batch size	Mean throughput (vectors/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	40.96	25 ms	—	8.25	119.3	1	65a1050f ↗
B4	100.37	40 ms	—	8.25	149.9	1	65a1050f ↗
B16	228.92	70 ms	—	8.25	178.1	1	65a1050f ↗

Qwen3-Embedding-0.6B

Qwen · 1 tested configuration

fp16 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s

Throughput · Batch 1: 95.3vectors/sec
Best measured · Batch 16: 771vectors/sec
Observed GPU occupancy: 1.7GiB
P95 latency: 11 ms

Use this recipe · Batch 16 ↗Evidence report and run availability ↗Same test on other GPUs

Test details · 3 batch sizes

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

Engine: sentence-transformers · PyTorch 2.12.1+cu126
Protocol: native-forward-v1
Dataset: embedding-mteb-tasks-v1
Workload: embedding

Batch size	Mean throughput (vectors/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	95.33	11 ms	—	1.75	70.8	1	f91802d1 ↗
B4	308.67	13 ms	—	1.75	90.1	1	f91802d1 ↗
B16	771.26	21 ms	—	1.75	99.2	1	f91802d1 ↗

Kokoro-82M
hexgrad · 2 tested configurations
- fp32 · Text → Audio · kokoro-english-v1 · Test duration 6 s
  Throughput · Batch 1
  35.8audio-sec/sec
  Best measured · Batch 2
  37.6audio-sec/sec
  Observed GPU occupancy
  1.4GiB
  P95 latency
  105 ms
  Evidence report and run availability ↗Same test on other GPUs
  Test details · 2 batch sizes
  Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
  Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
  Engine
  kokoro · PyTorch 2.12.1+cu126
  Protocol
  audio-native-v1
  Dataset
  kokoro-english-v1
  Workload
  tts
  Batch size Mean throughput (audio-sec/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
  B1 35.82 105 ms — 1.45 76.2 10 b40f8533 ↗
  B2 37.6 239 ms — 1.45 82.3 10 b40f8533 ↗
- 1 other test configurations on this GPU
  - fp32 · Text → Audio · kokoro-english-v1 · Test duration 6 s
    Throughput · Batch 1
    37.2audio-sec/sec
    Best measured · Batch 2
    40.1audio-sec/sec
    Observed GPU occupancy
    1.4GiB
    P95 latency
    100 ms
    Evidence report and run availability ↗Same test on other GPUs
    Test details · 2 batch sizes
    Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
    Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
    Engine
    kokoro · PyTorch 2.12.1+cu126
    Protocol
    audio-native-v1
    Dataset
    kokoro-english-v1
    Workload
    tts
    Batch size Mean throughput (audio-sec/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
    B1 37.23 100 ms — 1.45 77.6 1 baa57457 ↗
    B2 40.1 218 ms — 1.45 85.3 1 baa57457 ↗
faster-whisper-tiny
Systran · 2 tested configurations
- fp16 · Audio → Text · asr_prompts · Test duration 5 s
  Throughput · Batch 1
  90.9audio-sec/sec
  Best measured · Batch 4
  132audio-sec/sec
  Observed GPU occupancy
  0.5GiB
  P95 latency
  Not recorded
  Evidence report and run availability ↗Same test on other GPUs
  Test details · 2 batch sizes
  Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
  Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
  Engine
  faster-whisper · PyTorch 2.12.1+cu126
  Protocol
  audio-native-v1
  Dataset
  asr_prompts
  Workload
  asr
  Batch size Mean throughput (audio-sec/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
  B1 90.94 — — 0.51 51.5 10 e4ef00c1 ↗
  B4 132.16 — — 0.52 51.2 10 e4ef00c1 ↗
- 1 other test configurations on this GPU
  - fp16 · Audio → Text · asr_prompts · Test duration 1 s
    Throughput · Batch 1
    105audio-sec/sec
    Best measured · Batch 4
    156audio-sec/sec
    Observed GPU occupancy
    0.5GiB
    P95 latency
    Not recorded
    Evidence report and run availability ↗Same test on other GPUs
    Test details · 2 batch sizes
    Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
    Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
    Engine
    faster-whisper · PyTorch 2.12.1+cu126
    Protocol
    audio-native-v1
    Dataset
    asr_prompts
    Workload
    asr
    Batch size Mean throughput (audio-sec/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
    B1 104.96 — — 0.51 51.5 1 64035c7b ↗
    B4 155.93 — — 0.52 50.1 1 64035c7b ↗

Batch size	Mean throughput (audio-sec/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	35.82	105 ms	—	1.45	76.2	10	b40f8533 ↗
B2	37.6	239 ms	—	1.45	82.3	10	b40f8533 ↗

Batch size	Mean throughput (audio-sec/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	37.23	100 ms	—	1.45	77.6	1	baa57457 ↗
B2	40.1	218 ms	—	1.45	85.3	1	baa57457 ↗

Batch size	Mean throughput (audio-sec/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	90.94	—	—	0.51	51.5	10	e4ef00c1 ↗
B4	132.16	—	—	0.52	51.2	10	e4ef00c1 ↗

Batch size	Mean throughput (audio-sec/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	104.96	—	—	0.51	51.5	1	64035c7b ↗
B4	155.93	—	—	0.52	50.1	1	64035c7b ↗

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Sortable tables by exact test condition · 17 rows

C: concurrency · B: batch · —: not measured

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 9 s · Batch 1 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	40.96	40.96	—	0.025	8.25	119.3	40.96	0.34	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 9 s · Batch 4 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	100.37	100.37	—	0.04	8.25	149.9	100.37	0.67	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 9 s · Batch 16 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	228.92	228.92	—	0.07	8.25	178.1	228.92	1.29	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 1 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	40.94	40.94	—	0.025	8.25	123.4	40.94	0.33	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 1 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-0.6BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	95.33	95.33	—	0.011	1.75	70.8	95.33	1.35	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 4 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	100.23	100.23	—	0.04	8.25	146.9	100.23	0.68	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 4 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-0.6BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	308.67	308.67	—	0.013	1.75	90.1	308.67	3.43	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 16 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-4BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	228.35	228.35	—	0.07	8.25	180.4	228.35	1.27	1 ↗

Embedding · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s · Batch 16 · vectors/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Qwen/Qwen3-Embedding-0.6BScaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	771.26	771.26	—	0.021	1.75	99.2	771.26	7.78	1 ↗

Speech recognition · Audio → Text · asr_prompts · Test duration 1 s · Batch 1 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Systran/faster-whisper-tinyScaling curves ↗	fp16	faster-whisper	asraudio-native-v1	B1	104.96	104.96	—	—	0.51	51.5	104.96	2.04	1 ↗

Speech recognition · Audio → Text · asr_prompts · Test duration 1 s · Batch 4 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Systran/faster-whisper-tinyScaling curves ↗	fp16	faster-whisper	asraudio-native-v1	B4	155.93	155.93	—	—	0.52	50.1	155.93	3.11	1 ↗

Speech recognition · Audio → Text · asr_prompts · Test duration 5 s · Batch 1 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Systran/faster-whisper-tinyScaling curves ↗	fp16	faster-whisper	asraudio-native-v1	B1	90.94	96.73	—	—	0.51	51.5	86.43	1.76	10 ↗

Speech recognition · Audio → Text · asr_prompts · Test duration 5 s · Batch 4 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
Systran/faster-whisper-tinyScaling curves ↗	fp16	faster-whisper	asraudio-native-v1	B4	132.16	138.53	—	—	0.52	51.2	123.89	2.58	10 ↗

Speech synthesis · Text → Audio · kokoro-english-v1 · Test duration 6 s · Batch 1 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
hexgrad/Kokoro-82MScaling curves ↗	fp32	kokoro	ttsaudio-native-v1	B1	35.82	37.2	—	0.105	1.45	76.2	31.11	0.47	10 ↗

Speech synthesis · Text → Audio · kokoro-english-v1 · Test duration 6 s · Batch 1 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
hexgrad/Kokoro-82MScaling curves ↗	fp32	kokoro	ttsaudio-native-v1	B1	37.23	37.23	—	0.1	1.45	77.6	37.23	0.48	1 ↗

Speech synthesis · Text → Audio · kokoro-english-v1 · Test duration 6 s · Batch 2 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
hexgrad/Kokoro-82MScaling curves ↗	fp32	kokoro	ttsaudio-native-v1	B2	37.6	39.92	—	0.239	1.45	82.3	29.89	0.46	10 ↗

Speech synthesis · Text → Audio · kokoro-english-v1 · Test duration 6 s · Batch 2 · audio-sec/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
hexgrad/Kokoro-82MScaling curves ↗	fp32	kokoro	ttsaudio-native-v1	B2	40.1	40.1	—	0.218	1.45	85.3	40.1	0.47	1 ↗

Scaling curves and individual runs

One model and test protocol per curve; different workloads are never joined into one line.

Qwen/Qwen3-Embedding-4B · fp16 · sentence-transformers · embedding · native-forward-v1

Mean throughput vectors/sec

Measured points only. Select a point to open its source report.

Mean run P95 latency s

Measured points only. Select a point to open its source report.

Mean GPU memory used GB

Measured points only. Select a point to open its source report.

Each point is the mean of recorded runs at that level. A source report may differ from the mean; all runs are listed below.

Runs behind the curve · Qwen/Qwen3-Embedding-4B · vectors/sec · 3 rows
Batch size	Throughput	Decode (tok/s)	Uncached prefill (tok/s)	GPU power (W)	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	Test date	View report
1	40.94 vectors/sec	—	Uncached prefill unavailableView report	123.45	—	0.025	8.25	10/8/2026, 4:17:03 PM	cd3861d6 ↗
4	100.23 vectors/sec	—	Uncached prefill unavailableView report	146.93	—	0.04	8.25	10/8/2026, 4:17:03 PM	cd3861d6 ↗
16	228.35 vectors/sec	—	Uncached prefill unavailableView report	180.42	—	0.07	8.25	10/8/2026, 4:17:03 PM	cd3861d6 ↗

Original reports

Every report for this task, including quality, diagnostic and legacy runs that do not enter comparisons.

Original reports · 25 rows
Model	Test task	Format	Engine	Workload	Protocol	Quality score	Test date
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 3:48:11 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 3:49:37 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 3:55:44 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 3:58:04 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:00:10 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:02:22 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:04:55 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:06:57 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:08:57 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:11:08 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:13:38 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:16:07 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:18:18 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:20:29 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:22:56 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:24:56 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:27:05 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:29:45 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:32:10 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:35:06 AM
faster-whisper-tiny ↗	Speech recognition	fp16	faster-whisper	asr	audio-native-v1	—	10/8/2026, 4:37:16 AM
Kokoro-82M ↗	Speech synthesis	fp32	kokoro	tts	audio-native-v1	—	10/8/2026, 4:39:35 AM
Qwen3-Embedding-4B ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/8/2026, 10:23:19 AM
Qwen3-Embedding-0.6B ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/8/2026, 3:50:00 PM
Qwen3-Embedding-4B ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/8/2026, 4:17:03 PM

Running benchmarks

0Updates automatically

No benchmarks are currently reported as running.

Methodology

Average of error-free runs for the selected recorded configuration. Missing methods and fallback runs are not ranked. Performance index = average / fastest matching hardware × 100. Value index = performance per purchase cost / best comparable value × 100. Quality is separate by dataset and protocol. Prices are dated seller listings, not completed transactions or market averages.

Updated 11/5/2026, 11:07:29 PM

All tasks · Qualified results

Qwen3-Embedding-4B

Qwen3-Embedding-0.6B

Kokoro-82M

faster-whisper-tiny

Original reports

Running benchmarks