16 GB · 37 reports

T4 · 16 GB

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

Embedding · Qualified results

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

embeddinggemma-2

unsloth · 2 tested configurations

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

Throughput · Batch 1: 109vectors/sec
Best measured · Batch 16: 1,002vectors/sec
Observed GPU occupancy: 0.9GiB
P95 latency: 9 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	108.76	9 ms	—	0.89	50.3	2	cfe051ff ↗
B4	365.82	11 ms	—	0.89	58.1	2	cfe051ff ↗
B16	1,002.27	16 ms	—	0.89	63.9	2	cfe051ff ↗

1 other test configurations on this GPU

nf4 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s

Throughput · Batch 1: 90.1vectors/sec
Best measured · Batch 16: 809vectors/sec
Observed GPU occupancy: 0.7GiB
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	90.07	11 ms	—	0.75	56.2	2	2f9465c4 ↗
B4	302.9	13 ms	—	0.75	61.5	2	2f9465c4 ↗
B16	809.43	20 ms	—	0.75	64	2	2f9465c4 ↗

embeddinggemma-2

google · 2 tested configurations

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

Throughput · Batch 1: 109vectors/sec
Best measured · Batch 16: 1,004vectors/sec
Observed GPU occupancy: 0.9GiB
P95 latency: 9 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	108.65	9 ms	—	0.89	49.3	2	2846de88 ↗
B4	365.65	11 ms	—	0.89	57.9	2	2846de88 ↗
B16	1,004.11	16 ms	—	0.89	63.9	2	2846de88 ↗

1 other test configurations on this GPU

nf4 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s

Throughput · Batch 1: 90vectors/sec
Best measured · Batch 16: 815vectors/sec
Observed GPU occupancy: 0.7GiB
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	90.02	11 ms	—	0.75	54.6	2	43c67a74 ↗
B4	303.35	13 ms	—	0.75	60.1	2	43c67a74 ↗
B16	814.64	20 ms	—	0.75	64.1	2	43c67a74 ↗

bge-small-zh-v1.5

BAAI · 1 tested configuration

nf4 · Text → Vectors · embedding-mteb-tasks-v1 · Test duration 90 s

Throughput · Batch 1: 1,170vectors/sec
Best measured · Batch 16: 5,188vectors/sec
Observed GPU occupancy: 0.3GiB
P95 latency: 1 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	1,169.81	1 ms	—	0.29	39	1	0fb03772 ↗
B4	2,301.06	2 ms	—	0.29	32.5	1	0fb03772 ↗
B16	5,188.01	3 ms	—	0.29	32.5	1	0fb03772 ↗

granite-embedding-97m-multilingual-r2

ibm-granite · 1 tested configuration

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

Throughput · Batch 1: 464vectors/sec
Best measured · Batch 16: 4,084vectors/sec
Observed GPU occupancy: 0.5GiB
P95 latency: 2 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	464.26	2 ms	—	0.47	45.4	1	37c260e4 ↗
B4	1,713.26	2 ms	—	0.47	49.4	1	37c260e4 ↗
B16	4,083.56	4 ms	—	0.47	38.3	1	37c260e4 ↗

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

Sortable tables by exact test condition · 18 rows

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

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
unsloth/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	108.76	108.81	—	0.009	0.89	50.3	108.71	2.16	2 ↗

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
unsloth/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B1	90.07	90.08	—	0.011	0.75	56.2	90.06	1.6	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	108.65	108.66	—	0.009	0.89	49.3	108.64	2.2	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B1	90.02	90.07	—	0.011	0.75	54.6	89.97	1.65	2 ↗

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
BAAI/bge-small-zh-v1.5Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B1	1,169.81	1,169.81	—	0.001	0.29	39	1,169.81	29.97	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
ibm-granite/granite-embedding-97m-multilingual-r2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B1	464.26	464.26	—	0.002	0.47	45.4	464.26	10.22	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
unsloth/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	365.82	366	—	0.011	0.89	58.1	365.65	6.3	2 ↗

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
unsloth/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B4	302.9	302.94	—	0.013	0.75	61.5	302.87	4.93	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	365.65	365.65	—	0.011	0.89	57.9	365.65	6.32	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B4	303.35	303.62	—	0.013	0.75	60.1	303.08	5.05	2 ↗

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
BAAI/bge-small-zh-v1.5Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B4	2,301.06	2,301.06	—	0.002	0.29	32.5	2,301.06	70.76	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
ibm-granite/granite-embedding-97m-multilingual-r2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B4	1,713.26	1,713.26	—	0.002	0.47	49.4	1,713.26	34.7	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
unsloth/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	1,002.27	1,002.27	—	0.016	0.89	63.9	1,002.27	15.67	2 ↗

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
unsloth/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B16	809.43	809.77	—	0.02	0.75	64	809.1	12.65	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	1,004.11	1,006.76	—	0.016	0.89	63.9	1,001.46	15.7	2 ↗

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
google/embeddinggemma-2Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B16	814.64	817.98	—	0.02	0.75	64.1	811.3	12.7	2 ↗

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
BAAI/bge-small-zh-v1.5Scaling curves ↗	nf4	sentence-transformers	embeddingnative-forward-v1	B16	5,188.01	5,188.01	—	0.003	0.29	32.5	5,188.01	159.61	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
ibm-granite/granite-embedding-97m-multilingual-r2Scaling curves ↗	fp16	sentence-transformers	embeddingnative-forward-v1	B16	4,083.56	4,083.56	—	0.004	0.47	38.3	4,083.56	106.72	1 ↗

Scaling curves and individual runs

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

unsloth/embeddinggemma-2 · 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 · unsloth/embeddinggemma-2 · vectors/sec · 6 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	108.81 vectors/sec	—	Uncached prefill unavailableView report	50.36	—	0.009	0.89	10/19/2026, 6:35:05 AM	76c86f7b ↗
1	108.71 vectors/sec	—	Uncached prefill unavailableView report	50.17	—	0.009	0.89	10/19/2026, 6:55:37 AM	cfe051ff ↗
4	366 vectors/sec	—	Uncached prefill unavailableView report	58.58	—	0.011	0.89	10/19/2026, 6:35:05 AM	76c86f7b ↗
4	365.65 vectors/sec	—	Uncached prefill unavailableView report	57.65	—	0.011	0.89	10/19/2026, 6:55:37 AM	cfe051ff ↗
16	1,002.27 vectors/sec	—	Uncached prefill unavailableView report	63.7	—	0.016	0.89	10/19/2026, 6:55:37 AM	cfe051ff ↗
16	1,002.27 vectors/sec	—	Uncached prefill unavailableView report	64.2	—	0.016	0.89	10/19/2026, 6:35:05 AM	76c86f7b ↗

Original reports

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

Original reports · 14 rows
Model	Test task	Format	Engine	Workload	Protocol	Quality score	Test date
bge-small-zh-v1.5 ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/18/2026, 5:50:19 PM
embeddinggemma-2 ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:55:37 AM
embeddinggemma-2 ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:50:06 AM
embeddinggemma-2 ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:45:33 AM
embeddinggemma-2 ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:40:07 AM
embeddinggemma-2 ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:35:05 AM
embeddinggemma-2 ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:29:40 AM
embeddinggemma-2 ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:24:52 AM
embeddinggemma-2 ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 6:18:38 AM
granite-embedding-97m-multilingual-r2 ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/18/2026, 5:47:32 PM
WeMM-Embedding-2B ↗	Embedding	fp16	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 7:05:43 AM
WeMM-Embedding-2B ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 7:00:33 AM
WeMM-Embedding-4B ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 7:11:34 AM
WeMM-Embedding-9B ↗	Embedding	nf4	sentence-transformers	embedding	native-forward-v1	—	10/19/2026, 7:22:08 AM

Running benchmarks

0Updates automatically

No benchmarks are currently reported as running.

T4 · 16 GB · 22 queued · 0 running · 0 awaiting reassignment · 3 succeeded · 8 failed or exhausted

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, 10:57:57 PM