96 GB · 196 reports

RTX PRO 6000 Blackwell Server Edition · 96 GB

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

Decision · Qualified results

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

clef-flash

Cloudflare · 2 tested configurations

bf16 · Text → Structured data · typed-invoices-v1 · Test duration 6 s
Throughput · Batch 16
47.7decisions/sec
Observed GPU occupancy
19.5GiB
P95 latency
338 ms
Use this recipe · Batch 16 ↗Evidence report and run availability ↗Same test on other GPUs
Test details
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
clef-native · PyTorch 2.11.0+cu130
Protocol
native-forward-v1
Dataset
typed-invoices-v1
Workload
decision
Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
B16 47.71 338 ms — 19.55 499.5 2 82d98b36 ↗

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B16	47.71	338 ms	—	19.55	499.5	2	82d98b36 ↗

1 other test configurations on this GPU

bf16 · Text → Structured data · typed-invoices-v1 · Test duration 33 s

Throughput · Batch 1: 31.8decisions/sec
Best measured · Batch 4: 39.3decisions/sec
Observed GPU occupancy: 18.9GiB
P95 latency: 32 ms

Use this recipe · Batch 4 ↗Evidence report and run availability ↗Same test on other GPUs

Test details · 4 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: clef-native · PyTorch 2.11.0
Protocol: native-forward-v1
Dataset: typed-invoices-v1
Workload: decision

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	31.83	32 ms	—	18.95	388.5	1	10c64aff ↗
B4	39.27	93 ms	—	18.95	449.5	1	10c64aff ↗
B16	5.41	8.21 s	—	19.28	233.4	1	10c64aff ↗
B64	15.43	9.66 s	—	21.89	324.3	1	10c64aff ↗

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

Sortable tables by exact test condition · 5 rows

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

Decision · Text → Structured data · typed-invoices-v1 · Test duration 6 s · Batch 16 · decisions/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
Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B16	47.71	47.74	—	0.338	19.55	499.5	47.67	0.1	2 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 33 s · Batch 1 · decisions/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
Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B1	31.83	31.83	—	0.032	18.95	388.5	31.83	0.08	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 33 s · Batch 4 · decisions/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
Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B4	39.27	39.27	—	0.093	18.95	449.5	39.27	0.09	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 33 s · Batch 16 · decisions/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
Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B16	5.41	5.41	—	8.208	19.28	233.4	5.41	0.02	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 33 s · Batch 64 · decisions/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
Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B64	15.43	15.43	—	9.663	21.89	324.3	15.43	0.05	1 ↗

Scaling curves and individual runs

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

Cloudflare/clef-flash · bf16 · clef-native · decision · native-forward-v1

Mean throughput decisions/sec

One measured point; more levels are needed for a curve.

Mean run P95 latency s

One measured point; more levels are needed for a curve.

Mean GPU memory used GB

One measured point; more levels are needed for a curve.

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 · Cloudflare/clef-flash · decisions/sec · 2 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
16	47.74 decisions/sec	—	Uncached prefill unavailableView report	499.04	—	0.336	19.65	10/23/2026, 12:06:35 PM	82d98b36 ↗
16	47.67 decisions/sec	—	Uncached prefill unavailableView report	499.9	—	0.34	19.45	10/23/2026, 12:02:11 PM	66fc25b0 ↗

Original reports

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

Original reports · 5 rows
Model	Test task	Format	Engine	Workload	Protocol	Quality score	Test date
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/23/2026, 12:06:35 PM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/23/2026, 12:02:11 PM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/23/2026, 9:04:07 AM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/14/2026, 8:26:33 AM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/14/2026, 6:35:30 AM

Running benchmarks

0Updates automatically

No benchmarks are currently reported as running.

RTX PRO 6000 Blackwell Server Edition · 96 GB · 403 queued · 0 running · 0 awaiting reassignment · 6 succeeded · 3 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 11/6/2026, 12:34:56 PM