Datasets:

Intelion
/

gpuark-gpu-dataset

gpu_id int64 195 11.9k	test_name stringlengths 8 106	score float64 0.41 180B	unit stringclasses 28 values	source__name stringclasses 3 values
10,819	Hashcat: Benchmark: SHA-256	9,853,600,000	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: NTLM	10,387,833,333	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: SHA-512	250,700,000	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: 7-Zip	80,162	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: LastPass	3,006	H/s	OpenBenchmarking: Hashcat
10,335	Llama.cpp: Backend: CPU BLAS - Model: DeepSeek-R1-Distill-Llama-8B-Q8_0 - Test: Prompt Processing 2048	87.46	Tokens Per Second	OpenBenchmarking: Llama.cpp
10,335	Llama.cpp: Backend: CPU BLAS - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128	15.67	Tokens Per Second	OpenBenchmarking: Llama.cpp
10,335	Llama.cpp: Backend: CPU BLAS - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512	357.2	Tokens Per Second	OpenBenchmarking: Llama.cpp
3,379	Hashcat: Benchmark: SHA1	1,965,200,000	H/s	OpenBenchmarking: Hashcat
10,335	vLLM CPU: Test: deepseek-moe-16b-chat Latency	23.09	99p Latency - Seconds	OpenBenchmarking: Llama.cpp
3,379	Hashcat: Benchmark: MD5	6,116,766,667	H/s	OpenBenchmarking: Hashcat
10,335	Llama.cpp: Backend: CPU BLAS - Model: MiniMax-M2.5-UD-TQ1_0 - Test: Prompt Processing 1024	38.24	Tokens Per Second	OpenBenchmarking: Llama.cpp
3,379	Hashcat: Benchmark: SHA-256	450,900,000	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: TrueCrypt RIPEMD160 + XTS	69,255	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: Kerberos 5 + etype 23 + TGS-REP	93,845,167	H/s	OpenBenchmarking: Hashcat
3,379	Hashcat: Benchmark: LUKS v2 argon2id + SHA-256 + AES	2	H/s	OpenBenchmarking: Hashcat
1,094	Hashcat: Benchmark: WPA-PBKDF2-PMKID+EAPOL	240,200	H/s	OpenBenchmarking: Hashcat
10,335	Hashcat: Benchmark: SHA-256	604,199,142	H/s	OpenBenchmarking: Hashcat
10,335	vLLM CPU: Test: Hermes-3-Llama-3.2-3B Latency	9.92	99p Latency - Seconds	OpenBenchmarking: Llama.cpp
8,929	Llama.cpp: Backend: Vulkan - Model: MiniMax-M2.5-UD-TQ1_0 - Test: Prompt Processing 512	225.25	Tokens Per Second	OpenBenchmarking: Llama.cpp
10,335	Llama.cpp: Backend: CPU BLAS - Model: DeepSeek-R1-Distill-Llama-8B-Q8_0 - Test: Prompt Processing 1024	84.71	Tokens Per Second	OpenBenchmarking: Llama.cpp
8,929	Llama.cpp: Backend: AMD ROCm HIP - Model: gpt-oss-20b-Q8_0 - Test: Prompt Processing 512	1,726.09	Tokens Per Second	OpenBenchmarking: Llama.cpp
8,929	Llama.cpp: Backend: CPU BLAS - Model: GLM-4.7-Flash-IQ4_XS - Test: Prompt Processing 512	90.28	Tokens Per Second	OpenBenchmarking: Llama.cpp
8,929	Llama.cpp: Backend: Vulkan - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128	26.17	Tokens Per Second	OpenBenchmarking: Llama.cpp
3,379	Hashcat: Benchmark: bcrypt	5,430	H/s	OpenBenchmarking: Hashcat
10,910	vkpeak: fp32-scalar	34,789.41	GFLOPS	OpenBenchmarking: Hashcat
10,910	vkpeak: fp32-vec4	34,412.58	GFLOPS	OpenBenchmarking: Hashcat
10,910	vkpeak: fp16-scalar	25,965.34	GFLOPS	OpenBenchmarking: Hashcat
10,910	vkpeak: fp16-vec4	25,822.58	GFLOPS	OpenBenchmarking: Hashcat
10,910	vkpeak: fp16-matrix	205,769.65	GFLOPS	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: NTLM	179,533,333,333	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: SHA1	38,457,866,667	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: 7-Zip	1,767,267	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: bcrypt	133,400	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: SHA-256	13,393,933,333	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: SHA-512	4,443,333,333	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: LastPass	55,862	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: WPA-PBKDF2-PMKID+EAPOL	1,761,767	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: TrueCrypt RIPEMD160 + XTS	1,281,400	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: Kerberos 5 + etype 23 + TGS-REP	2,110,933,333	H/s	OpenBenchmarking: Hashcat
10,761	Hashcat: Benchmark: LUKS v2 argon2id + SHA-256 + AES	12	H/s	OpenBenchmarking: Hashcat
5,057	LLM Inference: LLM Perf: Phi-3-mini-4k-instruct (awq)	24.443189	tokens/s	LLM Inference
10,938	LLM Inference: Geekbench AI: Geekbench AI Quantized (int8)	33,825	points	LLM Inference
5,057	LLM Inference: LLM Perf: Qwen1.5-7B (gptq)	28.810004	tokens/s	LLM Inference
1,814	LLM Inference: llama.cpp Bench: llama-2-7b-Q4_0 (Q4_0)	9.49	tokens/s	LLM Inference
5,057	LLM Inference: LLM Perf: Meta-Llama-3-8B-Instruct (awq)	22.274933	tokens/s	LLM Inference
10,910	Hashcat: Benchmark: SHA-512	4,112,433,333	H/s	OpenBenchmarking: Hashcat
10,910	Hashcat: Benchmark: LastPass	47,630	H/s	OpenBenchmarking: Hashcat
10,910	Hashcat: Benchmark: WPA-PBKDF2-PMKID+EAPOL	1,400,167	H/s	OpenBenchmarking: Hashcat
10,910	Hashcat: Benchmark: TrueCrypt RIPEMD160 + XTS	1,159,267	H/s	OpenBenchmarking: Hashcat
10,910	Hashcat: Benchmark: LUKS v2 argon2id + SHA-256 + AES	13	H/s	OpenBenchmarking: Hashcat
10,910	cl-mem: Benchmark: Read	793.5	GB/s	OpenBenchmarking: Hashcat
10,910	cl-mem: Benchmark: Write	775.8	GB/s	OpenBenchmarking: Hashcat
10,910	NAMD CUDA: Input: ATPase with 327,506 Atoms	11.88743	ns/day	OpenBenchmarking: Hashcat
10,910	NAMD CUDA: Input: STMV with 1,066,628 Atoms	3.51814	ns/day	OpenBenchmarking: Hashcat
10,910	VkResample: Upscale: 2x - Precision: Single	11.643	ms	OpenBenchmarking: Hashcat
10,910	FAHBench	389.0836	Ns Per Day	OpenBenchmarking: Hashcat
10,910	clpeak: OpenCL Test: Double-Precision Compute	812.66	GFLOPS	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - sAXPY	449	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - sDOT	367	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dCOPY	546	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dAXPY	683	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dDOT	626	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMV-N	184	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMV-T	345	GB/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMM-NN	664	GFLOPs/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMM-NT	711	GFLOPs/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMM-TN	712	GFLOPs/s	OpenBenchmarking: Hashcat
10,910	ViennaCL: Test: OpenCL BLAS - dGEMM-TT	776	GFLOPs/s	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: mobilenet	25.89	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU-v2-v2 - Model: mobilenet-v2	5.94	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU-v3-v3 - Model: mobilenet-v3	4.05	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: shufflenet-v2	3.65	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: mnasnet	4.97	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: efficientnet-b0	7.94	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: blazeface	2.32	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: googlenet	8.95	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: vgg16	37.44	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: resnet18	6.25	ms	OpenBenchmarking: Hashcat
10,910	NCNN: Target: Vulkan GPU - Model: alexnet	6.21	ms	OpenBenchmarking: Hashcat
11,889	vkpeak: fp32-vec4	34,412.58	GFLOPS	OpenBenchmarking: Hashcat
10,938	LLM Inference: Geekbench AI: Geekbench AI Single Precision (fp32)	45,417	points	LLM Inference
11,889	vkpeak: fp16-scalar	25,965.34	GFLOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: fp16-vec4	25,822.58	GFLOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: fp16-matrix	205,769.65	GFLOPS	OpenBenchmarking: Hashcat
10,938	LLM Inference: Geekbench AI: Geekbench AI Half Precision (fp16)	66,366	points	LLM Inference
10,335	OpenRadioss: Model: Rubber O-Ring Seal Installation	50.75	Seconds	OpenBenchmarking: Llama.cpp
10,335	QMCPACK: Input: H4_ae	10.72	Total Execution Time - Seconds	OpenBenchmarking: Llama.cpp
10,335	OpenRadioss: Model: Bird Strike on Windshield	105.37	Seconds	OpenBenchmarking: Llama.cpp
10,335	ClickHouse: 100M Rows Hits Dataset, Latency, Second Run	861.82	Queries Per Minute, Geo Mean	OpenBenchmarking: Llama.cpp
10,335	OpenRadioss: Model: Bumper Beam	77.15	Seconds	OpenBenchmarking: Llama.cpp
11,889	vkpeak: fp64-vec4	812.64	GFLOPS	OpenBenchmarking: Hashcat
10,335	QMCPACK: Input: Li2_STO_ae	97.311	Total Execution Time - Seconds	OpenBenchmarking: Llama.cpp
10,335	ClickHouse: 100M Rows Hits Dataset, Throughput	245.09	Queries Per Minute, Geo Mean	OpenBenchmarking: Llama.cpp
11,889	vkpeak: int16-vec4	24,589.36	GIOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: int8-dotprod	32,808.08	GIOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: int8-matrix	409,762.38	GIOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: int32-scalar	25,964.43	GIOPS	OpenBenchmarking: Hashcat
11,889	vkpeak: bf16-matrix	101,289.72	GFLOPS	OpenBenchmarking: Hashcat
11,889	RealSR-NCNN: Scale: 4x - TAA: No	5.589	Seconds	OpenBenchmarking: Hashcat

End of preview. Expand in Data Studio

GPU Ark — open GPU specifications & benchmarks dataset

Specifications of 13,566 GPUs released between 1999 and 2025 — from the GeForce 256 to NVIDIA Blackwell and AMD Instinct MI355X — plus 993 third-party benchmark results. Curated and maintained by GPU Ark (a GPU catalog & price comparison project). Canonical source and always-fresh copy: https://gpuark.com/datasets/.

Files

File	Rows	What
`gpuark-gpu-specs.csv`	13,566	One row per GPU — public spec columns
`gpuark-benchmarks.csv`	993	Third-party benchmark results, join on `gpu_id`
`gpuark-gpu-dataset.sqlite`	—	Both tables (`gpu_specs`, `benchmarks`) for SQL

Key columns (`gpu_specs`)

id, name, slug, vendor (nvd/amd/int), manufacturer, arch_name, card_release_date, proc_foundry, proc_process_size, proc_transistors, cores, tensor_cores, base_clock, boost_clock, ram, ram_type, bus_width, ram_bandwidth, fp16/fp32/fp64/bf16/tf32/int8_performance, tdp, multi_gpu, api_cuda, is_retail_board, gpi_value.

Quick start

import pandas as pd
df = pd.read_csv("gpuark-gpu-specs.csv", parse_dates=["card_release_date"])
nv = df[df.vendor == "nvd"]
# peak FP32 flagship per year
print(nv.groupby(nv.card_release_date.dt.year).fp32_performance.max())

Known issues (read before drawing conclusions)

vendor is set for ~2,360 of ~13,566 rows (nvd/amd/int); the rest are mostly partner/OEM board variants without a chip-vendor tag. Filter on vendor for vendor-level work.
ram (VRAM) unit is inconsistent across eras — older cards store MB, newer store GB (a value ≥ 256 on a pre-2018 card is almost certainly MB).
fp16/bf16/int8 are sparse and not consistently tensor-vs-non-tensor across vendors (NVIDIA Ampere+ tensor figures are often listed with structured sparsity = 2× dense). Don't compare low-precision peaks cross-vendor without checking the card.
card_release_date has a handful of implausible years — filter to 1998..2025.
is_retail_board=True = AIB/OEM editions of a reference chip (near-duplicates).

License & attribution

CC BY 4.0 — free to use with attribution to gpuark.com.

Citation

GPU Ark (2026). GPU specifications & benchmarks dataset. https://gpuark.com/datasets/

Downloads last month: 8