inference/research

Accelerators (236)

Memory capacity, bandwidth, tensor-core FLOPS per precision, TDP and fabric. This is the hardware-selection table: compare two parts on the Compare page before reading a benchmark that names one.

contested 16draft 88verified 131deprecated 1→ sources (2562)
AMD Instinct MI355X
amd CDNA4 (gfx950) 2025288 GB hbm3e8000 GB/s1400 Waccelerators/amd-instinct-mi355x
contestedhigh · 0.90
Biren BR100 GPGPU (2-die OAM training accelerator)
asic-other Biren SPC (2-die chiplet, CoWoS packaging) 202264 GB hbm2e1600 GB/s550 Waccelerators/biren-br100
contestedlow · 0.50
Cambricon MLU270-S4
cambricon MLUv02 201916 GB GDDR5102.4 GB/s70 Waccelerators/cambricon-mlu270
contestedlow · 0.40
Cambricon MLU590 (Siyuan 590) training accelerator
cambricon mluarch05 2023192 GB hbm2e2400 GB/s350 Waccelerators/cambricon-mlu590
contestedlow · 0.30
Cerebras WSE-3 (CS-3 system)
cerebras wafer-scale engine (WSE-3) 202427000 Waccelerators/cerebras-wse-3
contestedlow · 0.55
Groq LPU (Language Processing Unit, Tensor Streaming Processor)
groq Tensor Streaming Processor / LPU 2019accelerators/groq-lpu
contestedlow · 0.55
Huawei Ascend 910B
asic-other da-vinci (Ascend 910B) 202364 GB hbm2e1400 GB/s400 Waccelerators/huawei-ascend-910b
contestedlow · 0.45
Huawei Ascend 910C (dual-910B package)
asic-other da-vinci (dual 910B chiplet) 2025128 GB hbm2e3200 GB/s600 Waccelerators/huawei-ascend-910c
contestedlow · 0.45
Huawei Ascend 950DT NPU (per-accelerator, Atlas 650E / 950 SuperPoD)
asic-other da-vinci (950-series) 202696 GB on-chip memory (Huawei-developed; not HBM)500 GB/saccelerators/huawei-ascend-950dt
contestedmedium · 0.70
Intel Data Center GPU Flex 170
intel alchemist (Xe-HPG, ACM-G11) 202216 GB gddr6576 GB/s150 Waccelerators/xpu-intel-dc-gpu-flex-170
contestedhigh · 0.85
Intel Data Center GPU Max 1550
intel ponte-vecchio (Xe-HPC, PVC) 2023128 GB hbm2e3276 GB/s600 Waccelerators/xpu-intel-dc-gpu-max-1550
contestedhigh · 0.85
NVIDIA B200
nvidia blackwell 2024180 GB HBM3e7700 GB/s1000 Waccelerators/nvidia-b200
contestedmedium · 0.75
NVIDIA B300 (Blackwell Ultra)
nvidia blackwell-ultra 2025270 GB HBM3e7700 GB/s1100 Waccelerators/nvidia-b300
contestedlow · 0.60
NVIDIA GB300 (Blackwell Ultra GPU as fitted in the GB300 NVL72 rack, per-GPU figures)
nvidia blackwell-ultra 2025279 GB hbm3e8000 GB/s1400 Waccelerators/nvidia-gb300
contestedmedium · 0.65
NVIDIA H800 80GB SXM (China export SKU)
nvidia hopper 2023accelerators/nvidia-h800-80gb
contestedlow · 0.40
Xilinx Alveo U250/U280 (used FPGA accelerator card)
amd FPGA (UltraScale+ / Alveo) 20188 GB HBM2 (8 GB) + DDR4 RDIMM (32 GB, two 16 GB 2400 MT/s DIMMs)460 GB/s225 Waccelerators/xilinx-alveo-u250-u280-used-fpga-accel-card
contestedlow · 0.60
AMD Radeon 8060S (Ryzen AI Max+ integrated GPU)
amd rdna3.5 (Strix Halo iGPU, gfx1151) 2025256 GB/saccelerators/amd-radeon-8060s-ryzen-ai-max
draftlow · 0.60
AMD Radeon 880M
amd rdna3.5 2024accelerators/amd-radeon-880m
draftlow · 0.55
AMD Radeon Instinct MI25
amd vega (gfx900) 201716 GB hbm2484 GB/s300 Waccelerators/amd-instinct-mi25
draftlow · 0.55
AMD Radeon PRO V620
amd rdna2 (Navi 21, gfx1030) 202132 GB gddr6 with ECC512 GB/s300 Waccelerators/amd-radeon-pro-v620
draftlow · 0.60
AMD Radeon RX 5700 XT
amd rdna1 (Navi 10, gfx1010) 20198 GB gddr6448 GB/s225 Waccelerators/amd-radeon-rx-5700-xt
draftlow · 0.60
AMD Radeon RX 6600
amd rdna2 (Navi 23) 20218 GB gddr6224 GB/s132 Waccelerators/amd-radeon-rx-6600
draftmedium · 0.70
AMD Radeon RX 6600 XT
amd rdna2 (Navi 23) 20218 GB gddr6256 GB/s160 Waccelerators/amd-radeon-rx-6600-xt
draftmedium · 0.70
AMD Radeon RX 6650 XT
amd rdna2 (Navi 23) 20228 GB gddr6280 GB/s180 Waccelerators/amd-radeon-rx-6650-xt
draftmedium · 0.70
AMD Radeon RX 6700 XT
amd rdna2 (Navi 22) 202112 GB gddr6384 GB/s230 Waccelerators/amd-radeon-rx-6700-xt
draftmedium · 0.70
AMD Radeon RX 6750 XT
amd rdna2 (Navi 22) 202212 GB gddr6432 GB/s250 Waccelerators/amd-radeon-rx-6750-xt
draftmedium · 0.70
AMD Radeon RX 6800
amd rdna2 (Navi 21) 202016 GB gddr6512 GB/s250 Waccelerators/amd-radeon-rx-6800
draftmedium · 0.70
AMD Radeon RX 6800 XT
amd rdna2 (Navi 21, gfx1030) 202016 GB gddr6512 GB/s300 Waccelerators/amd-radeon-rx-6800-xt
draftlow · 0.60
AMD Radeon RX 6900 XT
amd rdna2 (Navi 21, gfx1030) 202016 GB gddr6512 GB/s300 Waccelerators/amd-radeon-rx-6900-xt
draftlow · 0.60
AMD Radeon RX 7600
amd rdna3 (Navi 33) 20238 GB gddr6288 GB/s165 Waccelerators/amd-radeon-rx-7600
draftmedium · 0.70
AMD Radeon RX 7600 XT
amd rdna3 (Navi 33, gfx1103) 202416 GB gddr6288 GB/s190 Waccelerators/amd-radeon-rx-7600-xt
draftlow · 0.60
AMD Radeon RX 7800 XT
amd rdna3 (Navi 32, gfx1101) 202316 GB gddr6624 GB/s263 Waccelerators/amd-radeon-rx-7800-xt
draftlow · 0.60
AMD Radeon RX 7900 GRE
amd rdna3 (Navi 31, gfx1100) 202416 GB gddr6576 GB/s260 Waccelerators/amd-radeon-rx-7900-gre
draftlow · 0.60
AMD Radeon RX 7900 XT
amd rdna3 (Navi 31, gfx1100) 202220 GB gddr6800 GB/s315 Waccelerators/amd-radeon-rx-7900-xt
draftlow · 0.60
AMD Radeon RX 9060 XT 16GB
amd rdna4 (Navi 44, gfx1200) 202516 GB gddr6320 GB/s160 Waccelerators/amd-radeon-rx-9060-xt
draftlow · 0.60
AMD Radeon RX 9070
amd rdna4 (Navi 48, gfx1201) 202516 GB gddr6640 GB/s220 Waccelerators/amd-radeon-rx-9070
draftlow · 0.60
Apple M1 Max
apple apple m1 max 202164 GB lpddr5 unified400 GB/s60 Waccelerators/apple-m1-max
draftlow · 0.60
Apple M4 (base, Mac mini)
apple apple-m4 202424 GB lpddr5x (unified)120 GB/saccelerators/local-apple-m4
draftlow · 0.60
Apple M6 (base, Mac mini)
apple apple-m6 202632 GB lpddr5x (unified)170 GB/saccelerators/local-apple-m6
draftlow · 0.60
AWS Inferentia (first generation, Inf1)
amazon NeuronCore (1st generation, 4 cores per chip) 20208 GB ddr4 (plus a separate, unquantified on-chip memory tier)accelerators/aws-inferentia1
draftlow · 0.60
AWS Trainium (first generation, Trn1 / Trn1n)
amazon NeuronCore (2nd generation, 2 cores per chip) 202032 GB hbm800 GB/saccelerators/aws-trainium1
draftmedium · 0.65
Biren BILI 166 series (166C inference card / 166M / 166L OAM modules)
asic-other Biren proprietary GPGPU 2025300 Waccelerators/biren-bili-166
draftlow · 0.35
Biren BR104 GPGPU (single-die inference accelerator)
asic-other Biren SPC (single die, half-size sibling of BR100) 202332 GB hbm2e300 Waccelerators/biren-br104
draftlow · 0.35
Cerebras WSE-3 Turbo (WSE-3T, the wafer inside the CS-4 system)
cerebras wafer-scale engine, Turbo generation (WSE-3T) 2025accelerators/cerebras-wse-3t
draftlow · 0.60
Enflame S60 (SCORPIO-AO) AI accelerator
asic-other Enflame SuiSi / SCORPIO chiplet 2025accelerators/enflame-s60
draftlow · 0.30
Horizon Robotics Journey 6P (Nash BPU automotive SoC)
asic-other BPU 3.0 Nash 2024204 GB/saccelerators/horizon-journey-6p
draftlow · 0.50
Huawei Ascend 310P (Da Vinci edge inference NPU)
asic-other da-vinci (310-series) 202224 GB lpddr4x204.8 GB/s72 Waccelerators/huawei-ascend-310p
draftlow · 0.40
Huawei Ascend 910D
asic-other da-vinci (910-series successor to 910C) 2025accelerators/huawei-ascend-910d
draftlow · 0.30
Iluvatar CoreX BI-V100 (Tiangai 100) training card
asic-other Iluvatar general GPU (2nd gen) 202132 GB hbm2250 Waccelerators/iluvatar-bi-v100-mr-v100
draftlow · 0.50
Iluvatar CoreX BI-V150 (Tiangai 150) cloud GPGPU training accelerator
asic-other ivcore11 (Iluvatar self-developed general-purpose GPU architecture) 2023350 Waccelerators/iluvatar-bi-v150
draftlow · 0.30
Iluvatar CoreX BI-V250 (Tiangai 250) cloud GPGPU
asic-other ivcore11+ (Iluvatar self-developed general-purpose GPU architecture)accelerators/iluvatar-bi-v250
draftlow · 0.15
Iluvatar CoreX MR-V100 (Zhikai 100) HBM inference card
asic-other Iluvatar general GPU (2nd gen) 202232 GB hbm2e150 Waccelerators/iluvatar-mr-v100-zhikai100
draftlow · 0.50
Intel Arc B570
intel battlemage (Xe2) 2024accelerators/intel-arc-b570
draftlow · 0.45
MetaX C550 (flagship OAM GPGPU, training and inference)
asic-other MetaX XCORE (proprietary GPGPU core IP) 202464 GB hbm2e450 Waccelerators/metax-c550
draftlow · 0.55
MetaX MXC500 / MXN100 GPGPU (training and inference)
asic-other MetaX proprietary GPGPU core IP 2023accelerators/metax-c500-n100
draftlow · 0.45
Moore Threads MTT S5000 (PH100, Pinghu architecture)
asic-other MUSA (4th gen) / Pinghu 2026accelerators/moore-threads-s5000
draftlow · 0.60
Mythic M1108
asic-other analog compute-in-memory (Flash memory cell array) 20194 Waccelerators/mythic-m1108
draftlow · 0.60
NVIDIA A40
nvidia ampere (GA102) 202048 GB gddr6696 GB/s300 Waccelerators/nvidia-a40
draftlow · 0.60
NVIDIA GB10 Grace Blackwell (DGX Spark)
nvidia blackwell + grace (GB10) 2025128 GB lpddr5x unified273 GB/s140 Waccelerators/nvidia-gb10
draftlow · 0.60
NVIDIA GeForce GTX 1060 6GB
nvidia pascal (GP106) 20166 GB gddr5192 GB/s120 Waccelerators/nvidia-gtx-1060-6gb
draftlow · 0.55
NVIDIA GeForce GTX 1070 Ti
nvidia pascal (GP104) 20178 GB gddr5256 GB/s180 Waccelerators/nvidia-gtx-1070-ti
draftlow · 0.55
NVIDIA GeForce GTX 1080 Ti
nvidia pascal (GP102) 201711 GB gddr5xaccelerators/nvidia-gtx-1080-ti
draftlow · 0.45
NVIDIA GeForce GTX 1660
nvidia turing (TU116) 20196 GB gddr5192 GB/s120 Waccelerators/nvidia-gtx-1660
draftlow · 0.55
NVIDIA GeForce RTX 2060 SUPER
nvidia turing (TU106) 20198 GB gddr6448 GB/s175 Waccelerators/nvidia-rtx-2060-super
draftlow · 0.55
NVIDIA GeForce RTX 2070 SUPER
nvidia turing (TU104) 20198 GB gddr6448 GB/s215 Waccelerators/nvidia-rtx-2070-super
draftmedium · 0.70
NVIDIA GeForce RTX 2080 Ti
nvidia turing (TU102) 201811 GB gddr6616 GB/s250 Waccelerators/nvidia-rtx-2080-ti
draftlow · 0.60
NVIDIA GeForce RTX 3060 12GB
nvidia ampere (GA106) 202112 GB gddr6360 GB/s170 Waccelerators/nvidia-rtx-3060
draftlow · 0.60
NVIDIA GeForce RTX 3060 Ti
nvidia ampere (GA104) 20208 GB gddr6448 GB/s200 Waccelerators/nvidia-rtx-3060-ti
draftmedium · 0.70
NVIDIA GeForce RTX 3070
nvidia ampere (GA104) 20208 GB gddr6448 GB/s220 Waccelerators/nvidia-rtx-3070
draftlow · 0.60
NVIDIA GeForce RTX 3080 10GB
nvidia ampere (GA102) 202010 GB gddr6x760 GB/s320 Waccelerators/nvidia-rtx-3080-10gb
draftlow · 0.60
NVIDIA GeForce RTX 3080 Ti
nvidia ampere (GA102) 202112 GB gddr6x912 GB/s350 Waccelerators/nvidia-rtx-3080-ti
draftlow · 0.60
NVIDIA GeForce RTX 4050 Laptop GPU
nvidia ada lovelace 20236 GB gddr6192 GB/s115 Waccelerators/nvidia-rtx-4050-laptop
draftlow · 0.55
NVIDIA GeForce RTX 4060 Ti
nvidia ada (AD106) 202316 GB gddr6288 GB/s160 Waccelerators/nvidia-rtx-4060-ti
draftlow · 0.60
NVIDIA GeForce RTX 4070
nvidia ada (AD104) 202312 GB gddr6x504 GB/s200 Waccelerators/nvidia-rtx-4070
draftlow · 0.60
NVIDIA GeForce RTX 4070 SUPER
nvidia ada (AD104) 202412 GB gddr6x504 GB/s220 Waccelerators/nvidia-rtx-4070-super
draftlow · 0.60
NVIDIA GeForce RTX 4070 Ti
nvidia ada (AD104) 202312 GB gddr6x504 GB/s285 Waccelerators/nvidia-rtx-4070-ti
draftlow · 0.60
NVIDIA GeForce RTX 4070 Ti SUPER
nvidia ada (AD103) 202416 GB gddr6x672 GB/s285 Waccelerators/nvidia-rtx-4070-ti-super
draftlow · 0.60
NVIDIA GeForce RTX 4080 SUPER
nvidia ada (AD103) 202416 GB gddr6x736 GB/s320 Waccelerators/nvidia-rtx-4080-super
draftlow · 0.60
NVIDIA GeForce RTX 5060 Ti 16GB
nvidia blackwell (GB206) 202516 GB gddr7448 GB/s180 Waccelerators/nvidia-rtx-5060-ti
draftlow · 0.60
NVIDIA GH200 Grace Hopper (96GB HBM3)
nvidia hopper + grace (GH100/Grace) 202396 GB hbm3 + lpddr5x unified4000 GB/saccelerators/nvidia-gh200
draftlow · 0.60
NVIDIA H20
nvidia hopper 202496 GB HBM34000 GB/s400 Waccelerators/nvidia-h20
draftlow · 0.50
NVIDIA Jetson AGX Orin 64GB
nvidia ampere (Orin) 202264 GB lpddr5204.8 GB/s60 Waccelerators/nvidia-jetson-agx-orin-64gb
draftlow · 0.55
NVIDIA Jetson AGX Thor 128GB
nvidia blackwell (jetson thor) 2025128 GB lpddr5x unified273 GB/s130 Waccelerators/nvidia-jetson-agx-thor
draftlow · 0.60
NVIDIA L40 (Ada, 48 GB)
nvidia ada (AD102) 202248 GB gddr6 with ECC864 GB/s300 Waccelerators/nvidia-l40
draftlow · 0.60
NVIDIA P106-100
nvidia pascal (GP106) 20176 GB gddr5192 GB/s120 Waccelerators/nvidia-p106-100
draftlow · 0.55
NVIDIA Quadro P1000
nvidia pascal (GP107) 20174 GB gddr580 GB/s47 Waccelerators/nvidia-quadro-p1000
draftlow · 0.55
NVIDIA Quadro P2000
nvidia pascal (GP106) 20175 GB gddr5140 GB/s75 Waccelerators/nvidia-quadro-p2000
draftlow · 0.55
NVIDIA Quadro RTX 4000
nvidia turing (TU104) 20188 GB gddr6416 GB/s160 Waccelerators/nvidia-quadro-rtx-4000
draftlow · 0.55
NVIDIA Quadro T1000
nvidia turing (TU117) 20194 GB gddr6160 GB/s50 Waccelerators/nvidia-quadro-t1000
draftlow · 0.55
NVIDIA RTX 2000 Ada Generation
nvidia ada (AD107) 202416 GB gddr6 with ECC224 GB/s70 Waccelerators/nvidia-rtx-2000-ada
draftlow · 0.60
NVIDIA RTX 4000 Ada Generation
nvidia ada (AD104) 202320 GB gddr6 with ECC360 GB/s130 Waccelerators/nvidia-rtx-4000-ada
draftlow · 0.60
NVIDIA RTX 5000 Ada Generation
nvidia ada (AD102) 202332 GB gddr6 with ECC576 GB/s250 Waccelerators/nvidia-rtx-5000-ada
draftlow · 0.60
NVIDIA RTX A2000 12GB
nvidia ampere (GA106) 202112 GB gddr670 Waccelerators/nvidia-rtx-a2000
draftlow · 0.45
NVIDIA RTX A4000
nvidia ampere (GA104) 202116 GB gddr6 with ECC448 GB/s140 Waccelerators/nvidia-rtx-a4000
draftlow · 0.60
NVIDIA RTX A4500
nvidia ampere (GA102) 202120 GB gddr6 with ECC640 GB/s200 Waccelerators/nvidia-rtx-a4500
draftlow · 0.60
NVIDIA RTX A5000
nvidia ampere (GA102) 202124 GB gddr6 with ECC768 GB/s230 Waccelerators/nvidia-rtx-a5000
draftlow · 0.60
NVIDIA RTX A6000
nvidia ampere (GA102) 202048 GB gddr6 with ECC768 GB/s300 Waccelerators/nvidia-rtx-a6000
draftlow · 0.60
NVIDIA RTX PRO 4000 Blackwell
nvidia blackwell 202524 GB gddr7 with ECC672 GB/s140 Waccelerators/nvidia-rtx-pro-4000-blackwell
draftlow · 0.60
NVIDIA Tesla K80
nvidia kepler (GK210, dual GPU) 201424 GB gddr5480 GB/s300 Waccelerators/nvidia-tesla-k80
draftlow · 0.55
NVIDIA Tesla P4
nvidia pascal (GP104) 20168 GB gddr5192 GB/s75 Waccelerators/nvidia-tesla-p4
draftlow · 0.55
NVIDIA Tesla T4
nvidia turing (TU104) 201816 GB gddr6320 GB/s70 Waccelerators/nvidia-tesla-t4
draftlow · 0.60
NVIDIA TITAN Xp
nvidia pascal (GP102) 201712 GB gddr5x547.7 GB/s250 Waccelerators/nvidia-titan-xp
draftlow · 0.55
Qualcomm Cloud AI 100 (family: HHHL-Pro / HHHL-Std / HHHL-Lite / Dual M.2 / Dual M.2e)
qualcomm qualcomm ai core (scalar 4-way VLIW with vector/tensor units) 202032 GB lpddr4x137 GB/s75 Waccelerators/qualcomm-ai100
draftmedium · 0.70
Qualcomm Cloud AI 100 Ultra (AI200 accelerator)
qualcomm qualcomm ai core (successor to Cloud AI 100; AI200) 2023128 GB lpddr4x548 GB/s150 Waccelerators/qualcomm-ai200
draftmedium · 0.70
AMD EPYC 9334 (Genoa, 32-core, dual-socket, 12-channel DDR5 host CPU)
amd Zen 4 (Genoa, SP5) 2022921.6 GB/s420 Waccelerators/amd-epyc-9334-genoa-dual
verifiedmedium · 0.72
AMD EPYC 9454P (Genoa, 48-core, 12-channel DDR5 host CPU)
amd Zen 4 (Genoa, SP5) 2022460.8 GB/s290 Waccelerators/amd-epyc-9454p-genoa
verifiedmedium · 0.78
AMD EPYC 9554 (Genoa, 64-core, 12-channel DDR5 host CPU)
amd Zen 4 (Genoa, SP5) 2022460.8 GB/s360 Waccelerators/amd-epyc-9554-genoa
verifiedmedium · 0.80
AMD Instinct MI100
amd CDNA1 (gfx908) 202032 GB hbm21228 GB/s300 Waccelerators/amd-instinct-mi100
verifiedhigh · 0.90
AMD Instinct MI210
amd CDNA2 (gfx90a) 202264 GB hbm2e1600 GB/saccelerators/amd-instinct-mi210
verifiedhigh · 0.85
AMD Instinct MI250
amd CDNA2 (gfx90a) 2021128 GB hbm2e3200 GB/s560 Waccelerators/amd-instinct-mi250
verifiedhigh · 0.90
AMD Instinct MI250X
amd CDNA2 (gfx90a) 2021128 GB hbm2e3200 GB/s560 Waccelerators/amd-instinct-mi250x
verifiedhigh · 0.90
AMD Instinct MI300A
amd CDNA3 (gfx942) 2023128 GB hbm35300 GB/s550 Waccelerators/amd-instinct-mi300a
verifiedhigh · 0.90
AMD Instinct MI300X
amd CDNA3 (gfx942) 2023192 GB hbm35300 GB/s750 Waccelerators/amd-instinct-mi300x
verifiedhigh · 0.95
AMD Instinct MI325X
amd CDNA3 (gfx942) 2024256 GB hbm3e6000 GB/s1000 Waccelerators/amd-instinct-mi325x
verifiedhigh · 0.90
AMD Instinct MI350P
amd CDNA4 (gfx950) 2025144 GB hbm3e4000 GB/s600 Waccelerators/amd-instinct-mi350p
verifiedhigh · 0.85
AMD Instinct MI350X
amd CDNA4 (gfx950) 2025288 GB hbm3e8000 GB/s1000 Waccelerators/amd-instinct-mi350x
verifiedhigh · 0.90
AMD Radeon AI PRO R9700
amd rdna4 (Navi 48, gfx1201) 202532 GB gddr6640 GB/s300 Waccelerators/amd-radeon-ai-pro-r9700
verifiedhigh · 0.90
AMD Radeon Instinct MI50X
amd GCN5.1 (gfx906, Vega20) 201832 GB hbm21024 GB/s300 Waccelerators/amd-radeon-instinct-mi50x
verifiedmedium · 0.80
AMD Radeon Instinct MI60
amd GCN5.1 (gfx906, Vega20) 201832 GB hbm21024 GB/s300 Waccelerators/amd-radeon-instinct-mi60
verifiedmedium · 0.80
AMD Radeon Pro VII
amd GCN5.1 (gfx906, Vega20) 201916 GB hbm21024 GB/s250 Waccelerators/amd-radeon-pro-vii
verifiedhigh · 0.90
AMD Radeon Pro W5700
amd RDNA1 (Navi 10) 20198.0 GB gddr6448 GB/s205 Waccelerators/amd-radeon-pro-w5700
verifiedmedium · 0.70
AMD Radeon PRO W7900
amd rdna3 202348 GB gddr6864 GB/s295 Waccelerators/amd-radeon-pro-w7900
verifiedhigh · 0.90
AMD Radeon RX 7900 XTX
amd rdna3 (Navi 31, gfx1100) 202224 GB gddr6960 GB/s355 Waccelerators/amd-radeon-rx-7900-xtx
verifiedhigh · 0.90
AMD Radeon RX 9070 XT
amd rdna4 (Navi 48, gfx1201) 202516 GB gddr6640 GB/s304 Waccelerators/amd-radeon-rx-9070-xt
verifiedhigh · 0.90
AMD Radeon RX Vega 56
amd vega 10 (GCN 5, gfx900) 20178 GB hbm2410 GB/s210 Waccelerators/amd-radeon-rx-vega-56
verifiedmedium · 0.75
AMD Radeon RX Vega 64
amd vega 10 (GCN 5, gfx900) 20178 GB hbm2484 GB/s295 Waccelerators/amd-radeon-rx-vega-64
verifiedmedium · 0.80
AMD Radeon VII
amd vega 20 (GCN 5.1, gfx906) 201916 GB hbm21024 GB/s300 Waccelerators/amd-radeon-vii
verifiedmedium · 0.75
AMD Ryzen AI 9 HX 370 (Strix Point, dual-channel LPDDR5X)
amd Zen 5 + RDNA 3.5 (Strix Point APU) 2024120 GB/s54 Waccelerators/amd-ryzen-ai-9-hx-370
verifiedmedium · 0.75
AMD Ryzen AI Max+ 395 (Framework Desktop, Strix Halo)
amd Zen 5 + RDNA 3.5 (Strix Halo APU) 2025128 GB LPDDR5X-8000 unified256 GB/s140 Waccelerators/amd-ryzen-ai-max-395-framework-desktop
verifiedhigh · 0.85
Apple M1 (base, 7- or 8-core GPU)
apple m-series (M1) 202016 GB lpddr4x unified68 GB/saccelerators/local-apple-m1
verifiedmedium · 0.80
Apple M1 Pro (14- or 16-core GPU)
apple m-series (M1 Pro) 202132 GB lpddr5 unified200 GB/saccelerators/local-apple-m1-pro
verifiedmedium · 0.80
Apple M1 Ultra (Mac Studio, 2022)
apple m-series (M1 Ultra) 2022128 GB lpddr5 unified800 GB/s370 Waccelerators/apple-m1-ultra
verifiedhigh · 0.85
Apple M2 (base, 8- or 10-core GPU)
apple m-series (M2) 202224 GB lpddr5 unified100 GB/saccelerators/local-apple-m2
verifiedmedium · 0.80
Apple M2 Max (MacBook Pro 14/16-inch & Mac Studio, 2023)
apple m-series (M2 Max) 202396 GB lpddr5 unified400 GB/saccelerators/apple-m2-max
verifiedmedium · 0.80
Apple M2 Pro (16- or 19-core GPU)
apple m-series (M2 Pro) 202332 GB lpddr5 unified200 GB/saccelerators/local-apple-m2-pro
verifiedmedium · 0.80
Apple M2 Ultra
apple Apple Silicon (M2 Ultra, unified memory) 2023192 GB LPDDR5 (unified)800 GB/s215 Waccelerators/apple-m2-ultra
verifiedmedium · 0.78
Apple M3 (base, 10-core GPU)
apple m-series (M3) 202324 GB lpddr5 unified100 GB/saccelerators/local-apple-m3
verifiedmedium · 0.80
Apple M3 Max (MacBook Pro 16-inch, Nov 2023)
apple m-series (M3 Max) 2023128 GB lpddr5 unified400 GB/saccelerators/apple-m3-max
verifiedhigh · 0.85
Apple M3 Pro (14-core GPU bin)
apple m-series (M3 Pro) 202336 GB lpddr5 unified150 GB/saccelerators/local-apple-m3-pro-14core
verifiedmedium · 0.80
Apple M3 Pro (MacBook Pro 16-inch, Nov 2023)
apple m-series (M3 Pro) 202336 GB lpddr5 unified150 GB/saccelerators/local-apple-m3-pro
verifiedhigh · 0.85
Apple M3 Ultra (Mac Studio, 2025)
apple m-series (M3 Ultra) 2025512 GB lpddr5x unified819 GB/s480 Waccelerators/apple-m3-ultra
verifiedhigh · 0.85
Apple M4 Max (MacBook Pro 14-inch, 2024)
apple m-series (M4 Max) 2024128 GB lpddr5x unified546 GB/saccelerators/apple-m4-max
verifiedhigh · 0.85
Apple M4 Pro (MacBook Pro 14-inch, 2024)
apple m-series (M4 Pro) 202448 GB lpddr5x unified273 GB/saccelerators/local-apple-m4-pro
verifiedhigh · 0.85
Apple M4 Ultra (Mac Studio, 2025)
apple Apple M4 Ultra (UltraFusion, 2x M4 Max) 2025192 GB LPDDR5X unified819 GB/saccelerators/apple-m4-ultra
verifiedmedium · 0.70
Apple M5 (MacBook Air / MacBook Pro 14-inch, 2026)
apple m-series (M5) 202632 GB lpddr5x unified153 GB/saccelerators/local-apple-m5
verifiedhigh · 0.85
Apple M5 Max (MacBook Pro 14-inch, 2026)
apple m-series (M5 Max) 2026128 GB lpddr5x unified614 GB/saccelerators/local-apple-m5-max
verifiedhigh · 0.85
Apple M5 Pro (MacBook Pro 14-inch, 2026)
apple m-series (M5 Pro) 202664 GB lpddr5x unified307 GB/saccelerators/local-apple-m5-pro
verifiedhigh · 0.85
Apple M5 Ultra (Mac Studio, 2026)
apple m-series (M5 Ultra) 2026512 GB lpddr5x unified1200 GB/s480 Waccelerators/local-apple-m5-ultra
verifiedhigh · 0.85
AWS Inferentia2
amazon NeuronCore (2nd generation) 202232 GB hbm820 GB/saccelerators/aws-inferentia2
verifiedmedium · 0.70
AWS Trainium2
amazon NeuronCore (4th generation) 202496 GB hbm32875 GB/saccelerators/aws-trainium2
verifiedmedium · 0.75
AWS Trainium3
amazon NeuronCore-v4 (8 cores per chip) 2025144 GB hbm3e4900 GB/saccelerators/aws-trainium3
verifiedhigh · 0.85
Axelera AI Metis AIPU
asic-other Digital In-Memory Computing (D-IMC) with RISC-V 20241.0 GB DRAM (1 GB dedicated DRAM per Metis AIPU on Embedded 110m module)10.0 Waccelerators/axelera-metis
verifiedmedium · 0.80
Cambricon MLU370 (Siyuan 370) AIDC accelerators
cambricon mluarch03 202148 GB lpddr5307.2 GB/s75 Waccelerators/cambricon-mlu370
verifiedhigh · 0.85
Cambricon MLU370-X8 AIDC accelerator (dual Siyuan 370, train+infer)
cambricon mluarch03 (dual Siyuan 370 chiplet) 202248 GB lpddr5614.4 GB/s250 Waccelerators/cambricon-mlu370-x8
verifiedmedium · 0.80
d-Matrix Corsair
asic-other Corsair (digital in-memory compute, chiplet-based) 2025256 GB two-tier: on-chip integrated Performance Memory (SRAM-class, capacity in onchip_cache) + off-chip LPDDR5 Capacity Memory (up to 256 GB per card, vendor ceiling)accelerators/d-matrix-corsair
verifiedmedium · 0.80
FuriosaAI RNGD (Renegade)
asic-other RNGD (Renegade) LLM NPU 202648 GB hbm31500 GB/s180 Waccelerators/furiosaai-rngd-renegade
verifiedmedium · 0.80
FuriosaAI Warboy
asic-other Warboy (Gen 1 Vision NPU) 202116 GB lpddr4x66 GB/s40 Waccelerators/furiosaai-warboy
verifiedmedium · 0.75
Google Cloud TPU v2
google v2 201716 GB hbm700 GB/s280 Waccelerators/google-tpu-v2
verifiedhigh · 0.85
Google Cloud TPU v3
google v3 201832 GB hbm2900 GB/s262 Waccelerators/google-tpu-v3
verifiedhigh · 0.90
Google Cloud TPU v4
google v4 202132 GB hbm21200 GB/s192 Waccelerators/google-tpu-v4
verifiedhigh · 0.95
Google Cloud TPU v5e
google v5e 202316 GB hbm2819 GB/saccelerators/google-tpu-v5e
verifiedhigh · 0.95
Google Cloud TPU v5p
google v5p 202395 GB hbm2e2765 GB/saccelerators/google-tpu-v5p
verifiedhigh · 0.90
Google Cloud TPU v6e (Trillium)
google v6e (Trillium) 202432 GB hbm1638 GB/saccelerators/google-tpu-v6e
verifiedhigh · 0.95
Google Coral Edge TPU
google Edge TPU (vision ASIC) 2019accelerators/google-coral-edge-tpu
verifiedhigh · 0.85
Google TPU7x (Ironwood)
google TPU7x (Ironwood) 2025192 GB hbm7380 GB/saccelerators/google-tpu-ironwood
verifiedhigh · 0.90
Graphcore Bow IPU (Bow-2000)
asic-other colossus-bow 202265250 GB/s425 Waccelerators/graphcore-bow-ipu
verifiedhigh · 0.95
Hailo-10H Generative AI Processor
asic-other Hailo Structure-Driven Dataflow Architecture (Gen-2) 20248 GB LPDDR4X (Hailo-10H M.2 module: 4GB or 8GB LPDDR4X)accelerators/hailo-10h
verifiedmedium · 0.75
Hailo-15H
asic-other Hailo Vision Processor SoC 20235.0 Waccelerators/hailo-15h
verifiedhigh · 0.85
Hailo-8 AI Processor
asic-other Hailo Structure-Driven Dataflow Architecture 2020accelerators/hailo-8
verifiedmedium · 0.75
Horizon Robotics Journey 5 (J5) automotive AI SoC
asic-other BPU 3.0 Bayes (dual-core) 2021accelerators/horizon-journey-5
verifiedmedium · 0.80
Huawei Atlas 300I Duo inference card (dual 310-series, 48/96 GB)
asic-other da-vinci (310-series, dual) 202296 GB lpddr4x408 GB/s150 Waccelerators/huawei-atlas-300i-duo
verifiedmedium · 0.80
Huawei Atlas 300I Pro inference card (24 GB, 310-series)
asic-other da-vinci (310-series) 202124 GB lpddr4x204.8 GB/s72 Waccelerators/huawei-atlas-300i-pro
verifiedmedium · 0.75
Huawei Atlas 800I A2 inference server (8x Ascend 310-series)
asic-other da-vinci (310-series, 8x) 202232 GB lpddr4x204.8 GB/s350 Waccelerators/huawei-atlas-800i-a2
verifiedmedium · 0.65
Huawei CloudMatrix 384 (CM384) supernode — 384x Ascend 910C
asic-other da-vinci (dual 910B chiplet per NPU, 384 NPUs) 202549152 GB hbm2e1228800 GB/s559000 Waccelerators/huawei-ascend-910c-cloudmatrix-384
verifiedmedium · 0.75
Intel Arc A750 8GB
intel alchemist (Xe-HPG, ACM-G10) 20228.0 GB gddr6512 GB/s225 Waccelerators/intel-arc-a750
verifiedhigh · 0.90
Intel Arc A770 16GB
intel alchemist (Xe-HPG, ACM) 202216 GB gddr6560 GB/s225 Waccelerators/intel-arc-a770
verifiedhigh · 0.90
Intel Arc B580
intel battlemage (Xe2, BMG) 202412 GB gddr6456 GB/s190 Waccelerators/intel-arc-b580
verifiedhigh · 0.95
Intel Arc Pro B50
intel battlemage (Xe2 HPG, BMG) 202516 GB gddr6224 GB/s70 Waccelerators/intel-arc-pro-b50
verifiedhigh · 0.90
Intel Arc Pro B60
intel battlemage (Xe2 HPG, BMG G21) 202524 GB gddr6456 GB/s200 Waccelerators/intel-arc-pro-b60
verifiedhigh · 0.95
Intel Arc Pro B70
intel battlemage (Xe2 HPG, BMG G21) 202632 GB gddr6608 GB/s230 Waccelerators/intel-arc-pro-b70
verifiedhigh · 0.92
Intel Gaudi 2
intel gaudi-2 202296 GB hbm2e2460 GB/s600 Waccelerators/intel-gaudi2
verifiedhigh · 0.90
Intel Gaudi 3 (HL-325 OAM / HL-338 PCIe)
intel gaudi-3 2024128 GB hbm2e3700 GB/s900 Waccelerators/intel-gaudi3
verifiedhigh · 0.90
Intel Xeon 6787P (Granite Rapids-SP, 86-core, 8-channel DDR5 host CPU)
intel Redwood Cove (Granite Rapids-SP, Xeon 6, LGA 4710) 2025409.6 GB/s350 Waccelerators/intel-xeon-6787p-granite-rapids
verifiedhigh · 0.85
Intel Xeon 6788P (Granite Rapids-SP, 86-core, 8-channel DDR5 host CPU)
intel Redwood Cove (Granite Rapids-SP, Xeon 6, LGA 4710) 2025409.6 GB/s350 Waccelerators/intel-xeon-6788p-granite-rapids
verifiedhigh · 0.85
Intel Xeon 6972P (Granite Rapids-AP, 96-core, 12-channel DDR5 host CPU)
intel Redwood Cove (Granite Rapids-AP, Xeon 6, LGA 7529) 2024614.4 GB/s500 Waccelerators/intel-xeon-6972p-granite-rapids
verifiedhigh · 0.85
Intel Xeon 6980P (Granite Rapids-AP, 128-core, 12-channel DDR5 host CPU)
intel Redwood Cove (Granite Rapids-AP, Xeon 6, LGA 7529) 2024614.4 GB/s500 Waccelerators/intel-xeon-6980p-granite-rapids
verifiedhigh · 0.85
Kinara Ara240
asic-other Programmable RISC-V Dataflow Discrete NPU 202416 GBaccelerators/kinara-ara240
verifiedhigh · 0.85
Moore Threads MTT S4000 LLM training/inference accelerator
asic-other MUSA (3rd gen) / Sudi 202448 GB gddr6768 GB/s450 Waccelerators/moore-threads-s4000
verifiedmedium · 0.80
NVIDIA A10
nvidia ampere 202124 GB GDDR6600 GB/s150 Waccelerators/nvidia-a10
verifiedhigh · 0.90
NVIDIA A100 40GB PCIe
nvidia ampere 202040 GB HBM21555 GB/s250 Waccelerators/nvidia-a100-40gb-pcie
verifiedhigh · 0.85
NVIDIA A100 40GB SXM4
nvidia ampere 202040 GB HBM21555 GB/s400 Waccelerators/nvidia-a100-40gb-sxm4
verifiedhigh · 0.90
NVIDIA A100 80GB PCIe
nvidia ampere 202080 GB HBM2e1935 GB/s300 Waccelerators/nvidia-a100-80gb-pcie
verifiedhigh · 0.90
NVIDIA A100 80GB SXM4
nvidia ampere 202080 GB HBM2e2039 GB/s400 Waccelerators/nvidia-a100-80gb-sxm4
verifiedhigh · 0.90
NVIDIA A100 SXM (capacity unstated)
nvidia ampere 2020400 Waccelerators/nvidia-a100-sxm
verifiedhigh · 0.85
NVIDIA A30
nvidia ampere 202124 GB HBM2933 GB/s165 Waccelerators/nvidia-a30
verifiedhigh · 0.90
NVIDIA DGX-1 (8x Tesla V100 32GB SXM2, NVLink)
nvidia Volta (8x Tesla V100 32GB SXM2 on NVLink plane) 2017256 GB HBM2 (8x 32GB)900 GB/s3500 Waccelerators/nvidia-dgx-1-8x-v100
verifiedmedium · 0.70
NVIDIA GB200 NVL72 (per-accelerator)
nvidia blackwell 2024186 GB HBM3e8000 GB/s1200 Waccelerators/nvidia-gb200-nvl72
verifiedmedium · 0.70
NVIDIA GeForce RTX 3090
nvidia ampere (GA102) 202024 GB gddr6x936 GB/s350 Waccelerators/nvidia-rtx-3090
verifiedhigh · 0.90
NVIDIA GeForce RTX 3090 Ti
nvidia ampere (GA102) 202124 GB gddr6x1008 GB/s450 Waccelerators/nvidia-rtx-3090-ti
verifiedhigh · 0.90
NVIDIA GeForce RTX 4080 (16 GB)
nvidia ada (AD103) 202216 GB gddr6x716.8 GB/s320 Waccelerators/nvidia-rtx-4080
verifiedhigh · 0.90
NVIDIA GeForce RTX 4090
nvidia ada (AD102) 202224 GB gddr6x1008 GB/s450 Waccelerators/nvidia-rtx-4090
verifiedhigh · 0.95
NVIDIA GeForce RTX 4090 D (China, AD102-250)
nvidia ada (AD102-250) 202324 GB gddr6x1008 GB/s425 Waccelerators/nvidia-rtx-4090-d
verifiedmedium · 0.70
NVIDIA GeForce RTX 5070
nvidia blackwell (GB203) 202512 GB gddr7672 GB/s250 Waccelerators/local-nvidia-rtx-5070
verifiedhigh · 0.90
NVIDIA GeForce RTX 5070 Ti
nvidia blackwell (GB203) 202516 GB gddr7896 GB/s300 Waccelerators/nvidia-rtx-5070-ti
verifiedhigh · 0.90
NVIDIA GeForce RTX 5080
nvidia blackwell (GB203) 202516 GB gddr7960 GB/s360 Waccelerators/nvidia-rtx-5080
verifiedhigh · 0.93
NVIDIA GeForce RTX 5090
nvidia blackwell (GB202) 202532 GB gddr71792 GB/s575 Waccelerators/nvidia-rtx-5090
verifiedhigh · 0.95
NVIDIA GeForce RTX 5090 Laptop GPU (GB203, 24 GB)
nvidia blackwell (GB203) 202524 GB gddr7896 GB/s175 Waccelerators/nvidia-rtx-5090-laptop
verifiedmedium · 0.65
NVIDIA H100 NVL
nvidia hopper 202494 GB HBM33938 GB/s400 Waccelerators/nvidia-h100-nvl
verifiedhigh · 0.85
NVIDIA H100 PCIe
nvidia hopper 202280 GB HBM2e2000 GB/s350 Waccelerators/nvidia-h100-pcie
verifiedhigh · 0.90
NVIDIA H100 SXM
nvidia hopper 202280 GB HBM33350 GB/s700 Waccelerators/nvidia-h100-sxm
verifiedhigh · 0.90
NVIDIA H200 NVL
nvidia hopper 2024141 GB HBM3e4800 GB/s600 Waccelerators/nvidia-h200-nvl
verifiedmedium · 0.80
NVIDIA H200 SXM
nvidia hopper 2024141 GB HBM3e4800 GB/s700 Waccelerators/nvidia-h200-sxm
verifiedhigh · 0.85
NVIDIA L4
nvidia ada 202324 GB GDDR6300 GB/s72 Waccelerators/nvidia-l4
verifiedhigh · 0.85
NVIDIA L40S
nvidia ada 202348 GB GDDR6864 GB/s350 Waccelerators/nvidia-l40s
verifiedhigh · 0.85
NVIDIA Quadro RTX 6000 (Turing, 24 GB)
nvidia turing (TU102) 201824 GB gddr6672 GB/s260 Waccelerators/nvidia-quadro-rtx-6000
verifiedhigh · 0.85
NVIDIA Quadro RTX 8000 (Turing, 48 GB)
nvidia turing (TU102) 201848 GB gddr6672 GB/s295 Waccelerators/nvidia-quadro-rtx-8000
verifiedhigh · 0.85
NVIDIA RTX 6000 Ada Generation
nvidia ada (AD102) 202248 GB gddr6 with ECC960 GB/s300 Waccelerators/nvidia-rtx-6000-ada
verifiedhigh · 0.90
NVIDIA RTX PRO 2000 Blackwell
nvidia blackwell 202516 GB gddr7 with ECC256 GB/s70 Waccelerators/nvidia-rtx-pro-2000-blackwell
verifiedmedium · 0.75
NVIDIA RTX PRO 4500 Blackwell Workstation Edition
nvidia blackwell (GB203-class) 202632 GB gddr7 with ECC896 GB/s200 Waccelerators/nvidia-rtx-pro-4500-blackwell
verifiedmedium · 0.80
NVIDIA RTX PRO 5000 72GB Blackwell Workstation Edition
nvidia blackwell 202572 GB gddr7 with ECC1344 GB/s300 Waccelerators/nvidia-rtx-pro-5000-72gb-blackwell
verifiedmedium · 0.80
NVIDIA RTX PRO 5000 Blackwell Workstation Edition
nvidia blackwell 202548 GB gddr7 with ECC1344 GB/s300 Waccelerators/nvidia-rtx-pro-5000-blackwell
verifiedmedium · 0.80
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
nvidia blackwell (GB202) 202596 GB gddr7 with ECC1792 GB/s600 Waccelerators/nvidia-rtx-pro-6000-blackwell
verifiedhigh · 0.90
NVIDIA Tesla M40 24GB
nvidia maxwell 201524 GB GDDR5288 GB/s250 Waccelerators/nvidia-tesla-m40
verifiedmedium · 0.72
NVIDIA Tesla P100 SXM2 16GB (GP100)
nvidia pascal 201616 GB hbm2732 GB/s300 Waccelerators/nvidia-tesla-p100
verifiedhigh · 0.90
NVIDIA Tesla P40 24GB
nvidia pascal 201624 GB GDDR5347 GB/s250 Waccelerators/nvidia-tesla-p40
verifiedmedium · 0.80
NVIDIA TITAN V
nvidia volta (GV100) 201712 GB hbm2accelerators/nvidia-titan-v
verifiedmedium · 0.80
NVIDIA V100 32GB PCIe
nvidia volta 201732 GB HBM2900 GB/s250 Waccelerators/nvidia-v100-pcie-32gb
verifiedmedium · 0.80
NVIDIA V100 SXM2 32GB
nvidia volta 201732 GB HBM2900 GB/s300 Waccelerators/nvidia-v100-sxm2-32gb
verifiedmedium · 0.80
Rebellions Rebel100
asic-other rebel100 (4-chiplet UCIe-Advanced) 2025144 GB hbm3e4800 GB/s600 Waccelerators/rebellions-rebel100
verifiedmedium · 0.70
Rockchip RK3588 NPU (Orange Pi / SBC)
other RK3588 (ARM SoC + 6 TOPS NPU) 2022accelerators/rockchip-rk3588-npu
verifiedmedium · 0.80
SambaNova SN40L Reconfigurable Dataflow Unit (RDU)
samba Reconfigurable Dataflow Architecture (RDU) 202464.0 GB three-tier: 520 MiB on-chip PMU SRAM, 64 GiB co-packaged HBM, up to 1.5 TiB DDR DRAMaccelerators/sambanova-sn40l
verifiedhigh · 0.85
SiMa.ai MLSoC Modalix
asic-other MLSoC (Machine Learning System-on-Chip) 202310.0 Waccelerators/simaai-mlsoc
verifiedhigh · 0.85
Tenstorrent Blackhole Tensix (p100a / p150a / p150b)
tenstorrent blackhole Tensix 202432 GB gddr6512 GB/s300 Waccelerators/tenstorrent-blackhole-p150
verifiedhigh · 0.85
Tenstorrent Wormhole n300
tenstorrent Wormhole (Tensix) 202324 GB GDDR6576 GB/s300 Waccelerators/tenstorrent-wormhole-n300
verifiedhigh · 0.85
Tenstorrent Wormhole Tensix (n150 / n300)
tenstorrent wormhole Tensix 202312 GB gddr6288 GB/s160 Waccelerators/tenstorrent-wormhole-n150
verifiedhigh · 0.90
Tenstorrent Grayskull Tensix (e75 / e150)
tenstorrent grayskull Tensix 20218 GB lpddr4118 GB/s200 Waccelerators/tenstorrent-grayskull-e150
deprecatedhigh · 0.85

Most referenced

The records other records point at. Useful as an entry point: a heavily referenced accelerator or engine is one many decisions in this repo assume.

RecordReferenced byCites
AMD Radeon 8060S (Ryzen AI Max+ integrated GPU)6163
NVIDIA H100 SXM4406
NVIDIA A100 80GB SXM43193
AMD Instinct MI300X2276
NVIDIA GeForce RTX 40902244
NVIDIA A100 40GB SXM41956
NVIDIA B2001933
NVIDIA GeForce RTX 50901895