| 1 | NVIDIA B200 SXM (192GB) Blackwell•HBM3e•8192-bitBlackwell AI Superchip | | 8000 GB/s100% | 671B MoE | S+ | $45,000 | |
| 2 | NVIDIA B100 SXM (192GB) Blackwell•HBM3e•8192-bitBlackwell 700W Accelerator | | 8000 GB/s100% | 671B MoE | S+ | $32,000 | |
| 3 | AMD Instinct MI325X (256GB) CDNA 3•HBM3e•8192-bit256GB HBM3e AI Supercluster Accelerator | | 6000 GB/s75% | 671B MoE | S+ | $24,000 | |
| #4 | AMD Instinct MI300X (192GB) CDNA 3•HBM3•8192-bit192GB CDNA3 Flagship Accelerator | | 5300 GB/s66% | 671B MoE | S+ | $20,000 | |
| #5 | NVIDIA H200 SXM (141GB) Hopper•HBM3e•6144-bitUltimate 141GB Inference King | | 4800 GB/s60% | 671B MoE | S+ | $38,000 | |
| #6 | AMD Instinct MI300A (128GB APU) CDNA 3•HBM3•8192-bit128GB APU HPC/AI Workhorse | 统一内存 | 5300 GB/s66% | 70B / 123B | S+ | $18,000 | |
| #7 | Intel Gaudi 3 (128GB) Gaudi•HBM2e•8192-bit128GB Intel AI Accelerator | | 3700 GB/s46% | 70B / 123B | S+ | $15,000 | |
| #8 | AMD Instinct MI250X (128GB) CDNA 2•HBM2e•8192-bitDual-GCD 128GB CDNA2 | | 3200 GB/s40% | 70B / 123B | S+ | $12,000 | |
| #9 | NVIDIA H100 NVL (94GB) Hopper•HBM3•5120-bitDual-GPU 94GB LLM Engine | | 3900 GB/s49% | 70B / 123B | S+ | $32,000 | |
| #10 | NVIDIA H100 SXM5 (80GB) Hopper•HBM3•5120-bitData Center Speed Sovereign | | 3350 GB/s42% | 70B / 123B | S+ | $30,000 | |
| #11 | NVIDIA H800 SXM (80GB) Hopper•HBM3•5120-bitHopper 80GB Compliance Variant | | 3350 GB/s42% | 70B / 123B | S+ | $28,000 | |
| #12 | Intel Gaudi 2 (96GB) Gaudi•HBM2e•6144-bit96GB Intel Enterprise AI Accelerator | | 2450 GB/s31% | 70B / 123B | S | $9,000 | |
| #13 | Apple M2 Ultra (192GB Unified) Apple Silicon•Unified LPDDR5•800-bitRuns DeepSeek 671B Full MoE | 统一内存 | 800 GB/s10% | 671B MoE | S+ | $6,999 | |
| #14 | NVIDIA A100 SXM4 (80GB) Ampere•HBM2e•5120-bitEnterprise LLM Workhorse | | 2039 GB/s25% | 70B / 123B | S+ | $12,000 | |
| #15 | NVIDIA A800 SXM4 (80GB) Ampere•HBM2e•5120-bitAmpere 80GB Compliance Variant | | 2039 GB/s25% | 70B / 123B | S+ | $11,000 | |
| #16 | NVIDIA A100 PCIe (80GB) Ampere•HBM2e•5120-bitStandard PCIe Enterprise 80GB | | 1935 GB/s24% | 70B / 123B | S+ | $14,000 | |
| #17 | NVIDIA A800 PCIe (80GB) Ampere•HBM2e•5120-bitAmpere PCIe 80GB Compliance | | 1935 GB/s24% | 70B / 123B | S+ | $9,500 | |
| #18 | Apple M2 Ultra (128GB Unified) Apple Silicon•Unified LPDDR5•800-bit128GB Mac Studio Ultra | 统一内存 | 800 GB/s10% | 70B / 123B | S+ | $5,499 | |
| #19 | Apple M1 Ultra (128GB Unified) Apple Silicon•Unified LPDDR5•800-bitBudget 128GB Mac Studio Workhorse | 统一内存 | 800 GB/s10% | 70B / 123B | S+ | $3,299 | |
| #20 | AMD Instinct MI210 (64GB) CDNA 2•HBM2e•4096-bit64GB CDNA2 Accelerator | | 1600 GB/s20% | 70B / 123B | S | $6,500 | |
| #21 | Apple M4 Max (128GB Unified) Apple Silicon•Unified LPDDR5X•512-bit70B Mobile Powerhouse | 统一内存 | 546 GB/s7% | 70B / 123B | S+ | $3,999 | |
| #22 | Apple M3 Max (128GB Unified) Apple Silicon•Unified LPDDR5•512-bitGiant Local VRAM Pool | 统一内存 | 400 GB/s5% | 70B / 123B | S+ | $3,499 | |
| #23 | NVIDIA GeForce RTX 5090 Blackwell•GDDR7•512-bitNext-Gen Flagship Titan | | 1792 GB/s22% | 32B / 70B* | S+ | $1,999 | |
| #24 | NVIDIA A100 PCIe (40GB) Ampere•HBM2•5120-bit40GB High Bandwidth | | 1555 GB/s19% | 32B / 70B* | S | $6,500 | |
| #25 | Apple M3 Max (96GB Unified) Apple Silicon•Unified LPDDR5•512-bit96GB High-Capacity M3 Max | 统一内存 | 400 GB/s5% | 70B / 123B | S+ | $3,199 | |
| #26 | Apple M2 Max (96GB Unified) Apple Silicon•Unified LPDDR5•512-bit96GB Maxed-Out M2 Max | 统一内存 | 400 GB/s5% | 70B / 123B | S | $3,299 | |
| #27 | NVIDIA RTX 6000 Ada (48GB) Ada Lovelace•GDDR6 ECC•384-bitFlagship 48GB Workstation Titan | | 960 GB/s12% | 70B / 123B | S+ | $6,800 | |
| #28 | NVIDIA L40S (48GB) Ada Lovelace•GDDR6 ECC•384-bitHigh-Throughput Inference Champion | | 864 GB/s11% | 70B / 123B | S+ | $7,200 | |
| #29 | NVIDIA L40 (48GB) Ada Lovelace•GDDR6 ECC•384-bit48GB Ada Lovelace Datacenter | | 864 GB/s11% | 70B / 123B | S+ | $6,800 | |
| #30 | AMD Radeon PRO W7900 (48GB) RDNA 3•GDDR6 ECC•384-bit48GB RDNA3 Workstation Flagship | | 864 GB/s11% | 70B / 123B | S | $3,999 | |
| #31 | Apple M2 Ultra (64GB Unified) Apple Silicon•Unified LPDDR5•800-bit64GB High-Bandwidth Studio | 统一内存 | 800 GB/s10% | 70B / 123B | S | $3,999 | |
| #32 | Apple M1 Ultra (64GB Unified) Apple Silicon•Unified LPDDR5•800-bit64GB M1 Ultra Studio | 统一内存 | 800 GB/s10% | 70B / 123B | S | $2,399 | |
| #33 | AMD Instinct MI100 (32GB) CDNA 1•HBM2•4096-bit32GB CDNA1 Enterprise | | 1228 GB/s15% | 32B / 70B* | A | $3,500 | |
| #34 | NVIDIA RTX A6000 (48GB) Ampere•GDDR6 ECC•384-bit48GB Ampere Workstation Standard | | 768 GB/s10% | 70B / 123B | S+ | $4,200 | |
| #35 | NVIDIA A40 (48GB) Ampere•GDDR6 ECC•384-bit48GB Cloud Standard (RunPod/Lambda) | | 696 GB/s9% | 70B / 123B | S | $4,500 | |
| #36 | NVIDIA Quadro RTX 8000 (48GB) Turing•GDDR6 ECC•384-bit48GB Turing Single-GPU Legend | | 672 GB/s8% | 70B / 123B | S | $1,900 | |
| #37 | AMD Instinct MI50 (32GB) Vega•HBM2•4096-bit32GB HBM2 1TB/s ROCm Budget Hero | | 1024 GB/s13% | 32B / 70B* | A | $260 | |
| #38 | NVIDIA Tesla V100 SXM2 (32GB) Volta•HBM2•4096-bit32GB HBM2 Volta SXM2 | | 900 GB/s11% | 32B / 70B* | A | $550 | |
| #39 | NVIDIA Tesla V100 PCIe (32GB) Volta•HBM2•4096-bit32GB HBM2 Volta PCIe | | 900 GB/s11% | 32B / 70B* | A | $480 | |
| #40 | Apple M4 Max (64GB Unified) Apple Silicon•Unified LPDDR5X•384-bit64GB Unified M4 Max | 统一内存 | 410 GB/s5% | 70B / 123B | S | $3,499 | |
| #41 | Apple M2 Max (64GB Unified) Apple Silicon•Unified LPDDR5•512-bit64GB M2 Max Studio | 统一内存 | 400 GB/s5% | 70B / 123B | S | $2,799 | |
| #42 | Apple M1 Max (64GB Unified) Apple Silicon•Unified LPDDR5•512-bit64GB M1 Max Studio Value | 统一内存 | 400 GB/s5% | 70B / 123B | S | $1,799 | |
| #43 | NVIDIA GeForce RTX 4090 Ada Lovelace•GDDR6X•384-bitConsumer AI King | | 1008 GB/s13% | 32B / 70B* | S+ | $1,599 | |
| #44 | NVIDIA GeForce RTX 3090 Ti Ampere•GDDR6X•384-bitMaxed-Out Ampere 24GB | | 1008 GB/s13% | 32B / 70B* | S+ | $899 | |
| #45 | AMD Radeon RX 7900 XTX RDNA 3•GDDR6•384-bit24GB High Bandwidth AMD Flagship | | 960 GB/s12% | 32B / 70B* | A | $899 | |
| #46 | NVIDIA GeForce RTX 3090 Ampere•GDDR6X•384-bitLegendary 24GB AI Value King | | 936 GB/s12% | 32B / 70B* | S+ | $699 | |
| #47 | Apple M3 Max (64GB Unified) Apple Silicon•Unified LPDDR5•384-bit64GB Unified M3 Max | 统一内存 | 300 GB/s4% | 70B / 123B | S | $2,899 | |
| #48 | NVIDIA A30 (24GB) Ampere•HBM2•3072-bit24GB Low-Profile HBM2 Accelerator | | 933 GB/s12% | 32B / 70B* | A | $2,800 | |
| #49 | Apple M4 Pro (64GB Unified) Apple Silicon•Unified LPDDR5X•256-bit64GB M4 Pro Workstation | 统一内存 | 273 GB/s3% | 70B / 123B | S | $2,599 | |
| #50 | NVIDIA GeForce RTX 5080 Blackwell•GDDR7•256-bitHigh Bandwidth 16GB | | 1024 GB/s13% | 14B / 32B* | A | $999 | |
| #51 | AMD Instinct MI50 (16GB) Vega•HBM2•4096-bit16GB HBM2 1TB/s ROCm Budget | | 1024 GB/s13% | 14B / 32B* | B | $150 | |
| #52 | AMD Radeon VII (16GB HBM2) Vega•HBM2•4096-bit16GB 1TB/s Vega 20 Legend | | 1024 GB/s13% | 14B / 32B* | B | $220 | |
| #53 | NVIDIA RTX 5000 Ada (32GB) Ada Lovelace•GDDR6 ECC•256-bit32GB ECC Ada Workstation | | 576 GB/s7% | 32B / 70B* | S | $4,000 | |
| #54 | AMD Radeon PRO W7800 (32GB) RDNA 3•GDDR6 ECC•256-bit32GB RDNA3 Pro Workstation | | 576 GB/s7% | 32B / 70B* | A | $2,499 | |
| #55 | NVIDIA RTX A5500 (24GB) Ampere•GDDR6 ECC•384-bit24GB High-Bandwidth Ampere Pro | | 768 GB/s10% | 32B / 70B* | S | $2,600 | |
| #56 | NVIDIA RTX A5000 (24GB) Ampere•GDDR6 ECC•384-bit24GB Ampere Enterprise | | 768 GB/s10% | 32B / 70B* | S | $2,200 | |
| #57 | AMD Radeon PRO W6800 (32GB) RDNA 2•GDDR6 ECC•256-bit32GB RDNA2 Workstation Classic | | 512 GB/s6% | 32B / 70B* | A | $1,400 | |
| #58 | AMD Radeon RX 7900 XT RDNA 3•GDDR6•320-bit20GB Value AMD Option | | 800 GB/s10% | 14B / 32B* | B | $699 | |
| #59 | NVIDIA Tesla V100 PCIe (16GB) Volta•HBM2•4096-bit16GB HBM2 Volta PCIe | | 900 GB/s11% | 14B / 32B* | B | $280 | |
| #60 | NVIDIA GeForce RTX 5070 Ti Blackwell•GDDR7•256-bit16GB Blackwell Value | | 896 GB/s11% | 14B / 32B* | A | $749 | |
| #61 | NVIDIA TITAN RTX (24GB) Turing•GDDR6•384-bit24GB Turing Pro Titan | | 672 GB/s8% | 32B / 70B* | A | $750 | |
| #62 | NVIDIA Quadro RTX 6000 (24GB) Turing•GDDR6 ECC•384-bit24GB Turing Pro Enterprise | | 672 GB/s8% | 32B / 70B* | A | $950 | |
| #63 | Apple M3 Max (48GB Unified) Apple Silicon•Unified LPDDR5•384-bit48GB Balanced M3 Max | 统一内存 | 300 GB/s4% | 32B / 70B* | S | $2,699 | |
| #64 | NVIDIA GeForce RTX 3080 Ti Ampere•GDDR6X•384-bitFast Bandwidth 12GB Ampere | | 912 GB/s11% | 8B / 14B* | B | $499 | |
| #65 | NVIDIA GeForce RTX 3080 (12GB) Ampere•GDDR6X•384-bit384-bit 12GB Ampere | | 912 GB/s11% | 8B / 14B* | B | $449 | |
| #66 | NVIDIA A10G (24GB) Ampere•GDDR6•384-bitStandard AWS 24GB Cloud Workhorse | | 600 GB/s8% | 32B / 70B* | A | $1,800 | |
| #67 | Apple M4 Pro (48GB Unified) Apple Silicon•Unified LPDDR5X•256-bit32B Reasoning Studio | 统一内存 | 273 GB/s3% | 32B / 70B* | S | $2,399 | |
| #68 | NVIDIA GeForce RTX 2080 Ti (22GB Mod) Turing•GDDR6•352-bitCommunity 22GB Modded Legend | | 616 GB/s8% | 32B / 70B* | A | $420 | |
| #69 | NVIDIA RTX A4500 (20GB) Ampere•GDDR6 ECC•320-bit20GB 320-bit Ampere Sweet Spot | | 640 GB/s8% | 14B / 32B* | A | $1,600 | |
| #70 | NVIDIA GeForce RTX 4080 Super Ada Lovelace•GDDR6X•256-bit16GB Ada Super | | 736 GB/s9% | 14B / 32B* | A | $999 | |
| #71 | NVIDIA Tesla P100 PCIe (16GB) Pascal•HBM2•4096-bit16GB HBM2 Pascal Workhorse | | 732 GB/s9% | 14B / 32B* | C | $120 | |
| #72 | NVIDIA GeForce RTX 4080 Ada Lovelace•GDDR6X•256-bit16GB Ada Classic | | 717 GB/s9% | 14B / 32B* | A | $899 | |
| #73 | Apple M4 Max (36GB Unified) Apple Silicon•Unified LPDDR5X•384-bit32B Fast Reasoning Unified | 统一内存 | 410 GB/s5% | 32B / 70B* | S | $3,199 | |
| #74 | NVIDIA Tesla K80 (24GB) Kepler•GDDR5•384-bitDual-GPU 24GB Kepler Vintage | | 480 GB/s6% | 32B / 70B* | F | $60 | |
| #75 | NVIDIA GeForce RTX 4070 Ti Super Ada Lovelace•GDDR6X•256-bitBest Value 16GB Ada | | 672 GB/s8% | 14B / 32B* | A | $799 | |
| #76 | NVIDIA RTX 4500 Ada (24GB) Ada Lovelace•GDDR6 ECC•192-bit24GB ECC Workstation | | 432 GB/s5% | 32B / 70B* | A | $2,250 | |
| #77 | NVIDIA Quadro P6000 (24GB) Pascal•GDDR5X•384-bit24GB Pascal Workstation | | 432 GB/s5% | 32B / 70B* | C | $500 | |
| #78 | AMD Radeon RX 7800 XT RDNA 3•GDDR6•256-bit16GB Affordable RDNA3 | | 624 GB/s8% | 14B / 32B* | B | $489 | |
| #79 | Apple M2 Max (32GB Unified) Apple Silicon•Unified LPDDR5•512-bit32GB High-Bandwidth M2 Max | 统一内存 | 400 GB/s5% | 32B / 70B* | A | $1,999 | |
| #80 | Apple M1 Max (32GB Unified) Apple Silicon•Unified LPDDR5•512-bit32GB M1 Max High Bandwidth | 统一内存 | 400 GB/s5% | 32B / 70B* | A | $1,399 | |
| #81 | NVIDIA GeForce RTX 3080 (10GB) Ampere•GDDR6X•320-bitAwkward 10GB Capacity | | 760 GB/s10% | 8B / 14B* | C | $399 | |
| #82 | Apple M3 Max (36GB Unified) Apple Silicon•Unified LPDDR5•384-bit36GB Fast Inference M3 Max | 统一内存 | 300 GB/s4% | 32B / 70B* | A | $2,499 | |
| #83 | NVIDIA RTX 4090 Laptop (16GB) Ada Lovelace•GDDR6•256-bitTop-Tier Mobile 16GB | | 576 GB/s7% | 14B / 32B* | A | $2,200 | |
| #84 | AMD Radeon RX 7900 GRE RDNA 3•GDDR6•256-bit16GB Mid-Range AMD | | 576 GB/s7% | 14B / 32B* | B | $549 | |
| #85 | AMD Radeon RX 6950 XT RDNA 2•GDDR6•256-bit16GB RDNA2 High Performance | | 576 GB/s7% | 14B / 32B* | A | $520 | |
| #86 | NVIDIA GeForce RTX 5070 Blackwell•GDDR7•192-bit12GB Blackwell Fast Memory | | 672 GB/s8% | 8B / 14B* | B | $549 | |
| #87 | Intel Arc A770 (16GB) Alchemist•GDDR6•256-bitSub-$300 16GB High Bandwidth | | 560 GB/s7% | 14B / 32B* | A | $289 | |
| #88 | NVIDIA Tesla P40 (24GB) Pascal•GDDR5•384-bitHomelab #1 Budget 24GB GGUF Hero | | 346 GB/s4% | 32B / 70B* | C | $160 | |
| #89 | NVIDIA TITAN V (12GB HBM2) Volta•HBM2•3072-bit12GB HBM2 Volta Flagship | | 653 GB/s8% | 8B / 14B* | B | $380 | |
| #90 | NVIDIA RTX 3080 Ti Laptop (16GB) Ampere•GDDR6•256-bit16GB Ampere Laptop Flagship | | 512 GB/s6% | 14B / 32B* | A | $1,400 | |
| #91 | AMD Radeon RX 6900 XT RDNA 2•GDDR6•256-bit16GB RDNA2 Former Flagship | | 512 GB/s6% | 14B / 32B* | A | $440 | |
| #92 | AMD Radeon RX 6800 XT RDNA 2•GDDR6•256-bit16GB Last-Gen Value | | 512 GB/s6% | 14B / 32B* | B | $429 | |
| #93 | AMD Radeon RX 6800 (16GB) RDNA 2•GDDR6•256-bit16GB 256-bit Used Market Gem | | 512 GB/s6% | 14B / 32B* | B | $379 | |
| #94 | NVIDIA L4 (24GB) Ada Lovelace•GDDR6•192-bitLow-Power 72W 24GB Cloud GPU | | 300 GB/s4% | 32B / 70B* | A | $2,500 | |
| #95 | NVIDIA GeForce RTX 2080 Ti Turing•GDDR6•352-bit11GB Turing Legacy | | 616 GB/s8% | 8B / 14B* | C | $280 | |
| #96 | NVIDIA RTX 4000 Ada (20GB) Ada Lovelace•GDDR6 ECC•160-bitSingle-Slot 20GB Powerhouse | | 360 GB/s5% | 14B / 32B* | A | $1,250 | |
| #97 | NVIDIA RTX 3080 Laptop (16GB) Ampere•GDDR6•256-bit16GB Ampere Mobile Classic | | 448 GB/s6% | 14B / 32B* | B | $1,100 | |
| #98 | NVIDIA RTX A4000 (16GB) Ampere•GDDR6 ECC•256-bitSingle-Slot 16GB Classic | | 448 GB/s6% | 14B / 32B* | A | $1,000 | |
| #99 | NVIDIA Quadro RTX 5000 (16GB) Turing•GDDR6 ECC•256-bit16GB Turing Workstation | | 448 GB/s6% | 14B / 32B* | B | $550 | |
| #100 | NVIDIA TITAN Xp (12GB) Pascal•GDDR5X•384-bit12GB Pascal Flagship | | 548 GB/s7% | 8B / 14B* | C | $220 | |
| #101 | Apple M3 Pro (36GB Unified) Apple Silicon•Unified LPDDR5•192-bit36GB M3 Pro High-VRAM | 统一内存 | 150 GB/s2% | 32B / 70B* | B | $2,199 | |
| #102 | Apple M2 Pro (32GB Unified) Apple Silicon•Unified LPDDR5•256-bit32GB M2 Pro Studio | 统一内存 | 200 GB/s3% | 32B / 70B* | B | $1,699 | |
| #103 | Apple M1 Pro (32GB Unified) Apple Silicon•Unified LPDDR5•256-bit32GB M1 Pro MacBook | 统一内存 | 200 GB/s3% | 32B / 70B* | B | $1,100 | |
| #104 | NVIDIA GeForce RTX 3070 Ti Ampere•GDDR6X•256-bitFast Memory but 8GB Trap | | 608 GB/s8% | 7B (Tight) | F | $329 | |
| #105 | NVIDIA GeForce RTX 4070 Ti Ada Lovelace•GDDR6X•192-bit12GB 192-bit Fast Compute | | 504 GB/s6% | 8B / 14B* | B | $649 | |
| #106 | NVIDIA GeForce RTX 4070 Super Ada Lovelace•GDDR6X•192-bit12GB High Efficiency | | 504 GB/s6% | 8B / 14B* | B | $599 | |
| #107 | NVIDIA GeForce RTX 4070 Ada Lovelace•GDDR6X•192-bit12GB Budget Ada | | 504 GB/s6% | 8B / 14B* | B | $549 | |
| #108 | NVIDIA RTX 4000 SFF Ada (20GB) Ada Lovelace•GDDR6 ECC•160-bitLow-Profile 70W 20GB SFF | | 280 GB/s4% | 14B / 32B* | A | $1,250 | |
| #109 | NVIDIA TITAN X Pascal (12GB) Pascal•GDDR5X•384-bit12GB Pascal Titan | | 480 GB/s6% | 8B / 14B* | C | $180 | |
| #110 | NVIDIA GeForce GTX 1080 Ti Pascal•GDDR5X•352-bit11GB Pascal Vintage Legend | | 484 GB/s6% | 8B / 14B* | C | $160 | |
| #111 | Intel Arc B580 (12GB Battlemage) Battlemage•GDDR6•192-bit2025 $249 12GB Budget Hero | | 456 GB/s6% | 8B / 14B* | S | $249 | |
| #112 | Apple M4 (32GB Unified) Apple Silicon•Unified LPDDR5X•128-bit32GB High-VRAM Mini/Air | 统一内存 | 120 GB/s2% | 32B / 70B* | B | $1,599 | |
| #113 | Apple M4 Pro (24GB Unified) Apple Silicon•Unified LPDDR5X•256-bit24GB Base M4 Pro Studio | 统一内存 | 273 GB/s3% | 14B / 32B* | A | $1,999 | |
| #114 | NVIDIA RTX 4080 Laptop (12GB) Ada Lovelace•GDDR6•192-bit12GB Mobile High Performance | | 432 GB/s5% | 8B / 14B* | B | $1,700 | |
| #115 | AMD Radeon RX 7700 XT RDNA 3•GDDR6•192-bit12GB RDNA3 Mid-Range | | 432 GB/s5% | 8B / 14B* | B | $399 | |
| #116 | AMD Radeon RX 6750 XT RDNA 2•GDDR6•192-bit12GB Fast Clocked RDNA2 | | 432 GB/s5% | 8B / 14B* | B | $349 | |
| #117 | NVIDIA Tesla T4 (16GB) Turing•GDDR6•256-bitLegacy 70W 16GB Cloud Baseline | | 320 GB/s4% | 14B / 32B* | B | $900 | |
| #118 | Intel Arc A750 (8GB) Alchemist•GDDR6•256-bitFast Bandwidth but 8GB VRAM Limit | | 512 GB/s6% | 7B (Tight) | F | $199 | |
| #119 | Intel Arc A580 (8GB) Alchemist•GDDR6•256-bitBudget 256-bit Arc | | 512 GB/s6% | 7B (Tight) | F | $169 | |
| #120 | NVIDIA GeForce RTX 2080 Super Turing•GDDR6•256-bit8GB Fast Turing | | 496 GB/s6% | 7B (Tight) | C | $210 | |
| #121 | NVIDIA GeForce RTX 4060 Ti (16GB) Ada Lovelace•GDDR6•128-bitAffordable 16GB VRAM Champion | | 288 GB/s4% | 14B / 32B* | A | $449 | |
| #122 | AMD Radeon PRO W7700 (16GB) RDNA 3•GDDR6 ECC•128-bit16GB RDNA3 Workstation | | 288 GB/s4% | 14B / 32B* | B | $999 | |
| #123 | AMD Radeon RX 7600 XT RDNA 3•GDDR6•128-bitBudget 16GB VRAM Entry Card | | 288 GB/s4% | 14B / 32B* | A | $329 | |
| #124 | AMD Radeon RX 6700 XT RDNA 2•GDDR6•192-bit12GB RDNA2 192-bit Budget | | 384 GB/s5% | 8B / 14B* | B | $319 | |
| #125 | Intel Arc Pro A60 (12GB) Alchemist•GDDR6•192-bit12GB Intel Pro Workstation | | 384 GB/s5% | 8B / 14B* | B | $450 | |
| #126 | NVIDIA GeForce RTX 3060 (12GB) Ampere•GDDR6•192-bitAll-Time Budget LLM Champion | | 360 GB/s5% | 8B / 14B* | S | $279 | |
| #127 | NVIDIA RTX 3070 Ti Laptop (8GB) Ampere•GDDR6•256-bit8GB Ampere High-End Laptop | | 448 GB/s6% | 7B (Tight) | F | $850 | |
| #128 | NVIDIA RTX 3070 Laptop (8GB) Ampere•GDDR6•256-bit8GB Ampere Laptop Standard | | 448 GB/s6% | 7B (Tight) | F | $750 | |
| #129 | NVIDIA GeForce RTX 3070 Ampere•GDDR6•256-bit8GB VRAM Trap | | 448 GB/s6% | 7B (Tight) | F | $289 | |
| #130 | NVIDIA GeForce RTX 2080 Turing•GDDR6•256-bit8GB Turing Base | | 448 GB/s6% | 7B (Tight) | C | $180 | |
| #131 | NVIDIA GeForce RTX 2070 Super Turing•GDDR6•256-bit8GB Turing Workhorse | | 448 GB/s6% | 7B (Tight) | C | $170 | |
| #132 | NVIDIA GeForce RTX 2070 Turing•GDDR6•256-bit8GB Turing Classic | | 448 GB/s6% | 7B (Tight) | C | $150 | |
| #133 | NVIDIA GeForce RTX 2060 Super Turing•GDDR6•256-bit8GB 256-bit Budget Turing | | 448 GB/s6% | 7B (Tight) | C | $160 | |
| #134 | NVIDIA GeForce RTX 2060 (12GB) Turing•GDDR6•192-bit12GB Budget Turing Hero | | 336 GB/s4% | 8B / 14B* | B | $170 | |
| #135 | Intel Arc B570 (10GB Battlemage) Battlemage•GDDR6•160-bitSub-$220 Battlemage Entry | | 380 GB/s5% | 8B / 14B* | B | $219 | |
| #136 | NVIDIA Quadro RTX 4000 (8GB) Turing•GDDR6•256-bitSingle-Slot 8GB Turing | | 416 GB/s5% | 7B (Tight) | C | $320 | |
| #137 | NVIDIA RTX A2000 (12GB) Ampere•GDDR6 ECC•192-bit70W Low-Profile 12GB SFF Hero | | 288 GB/s4% | 8B / 14B* | B | $460 | |
| #138 | Apple M4 (24GB Unified) Apple Silicon•Unified LPDDR5X•128-bit24GB M4 Balanced | 统一内存 | 120 GB/s2% | 14B / 32B* | B | $1,299 | |
| #139 | AMD Radeon RX 6700 (10GB) RDNA 2•GDDR6•160-bit10GB RDNA2 160-bit | | 320 GB/s4% | 8B / 14B* | C | $240 | |
| #140 | Apple M3 (24GB Unified) Apple Silicon•Unified LPDDR5•128-bit24GB M3 Mini/Air | 统一内存 | 100 GB/s1% | 14B / 32B* | B | $1,399 | |
| #141 | Apple M2 (24GB Unified) Apple Silicon•Unified LPDDR5•128-bit24GB M2 Mini/MacBook | 统一内存 | 100 GB/s1% | 14B / 32B* | B | $1,199 | |
| #142 | NVIDIA GeForce GTX 1080 Pascal•GDDR5X•256-bit8GB Pascal Classic | | 320 GB/s4% | 7B (Tight) | F | $110 | |
| #143 | Apple M2 Pro (16GB Unified) Apple Silicon•Unified LPDDR5•256-bit16GB M2 Pro Mini | 统一内存 | 200 GB/s3% | 8B / 14B* | C | $1,299 | |
| #144 | Apple M1 Pro (16GB Unified) Apple Silicon•Unified LPDDR5•256-bit16GB M1 Pro MacBook | 统一内存 | 200 GB/s3% | 8B / 14B* | C | $850 | |
| #145 | Apple M3 Pro (18GB Unified) Apple Silicon•Unified LPDDR5•192-bit18GB Base M3 Pro | 统一内存 | 150 GB/s2% | 8B / 14B* | B | $1,799 | |
| #146 | NVIDIA GeForce RTX 4060 Ti (8GB) Ada Lovelace•GDDR6•128-bit8GB VRAM Bottleneck Trap | | 288 GB/s4% | 7B (Tight) | F | $369 | |
| #147 | AMD Radeon RX 7600 (8GB) RDNA 3•GDDR6•128-bit8GB RDNA3 Entry | | 288 GB/s4% | 7B (Tight) | C | $269 | |
| #148 | NVIDIA RTX 3060 Laptop (6GB) Ampere•GDDR6•192-bit6GB Mobile Ampere | | 336 GB/s4% | 7B (Tight) | F | $600 | |
| #149 | NVIDIA GeForce RTX 2060 (6GB) Turing•GDDR6•192-bit6GB Turing Entry | | 336 GB/s4% | 7B (Tight) | F | $120 | |
| #150 | NVIDIA GeForce GTX 1660 Super Turing•GDDR6•192-bit6GB Turing without Tensor | | 336 GB/s4% | 7B (Tight) | F | $100 | |
| #151 | AMD Radeon RX 6650 XT RDNA 2•GDDR6•128-bit8GB RDNA2 Refreshed | | 280 GB/s4% | 7B (Tight) | C | $229 | |
| #152 | NVIDIA GeForce RTX 4060 (8GB) Ada Lovelace•GDDR6•128-bit8GB Entry Ada | | 272 GB/s3% | 7B (Tight) | F | $299 | |
| #153 | NVIDIA RTX 4070 Laptop (8GB) Ada Lovelace•GDDR6•128-bit8GB Mobile Ada (Limited for LLM) | | 256 GB/s3% | 7B (Tight) | F | $1,200 | |
| #154 | NVIDIA RTX 4060 Laptop (8GB) Ada Lovelace•GDDR6•128-bit8GB Mainstream Mobile | | 256 GB/s3% | 7B (Tight) | F | $950 | |
| #155 | NVIDIA GeForce GTX 1070 Ti Pascal•GDDR5•256-bit8GB Pascal Mid-Tier | | 256 GB/s3% | 7B (Tight) | F | $95 | |
| #156 | NVIDIA GeForce GTX 1070 Pascal•GDDR5•256-bit8GB 150W Pascal | | 256 GB/s3% | 7B (Tight) | F | $85 | |
| #157 | AMD Radeon RX 6600 XT RDNA 2•GDDR6•128-bit8GB RDNA2 Standard | | 256 GB/s3% | 7B (Tight) | C | $199 | |
| #158 | NVIDIA GeForce RTX 3060 (8GB) Ampere•GDDR6•128-bit128-bit Cutdown Trap (Avoid) | | 240 GB/s3% | 7B (Tight) | F | $219 | |
| #159 | NVIDIA GeForce GTX 1660 Ti Turing•GDDR6•192-bit6GB Turing 120W | | 288 GB/s4% | 7B (Tight) | F | $95 | |
| #160 | NVIDIA RTX A2000 (6GB) Ampere•GDDR6 ECC•192-bit70W Low-Profile 6GB SFF | | 288 GB/s4% | 7B (Tight) | C | $280 | |
| #161 | Apple M4 (16GB Unified) Apple Silicon•Unified LPDDR5X•128-bitBase 16GB Entry Silicon | 统一内存 | 120 GB/s2% | 8B / 14B* | B | $999 | |
| #162 | NVIDIA GeForce RTX 3050 (8GB) Ampere•GDDR6•128-bit8GB Entry Ampere | | 224 GB/s3% | 7B (Tight) | F | $179 | |
| #163 | AMD Radeon RX 6600 (8GB) RDNA 2•GDDR6•128-bit8GB 132W Low Power RDNA2 | | 224 GB/s3% | 7B (Tight) | C | $189 | |
| #164 | Apple M3 (16GB Unified) Apple Silicon•Unified LPDDR5•128-bit16GB M3 Entry Level | 统一内存 | 100 GB/s1% | 8B / 14B* | C | $1,199 | |
| #165 | Apple M2 (16GB Unified) Apple Silicon•Unified LPDDR5•128-bit16GB M2 Entry | 统一内存 | 100 GB/s1% | 8B / 14B* | C | $999 | |
| #166 | Apple M1 (16GB Unified) Apple Silicon•Unified LPDDR4X•128-bit16GB Original Apple Silicon | 统一内存 | 68.25 GB/s1% | 8B / 14B* | C | $499 | |
| #167 | NVIDIA RTX 4050 Laptop (6GB) Ada Lovelace•GDDR6•96-bit6GB Budget Mobile | | 192 GB/s2% | 7B (Tight) | F | $750 | |
| #168 | NVIDIA GeForce GTX 1060 (6GB) Pascal•GDDR5•192-bit6GB Pascal Vintage Entry | | 192 GB/s2% | 7B (Tight) | F | $70 | |
| #169 | Intel Arc Pro A50 (6GB) Alchemist•GDDR6•96-bit75W Single-Slot 6GB Intel Pro | | 192 GB/s2% | 7B (Tight) | C | $299 | |
| #170 | Intel Arc Pro A40 (6GB) Alchemist•GDDR6•96-bit50W Low-Profile 6GB Intel Pro | | 192 GB/s2% | 7B (Tight) | C | $249 | |
| #171 | Intel Arc A380 (6GB) Alchemist•GDDR6•96-bit75W Sub-$130 6GB AV1 / AI Entry | | 186 GB/s2% | 7B (Tight) | C | $129 | |
| #172 | NVIDIA GeForce RTX 3050 (6GB) Ampere•GDDR6•96-bit70W Low-Power 6GB Entry | | 168 GB/s2% | 7B (Tight) | F | $149 | |
| #173 | Apple M3 (8GB Unified) Apple Silicon•Unified LPDDR5•128-bit8GB Unified Memory Bottleneck | 统一内存 | 100 GB/s1% | 7B (Tight) | F | $899 | |
| #174 | Apple M2 (8GB Unified) Apple Silicon•Unified LPDDR5•128-bit8GB Legacy Baseline | 统一内存 | 100 GB/s1% | 7B (Tight) | F | $699 | |
| #175 | Intel Arc A310 (4GB) Alchemist•GDDR6•64-bit4GB Ultra-Budget Video/Inference | | 124 GB/s2% | 7B (Tight) | F | $99 | |
| #176 | Apple M1 (8GB Unified) Apple Silicon•Unified LPDDR4X•128-bit8GB Original Silicon Entry | 统一内存 | 68.25 GB/s1% | 7B (Tight) | F | $350 | |