NVIDIA B200 vs NVIDIA Tesla V100 PCIe 16 GB Comparison

Where Each One Wins

The benchmark data splits cleanly along generational lines. The NVIDIA B200 wins the only shared benchmark test, Geekbench OpenCL, with a score of 345,482 against the Tesla V100 PCIe 16 GB's 163,063. That is a 111.9% advantage, meaning the B200 more than doubles the V100's raw compute throughput in this workload. The B200 also holds a Geekbench Vulkan score of 113,062 from the V100, though the V100 has no Vulkan result in the data, so the comparison is incomplete. Still, the B200's overall benchmark percentile of 100 versus the V100's 96 shows it sits at the absolute top of the database, while the V100, despite being end-of-life, remains in the 96th percentile—a signal of how far ahead it was at launch.

The V100's wins are not about speed; they are about legacy and compatibility. It has a Geekbench Vulkan score of 113,062, which the B200 cannot match because the B200 has no Vulkan API support listed. For any workload that depends on Vulkan compute, the V100 is the only option of the two. The V100 also carries a DirectX 12 (12_1) API entry, while the B200 lists no DirectX support, making the older card the only choice for DX12-based applications. The V100's dual-slot PCIe form factor with 2x 8-pin power connectors and a 300 W TDP also makes it far easier to slot into existing servers, whereas the B200's SXM module form factor with a 1000 W TDP demands a specific chassis and power infrastructure. In raw compute, the B200 wins everywhere; in practical deployability and legacy software support, the V100 has the edge.

Architecture Differences

The two GPUs come from completely different design philosophies separated by two major architecture generations. The B200 uses the GB100 chip built on Blackwell architecture, fabricated on a 5 nm process at TSMC with 104,000 million transistors. The V100 uses the GV100 chip built on Volta architecture, fabricated on a 12 nm process at the same foundry with 21,100 million transistors. That is roughly a 5x difference in transistor count, packed into a die that the B200 does not report dimensions for, while the V100 has a die size of 815 mm² and a transistor density of 25.9M per mm².

The memory subsystems are radically different. The B200 pairs 90 GB of HBM3e memory with a 4096-bit bus, delivering 4.10 TB/s of bandwidth. The V100 offers 16 GB of HBM2 on the same 4096-bit bus width, but bandwidth drops to 897.0 GB/s. The B200's memory clock is listed at 2000 MHz with 8 Gbps effective, while the V100 runs at 876 MHz with 1752 Mbps effective. The B200 also uses a PCIe 5.0 x16 interface, whereas the V100 uses PCIe 3.0 x16—a two-generation gap in host connectivity.

Compute resources differ just as dramatically. The B200 has 18,944 shading units, 592 TMUs, and 24 ROPs, plus 592 tensor cores. The V100 has 5,120 shading units, 320 TMUs, 128 ROPs, and 640 tensor cores. Interestingly, the V100 has more ROPs and more tensor cores, but the B200's tensor cores are built for Blackwell's 16:1 FP16 ratio, while the V100's are Volta-era units at 2:1. The B200's FP16 throughput is 1,191.2 TFLOPS (16:1), against the V100's 28.26 TFLOPS (2:1). The B200 also has a much higher boost clock of 1965 MHz versus the V100's 1380 MHz, though the B200's base clock is lower at 700 MHz versus 1245 MHz. Both cards lack display outputs, and both have no RT cores listed. The B200 is marked as Active production, while the V100 is End-of-life, released on 2017-06-20.

Head-to-Head Benchmarks

The only direct head-to-head benchmark available is Geekbench OpenCL, and the result is lopsided. The B200 scores 345,482, while the V100 scores 163,063, giving the B200 a 111.9% delta. To put that in perspective, the B200's nearest rivals in the database are the NVIDIA B300 SXM6 AC at 369,831 (6.6% faster than the B200), the NVIDIA H200 NVL at 334,891 (3.2% slower), the AMD Instinct MI300X at 317,994 (8.6% slower), and the NVIDIA L40S at 295,763 (16.8% slower). The V100, by contrast, sits among much older and weaker competitors: the NVIDIA Tesla V100 SXM2 32 GB at 137,731 (0.2% slower), the AMD Instinct MI100 at 139,035 (0.7% faster), the AMD Radeon PRO V620 at 136,472 (1.2% slower), and the AMD Radeon Pro W6800X Duo at 135,774 (1.7% slower). The B200 is in a different performance class entirely, competing with the current flagship accelerators, while the V100 trades blows with cards from the same era.

The B200's FP32 throughput is 74.45 TFLOPS, versus the V100's 14.13 TFLOPS—a 5.3x difference. Pixel rate is the one metric where the V100 wins: 176.6 GPixel/s versus the B200's 47.16 GPixel/s, because the V100 has 128 ROPs against the B200's 24. Texture rate favors the B200 at 1,163.3 GTexel/s versus 441.6 GTexel/s. The B200's memory bandwidth of 4.10 TB/s is 4.6x the V100's 897.0 GB/s. On paper, the B200 is the clear winner in every compute-heavy metric, with the V100 only leading in pixel fill rate—a metric that matters little for the AI and HPC workloads these cards target.

Specification Differences

| Specification | NVIDIA B200 | NVIDIA Tesla V100 PCIe 16 GB |

|---|---|---|

| Chip | GB100 | GV100 |

| Architecture | Blackwell | Volta |

| Generation | Server Blackwell (Bxx) | Tesla Volta (Vxx) |

| Process Node | 5 nm | 12 nm |

| Transistors | 104,000 million | 21,100 million |

| Die Size | Not listed | 815 mm² |

| Transistor Density | Not listed | 25.9M / mm² |

| Base Clock | 700 MHz | 1245 MHz |

| Boost Clock | 1965 MHz | 1380 MHz |

| Memory Clock | 2000 MHz (8 Gbps effective) | 876 MHz (1752 Mbps effective) |

| Memory Size | 90 GB | 16 GB |

| Memory Type | HBM3e | HBM2 |

| Memory Bus | 4096 bit | 4096 bit |

| Memory Bandwidth | 4.10 TB/s | 897.0 GB/s |

| Shading Units | 18,944 | 5,120 |

| TMUs | 592 | 320 |

| ROPs | 24 | 128 |

| Tensor Cores | 592 | 640 |

| FP32 | 74.45 TFLOPS | 14.13 TFLOPS |

| FP16 | 1,191.2 TFLOPS (16:1) | 28.26 TFLOPS (2:1) |

| Pixel Rate | 47.16 GPixel/s | 176.6 GPixel/s |

| Texture Rate | 1,163.3 GTexel/s | 441.6 GTexel/s |

| TDP | 1000 W | 300 W |

| Slot Width | SXM Module | Dual-slot |

| Power Connectors | Not listed | 2x 8-pin |

| Suggested PSU | 1400 W | 700 W |

| Bus Interface | PCIe 5.0 x16 | PCIe 3.0 x16 |

| DirectX | Not listed | 12 (12_1) |

| OpenGL | Not listed | 4.6 |

| Vulkan | Not listed | 1.4 |

| Production Status | Active | End-of-life |

| Release Date | Not listed | 2017-06-20 |

| Predecessor | Server Hopper | Tesla Pascal |

| Successor | Server Rubin | Tesla Turing |

FAQ

Q: Which GPU is faster in OpenCL compute?

A: The NVIDIA B200 scores 345,482 in Geekbench OpenCL, which is 111.9% higher than the Tesla V100 PCIe 16 GB's 163,063. The B200 more than doubles the V100's score.

Q: Can the Tesla V100 run Vulkan workloads that the B200 cannot?

A: Yes. The V100 lists Vulkan 1.4 API support and has a Geekbench Vulkan score of 113,062. The B200 lists no Vulkan API support and has no Vulkan benchmark result.

Q: How does the B200's memory bandwidth compare to the V100's?

A: The B200 has 4.10 TB/s of bandwidth from 90 GB of HBM3e memory, while the V100 has 897.0 GB/s from 16 GB of HBM2. The B200 delivers roughly 4.6x the bandwidth.

Q: Why does the V100 have a higher pixel rate than the B200?

A: The V100 has 128 ROPs compared to the B200's 24 ROPs, giving it a pixel rate of 176.6 GPixel/s versus the B200's 47.16 GPixel/s. This is an artifact of the V100's older, more rasterization-focused design.

Q: What are the power requirements for each card?

A: The B200 has a 1000 W TDP and requires a 1400 W suggested PSU, while the V100 has a 300 W TDP and requires a 700 W suggested PSU. The B200 uses an SXM module form factor, whereas the V100 is a dual-slot PCIe card with 2x 8-pin power connectors.

Q: Is the V100 still in production?

A: No. The V100 is marked as End-of-life and was released on 2017-06-20. The B200 is marked as Active production.

The Verdict

The data is unambiguous for raw compute: the NVIDIA B200 is the superior accelerator by every measurable compute metric. It more than doubles the V100's OpenCL score (345,482 vs 163,063), offers 5.3x the FP32 throughput (74.45 TFLOPS vs 14.13 TFLOPS), and provides 4.6x the memory bandwidth (4.10 TB/s vs 897.0 GB/s). Its 90 GB of HBM3e memory versus the V100's 16 GB of HBM2 means it can handle far larger models and datasets without spilling to host memory. The B200's nearest rivals are current flagship accelerators like the H200 NVL and MI300X, while the V100's rivals are older cards like the MI100 and Radeon PRO V620. If you are building a new compute node and the workload is memory-bandwidth-bound or FP16-heavy, the B200 is the only rational choice.

However, the V100 is not without a niche. It is the only one of the two with Vulkan support (1.4), DirectX 12 (12_1) support, and a Geekbench Vulkan score of 113,062. It has a 300 W TDP, dual-slot PCIe form factor, and 2x 8-pin power connectors, making it drop-in compatible with older servers that lack the power delivery and chassis requirements of an SXM module. The V100's 128 ROPs give it a pixel rate of 176.6 GPixel/s versus the B200's 47.16 GPixel/s, which could matter for unusual rasterization workloads, though neither card has display outputs. For legacy software that only supports Volta's compute capabilities, or for systems where a 1000 W SXM module is physically impossible, the V100 remains functional. But the data says it clearly: the B200 is a generation-defining leap, and the V100 is a relic whose only advantages are compatibility and power draw. Choose the B200 for performance, the V100 only if your software or hardware constraints leave you no alternative.