NVIDIA B200 vs NVIDIA Tesla P100 PCIe 16 GB Comparison

FAQ

Q: How does the NVIDIA B200 compare to the Tesla P100 PCIe 16 GB in the recorded benchmark?

A: The database contains one head-to-head benchmark, Geekbench OpenCL, where the B200 scores 345,482 against the Tesla P100’s 79,605. That is a 334% higher score, making the B200 the outright winner in this comparison.

Q: What is the performance gap between the B200 and its closest rival?

A: The B200’s average benchmark score is 345,482. Its nearest rival, the NVIDIA H200 NVL, scores 334,891, which is 3.2% lower. The B300 SXM6 AC sits 6.6% above the B200 with 369,831, while the AMD Instinct MI300X trails by 8.6% at 317,994.

Q: Where does the Tesla P100 rank among all GPUs?

A: The Tesla P100 sits at the 92nd percentile of all GPUs in the database, with an average score of 79,605. Its closest competitor, the Tesla P100 PCIe 12 GB, is nearly identical at 79,396, only 0.3% behind.

Q: What architecture generation does each card belong to?

A: The B200 uses the Blackwell architecture on a 5 nm process, while the Tesla P100 uses the older Pascal architecture on a 16 nm process. The B200 is built on the GB100 chip; the Tesla P100 uses the GP100 chip.

Q: Are both cards still in production?

A: No. The B200 is listed as active in production, while the Tesla P100 is marked as end-of-life. The Tesla P100 was released in June 2016; the B200 has no release date recorded.

Q: Which card has more memory and bandwidth?

A: The B200 has 90 GB of HBM3e memory with a 4.10 TB/s bandwidth, while the Tesla P100 has 16 GB of HBM2 with 732.2 GB/s bandwidth. Both use a 4096-bit bus.

Architecture Differences

The B200 and Tesla P100 represent two distinct eras of NVIDIA server hardware. The B200 is built on the Blackwell architecture using the GB100 chip, fabricated on a 5 nm process at TSMC. It packs 104,000 million transistors. The Tesla P100, by contrast, uses the Pascal architecture with the GP100 chip on a 16 nm process, holding 15,300 million transistors. The B200’s transistor count is roughly 6.8 times higher, reflecting the generational leap in manufacturing density.

The B200 is a pure compute accelerator with no display outputs and no API support listed for DirectX, OpenGL, or Vulkan. The Tesla P100 also has no display outputs, but it does carry API support: DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. This difference matters for any workload that relies on graphics APIs, though both cards are primarily aimed at compute tasks.

In terms of compute resources, the B200 has 18,944 shading units, 592 TMUs, and 24 ROPs. The Tesla P100 has 3,584 shading units, 224 TMUs, and 96 ROPs. The B200 also includes 592 tensor cores, while the Tesla P100 has none. The B200’s tensor cores enable AI and deep learning acceleration that the older Pascal card simply cannot perform natively.

The B200’s clock speeds are unusual: a base clock of 700 MHz and a boost clock of 1,965 MHz. The Tesla P100 runs at 1,190 MHz base and 1,329 MHz boost. Despite the lower base clock, the B200’s massive shader count and higher boost clock deliver far greater throughput. The memory clocks also differ sharply: the B200 runs at 2,000 MHz with 8 Gbps effective, while the Tesla P100 runs at 715 MHz with 1,430 Mbps effective.

Power and physical design separate the two as well. The B200 is an SXM module with a 1000 W TDP and a suggested PSU of 1400 W. The Tesla P100 is a dual-slot PCIe card, 267 mm long, with a 250 W TDP, a single 8-pin power connector, and a suggested PSU of 600 W. The B200 uses PCIe 5.0 x16, while the Tesla P100 uses PCIe 3.0 x16.

Where Each One Wins

The B200 wins decisively in raw compute throughput, memory capacity, and memory bandwidth. Its 90 GB of HBM3e and 4.10 TB/s bandwidth dwarf the Tesla P100’s 16 GB and 732.2 GB/s. For workloads that load large models or datasets into VRAM, the B200 has a clear advantage. The 592 tensor cores also give it a lead in any matrix-multiply-heavy task, including neural network training and inference.

The Tesla P100’s wins are narrower but real. It has a higher pixel rate: 127.6 GPixel/s versus the B200’s 47.16 GPixel/s. It also has four times the ROPs (96 versus 24), which helps in rasterization-style tasks. The Tesla P100 carries graphics API support for DirectX 12, OpenGL 4.6, and Vulkan 1.3, which the B200 lacks entirely. If a workload requires those APIs, the Tesla P100 is the only option of the two.

The Tesla P100 also wins on power efficiency in a strict sense: its TDP is 250 W versus the B200’s 1000 W. The B200 delivers far more performance per watt, however, given its 334% higher benchmark score at only 4 times the power draw. The Tesla P100’s smaller footprint (dual-slot PCIe versus SXM module) makes it easier to slot into existing servers, but that is a physical constraint, not a performance advantage.

For legacy compatibility, the Tesla P100 supports PCIe 3.0, while the B200 uses PCIe 5.0. Older systems with PCIe 3.0 slots can still run the Tesla P100, whereas the B200 requires a newer platform. The Tesla P100’s release in June 2016 also means it has a long track record of driver maturity, though it is now end-of-life.

Specification Differences

| Specification | NVIDIA B200 | NVIDIA Tesla P100 PCIe 16 GB |

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

| Architecture | Blackwell | Pascal |

| Process Node | 5 nm | 16 nm |

| Transistors | 104,000 million | 15,300 million |

| Die Size | Not listed | 610 mm² |

| Base Clock | 700 MHz | 1,190 MHz |

| Boost Clock | 1,965 MHz | 1,329 MHz |

| Memory Size | 90 GB | 16 GB |

| Memory Type | HBM3e | HBM2 |

| Memory Clock | 2,000 MHz (8 Gbps effective) | 715 MHz (1,430 Mbps effective) |

| Memory Bandwidth | 4.10 TB/s | 732.2 GB/s |

| Shading Units | 18,944 | 3,584 |

| TMUs | 592 | 224 |

| ROPs | 24 | 96 |

| Tensor Cores | 592 | None |

| FP32 Performance | 74.45 TFLOPS | 9.526 TFLOPS |

| FP16 Performance | 1,191.2 TFLOPS (16:1) | 19.05 TFLOPS (2:1) |

| TDP | 1000 W | 250 W |

| Slot Width | SXM Module | Dual-slot |

| Power Connectors | Not listed | 1x 8-pin |

| Suggested PSU | 1400 W | 600 W |

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

| Display Outputs | No outputs | No outputs |

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

| OpenGL | Not listed | 4.6 |

| Vulkan | Not listed | 1.3 |

| Length | Not listed | 267 mm (10.5 inches) |

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

| Release Date | Not listed | 2016-06-19 |

| Predecessor | Server Hopper | Tesla Maxwell |

| Successor | Server Rubin | Tesla Volta |

Head-to-Head Benchmarks

The only recorded head-to-head test is Geekbench OpenCL. The B200 scores 345,482, and the Tesla P100 scores 79,605. The delta is 334% in favor of the B200. That is a massive gap, more than a fourfold difference in raw compute throughput. The B200 wins this comparison outright, with a 1-0 record in the database.

To put that score in context, the B200’s nearest rivals in the database are the H200 NVL at 334,891 (3.2% lower), the B300 SXM6 AC at 369,831 (6.6% higher), the AMD Instinct MI300X at 317,994 (8.6% lower), and the L40S at 295,763 (16.8% lower). The B200 sits at the 100th percentile of all GPUs, meaning no other GPU in the database scores higher on average.

The Tesla P100’s score of 79,605 places it at the 92nd percentile. Its nearest rivals are the Tesla P100 PCIe 12 GB at 79,396 (0.3% lower), the GeForce RTX 5090 at 79,842 (0.3% higher), the Radeon RX 6850M XT at 78,940 (0.8% lower), and the Radeon Pro Vega 64X at 80,959 (1.7% higher). The Tesla P100 is essentially in a dead heat with those cards, all within a 2% band.

The B200’s FP32 throughput is 74.45 TFLOPS versus the Tesla P100’s 9.526 TFLOPS, a ratio of roughly 7.8 to 1. In FP16, the B200 hits 1,191.2 TFLOPS with a 16:1 ratio, while the Tesla P100 manages 19.05 TFLOPS with a 2:1 ratio. The B200’s FP16 advantage is about 62.5 times, a reflection of its tensor core design for AI workloads.

Texture rate also favors the B200: 1,163.3 GTexel/s versus 297.7 GTexel/s. The pixel rate flips the other way: the Tesla P100 delivers 127.6 GPixel/s, while the B200 delivers 47.16 GPixel/s. That is one area where the older card leads, driven by its higher ROP count.

Memory bandwidth differences are dramatic: the B200’s 4.10 TB/s is 5.6 times the Tesla P100’s 732.2 GB/s. Combined with 90 GB versus 16 GB capacity, the B200 can hold larger working sets and feed them faster.

The Verdict

The data points to one clear conclusion: the NVIDIA B200 is in a different performance class from the Tesla P100 PCIe 16 GB. The 334% Geekbench OpenCL delta is the largest single benchmark gap in this comparison, and the B200’s 100th percentile rank confirms it as the top-scoring GPU in the database. The B200 wins on shading units, tensor cores, memory size, memory bandwidth, FP32, and FP16, and it uses a newer architecture on a smaller process node.

The Tesla P100 retains specific advantages: higher pixel rate, more ROPs, graphics API support, lower power draw, and a physical PCIe card form factor. It is also end-of-life, so new deployments would rely on existing stock or used units. Its 92nd percentile rank is respectable for a 2016 card, but it cannot match the B200 in any compute-heavy benchmark.

For buyers choosing between these two, the decision depends entirely on the workload. If the task involves large-scale AI training, dense matrix math, or massive memory footprints, the B200 is the only rational choice. Its 90 GB capacity and tensor cores are purpose-built for that. If the workload requires graphics API compatibility, lower power draw, or a legacy PCIe 3.0 slot, the Tesla P100 may still serve, but it will do so at a fraction of the performance.

The launch MSRP for the Tesla P100 is 5,699 USD, recorded once in the database. The B200 has no launch MSRP listed. Production status is the final differentiator: the B200 is active, the Tesla P100 is end-of-life. The verdict is straightforward: the B200 wins the compute crown, while the Tesla P100 is a legacy option for niche compatibility needs.