NVIDIA A10G vs NVIDIA Tesla P100 PCIe 12 GB Comparison

Where Each One Wins

The benchmark data splits cleanly between these two server accelerators. The NVIDIA A10G wins the only head-to-head comparison in the database, the Geekbench OpenCL test, by a wide margin. It posts a score of 158063 against 79396 for the Tesla P100 PCIe 12 GB, a delta of 99.1%. That result places the A10G at the 97th percentile of all GPUs in the database, while the P100 sits at the 92nd percentile.

The A10G is the clear choice for compute-heavy workloads that leverage OpenCL, the dominant general-purpose GPU compute API in the database. Its average benchmark score of 151963 reflects a card that sits in the top tier of accelerators. The P100's average score of 79396 is roughly half that figure, and its nearest rival, the Tesla P100 PCIe 16 GB, scores only 0.3% higher at 79605. The older card is not without merit, but the data does not show any benchmark category where the P100 emerges victorious.

For workloads that rely on ray tracing or tensor operations, the A10G has dedicated hardware: 72 RT cores and 288 tensor cores. The P100 has none. In the database's API support records, the A10G reaches DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the P100 tops out at DirectX 12 (12_1) and Vulkan 1.3. Any application that requires those newer API features will only run on the A10G.

The P100 does retain one practical advantage: its launch MSRP was 4,599 USD. That figure is historical, and the production status for both cards is end-of-life, so it should be read as a reference point rather than a current market price. The P100 also uses HBM2 memory, which offers a different performance profile than the A10G's GDDR6, though the recorded bandwidth figures favor the A10G in aggregate.

Architecture Differences

The two cards come from different architectural generations. The A10G is built on the Ampere architecture with the GA102 chip, fabricated on an 8 nm process at Samsung. The P100 uses the Pascal architecture with the GP100 chip, fabricated on a 16 nm process at TSMC. That process difference is substantial: the A10G packs 28,300 million transistors into a 628 mm² die, yielding a transistor density of 45.1 million per square millimeter. The P100 fits 15,300 million transistors into a 610 mm² die, a density of 25.1 million per square millimeter. The A10G therefore crams nearly twice the transistor density into a slightly larger die.

The compute resources differ dramatically. The A10G has 9216 shading units, 288 texture mapping units, and 96 ROPs. The P100 has 3584 shading units, 224 TMUs, and 96 ROPs. The A10G also adds 72 RT cores and 288 tensor cores; the P100 has neither. The A10G's FP32 throughput is 31.52 TFLOPS, more than three times the P100's 9.526 TFLOPS. In FP16, the A10G maintains 31.52 TFLOPS with a 1:1 ratio, while the P100 reaches 19.05 TFLOPS but at a 2:1 ratio, meaning its FP16 rate is achieved by pairing two FP32 operations.

Memory architecture differs as well. The A10G uses 24 GB of GDDR6 on a 384-bit bus, delivering 600.2 GB/s of bandwidth. The P100 uses 12 GB of HBM2 on a much wider 3072-bit bus but delivers only 549.1 GB/s. The A10G's memory clock is 1563 MHz (12.5 Gbps effective), while the P100's is 715 MHz (1430 Mbps effective). The A10G's smaller bus width is compensated by its much higher per-pin data rate.

Power and physical design also diverge. The A10G has a 150 W TDP and is a single-slot card with an 8-pin EPS power connector and a suggested 450 W PSU. The P100 has a 250 W TDP, occupies a dual-slot design, uses a single 8-pin connector, and suggests a 600 W PSU. Both are 267 mm long and 10.5 inches, and both have no display outputs. The bus interface differs: the A10G uses PCIe 4.0 x16, the P100 uses PCIe 3.0 x16.

Head-to-Head Benchmarks

The database contains one direct comparison: Geekbench OpenCL. The A10G scores 158063, the P100 scores 79396, and the delta is 99.1% in favor of the A10G. That is a near-doubling of the raw score. It is worth contextualizing this against the nearest rivals for each card. The A10G sits 1.1% above the Tesla V100 PCIe 32 GB (150305) and 5.4% below the AMD Radeon Pro W6800X (160671). The P100 sits 0.3% below the Tesla P100 PCIe 16 GB (79605) and 0.6% below the GeForce RTX 5090 (79842).

The delta between the two cards in this test is larger than the gap between the A10G and any of its listed rivals. The A10G's advantage over the P100 is 99.1%, while its closest rival gap is 1.1%. This indicates that the performance difference between the A10G and P100 is not a marginal generational bump but a fundamental shift in compute capability. The P100's closest rival, the RTX 5090, is only 0.6% ahead, showing the P100 still trades blows with modern consumer parts in this workload. But it cannot approach the A10G.

The A10G's FP32 figure of 31.52 TFLOPS, combined with its 600.2 GB/s memory bandwidth, explains the OpenCL result. The P100's 9.526 TFLOPS FP32 and 549.1 GB/s bandwidth cannot match that throughput. The A10G also has a higher pixel rate (164.2 GPixel/s vs 127.6 GPixel/s) and texture rate (492.5 GTexel/s vs 297.7 GTexel/s), which contributes to compute-heavy workloads that stress those units.

Specification Differences

| Specification | NVIDIA A10G | NVIDIA Tesla P100 PCIe 12 GB |

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

| Architecture | Ampere | Pascal |

| Chip | GA102 | GP100 |

| Process node | 8 nm | 16 nm |

| Foundry | Samsung | TSMC |

| Transistors | 28,300 million | 15,300 million |

| Die size | 628 mm² | 610 mm² |

| Transistor density | 45.1M / mm² | 25.1M / mm² |

| Base clock | 1320 MHz | 1190 MHz |

| Boost clock | 1710 MHz | 1329 MHz |

| Memory clock | 1563 MHz (12.5 Gbps effective) | 715 MHz (1430 Mbps effective) |

| Memory size | 24 GB | 12 GB |

| Memory type | GDDR6 | HBM2 |

| Memory bus | 384 bit | 3072 bit |

| Memory bandwidth | 600.2 GB/s | 549.1 GB/s |

| Shading units | 9216 | 3584 |

| TMUs | 288 | 224 |

| ROPs | 96 | 96 |

| RT cores | 72 | None |

| Tensor cores | 288 | None |

| Pixel rate | 164.2 GPixel/s | 127.6 GPixel/s |

| Texture rate | 492.5 GTexel/s | 297.7 GTexel/s |

| FP32 | 31.52 TFLOPS | 9.526 TFLOPS |

| FP16 | 31.52 TFLOPS (1:1) | 19.05 TFLOPS (2:1) |

| TDP | 150 W | 250 W |

| Slot width | Single-slot | Dual-slot |

| Power connector | 8-pin EPS | 1x 8-pin |

| Suggested PSU | 450 W | 600 W |

| Bus interface | PCIe 4.0 x16 | PCIe 3.0 x16 |

| DirectX | 12 Ultimate (12_2) | 12 (12_1) |

| Vulkan | 1.4 | 1.3 |

| Length | 267 mm (10.5 inches) | 267 mm (10.5 inches) |

| Release date | 2021-04-11 | 2016-06-19 |

| Launch MSRP | None listed | 4,599 USD |

FAQ

Q: Which card is faster in OpenCL compute?

A: The NVIDIA A10G scores 158063 in Geekbench OpenCL, which is 99.1% higher than the Tesla P100 PCIe 12 GB's 79396.

Q: Does the Tesla P100 support ray tracing or tensor operations?

A: No. The P100 has no RT cores and no tensor cores. The A10G has 72 RT cores and 288 tensor cores.

Q: How do the memory bandwidth figures compare?

A: The A10G delivers 600.2 GB/s from 24 GB of GDDR6 on a 384-bit bus. The P100 delivers 549.1 GB/s from 12 GB of HBM2 on a 3072-bit bus. The A10G has higher bandwidth despite a much narrower bus.

Q: What is the power draw difference?

A: The A10G has a 150 W TDP and suggests a 450 W PSU. The P100 has a 250 W TDP and suggests a 600 W PSU. The A10G uses an 8-pin EPS connector, the P100 uses a single 8-pin connector.

Q: Are these cards still in production?

A: Both are end-of-life. The A10G was released on 2021-04-11, and the P100 was released on 2016-06-19.

Q: How does the A10G compare to its nearest rivals?

A: The A10G's average score is 151963, which is 1.1% above the Tesla V100 PCIe 32 GB and 9.3% above the AMD Instinct MI100, but 5.4% below the AMD Radeon Pro W6800X and 6.5% below the NVIDIA A100 PCIe 40 GB.