Intel Iris Xe MAX Graphics vs NVIDIA P106-090 Comparison

# Head-to-Head Benchmarks

The single head-to-head benchmark available in the data is Geekbench OpenCL, and it delivers a decisive result: the NVIDIA P106-090 scores 21,304 against the Intel Iris Xe MAX Graphics' 14,315. That represents a 32.8% advantage for the NVIDIA part, a substantial margin that places the two products in different performance tiers despite their similar shading unit counts.

The Intel part's score of 14,315 places it within striking distance of its nearest rivals. The AMD Radeon Vega 11 scores 14,352, just 0.3% higher, while the NVIDIA GeForce GTX TITAN sits at 14,373, a 0.4% gap. The AMD Radeon RX Vega 11 leads that cluster at 14,385, 0.5% ahead. Interestingly, the NVIDIA GeForce GTX 1070 Ti is actually behind the Intel part in this particular test, scoring 14,277 for a 0.3% deficit. The Intel Iris Xe MAX is therefore not a weak performer in absolute terms — it sits comfortably in the mid-range of the GPU landscape, at the 56th percentile of all GPUs.

The NVIDIA P106-090, meanwhile, posts an OpenCL score of 21,304 that towers over its nearest listed rivals. The NVIDIA GeForce GTX 570 averages 13,515, a 0.3% gap, and the AMD Radeon Pro 555 scores 13,407, 0.5% behind. The AMD Radeon HD 7770M and AMD Radeon RX 9070 XT round out the comparison at 13,536 and 13,543 respectively. The P106-090's raw OpenCL score is roughly 57% higher than these rivals, indicating that the benchmark result is not merely a marginal edge but a substantial performance separation.

Yet the P106-090's average benchmark score across all tests — 13,470 — is actually lower than its OpenCL score, because it also includes a 3DMark Steel Nomad DX12 result of 509 and a Geekbench Vulkan score of 18,596. The Intel part's average is simply its single OpenCL score of 14,315. This means the two products' overall percentile rankings are close: Intel sits at the 56th percentile, NVIDIA at the 54th. The picture that emerges is nuanced — NVIDIA wins the head-to-head OpenCL test convincingly, but the broader benchmark portfolio suggests the Intel part has competitive strengths in specific workloads.

FAQ

Q: Which GPU wins the Geekbench OpenCL benchmark, and by how much?

A: The NVIDIA P106-090 wins decisively with a score of 21,304 versus the Intel Iris Xe MAX's 14,315, a 32.8% advantage.

Q: How does the Intel Iris Xe MAX compare to its nearest rivals?

A: The Intel part scores 14,315 in OpenCL, landing between the AMD Radeon Vega 11 (14,352, 0.3% higher) and the NVIDIA GeForce GTX 1070 Ti (14,277, 0.3% lower). It trails the NVIDIA GeForce GTX TITAN by 0.4% and the AMD Radeon RX Vega 11 by 0.5%.

Q: What explains the NVIDIA P106-090's lower average benchmark score despite its higher OpenCL result?

A: The P106-090's average of 13,470 includes a 3DMark Steel Nomad DX12 score of 509 and a Geekbench Vulkan score of 18,596, which drag the average below its OpenCL peak of 21,304. The Intel part has only one benchmark recorded, so its average equals its OpenCL score of 14,315.

Q: Are these GPUs aimed at different use cases based on the data?

A: The data suggests so. The Intel Iris Xe MAX is an integrated graphics processor (IGP) with a 25W TDP, while the NVIDIA P106-090 is a dual-slot discrete card with a 75W TDP and a 1x 6-pin power connector. The NVIDIA part's higher memory bandwidth (192.2 GB/s vs 68.26 GB/s) and larger bus width (192-bit vs 128-bit) point toward more demanding compute tasks.

Q: Which GPU has better raw compute throughput on paper?

A: The Intel part has a slight theoretical edge in FP32: 2.534 TFLOPS versus 2.352 TFLOPS for NVIDIA. However, the NVIDIA part has double the pixel fill rate (73.49 GPixel/s vs 39.60 GPixel/s) and a higher texture fill rate is not directly comparable since they differ (73.49 GTexel/s vs 79.20 GTexel/s).

Q: What do the percentile rankings tell us?

A: The Intel Iris Xe MAX sits at the 56th percentile of all GPUs, while the NVIDIA P106-090 sits at the 54th percentile. Despite the NVIDIA part's large OpenCL win, its inclusion of slower DX12 and Vulkan results brings its overall standing slightly below Intel's.

Architecture Differences

The two GPUs come from fundamentally different design philosophies. Intel's DG1 chip is built on a 10 nm process at Intel's own foundry, marking the company's entry into discrete-class graphics with its Generation 12.1 architecture. The die size is a compact 95 mm². NVIDIA's GP106, by contrast, is fabricated on a 16 nm process at TSMC, with a substantially larger die measuring 200 mm² and containing 4,400 million transistors. The transistor density figures underscore the process gap: Intel packs its transistors at a density that yields a 95 mm² die, while NVIDIA's larger, older process produces a density of 22.0M transistors per mm².

Memory architectures diverge sharply. The Intel Iris Xe MAX uses 4 GB of LPDDR4X on a 128-bit bus, delivering 68.26 GB/s of bandwidth. The NVIDIA P106-090 counters with 3 GB of GDDR5 on a 192-bit bus, achieving 192.2 GB/s — nearly three times the bandwidth. This difference is likely to matter in bandwidth-sensitive workloads like texture-heavy rendering or large data transfers.

The compute units are numerically identical in some respects: both have 768 shading units and 48 texture mapping units. But the render output units differ — Intel has 24, NVIDIA has 48. Clock speeds tell a telling story: Intel runs at a 300 MHz base and 1650 MHz boost, while NVIDIA runs at 1354 MHz base and 1531 MHz boost. Intel's boost clock is higher, but NVIDIA's base clock is over four times higher, suggesting NVIDIA maintains more consistent performance under load.

Memory clocks also differ: Intel's memory runs at 2133 MHz with 4.3 Gbps effective, while NVIDIA's runs at 2002 MHz with 8 Gbps effective. The effective data rate is nearly double on NVIDIA, which combined with the wider bus explains the bandwidth gap.

The FP16 compute rates reveal a striking architectural split. Intel achieves 5.069 TFLOPS FP16 via a 2:1 ratio, meaning its FP16 throughput is double its FP32. NVIDIA's FP16 is a minuscule 36.74 GFLOPS at a 1:64 ratio, indicating Pascal's near-total neglect of half-precision compute. This is a major differentiator for any workload that can exploit FP16.

Power and physical design differ as well. Intel's TDP is 25 W, and it's classified as an IGP (integrated graphics processor) with no display outputs. NVIDIA's TDP is 75 W, it's a dual-slot card requiring a 1x 6-pin power connector, and it measures 250 mm (9.8 inches) in length. The NVIDIA part also has no display outputs, consistent with its mining-oriented positioning. The bus interfaces differ: Intel uses PCIe 4.0 x8, while NVIDIA uses PCIe 1.0 x1 — a severe bandwidth limitation for the NVIDIA part that may bottleneck data transfer in some scenarios.

The API support is identical: both support DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. Both are end-of-life products, with Intel releasing on October 30, 2020, and NVIDIA on July 30, 2017.

The Verdict

The benchmark data paints a clear picture for raw OpenCL compute: the NVIDIA P106-090 is the superior choice, beating the Intel Iris Xe MAX by 32.8% in the only direct head-to-head test available. This is not a marginal win — it's a category-level difference that would translate to noticeably faster execution in OpenCL-based applications.

However, the data also shows that the Intel part has its own strengths. Its FP32 throughput (2.534 TFLOPS) exceeds NVIDIA's (2.352 TFLOPS), and its FP16 capability (5.069 TFLOPS) is in a different universe from NVIDIA's paltry 36.74 GFLOPS. For any workload leveraging half-precision arithmetic — common in machine learning inference or certain graphics effects — the Intel part would be dramatically faster.

The Intel part also benefits from a lower TDP (25 W vs 75 W), making it suitable for systems with tight power budgets. Its PCIe 4.0 x8 interface is far more modern than NVIDIA's PCIe 1.0 x1, which could bottleneck data transfer despite NVIDIA's higher memory bandwidth.

The percentile rankings suggest overall parity: Intel at the 56th percentile, NVIDIA at the 54th. The NVIDIA part's high OpenCL score is offset by its weaker DX12 and Vulkan results in the average calculation.

For users prioritizing OpenCL compute, the NVIDIA P106-090 is the obvious pick. For those who need FP16 performance, lower power draw, or a more modern PCIe interface, the Intel Iris Xe MAX has compelling advantages. The data does not support a universal winner — it supports a workload-dependent choice.

Specification Differences

| Specification | Intel Iris Xe MAX | NVIDIA P106-090 |

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

| Process node | 10 nm | 16 nm |

| Die size | 95 mm² | 200 mm² |

| Transistors | Not listed | 4,400 million |

| Transistor density | Not listed | 22.0M / mm² |

| Base clock | 300 MHz | 1354 MHz |

| Boost clock | 1650 MHz | 1531 MHz |

| Memory size | 4 GB | 3 GB |

| Memory type | LPDDR4X | GDDR5 |

| Memory bus width | 128 bit | 192 bit |

| Memory bandwidth | 68.26 GB/s | 192.2 GB/s |

| Memory clock | 2133 MHz / 4.3 Gbps effective | 2002 MHz / 8 Gbps effective |

| ROPs | 24 | 48 |

| Pixel rate | 39.60 GPixel/s | 73.49 GPixel/s |

| Texture rate | 79.20 GTexel/s | 73.49 GTexel/s |

| FP32 | 2.534 TFLOPS | 2.352 TFLOPS |

| FP16 | 5.069 TFLOPS (2:1) | 36.74 GFLOPS (1:64) |

| TDP | 25 W | 75 W |

| Slot width | IGP | Dual-slot |

| Power connectors | None | 1x 6-pin |

| Suggested PSU | 200 W | 250 W |

| Bus interface | PCIe 4.0 x8 | PCIe 1.0 x1 |

| Length | Not listed | 250 mm (9.8 inches) |

| Release date | 2020-10-30 | 2017-07-30 |

Where Each One Wins

Intel Iris Xe MAX wins on:

• FP32 compute (2.534 TFLOPS vs 2.352 TFLOPS)
• FP16 compute (5.069 TFLOPS vs 36.74 GFLOPS) — a massive advantage for half-precision workloads
• Power efficiency (25 W TDP vs 75 W TDP)
• Memory capacity (4 GB vs 3 GB)
• Modern PCIe interface (4.0 x8 vs 1.0 x1)
• Texture fill rate (79.20 GTexel/s vs 73.49 GTexel/s)
• Higher boost clock (1650 MHz vs 1531 MHz)

NVIDIA P106-090 wins on:

• OpenCL benchmark (21,304 vs 14,315, a 32.8% advantage)
• Memory bandwidth (192.2 GB/s vs 68.26 GB/s)
• Pixel fill rate (73.49 GPixel/s vs 39.60 GPixel/s)
• ROP count (48 vs 24)
• Higher base clock (1354 MHz vs 300 MHz)
• Larger die and more transistors (4,400 million vs not listed)
• Better percentile vs all GPUs in the head-to-head test

The data suggests a split personality: Intel's strength lies in modern compute features and efficiency, while NVIDIA's lies in raw memory throughput and pixel processing. The choice hinges on which workload profiles matter more. For bandwidth-hungry rasterization or large data movement, the P106-090's 192.2 GB/s is a decisive edge. For power-constrained systems or FP16-accelerated tasks, the Iris Xe MAX's architectural advantages are compelling. The 32.8% OpenCL gap is the single largest measured difference, and it favors NVIDIA.