GPU Comparison

AMD Radeon Pro 575 and NVIDIA P104-100 represent two very different interpretations of a professional GPU. The AMD card is a mobile workstation component built on the GCN 4.0 architecture, while the NVIDIA part is a mining-oriented board stripped of display outputs. The database shows a clear performance split: the NVIDIA P104-100 wins both recorded head-to-head benchmarks, with a 33.9% lead in Geekbench OpenCL and a 16.1% lead in Geekbench Vulkan. However, the average benchmark scores tell a more nuanced story, as the AMD Radeon Pro 575 posts a higher overall average of 39,555 compared to the NVIDIA card’s 32,982. This discrepancy stems from the fact that the two cards were tested across different benchmark suites, with the AMD card’s results drawn from Geekbench Metal, OpenCL, and Vulkan, while the NVIDIA card’s results include a 3DMark Steel Nomad DX12 run that significantly drags down its aggregate.

Where Each One Wins

The NVIDIA P104-100 dominates in raw compute throughput. Its FP32 performance is rated at 6.655 TFLOPS, which is substantially higher than the AMD Radeon Pro 575’s 4.489 TFLOPS. This advantage shows up directly in the Geekbench OpenCL test, where the NVIDIA card scores 52,368 against the AMD card’s 34,596. The gap is 33.9%, a decisive margin that reflects the underlying hardware differences. The NVIDIA card also wins the Geekbench Vulkan test, scoring 45,165 versus 37,878, a 16.1% advantage. In both shared tests, the NVIDIA P104-100 is the clear winner.

The AMD Radeon Pro 575, by contrast, wins in the broader compatibility and feature set. It supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3, while the NVIDIA card supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4. The AMD card has a higher transistor density at 24.6M per mm² versus 22.9M per mm², and it operates on a 14 nm process from GlobalFoundries, compared to the NVIDIA card’s 16 nm TSMC process. The AMD card also has a higher average benchmark score of 39,555, which places it in the 82nd percentile of all GPUs, while the NVIDIA P104-100 sits in the 77th percentile despite its higher raw compute.

The NVIDIA card’s 3DMark Steel Nomad DX12 score of 1,413 is its only non-Geekbench result, and it reveals a weakness in modern DirectX 12 workloads. The AMD Radeon Pro 575 has no equivalent 3DMark result in the database, so a direct comparison in that specific test is impossible. However, the NVIDIA card’s low score in that test contributes to its lower average benchmark score and percentile ranking.

Architecture Differences

The two cards are built on fundamentally different architectures. The AMD Radeon Pro 575 uses the Ellesmere chip, which is based on GCN 4.0 and manufactured on a 14 nm process at GlobalFoundries. The chip contains 5,700 million transistors on a 232 mm² die, resulting in a transistor density of 24.6M per mm². The NVIDIA P104-100 uses the GP104 chip, based on the Pascal architecture, manufactured on a 16 nm process at TSMC. This chip contains 7,200 million transistors on a 314 mm² die, giving it a lower transistor density of 22.9M per mm². The NVIDIA chip is physically larger and has more transistors, but the AMD chip packs them more densely.

Memory configurations are similar in capacity but differ in type and bandwidth. Both cards have 4 GB of memory, but the AMD card uses GDDR5 while the NVIDIA card uses GDDR5X. The NVIDIA card’s memory runs at 10 Gbps effective and delivers 320.3 GB/s of bandwidth, while the AMD card’s memory runs at 6.8 Gbps effective and delivers 217.0 GB/s. The NVIDIA card has a 62.8% bandwidth advantage, which contributes to its higher texture rate of 208.0 GTexel/s compared to the AMD card’s 140.3 GTexel/s. The NVIDIA card also has a much higher pixel rate at 110.9 GPixel/s versus 35.07 GPixel/s, a difference driven by its 64 ROPs compared to the AMD card’s 32.

The compute unit configurations also differ. The AMD card has 2,048 shading units, 128 TMUs, and 32 ROPs. The NVIDIA card has 1,920 shading units, 120 TMUs, and 64 ROPs. Despite having fewer shading units and TMUs, the NVIDIA card achieves higher FP32 throughput due to its higher clock speeds. The NVIDIA card has a base clock of 1607 MHz and a boost clock of 1733 MHz, while the AMD card has no recorded base or boost clock in the database, only a memory clock of 1695 MHz. The NVIDIA card’s FP16 performance is notably weak at 104.0 GFLOPS, running at a 1:64 ratio to FP32, while the AMD card offers full-rate FP16 at 4.489 TFLOPS, a 1:1 ratio.

The physical and interface differences are stark. The AMD Radeon Pro 575 is an MXM Module with no power connectors and a PCIe 3.0 x16 interface, designed for portable devices with display outputs that are dependent on the host system. The NVIDIA P104-100 is a dual-slot card, 267 mm long (10.5 inches), with a single 8-pin power connector, a suggested power supply of 200 W, and no display outputs at all. The NVIDIA card uses a PCIe 1.0 x4 interface, which severely limits its bandwidth to the host system. The AMD card has a TDP of 150 W, while the NVIDIA card has no recorded TDP in the database.

The Verdict

The data indicates that the NVIDIA P104-100 is the stronger raw performer in compute-heavy workloads. Its 33.9% lead in Geekbench OpenCL and 16.1% lead in Geekbench Vulkan are substantial, and its higher FP32 throughput, texture rate, and pixel rate support its position as the faster card in head-to-head comparisons. The NVIDIA card’s higher memory bandwidth of 320.3 GB/s and its larger ROP count give it clear advantages in tasks that saturate memory or require high fill rates.

However, the AMD Radeon Pro 575 is the better-rounded product for professional use. Its average benchmark score of 39,555 is 19.9% higher than the NVIDIA card’s 32,982, and it ranks in the 82nd percentile of all GPUs versus the NVIDIA card’s 77th. The AMD card’s support for display outputs, its MXM form factor, and its full-rate FP16 capability make it a more practical choice for mobile workstations. The NVIDIA P104-100 has no display outputs, making it unsuitable for any workload that requires visual output, and its PCIe 1.0 x4 interface is a severe bottleneck for data transfer.

Users who need maximum compute performance in OpenCL or Vulkan workloads should choose the NVIDIA P104-100, provided they do not need display output and can work around the PCIe 1.0 x4 interface limitation. Users who need a professional mobile GPU with display support, a higher average benchmark score, and better architectural efficiency should choose the AMD Radeon Pro 575. The NVIDIA card is a mining-oriented part that excels at raw throughput but lacks the versatility of the AMD card.

FAQ

Q: Which card has a higher average benchmark score?

A: The AMD Radeon Pro 575 has an average benchmark score of 39,555, while the NVIDIA P104-100 has an average of 32,982. This places the AMD card in the 82nd percentile of all GPUs and the NVIDIA card in the 77th percentile.

Q: How does the memory bandwidth compare between the two cards?

A: The NVIDIA P104-100 has 320.3 GB/s of memory bandwidth from its GDDR5X memory running at 10 Gbps effective. The AMD Radeon Pro 575 has 217.0 GB/s from its GDDR5 memory running at 6.8 Gbps effective.

Q: Do both cards support display outputs?

A: No. The AMD Radeon Pro 575 has display outputs that are dependent on the portable device it is installed in. The NVIDIA P104-100 has no display outputs at all.

Q: Which card is better for FP16 compute?

A: The AMD Radeon Pro 575 offers full-rate FP16 at 4.489 TFLOPS, a 1:1 ratio to its FP32 performance. The NVIDIA P104-100 has FP16 performance of 104.0 GFLOPS, which is a 1:64 ratio to its FP32 performance, making it far weaker in FP16 workloads.

Q: What are the power requirements for each card?

A: The AMD Radeon Pro 575 has a TDP of 150 W and uses no power connectors. The NVIDIA P104-100 has no recorded TDP but requires a single 8-pin power connector and a suggested power supply of 200 W.

Q: Which card has more shading units?

A: The AMD Radeon Pro 575 has 2,048 shading units, while the NVIDIA P104-100 has 1,920. Despite having fewer shading units, the NVIDIA card achieves higher FP32 throughput due to its higher clock speeds.

Head-to-Head Benchmarks

The two cards share two recorded benchmarks in the database: Geekbench OpenCL and Geekbench Vulkan. The NVIDIA P104-100 wins both, but the margins differ significantly.

In Geekbench OpenCL, the NVIDIA P104-100 scores 52,368 against the AMD Radeon Pro 575’s 34,596. The delta is 33.9% in favor of the NVIDIA card. This is the largest gap between the two cards in any shared test. The NVIDIA card’s FP32 throughput of 6.655 TFLOPS and its memory bandwidth of 320.3 GB/s are the likely drivers of this result. The AMD card’s FP32 throughput is 4.489 TFLOPS, and its memory bandwidth is 217.0 GB/s, both of which are lower.

In Geekbench Vulkan, the NVIDIA P104-100 scores 45,165 against the AMD Radeon Pro 575’s 37,878. The delta is 16.1% in favor of the NVIDIA card. The margin is smaller than in OpenCL, which suggests that the AMD card’s GCN 4.0 architecture handles Vulkan workloads relatively better than OpenCL ones. The AMD card supports Vulkan 1.3, while the NVIDIA card supports Vulkan 1.4, so the newer API version on the NVIDIA side may also play a role.

The AMD Radeon Pro 575 has an additional benchmark result that the NVIDIA card does not: Geekbench Metal, where it scores 46,192. This test is exclusive to Apple’s Metal API, and the NVIDIA P104-100 has no corresponding result in the database. This Metal score is the AMD card’s highest individual benchmark result and contributes to its higher average.

The NVIDIA P104-100 has an additional benchmark result that the AMD card does not: 3DMark Steel Nomad DX12, where it scores 1,413. This is the NVIDIA card’s lowest individual benchmark result and is significantly lower than its Geekbench scores. The low score in this modern DirectX 12 test suggests that the NVIDIA card’s Pascal architecture, despite its high raw compute, does not perform as well in this specific workload. The AMD Radeon Pro 575 has no corresponding 3DMark result in the database.

The average benchmark scores reflect these different test distributions. The AMD Radeon Pro 575’s average of 39,555 is based on three Geekbench results, all of which are above 34,000. The NVIDIA P104-100’s average of 32,982 is based on three results, one of which is the low 3DMark score of 1,413. If the 3DMark result were excluded, the NVIDIA card’s average would be higher, but the database records it as part of the card’s overall performance profile. The percentiles reflect the overall averages: the AMD card ranks in the 82nd percentile, while the NVIDIA card ranks in the 77th. The nearest rivals for the AMD card include the NVIDIA RTX A500 Mobile at 39,568, which is essentially identical, and the AMD Radeon Pro 575X at 39,116, which is 1.1% slower. The NVIDIA card’s nearest rivals include the NVIDIA T600 Mobile at 32,849, which is 0.4% faster, and the NVIDIA GeForce RTX 3050 Mobile at 33,170, which is 0.6% faster. These rival comparisons show that both cards sit in a crowded performance band, but the AMD card occupies a higher overall tier.