GPU Comparison

Head-to-Head Benchmarks

The benchmark data presents a tightly contested pairing, with each GPU claiming one victory in the two shared tests. The AMD Radeon Pro 570 and NVIDIA Tesla M60 are separated by margins that, while definitive in the results, are narrow enough to warrant careful scrutiny of the workloads involved.

In the Geekbench OpenCL test, the NVIDIA Tesla M60 takes the win with a score of 29,506 against the AMD Radeon Pro 570's 27,702. That represents a 6.1% advantage for the Tesla M60. This is the larger of the two deltas in this comparison, suggesting that the Maxwell architecture's compute scheduling handles this particular OpenCL workload with greater efficiency. The Tesla M60's higher shading unit count of 2,048 versus 1,792 on the Radeon Pro 570 likely contributes to this result, as does its superior peak FP32 throughput of 4.825 TFLOPS against the AMD part's 3.960 TFLOPS.

The Geekbench Vulkan test flips the outcome, but with a much smaller margin. The AMD Radeon Pro 570 scores 31,974, while the NVIDIA Tesla M60 trails at 31,473. The delta here is just 1.6% in favor of the AMD card. This is a statistically thin lead, but it is nonetheless a win. The Vulkan result is interesting because it shows the Radeon Pro 570 punching above its raw compute weight. Despite having fewer shaders, fewer texture units, and lower peak FP32, the GCN 4.0 architecture's driver overhead and async compute capabilities appear to give it an edge in this specific low-level API test.

When looking at the average benchmark scores across all tests each card participated in, the picture shifts slightly. The AMD Radeon Pro 570 averages 33,207 across its three benchmarks (Geekbench Metal, OpenCL, and Vulkan), placing it at the 78th percentile of all GPUs. The NVIDIA Tesla M60 averages 30,490 across its two benchmarks (OpenCL and Vulkan), placing it at the 75th percentile. The Radeon Pro 570's average is inflated by its strong Geekbench Metal score of 39,945, a test the Tesla M60 does not have a result for. The Tesla M60's nearest rival, the NVIDIA CMP 70HX, sits at a 0% delta with an average score of 30,476, while the AMD card's closest competitor, the NVIDIA GeForce RTX 3050 Mobile, is a mere 0.1% behind at 33,170.

The texture and pixel throughput figures tell a story of divergent design priorities. The Tesla M60 delivers 150.8 GTexel/s and 75.39 GPixel/s, significantly ahead of the Radeon Pro 570's 123.8 GTexel/s and 35.36 GPixel/s. The Tesla M60's 64 ROPs double the 32 found on the AMD chip, which explains its massive 2.13x advantage in pixel fill rate. This is a hardware specification that will manifest in rasterization-heavy workloads, though the benchmark suite here does not directly test this.

Memory bandwidth is another differentiator. The AMD Radeon Pro 570 utilizes 4 GB of GDDR5 on a 256-bit bus, achieving 217.0 GB/s. The NVIDIA Tesla M60 has 8 GB of GDDR5 on the same 256-bit bus width but only reaches 160.4 GB/s. The AMD card's 35% higher memory bandwidth is a clear advantage in bandwidth-bound scenarios, even though the Tesla M60 doubles the memory capacity. The Radeon Pro 570's memory clock runs at 6.8 Gbps effective, versus the Tesla M60's 5 Gbps effective, which is the direct cause of this bandwidth gap.

The Verdict

The data does not crown a single outright winner; instead, it reveals a split decision that depends entirely on the API and workload in question. The NVIDIA Tesla M60 wins the OpenCL compute test decisively, while the AMD Radeon Pro 570 edges ahead in Vulkan. For a buyer forced to choose based purely on these benchmark results, the decision hinges on which API dominates their target applications.

If the primary workload is OpenCL-based compute, the NVIDIA Tesla M60 is the data-backed choice. Its 6.1% lead in that test is the largest performance gap observed between these two cards. The Tesla M60 also offers double the memory capacity (8 GB versus 4 GB), which is a qualitative advantage for larger datasets even if the bandwidth is lower. The M60's 75th percentile standing among all GPUs, while lower than the AMD card's 78th, is still respectable for a professional-grade part.

If the workload is Vulkan-centric, the AMD Radeon Pro 570 is the safer pick. Its 1.6% Vulkan lead is slim but consistent with its overall higher average benchmark score of 33,207 versus 30,490. The Radeon Pro 570 also demonstrates superior memory bandwidth (217.0 GB/s versus 160.4 GB/s), which could translate to better performance in texture-heavy or data-streaming scenarios that the benchmarks may not fully capture.

The physical and power profiles are starkly different. The AMD Radeon Pro 570 is an IGP (integrated graphics processor) with a 150 W TDP and no power connectors, while the NVIDIA Tesla M60 is a dual-slot, 267 mm card with a 300 W TDP and a single 8-pin power connector. The Tesla M60 also carries a suggested PSU rating of 700 W, whereas the AMD part has no such requirement. For integration into constrained systems, the Radeon Pro 570 is clearly the more flexible option from a form-factor and power standpoint.

Where Each One Wins

The NVIDIA Tesla M60 wins in scenarios that favor raw compute throughput and high fill rates. Its 4.825 TFLOPS FP32 output exceeds the Radeon Pro 570's 3.960 TFLOPS, and its 64 ROPs provide a substantial fill-rate advantage. The M60's 2048 shading units and 128 TMUs are also higher than the AMD card's 1792 and 112, respectively. This makes the Tesla M60 the logical choice for applications that saturate compute units with parallel math, such as simulation, rendering, or heavy data-parallel workloads. The 8 GB memory capacity is another win for the M60, allowing it to hold larger working sets without spilling to system memory.

The AMD Radeon Pro 570 wins in scenarios that are sensitive to memory bandwidth and Vulkan API efficiency. The 217.0 GB/s bandwidth is a decisive advantage over the M60's 160.4 GB/s, which means the AMD card can feed its shaders with texture data or geometry faster in bandwidth-bound situations. The Vulkan win, even at 1.6%, indicates that the GCN 4.0 architecture handles modern low-level APIs with lower overhead. The Radeon Pro 570's 78th percentile ranking and higher average benchmark score (33,207) also suggest it is the more balanced performer across a broader range of tests, including the Geekbench Metal test where it scores 39,945.

The power efficiency also favors the AMD card. The Radeon Pro 570 achieves its performance at 150 W, exactly half the TDP of the Tesla M60's 300 W. This is not a direct performance metric, but it is a significant operational advantage for any deployment where thermal dissipation or energy consumption is a constraint.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro 570 has an average benchmark score of 33,207, which is higher than the NVIDIA Tesla M60's average of 30,490.

Q: What is the performance difference in OpenCL compute?

A: The NVIDIA Tesla M60 scores 29,506 in Geekbench OpenCL, which is 6.1% higher than the AMD Radeon Pro 570's score of 27,702.

Q: How do the two cards compare in Vulkan performance?

A: The AMD Radeon Pro 570 scores 31,974 in Geekbench Vulkan, beating the NVIDIA Tesla M60's 31,473 by a margin of 1.6%.

Q: Which card has more memory bandwidth?

A: The AMD Radeon Pro 570 has a memory bandwidth of 217.0 GB/s, which is significantly higher than the NVIDIA Tesla M60's 160.4 GB/s.

Q: What are the TDP requirements for each card?

A: The AMD Radeon Pro 570 has a TDP of 150 W and requires no power connectors, while the NVIDIA Tesla M60 has a TDP of 300 W and requires a single 8-pin power connector and a suggested 700 W PSU.

Q: Which GPU has a higher pixel fill rate?

A: The NVIDIA Tesla M60 has a pixel rate of 75.39 GPixel/s, which is more than double the AMD Radeon Pro 570's 35.36 GPixel/s.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process nodes. The AMD Radeon Pro 570 uses the Ellesmere chip based on GCN 4.0 architecture, fabricated on a 14 nm process at GlobalFoundries. The NVIDIA Tesla M60 uses the GM204 chip based on Maxwell 2.0 architecture, fabricated on an older 28 nm process at TSMC. This process gap is reflected in transistor density: the AMD chip packs 5,700 million transistors into a 232 mm² die, yielding 24.6 million transistors per mm², while the NVIDIA chip contains 5,200 million transistors spread across a larger 398 mm² die, for a density of just 13.1 million per mm².

The memory subsystems differ in size and speed. The AMD Radeon Pro 570 is equipped with 4 GB of GDDR5 memory running at 6.8 Gbps effective, while the NVIDIA Tesla M60 has 8 GB of GDDR5 memory running at 5 Gbps effective. Both use a 256-bit bus, but the AMD card's higher memory clock gives it a bandwidth advantage (217.0 GB/s versus 160.4 GB/s), while the NVIDIA card's larger capacity allows for bigger working sets.

Compute resources are allocated differently. The AMD Radeon Pro 570 has 1,792 shading units, 112 TMUs, and 32 ROPs. The NVIDIA Tesla M60 has 2,048 shading units, 128 TMUs, and 64 ROPs. The NVIDIA card's higher counts give it advantages in peak FP32 throughput (4.825 TFLOPS versus 3.960 TFLOPS) and texture/pixel fill rates. The AMD card supports FP16 at a 1:1 ratio (3.960 TFLOPS), while the Tesla M60 has no listed FP16 capability.

Clock behavior also differs. The AMD Radeon Pro 570 has a base clock of 1000 MHz and a boost clock of 1105 MHz. The NVIDIA Tesla M60 has a much lower base clock of 557 MHz but a significantly higher boost clock of 1178 MHz, indicating a wider dynamic range in its power management. The AMD card's memory runs at 1695 MHz (6.8 Gbps effective), while the NVIDIA card's memory runs at 1253 MHz (5 Gbps effective).

API support is broadly similar, with both supporting DirectX 12 (12_0 for AMD, 12_1 for NVIDIA), OpenGL 4.6, and Vulkan (1.3 for AMD, 1.4 for NVIDIA). The physical form factors are entirely different: the AMD Radeon Pro 570 is an IGP with no display outputs of its own (portable device dependent) and no power connectors, while the NVIDIA Tesla M60 is a dual-slot card measuring 267 mm with no display outputs and a single 8-pin power connector. The NVIDIA card is also part of a generation with explicit predecessors and successors (Tesla Kepler and Tesla Pascal), while the AMD card's generation is listed as Radeon Pro Mac (500 Series) with no predecessor or successor specified.