GPU Comparison

Head-to-Head Benchmarks

The benchmark data presents a single direct comparison between the NVIDIA Tesla K10 and the AMD Radeon 660M. In the Geekbench OpenCL test, the Tesla K10 scores 14,029 points, while the Radeon 660M scores 12,876 points. This yields a 9% delta in favor of the NVIDIA part. That is a decisive margin, though not overwhelming; it places the Tesla K10 clearly ahead in raw compute throughput as measured by this particular workload.

Context from the nearestRivals data sharpens the picture. The Tesla K10’s 14,029 OpenCL score sits just 0.9% below the NVIDIA GeForce GTX 680 (14,150) and 1.1% above the AMD Radeon RX 570X (13,871). The Radeon 660M, with its 12,876 OpenCL score, trails the same GTX 680 by roughly 9% and falls 1.5% behind the Tesla K10 in the average benchmark score comparison. Notably, the Radeon 660M’s average benchmark score of 13,812 is computed from two tests, OpenCL and Vulkan, whereas the Tesla K10’s average (14,029) is based solely on its OpenCL result. The Vulkan score of 14,748 for the Radeon 660M is actually higher than its OpenCL score, suggesting the AMD part performs relatively better under Vulkan, but that test was not run on the Tesla K10.

The deltaPct values in the rival lists reinforce the closeness of this matchup. From the Tesla K10’s perspective, the Radeon 660M is 1.6% behind. From the Radeon 660M’s perspective, the Tesla K10 is 1.5% ahead. These symmetric figures indicate that, in aggregate benchmark terms, the two GPUs are nearly equivalent, with the NVIDIA card holding a slight edge. The 9% OpenCL gap, however, is larger than the average gap suggests, implying that the Tesla K10’s advantage is concentrated in the OpenCL workload, while the Radeon 660M’s Vulkan result (14,748) exceeds the Tesla K10’s OpenCL score by 5.1%. This split means the winner depends heavily on the API being used.

Where Each One Wins

The Tesla K10 wins the only head-to-head benchmark available: Geekbench OpenCL, with a 9% advantage. This is a compute-oriented test that stresses raw floating-point throughput, shading units, and memory bandwidth. The Tesla K10’s architecture, with 1,536 shading units and a 256-bit GDDR5 interface delivering 160.0 GB/s, is well-suited to such workloads. Its FP32 throughput of 2.289 TFLOPS is substantially higher than the Radeon 660M’s 1,459.2 GFLOPS. In purely numerical compute tasks, such as OpenCL-based rendering, physics simulation, or general-purpose GPU computing, the Tesla K10 should hold a meaningful edge.

The Radeon 660M, however, demonstrates its strength in the Vulkan benchmark, scoring 14,748. While no direct Vulkan comparison exists for the Tesla K10, this score is 5.1% higher than the Tesla K10’s OpenCL result. This suggests the AMD part benefits from modern API optimizations, particularly given its RDNA 2.0 architecture and support for DirectX 12 Ultimate (12_2), which includes hardware ray tracing via its 6 RT cores. The Radeon 660M also has a significantly higher pixel rate (30.40 GPixel/s vs. 23.84 GPixel/s) and a much higher boost clock (1900 MHz vs. the Tesla K10’s unspecified clocks). For graphics-oriented tasks that leverage Vulkan or DirectX 12 Ultimate features, the Radeon 660M is likely the better performer, despite its lower raw FP32 count.

In terms of memory, the Tesla K10’s dedicated 4 GB of GDDR5 with 160.0 GB/s bandwidth is a clear advantage over the Radeon 660M’s System Shared memory, whose bandwidth is "System Dependent." For workloads that require consistent, high-bandwidth access to large datasets, the Tesla K10’s dedicated memory pool is preferable. On the other hand, the Radeon 660M’s 6 nm process node (vs. 28 nm) and 40 W TDP (vs. 225 W) make it far more power-efficient, though efficiency is not directly benchmarked. The Tesla K10 wins on raw compute; the Radeon 660M wins on modern API support and efficiency.

The Verdict

The data supports a clear but conditional conclusion. If the workload is OpenCL compute, the NVIDIA Tesla K10 is the choice, delivering 9% higher performance than the Radeon 660M. Its 2.289 TFLOPS FP32 throughput and 160.0 GB/s dedicated memory bandwidth are decisive in such scenarios. The Tesla K10 also sits in the 55th percentile of all GPUs, matching the Radeon 660M’s percentile, but with a single benchmark score that is higher than the AMD part’s OpenCL result.

However, if the use case involves modern graphics APIs, the Radeon 660M’s Vulkan score of 14,748, which exceeds any Tesla K10 benchmark result, makes it a more compelling option. The AMD part’s support for DirectX 12 Ultimate, Vulkan 1.4, and hardware ray tracing (6 RT cores) are features the Tesla K10 lacks entirely. The Tesla K10’s DirectX 12 support is limited to level 11_0, and it has no RT cores. For gaming or graphics applications that use these features, the Radeon 660M is clearly superior, despite its lower raw compute numbers.

The overall average benchmark scores are nearly identical: 14,029 for the Tesla K10 versus 13,812 for the Radeon 660M, a 1.6% difference. This indicates that, on average, the two GPUs are interchangeable. The decision hinges on the specific application. For pure compute in OpenCL, pick the Tesla K10. For Vulkan or DirectX 12 Ultimate workloads, pick the Radeon 660M. The Tesla K10’s 55th percentile ranking and end-of-life status suggest it is a legacy part, while the Radeon 660M, also end-of-life, is newer (release date 2022-01-03 vs. 2012-04-30) and more feature-rich.

FAQ

Q: Which GPU has the higher OpenCL benchmark score?

A: The NVIDIA Tesla K10 scores 14,029 in Geekbench OpenCL, which is 9% higher than the AMD Radeon 660M’s 12,876.

Q: Does the AMD Radeon 660M perform better in any benchmark?

A: Yes, the Radeon 660M scores 14,748 in Geekbench Vulkan, which is higher than its own OpenCL score and also higher than the Tesla K10’s OpenCL score of 14,029.

Q: How do the two GPUs compare in average benchmark score?

A: The Tesla K10 has an average score of 14,029, while the Radeon 660M averages 13,812. The Tesla K10 leads by 1.6%, based on the deltaPct values in the nearestRivals data.

Q: What is the memory configuration difference?

A: The Tesla K10 has 4 GB of dedicated GDDR5 memory with 160.0 GB/s bandwidth on a 256-bit bus. The Radeon 660M uses System Shared memory, with bandwidth described as "System Dependent."

Q: Which GPU is more power-efficient according to the specifications?

A: The Radeon 660M has a TDP of 40 W, compared to the Tesla K10’s 225 W, and uses a 6 nm process node versus 28 nm. This indicates significantly higher efficiency for the AMD part.

Q: What are the API support differences?

A: The Radeon 660M supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the Tesla K10 supports DirectX 12 (11_0) and Vulkan 1.2.175. The Radeon 660M also has 6 RT cores; the Tesla K10 has none.

Architecture Differences

The architectural gap between these two GPUs is substantial, reflecting a decade of design evolution. The NVIDIA Tesla K10 uses the GK104 chip on a 28 nm process from TSMC, with 3,540 million transistors on a 294 mm² die. This yields a transistor density of 12.0M / mm². The AMD Radeon 660M uses the Rembrandt chip on a 6 nm process, also from TSMC, with 13,100 million transistors on a 208 mm² die, achieving a transistor density of 63.0M / mm². The Radeon 660M packs more than 3.7 times the transistors into a smaller die, thanks to the newer process node.

The Tesla K10’s Kepler architecture is built around 1,536 shading units, 128 texture mapping units, and 32 ROPs. It delivers a pixel rate of 23.84 GPixel/s and a texture rate of 95.36 GTexel/s. Its FP32 performance is 2.289 TFLOPS. The Radeon 660M, based on RDNA 2.0, has 384 shading units, 24 TMUs, and 16 ROPs. Despite having fewer execution units, it achieves a higher pixel rate of 30.40 GPixel/s due to its higher clocks (boost up to 1900 MHz). Its texture rate is 45.60 GTexel/s, and FP32 performance is 1,459.2 GFLOPS. The Radeon 660M also supports FP16 at 2.918 TFLOPS (2:1 ratio), which the Tesla K10 does not list.

Memory architecture differs fundamentally. The Tesla K10 uses dedicated GDDR5 with 4 GB capacity, a 256-bit bus, and 160.0 GB/s bandwidth. The Radeon 660M relies on System Shared memory, with no dedicated VRAM. Its bandwidth is "System Dependent," meaning it varies with the host system’s memory configuration. The Tesla K10’s memory clock is 1250 MHz (5 Gbps effective), while the Radeon 660M’s memory clock is listed as "System Shared."

Feature support is another major divergence. The Radeon 660M includes 6 ray tracing cores, enabling hardware-accelerated ray tracing, and supports DirectX 12 Ultimate (12_2) and Vulkan 1.4. The Tesla K10 has no RT cores and is limited to DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6. The Radeon 660M’s bus interface is PCIe 4.0 x8, while the Tesla K10 uses PCIe 3.0 x16. The Tesla K10 is a dual-slot card with 1x 6-pin and 1x 8-pin power connectors and a 550 W suggested PSU, whereas the Radeon 660M is an IGP (integrated graphics processor) with no power connectors and no suggested PSU.

The Tesla K10’s display outputs are "No outputs," making it a compute-only accelerator. The Radeon 660M’s display outputs are "Portable Device Dependent," indicating it is designed for mobile integration. The Tesla K10 has a launch MSRP of 5,099 USD, while the Radeon 660M has no launch MSRP listed. The Tesla K10 is 272 mm long (10.7 inches); the Radeon 660M has no listed dimensions, consistent with its IGP form factor. Both GPUs are end-of-life, with the Tesla K10 released in 2012 and the Radeon 660M in 2022.