GPU Comparison

The Verdict

The NVIDIA GeForce GT 640 and Intel Arc Pro B60 occupy the same 20th percentile of all GPUs, yet they are separated by over a decade of architecture evolution. The data shows the GT 640, a Kepler-based card from 2012, delivers an average benchmark score of 3210, while the Arc Pro B60, built on Intel's Xe2-HPG Battlemage architecture, averages 3182. That is a 0.9% difference favoring the GT 640, which is within noise for most workloads.

For users constrained to legacy software environments or seeking a low-power card with no external power connectors, the GT 640 remains a functional choice. Its 65 W TDP, single-slot design, and 2 GB DDR3 memory make it suitable for basic display output and older DirectX 11 titles. The Arc Pro B60, however, is the only rational pick for modern workloads. It offers 24 GB of GDDR6 memory, 12.29 TFLOPS of FP32 compute, hardware ray tracing with 20 RT cores, and support for DirectX 12 Ultimate. The GT 640's 692.7 GFLOPS FP32 throughput is roughly 18 times lower, and it lacks any ray tracing capability. If the task involves contemporary content creation, AI inference, or high-resolution compute, the Arc Pro B60 wins decisively. The GT 640's only real advantage is its compatibility with older APIs and its negligible power draw.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA GeForce GT 640 scores 3210 on average, while the Intel Arc Pro B60 scores 3182. The GT 640 leads by 0.9%, though this margin is small enough to be considered effectively tied.

Q: How do the two cards compare in terms of memory capacity?

A: The Intel Arc Pro B60 ships with 24 GB of GDDR6 memory on a 192-bit bus, delivering 456.0 GB/s of bandwidth. The GT 640 has 2 GB of DDR3 on a 128-bit bus, providing only 28.51 GB/s. The Arc Pro B60 offers 12 times the capacity and 16 times the bandwidth.

Q: What are the power requirements for each card?

A: The GT 640 has a 65 W TDP and requires no power connectors, with a suggested power supply of 250 W. The Arc Pro B60 has a 200 W TDP, needs a single 8-pin connector, and recommends a 550 W power supply.

Q: Which card supports more modern graphics APIs?

A: The Intel Arc Pro B60 supports DirectX 12 Ultimate (12_2) and Vulkan 1.4. The GT 640 only reaches DirectX 12 (11_0) and Vulkan 1.2.175. Both cards support OpenGL 4.6.

Q: Is the GT 640 competitive in compute-heavy tasks?

A: No. The GT 640 delivers 692.7 GFLOPS FP32 and has no dedicated ray tracing or tensor cores. The Arc Pro B60 provides 12.29 TFLOPS FP32, 24.58 TFLOPS FP16, and 20 RT cores. The Arc Pro B60 is over 17 times faster in raw FP32 throughput.

Q: What is the production status of each card?

A: The GT 640 is end-of-life, having been released on June 4, 2012. The Arc Pro B60 is active in production, with a release date of September 4, 2025.

Architecture Differences

The GT 640 is built on NVIDIA's Kepler architecture using the GK107 chip manufactured on a 28 nm process at TSMC. This chip contains 1,270 million transistors on a die size of 118 mm², yielding a transistor density of 10.8 million per square millimeter. Kepler was designed for efficiency in its era, but it predates modern compute features.

The Arc Pro B60 uses Intel's Xe2-HPG architecture on the BMG-G21 chip, fabricated on a 5 nm TSMC process. This chip packs 19,600 million transistors into a 272 mm² die, achieving a density of 72.1 million transistors per mm², nearly seven times denser than the GT 640. Xe2-HPG introduces hardware ray tracing with 20 dedicated RT cores, a feature entirely absent from the GT 640. The architecture also supports DirectX 12 Ultimate, including mesh shaders and variable rate shading, which the GT 640's Kepler cannot handle.

Feature-wise, the GT 640 has no tensor cores, no RT cores, and its FP16 capability is not listed. The Arc Pro B60 also lacks tensor cores, but its FP16 throughput of 24.58 TFLOPS (2:1 ratio) indicates substantial compute flexibility. The memory subsystem differs drastically: GDDR6 with 456.0 GB/s versus DDR3 with 28.51 GB/s. The bus interface also diverges, the GT 640 uses PCIe 3.0 x16, while the Arc Pro B60 uses PCIe 5.0 x8, which provides comparable bandwidth in a narrower slot.

Specification Differences

The two cards differ across nearly every measurable specification.

• Process Node: 28 nm (GT 640) vs 5 nm (Arc Pro B60)
• Transistors: 1,270 million vs 19,600 million
• Die Size: 118 mm² vs 272 mm²
• Transistor Density: 10.8M / mm² vs 72.1M / mm²
• Base Clock: Not listed vs 2000 MHz
• Boost Clock: Not listed vs 2400 MHz
• Memory Clock: 891 MHz (1782 Mbps effective) vs 2375 MHz (19 Gbps effective)
• Memory Size: 2 GB vs 24 GB
• Memory Type: DDR3 vs GDDR6
• Memory Bus Width: 128 bit vs 192 bit
• Memory Bandwidth: 28.51 GB/s vs 456.0 GB/s
• Shading Units: 384 vs 2560
• Texture Mapping Units: 32 vs 160
• Render Output Units: 16 vs 80
• Ray Tracing Cores: None vs 20
• Pixel Rate: 7.216 GPixel/s vs 192.0 GPixel/s
• Texture Rate: 28.86 GTexel/s vs 384.0 GTexel/s
• FP32 Performance: 692.7 GFLOPS vs 12.29 TFLOPS
• FP16 Performance: Not listed vs 24.58 TFLOPS
• TDP: 65 W vs 200 W
• Slot Width: Single-slot vs Dual-slot
• Power Connectors: None vs 1x 8-pin
• Suggested PSU: 250 W vs 550 W
• Bus Interface: PCIe 3.0 x16 vs PCIe 5.0 x8
• Display Outputs: 1x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 vs 4x mini-DisplayPort 2.1
• DirectX Support: 12 (11_0) vs 12 Ultimate (12_2)
• Vulkan Support: 1.2.175 vs 1.4
• Dimensions: 145 mm length (5.7 inches) vs 167 mm length, 69 mm height, 40 mm width (6.6 x 2.7 x 1.6 inches)
• Release Date: June 4, 2012 vs September 4, 2025
• Production Status: End-of-life vs Active

Head-to-Head Benchmarks

The head-to-head benchmark list is empty, meaning no direct comparison tests were run between these two cards. Instead, the data relies on their individual benchmark suites and the nearestRivals data to establish relative performance.

The GT 640's benchmark results come from Geekbench compute tests. It scores 2086 in Geekbench Metal, 3736 in Geekbench OpenCL, and 3809 in Geekbench Vulkan. These scores aggregate to an average of 3210. The nearest rival, the Intel Arc Pro B60, averages 3182, which places the GT 640 ahead by 0.9%. Against other rivals, the GT 640 trails the NVIDIA GeForce 920M by 2.3% and the GT 730M by 3.2%, but leads the Quadro P1000 by 1.5%.

The Arc Pro B60's benchmark suite is more extensive, covering 3DMark and PassMark tests. Its 3DMark Steel Nomad DX12 score is 2646. PassMark results show 14580 in G3D, 763 in G2D, and 7029 in GPU compute. The DirectX-specific PassMark scores are 61 for DX10, 122 for DX11, 76 for DX12, and 179 for DX9. These produce an average of 3182. The Arc Pro B60's nearest rivals include the GT 640, which leads it by 0.9%, the Quadro P1000, which trails by 0.6%, and the GeForce 920M, which leads by 3.2%. Interestingly, the Arc Pro B60 is 3.5% ahead of the NVIDIA GeForce RTX 5080 SUPER, though that card's average score of 3075 appears anomalous given its expected performance class.

The biggest win for the GT 640 is its Vulkan score of 3809, which is respectable for a 2012 card. The biggest win for the Arc Pro B60 is its PassMark G3D score of 14580, which is over 4.5 times higher than the GT 640's entire average benchmark score across all tests. The Arc Pro B60's compute score of 7029 in PassMark GPU compute further underscores its superiority in parallel workloads. The GT 640's OpenCL score of 3736 is dwarfed by the Arc Pro B60's compute capabilities, though direct comparison is complicated by different test methodologies.

In practical terms, the 0.9% delta in average scores is misleading. The GT 640 achieves its score through legacy DirectX and OpenCL paths, while the Arc Pro B60's score reflects modern DX12 and compute workloads. For any current software, the Arc Pro B60 is overwhelmingly faster. The GT 640's only edge is in pure legacy compatibility, where its older driver stack and simpler architecture may avoid issues with outdated applications.